THE RISING SEA

Foundations of Algebraic Geometry

math216.wordpress.com

July 27, 2024 draft

© 2024 by Ravi Vakil.
As with so much else, this book is dedicated to Alice, Benjamin, and Jacob.
 Contents

Preface 13
0.1. For the reader 14
0.2. For the expert 17
0.3. Background and conventions 19
0.4. ⋆⋆ The goals of this book 20

Part I. Preliminaries 25

Chapter 1. Just enough category theory to be dangerous 27

1.1. Categories and functors 29
1.2. Universal properties determine an object up to unique isomorphism 35
1.3. Limits and colimits 43
1.4. Adjoints 48
1.5. An introduction to abelian categories 51
1.6. ⋆ Spectral sequences 62

Chapter 2. Sheaves 71
2.1. Motivating example: The sheaf of smooth functions 71
2.2. Definition of sheaf and presheaf 73
2.3. Morphisms of presheaves and sheaves 79
2.4. Properties determined at the level of stalks, and sheafification 82
2.5. Recovering sheaves from a “sheaf on a base” 86
2.6. Sheaves of abelian groups, and OX -modules, form abelian categories 89
2.7. The inverse image sheaf 92

Part II. Schemes 97

Chapter 3. Toward affine schemes: the underlying set, and topological space 99
3.1. Toward schemes 99
3.2. The underlying set of an affine scheme 101
3.3. Visualizing schemes: Generic points 113
3.4. The underlying topological space of an affine scheme 115
3.5. A base of the Zariski topology on Spec A: Distinguished open sets 118
3.6. Topological (and Noetherian) properties 119
3.7. The function I(·), taking subsets of Spec A to ideals of A 127

Chapter 4. The structure sheaf, and the definition of schemes in general 131
4.1. The structure sheaf of an affine scheme 131
4.2. Visualizing schemes: Nilpotents 135
4.3. Definition of schemes 139
5
 4.4. Three examples 142
4.5. Projective schemes, and the Proj construction 148

Chapter 5. Some properties of schemes 157

5.1. Topological properties 157
5.2. Reducedness and integrality 159
5.3. The Affine Communication Lemma, and properties of schemes that
can be checked “affine-locally” 161
5.4. Normality and factoriality 165

Chapter 6. Rings are to modules as schemes are to ... 171

6.1. Quasicoherent sheaves 171
6.2. Characterizing quasicoherence using the distinguished affine base 174
6.3. Quasicoherent sheaves form an abelian category 178
6.4. Finite type quasicoherent, finitely presented, and coherent sheaves 180
6.5. Algebraic interlude: The Jordan-Hölder package 183
6.6. Visualizing schemes: Associated points and zerodivisors 187
6.7. ⋆⋆ Coherent modules over non-Noetherian rings 197

Part III. Morphisms of schemes 201

Chapter 7. Morphisms of schemes 203

7.1. Motivations for the “right” definition of morphism of schemes 203
7.2. Morphisms of ringed spaces 204
7.3. From locally ringed spaces to morphisms of schemes 206
7.4. Maps of graded rings and maps of projective schemes 214
7.5. Rational maps from reduced schemes 216
7.6. ⋆ Representable functors and group schemes 222
7.7. ⋆⋆ The Grassmannian: First construction 227

Chapter 8. Useful classes of morphisms of schemes 229

8.1. “Reasonable” classes of morphisms (such as open embeddings) 229
8.2. Another algebraic interlude: Lying Over and Nakayama 232
8.3. A gazillion finiteness conditions on morphisms 236
8.4. Images of morphisms: Chevalley’s Theorem and elimination theory 245

Chapter 9. Closed embeddings and related notions 253

9.1. Closed embeddings and closed subschemes 253
9.2. Locally closed embeddings and locally closed subschemes 256
9.3. Important examples from projective geometry 259
9.4. The (closed sub)scheme-theoretic image 265
9.5. Slicing by effective Cartier divisors, regular sequences and regular
embeddings 269

Chapter 10. Fibered products of schemes, and base change 275

10.1. They exist 275
10.2. Computing fibered products in practice 281
10.3. Interpretations: Pulling back families, and fibers of morphisms 285
10.4. Properties preserved by base change 291
10.5. ⋆ Properties not preserved by base change, and how to fix them 292
 10.6. Products of projective schemes: The Segre embedding 301
10.7. Normalization 303

Chapter 11. Separated and proper morphisms, and (finally!) varieties 311
11.1. Fun with diagonal morphisms, and quasiseparatedness made easy 312
11.2. Separatedness, and varieties 314
11.3. The locus where two morphisms from X to Y agree, and the
“Reduced-to-Separated” Theorem 321
11.4. Proper morphisms 324

Part IV. “Geometric” properties of schemes 331

Chapter 12. Dimension 333

12.1. Dimension and codimension 333
12.2. Dimension, transcendence degree, and Noether normalization 338
12.3. Krull’s Theorems 346
12.4. Dimensions of fibers of morphisms of varieties 353

Chapter 13. Regularity and smoothness 359

13.1. The Zariski tangent space 359
13.2. Regularity, and smoothness over a field 365
13.3. Examples 371
13.4. Bertini’s Theorem 375
13.5. Discrete valuation rings, and Algebraic Hartogs’s Lemma 378
13.6. Smooth (and étale) morphisms: First definition 385
13.7. ⋆ Valuative criteria for separatedness and properness 389
13.8. ⋆ More sophisticated facts about regular local rings 393
13.9. ⋆ Filtered rings and modules, and the Artin-Rees Lemma 394

Part V. Quasicoherent sheaves on schemes, and their uses 397

Chapter 14. More on quasicoherent and coherent sheaves 399

14.1. Vector bundles “=” locally free sheaves 399
14.2. Locally free sheaves on schemes in particular 404
14.3. More pleasant properties of finite type and coherent sheaves 410
14.4. Pushforwards of quasicoherent sheaves 414
14.5. Pullbacks of quasicoherent sheaves: three different perspectives 416
14.6. The quasicoherent sheaf corresponding to a graded module 422

Chapter 15. Line bundles, maps to projective space, and divisors 425
15.1. Some line bundles on projective space 425
15.2. Line bundles and maps to projective space 429
15.3. The Curve-to-Projective Extension Theorem 437
15.4. Hard but important: Line bundles and Weil divisors 440
15.5. The pay-off: Many fun examples 447
15.6. Effective Cartier divisors “=” invertible ideal sheaves 451
15.7. The graded module corresponding to a quasicoherent sheaf 454

Chapter 16. Maps to projective space, and properties of line bundles 457
16.1. Globally generated quasicoherent sheaves 457
 16.2. Ample and very ample line bundles 459
16.3. Applications to curves 465
16.4. ⋆ The Grassmannian as a moduli space 471

Chapter 17. Projective morphisms, and relative versions of Spec and Proj 477
17.1. Relative Spec of a (quasicoherent) sheaf of algebras 477
17.2. Relative Proj of a (quasicoherent) sheaf of graded algebras 481
17.3. Projective morphisms 485

Chapter 18. Čech cohomology of quasicoherent sheaves 493

18.1. (Desired) properties of cohomology 493
18.2. Definitions and proofs of key properties 498
18.3. Cohomology of line bundles on projective space 504
18.4. Riemann-Roch, and arithmetic genus 507
18.5. A first glimpse of Serre duality 514
18.6. Hilbert functions, Hilbert polynomials, and genus 517
18.7. Higher pushforward (or direct image) sheaves 523
18.8. ⋆ Serre’s characterizations of ampleness and affineness 527
18.9. ⋆ From projective to proper hypotheses: Chow’s Lemma and
Grothendieck’s Coherence Theorem 530

Chapter 19. Application: Curves 535

19.1. A criterion for a morphism to be a closed embedding 535
19.2. A series of crucial tools 538
19.3. Curves of genus 0 541
19.4. Classical geometry arising from curves of positive genus 542
19.5. Hyperelliptic curves 545
19.6. Curves of genus 2 549
19.7. Curves of genus 3 550
19.8. Curves of genus 4 and 5 552
19.9. Curves of genus 1 555
19.10. Elliptic curves are group varieties 563
19.11. Counterexamples and pathologies using elliptic curves 569

Chapter 20. ⋆ Application: A glimpse of intersection theory 575

20.1. Intersecting n line bundles with an n-dimensional variety 575
20.2. Intersection theory on a surface 580
20.3. The Grothendieck group of coherent sheaves, and an algebraic
version of homology 587
20.4. ⋆⋆ The Nakai-Moishezon and Kleiman criteria for ampleness 589

Chapter 21. Differentials 595

21.1. Motivation and game plan 595
21.2. Definitions and first properties 596
21.3. Examples 611
21.4. The Riemann-Hurwitz Formula 616
21.5. Understanding smooth varieties using their cotangent bundles 622
21.6. Generic smoothness, and consequences 627
21.7. Unramified morphisms 631
Chapter 22. ⋆ Blowing up 635
22.1. Motivating example: Blowing up the origin in the plane 635
22.2. Blowing up, by universal property 637
22.3. The blow-up exists, and is projective 642
22.4. Examples and computations 647

Part VI. More cohomological tools 657

Chapter 23. Derived functors 659

23.1. The Tor functors 659
23.2. Derived functors in general 663
23.3. Derived functors and spectral sequences 667
23.4. Derived functor cohomology of O-modules 673
23.5. Čech cohomology and derived functor cohomology agree 676

Chapter 24. Flatness 685

24.1. Easier facts 687
24.2. Flatness through Tor 693
24.3. Ideal-theoretic criteria for flatness 696
24.4. ⋆⋆ Aside: The Koszul complex and the Hilbert Syzygy Theorem 704
24.5. Topological implications of flatness 706
24.6. Local criteria for flatness 711
24.7. Flatness implies constant Euler characteristic 717
24.8. Smooth and étale morphisms, and flatness 721

Chapter 25. Cohomology and base change theorems 729

25.1. Statements and applications 729
25.2. Proofs of cohomology and base change theorems 735
25.3. Applying cohomology and base change to moduli problems 741

Chapter 26. Depth and Cohen-Macaulayness 747

26.1. Depth 747
26.2. Cohen-Macaulay rings and schemes 751
26.3. Serre’s R1 + S2 criterion for normality 754

Chapter 27. The twenty-seven lines on a cubic surface 757

27.1. Preliminary facts 758
27.2. Every smooth cubic surface (over k) contains 27 lines 760
27.3. Every smooth cubic surface (over k) is a blown up plane 764

Chapter 28. Power series and the Theorem on Formal Functions 769
28.1. Algebraic preliminaries 769
28.2. Types of singularities 773
28.3. The Theorem on Formal Functions 776
28.4. Zariski’s Connectedness Lemma and Stein factorization 778
28.5. Zariski’s Main Theorem 781
28.6. Castelnuovo’s Criterion for contracting (−1)-curves 785
28.7. ⋆⋆ Proof of the Theorem on Formal Functions 28.3.2 789

Chapter 29. ⋆ Proof of Serre duality 793

 29.1. Desiderata 793
29.2. Ext groups and Ext sheaves for O-modules 798
29.3. Serre duality for projective k-schemes 802
29.4. The adjunction formula for the dualizing sheaf, and ωX = KX 805
Bibliography 811

Index 817
 © 2024 Ravi Vakil. Published by Princeton University Press. 11

Je pourrais illustrer la ... approche, en gardant l’image de la noix qu’il s’agit d’ouvrir.
La première parabole qui m’est venue à l’esprit tantôt, c’est qu’on plonge la noix dans
un liquide émollient, de l’eau simplement pourquoi pas, de temps en temps on frotte pour
qu’elle pénètre mieux, pour le reste on laisse faire le temps. La coque s’assouplit au fil des
semaines et des mois — quand le temps est mûr, une pression de la main suffit, la coque
s’ouvre comme celle d’un avocat mûr à point! . . .
L’image qui m’était venue il y a quelques semaines était différente encore, la chose
inconnue qu’il s’agit de connaı̂tre m’apparaissait comme quelque étendue de terre ou de
marnes compactes, réticente à se laisser pénétrer. ... La mer s’avance insensiblement et sans
bruit, rien ne semble se casser rien ne bouge l’eau est si loin on l’entend à peine... Pourtant
elle finit par entourer la substance rétive...
I can illustrate the ... approach with the ... image of a nut to be opened. The first
analogy that came to my mind is of immersing the nut in some softening liquid, and why
not simply water? From time to time you rub so the liquid penetrates better, and otherwise
you let time pass. The shell becomes more flexible through weeks and months — when the
time is ripe, hand pressure is enough, the shell opens like a perfectly ripened avocado! . . .
A different image came to me a few weeks ago. The unknown thing to be known
appeared to me as some stretch of earth or hard marl, resisting penetration ... the sea
advances insensibly in silence, nothing seems to happen, nothing moves, the water is so far
off you hardly hear it ... yet finally it surrounds the resistant substance.

— A. Grothendieck [Gr5, p. 552-3], translation by C. McLarty [Mc, p. 1]

 Preface

This book is intended to give a serious and reasonably complete introduction

to algebraic geometry, not just for (future) experts in the field. The exposition serves
a narrow set of goals (see §0.4), and necessarily takes a particular point of view on
the subject.
It has now been many decades since David Mumford wrote that algebraic
geometry “seems to have acquired the reputation of being esoteric, exclusive, and
very abstract, with adherents who are secretly plotting to take over all the rest
of mathematics! In one respect this last point is accurate ...” ([Mu4, preface] and
[Mu7, p. 227]). The revolution has now fully come to pass, and has fundamentally
changed how we think about many fields of pure mathematics. A remarkable
number of celebrated advances rely in some way on the insights and ideas first
articulated by Alexander Grothendieck, Jean-Pierre Serre, and others.
For a number of reasons, algebraic geometry has earned a reputation of being
inaccessible. The power of the subject comes from rather abstract heavy machinery,
and it is easy to lose sight of the intuitive nature of the objects and methods. Many
in nearby fields have only a vague sense of the fundamental ideas of the subject.
Algebraic geometry itself has fractured into many parts, and even within algebraic
geometry, new researchers are often unaware of the basic ideas in subfields removed
from their own.
But there is another more optimistic perspective to be taken. The ideas that
allow algebraic geometry to connect several parts of mathematics are fundamental,
and well-motivated. Many people in nearby fields would find it useful to develop
a working knowledge of the foundations of the subject, and not just at a superficial
level. Within algebraic geometry itself, there is a canon (at least for those approach-
ing the subject from this particular direction), that everyone in the field can and
should be familiar with. The rough edges of scheme theory have been sanded
down over the past half century, although there remains an inescapable need to
understand the subject on its own terms.

0.0.1. The importance of exercises. This book has a lot of exercises. I have found
that unless I have some problems I can think through, ideas don’t get fixed in my
mind. Some exercises are trivial — some experts find this offensive, but I find this
desirable. A very few necessary ones may be hard, but the reader should have
been given the background to deal with them — they are not just an excuse to push
hard material out of the text. The exercises are interspersed with the exposition,
not left to the end. Most have been extensively field-tested. The point of view here

This material is published by Princeton University Press as The Rising Sea: Foundations of Algebraic
Geometry by Ravi Vakil. This version is free to view and download for personal use only. Not for
redistribution, resale, or use in derivative works. © 2024 Ravi Vakil.

13
14 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

is one I explored with Kedlaya and Poonen in [KPV], a book that was ostensibly
about problems, but secretly a case for how one should learn and do and think
about mathematics. Most people learn by doing, rather than just passively reading.
Judiciously chosen problems can be the best way of guiding the learner toward
enlightenment.

0.0.2. Structure. You will quickly notice that everything is numbered by chapter and
section, and everything is numbered the same way after that (for ease of reference),
except exercises are indicated by letters (and are sprinkled throughout the text,
rather than at the end of sections). Individual paragraphs often get numbers for ease
of reference, or to indicate a new topic. Definitions are in bold, and are sometimes
given in passing.

0.1 For the reader

The contents of this book are intended to be a collection of communal wisdom,

necessarily distilled through an imperfect filter. I wish to say a few words on how
you might use it, although it is not clear to me if you should or will follow this
advice.
Before discussing details, I want to say clearly at the outset: the wonderful
machine of modern algebraic geometry was created to understand basic and naive
questions about geometry (broadly construed). The purpose of this book is to
give you a thorough foundation in these powerful ideas. Do not be seduced by the
lotus-eaters into infatuation with untethered abstraction. Hold tight to your geometric
motivation as you learn the formal structures which have proved to be so effective
in studying fundamental questions. When introduced to a new idea, always ask
why you should care. Do not expect an answer right away, but demand an answer
eventually. Try at least to apply any new abstraction to some concrete example you
can understand well. See if you can make a rough picture to capture the essence
of the idea. (I deliberately asked an uncoordinated and confused three-year-old to
make most of the figures in the book in order to show that even quick sketches can
enlighten and clarify.)
Understanding algebraic geometry is often thought to be hard because it con-
sists of large complicated pieces of machinery. In fact the opposite is true; to switch
metaphors, rather than being narrow and deep, algebraic geometry is shallow but
extremely broad. It is built out of a large number of very small parts, in keeping
with Grothendieck’s vision of mathematics. It is a challenge to hold the entire
organic structure, with its messy interconnections, in your head.
A reasonable place to start is with the idea of “affine complex varieties”: sub-
sets of Cn cut out by some polynomial equations. Your geometric intuition can
immediately come into play — you may already have some ideas or questions
about dimension, or smoothness, or solutions over subfields such as R or Q. Wiser
heads would counsel spending time understanding complex varieties in some
detail before learning about schemes. Instead, I encourage you to learn about
schemes immediately, learning about affine complex varieties as the central (but
not exclusive) example. This is not ideal, but can save time, and is surprisingly
 © 2024 Ravi Vakil. Published by Princeton University Press. 15

workable. An alternative is to learn about varieties elsewhere, and then come back
later.
The intuition for schemes can be built on the intuition for affine complex
varieties. Allen Knutson and Terry Tao have pointed out that this involves three
different simultaneous generalizations, which can be interpreted as three large
themes in mathematics. (i) We allow nilpotents in the ring of functions, which is
basically analysis (looking at near-solutions of equations instead of exact solutions).
(ii) We glue these affine schemes together, which is what we do in differential geometry
(looking at manifolds instead of coordinate patches). (iii) Instead of working over
C (or another algebraically closed field), we work more generally over a ring that
isn’t an algebraically closed field, or even a field at all, which is basically number
theory (solving equations over number fields, rings of integers, etc.).
Because our goal is to be comprehensive, and to understand everything one
should know after a first course, it will necessarily take longer to get to interesting
sample applications. You may be misled into thinking that one has to work this
hard to get to these applications — it is not true! You should deliberately keep an
eye out for examples you would have cared about before. This will take some time
and patience.
As you learn algebraic geometry, you should pay attention to crucial stepping
stones. Of course, the steps get bigger the farther you go.
Chapter 1. Category theory is only language, but it is language with an embed-
ded logic. Category theory is much easier once you realize that it is designed to
formalize and abstract things you already know. The initial chapter on category the-
ory prepares you to think cleanly. For example, when someone names something a
“cokernel” or a “product”, you should want to know why it deserves that name,
and what the name really should mean. The conceptual advantages of thinking
this way will gradually become apparent over time. Yoneda’s Lemma — and more
generally, the idea of understanding an object through the maps to it — will play
an important role.
Chapter 2. The theory of sheaves again abstracts something you already un-
derstand well (see the motivating example of §2.1), and what is difficult is under-
standing how one best packages and works with the information of a sheaf (stalks,
sheafification, sheaves on a base, etc.).
Chapters 1 and 2 are a risky gamble, and they attempt a delicate balance. Attempts
to explain algebraic geometry often leave such background to the reader, refer to
other sources the reader won’t read, or punt it to a telegraphic appendix. Instead,
this book attempts to explain everything necessary, but as little as possible, and
tries to get across how you should think about (and work with) these fundamental
ideas, and why they are more grounded than you might fear.
Chapters 3–5. Armed with this background, you will be able to think cleanly
about various sorts of “spaces” studied in different parts of geometry (including
real manifolds, topological spaces, and complex manifolds), as ringed spaces that
locally are of a certain form. A scheme is just another kind of “geometric space”,
and we are then ready to transport lots of intuition from “classical geometry” to
this new setting. (This also will set you up to later think about other geometric
kinds of spaces in algebraic geometry, such as complex analytic spaces, algebraic
16 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

spaces, orbifolds, stacks, rigid analytic spaces, and formal schemes.) The ways in
which schemes differ from your geometric intuition can be internalized, and your
intuition can be expanded to accommodate them. There are many properties you
will realize you will want, as well as other properties that will later prove important.
These all deserve names. Take your time becoming familiar with them.
Chapters 7–11. Thinking categorically will lead you to ask about morphisms
of schemes (and other spaces in geometry). One of Grothendieck’s fundamental
lessons is that the morphisms are central. Important geometric properties should
really be understood as properties of morphisms. There are many classes of mor-
phisms with special names, and in each case you should think through why that
class deserves a name.
Chapters 12–13. You will then be in a good position to think about fundamental
geometric properties of schemes: dimension and smoothness. You may be surprised
that these are subtle ideas, but you should keep in mind that they are subtle
everywhere in mathematics.
Chapters 14–22. Vector bundles are ubiquitous tools in geometry, and algebraic
geometry is no exception. They lead us to the more general notion of quasicoherent
sheaves, much as free modules over a ring lead us to modules more generally. We
study their properties next, including cohomology. Chapter 19, applying these
ideas to study curves, may help make clear how useful they are.
Chapters 23–29. With this in hand, you are ready to learn more advanced tools
widely used in the subject. Many examples of what you can do are given, and
the classical story of the 27 lines on a smooth cubic surface (Chapter 27) is a good
opportunity to see many ideas come together.
The rough logical dependencies among the chapters are shown in Figure 0.1.
(Caution: this should be taken with a grain of salt. For example, you can avoid
using much of Chapter 19 on curves in later chapters, but it is a crucial source of
examples, and a great way to consolidate your understanding.)
In general, I prefer to have as few hypotheses as possible. Certainly a hypothesis
that isn’t necessary to the proof is a red herring. But if a reasonable hypothesis can
make the proof cleaner and more memorable, I am willing to include it.
In particular, Noetherian hypotheses are handy when necessary, but are oth-
erwise misleading. Even Noetherian-minded readers (normal human beings) are
better off having the right hypotheses, as they will make clearer why things are
true.
We often state results particular to varieties, especially when there are tech-
niques unique to this situation that one should know. But restricting to algebraically
closed fields is useful surprisingly rarely. Geometers needn’t be afraid of arithmetic
examples or of algebraic examples; a central insight of algebraic geometry is that
the same formalism applies without change.
Pathological examples are useful to know. On mountain highways, there are
tall sticks on the sides of the road designed for bad weather. In winter, you cannot
see the road clearly, and the sticks serve as warning signs: if you cross this line, you
will die! Pathologies and (counter)examples serve a similar goal. They also serve as
a reality check, when confronting a new statement, theorem, or conjecture, whose
veracity you may doubt. (See, for example, §4.1.8.)
 © 2024 Ravi Vakil. Published by Princeton University Press. 17

F IGURE 0.1. Important logical dependences among chapters (or

more precisely, a directed graph showing which chapter should be
read before which other chapter)

When working through a book in algebraic geometry, it is particularly helpful

to have other algebraic geometry books at hand, to see different approaches and
to have alternative expositions when things become difficult. This book may
serve as a good secondary book. If it is your primary source, then two other
excellent books with what I consider a similar philosophy are [Liu] and [GW]. De
Jong’s encyclopedic online reference [Stacks] is peerless. There are many other
outstanding sources out there, perhaps one for each approach to the subject; you
should browse around and find one you find sympathetic.
If you are looking for a correct or complete history of the subject, you have
come to the wrong place. This book is not intended to be a complete guide to the
literature, and many important sources are ignored or left out, due to my own
ignorance and laziness.
Finally, if you attempt to read this without working through a significant
number of exercises (see §0.0.1), I will come to your house and pummel you with
[Gr-EGA] until you beg for mercy. It is important to not just have a vague sense of
what is true, but to be able to actually get your hands dirty. To quote Mark Kisin:
“You can wave your hands all you want, but it still won’t make you fly.” Note: The
hints may help you, but sometimes they may not.

0.2 For the expert

If you use this book for a course, you should of course adapt it to your own
point of view and your own interests. In particular, you should think about an
application or theorem you want to reach at the end of the course (which may
well not be in this book), and then work toward it. You should feel no compulsion
18 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

to sprint to the end; I advise instead taking more time, and ending at the right
place for your students. (Figure 0.1, showing large-scale dependencies among the
chapters, may help you map out a course.) I have found that the theory of curves
(Chapter 19) and the 27 lines on the cubic surface (Chapter 27) have served this
purpose well at the end of winter and spring quarters. This was true even if some
of the needed background was not covered, and had to be taken by students as
some sort of black box. For the first quarter, the goal is to build a common language
for many kinds of geometry and geometric spaces (not just algebraic geometry, but
also manifolds, complex geometry, some differential geometry, and more) — a sort
of archetypal form of a geometric space.
Faithfulness to the goals of §0.4 required a brutal triage, and I have made a
number of decisions you may wish to reverse. I will briefly describe some choices
made that may be controversial.
Decisions on how to describe things were made for the sake of the learners. If
there were two approaches, and one was “correct” from an advanced point of view,
and one was direct and natural from a naive point of view, I went with the latter.
On the other hand, the theory of varieties (over an algebraically closed field,
say) was not done first and separately. This choice brought me close to tears, but in
the end I am convinced that it can work well, if done in the right spirit.
Instead of spending the first part of the course on varieties, I spent the time
in a different way. It is tempting to assume that students will either arrive with
great comfort and experience with category theory and sheaf theory, or that they
should pick up these ideas on their own time. I would love to live in that world.
I encourage you to not skimp on these foundational issues. I have found that
although these first lectures felt painfully slow to me, they were revelatory to a
number of the students, and those with more experience were not bored and did
not waste their time. This investment paid off in spades when I was able to rely on
their ability to think cleanly and to use these tools in practice. Furthermore, if they
left the course with nothing more than hands-on experience with these ideas, the
world was still better off for it.
For the most part, we will state results in the maximal generality that the
proof justifies, but we will not give a much harder proof if the generality of the
stronger result will not be used. There are a few cases where we work harder to
prove a somewhat more general result that many readers may not appreciate. For
example, we prove a number of theorems for proper morphisms, not just projective
morphisms. But in such cases, readers are invited or encouraged to ignore the
subtleties required for the greater generality.
I consider line bundles (and maps to projective space) more fundamental than
divisors. General Cartier divisors are not seriously discussed (§15.6.4), although
effective Cartier divisors play an essential role.
Cohomology is done first using the Čech approach (as Serre first did), and de-
rived functor cohomology is introduced only later. I am well aware that Grothendieck
thinks of the fact that the agreement of Čech cohomology with derived functor
cohomology “should be considered as an accidental phenomenon”, and that “it is
important for technical reasons not to take as definition of cohomology the Čech
cohomology”, [Gr3, p. 108]. But I am convinced that this is the right way for most
people to see this kind of cohomology for the first time. (It is certainly true that
many topics in algebraic geometry are best understood in the language of derived
 © 2024 Ravi Vakil. Published by Princeton University Press. 19

functors. But this is a view from the mountaintop, looking down, and not the best
way to explore the forests. In order to appreciate derived functors appropriately,
one must understand the homological algebra behind it, and not just take it as a
black box.)
We restrict to the Noetherian case only when it is necessary, or (rarely) when it
really saves effort. In this way, non-Noetherian people will clearly see where they
should be careful, and Noetherian people will realize that non-Noetherian things
are not so terrible. Moreover, even if you are interested primarily in Noetherian
schemes, it helps to see “Noetherian” in the hypotheses of theorems only when
necessary, as it will help you remember how and when this property gets used.
There are some cases where Noetherian readers will suffer a little more than
they would otherwise. As an inflammatory example, instead of using Noetherian
hypotheses, the notion of quasiseparatedness comes up early and often. The
cost is that one extra word has to be remembered, on top of an overwhelming
number of other words. But once that is done, it is not hard to remember that
essentially every scheme anyone cares about is quasiseparated. Furthermore,
whenever the hypotheses “quasicompact and quasiseparated” turn up, the reader
will immediately guess a key idea of the proof. As another example, coherent
sheaves and finite type (quasicoherent) sheaves are the same in the Noetherian
situation, but are still worth distinguishing in statements of the theorems and
exercises, for the same reason: to be clearer on what is used in the proof.
Many important topics are not discussed. Valuative criteria are not proved (see
§13.7), and their statement is relegated to an optional section. Completely omitted:
dévissage, formal schemes, and cohomology with supports. Sorry!

0.3 Background and conventions

“Should you just be an algebraist or a geometer?” is like saying “Would you rather be
deaf or blind?”
— M. Atiyah, [At2, p. 659]
All rings are assumed to be commutative unless explicitly stated otherwise.
All rings are assumed to contain a unit, denoted 1. Maps of rings must send 1 to
1. We don’t require that 0 ̸= 1; in other words, the “0-ring” (with one element)
is a ring. (There is a ring map from any ring to the 0-ring; the 0-ring only maps
to itself. The 0-ring is the final object in the category of rings.) The definition of
“integral domain” includes 1 ̸= 0, so the 0-ring is not an integral domain. We accept
the Axiom of Choice, usually in the guise of Zorn’s Lemma. In particular, any
proper ideal in a ring is contained in a maximal ideal. (The Axiom of Choice also
arises in the argument that the category of A-modules has enough injectives, see
Exercise 23.2.G.)
The reader should be familiar with some basic notions in commutative ring
theory, in particular the notion of ideals (including prime and maximal ideals),
various types of rings (including integral domains, principal ideal domains, unique
factorization domains, and local rings), localization, and modules. Tensor products
and exact sequences of A-modules will be important. We will use the notation
(A, m) or (A, m, k) for local rings (rings with a unique maximal ideal) — A is the
20 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

ring, m its maximal ideal, and k = A/m its residue field. We will use the structure
theorem for finitely generated modules over a principal ideal domain A: any such
module can be written as the direct sum of principal modules A/(a). Some
experience with field theory will be important from time to time.
Manifolds will be brought up periodically as examples, but for the most part,
they are meant for motivation, so we will often not specify whether they are topo-
logical (real) manifolds, differentiable (C∞ , i.e., smooth) real manifolds, analytic
(real) manifolds, or complex (holomorphic) manifolds. Nonetheless, we will define
all four (Definition 4.3.9).

0.3.1. Caution about foundational issues. We will not concern ourselves with subtle
foundational issues (set-theoretic issues, universes, etc.). It is true that some people
should be careful about these issues. But is that really how you want to live your
life? (If you are one of these rare people, a good start is [KS2, §1.1].)

0.3.2. Further background. It may be helpful to have books on other subjects at

hand that you can dip into for specific facts, rather than reading them in advance.
In commutative algebra, [E] is good for this. Other popular choices are [AtM]
and [Mat2]. The book [Al] takes a point of view useful to algebraic geometry. For
homological algebra, [Weib] is simultaneously detailed and readable.
Background from other parts of mathematics (topology, geometry, complex
analysis, number theory, ...) will of course be helpful for intuition and grounding.
Some previous exposure to topology is certainly essential.

0.3.3. Nonmathematical conventions. “Unimportant” means “unimportant for the

current exposition”, not necessarily unimportant in the larger scheme of things.
Other words may be used idiosyncratically as well.
There are optional starred sections of topics worth knowing on a second or
third (but not first) reading. They are marked with a star: ⋆. Starred sections are
not necessarily harder, merely unimportant. You should not read double-starred
sections (⋆⋆) unless you really really want to, but you should be aware of their
existence. (It may be strange to have parts of a book that should not be read!)
Let’s now find out if you are taking my advice about double-starred sections.

0.4 ⋆⋆ The goals of this book

There are a number of possible introductions to the field of algebraic geome-

try: Riemann surfaces; complex geometry; the theory of varieties; a nonrigorous
examples-based introduction; algebraic geometry for number theorists; an abstract
functorial approach; and more. All have their place. Different approaches suit
different students (and different advisors). This book takes only one route.
Our intent is to cover a canon completely and rigorously, with enough examples
and calculations to help develop intuition for the machinery. This is often the
content of a “second course” in algebraic geometry, and in an ideal world, people
would learn this material over many years, after having background courses in
commutative algebra, algebraic topology, differential geometry, complex analysis,
homological algebra, number theory, and French literature. We do not live in
 © 2024 Ravi Vakil. Published by Princeton University Press. 21

an ideal world. For this reason, the book is written as a first introduction, but a
challenging one.
This book seeks to do a very few things, but to try to do them well. Our goals
and premises are as follows.
The core of the material should be digestible over a single year. After a
year of blood, sweat, and tears, readers should have a broad familiarity with
the foundations of the subject, and be ready to attend seminars, and learn more
advanced material. They should not just have a vague intuitive understanding of
the ideas of the subject; they should know interesting examples, know why they
are interesting, and be able to work through their details. Readers in other fields of
mathematics should know enough to understand the algebro-geometric ideas that
arise in their area of interest.
This means that this book is not encyclopedic, and even beyond that, hard
choices have to be made. (In particular, analytic aspects are essentially ignored,
and are at best dealt with in passing without proof. This is a book about algebraic
algebraic geometry.)
This book is usable (and has been used) for a course, but the course should (as
always) take on the personality of the instructor. With a good course, people should
be able to leave early and still get something useful from the experience. With this
book, it is possible to leave without regret after learning about category theory, or
about sheaves, or about geometric spaces, having become a better person.
The book is also usable (and has been used) for learning on your own. But most
mortals cannot learn algebraic geometry fully on their own; ideally you should
read in a group, and even if not, you should have someone you can ask questions
to (both stupid and smart questions).
There is certainly more than a year’s material here, but I have tried to make
clear which topics are essential, and which are not. Those teaching a class will
choose which “inessential” things are important for the point they wish to get
across, and use them.
There is a canon (at least for this particular approach to algebraic geometry). I
have been repeatedly surprised at how much people in different parts of algebraic
geometry agree on what every civilized algebraic geometer should know after
a first (serious) year. (There are of course different canons for different parts of
the subject, e.g., complex algebraic geometry, combinatorial algebraic geometry,
computational algebraic geometry, etc.) There are extra bells and whistles that
different instructors might add on, to prepare students for their particular part of
the field or their own point of view, but the core of the subject remains unified,
despite the diversity and richness of the subject. There are some serious and painful
compromises to be made to reconcile this goal with the previous one.
Algebraic geometry is for everyone (with the appropriate definition of “ev-
eryone”). Algebraic geometry courses tend to require a lot of background, which
makes them inaccessible to all but those who know they will go deeply into the
subject. Algebraic geometry is too important for that; it is essential that many of
those in nearby fields develop some serious familiarity with the foundational ideas
and tools of the subject, and not just at a superficial level. (Similarly, algebraic
geometers uninterested in any nearby field are necessarily arid, narrow thinkers.
Do not be such a person!)
22 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

For this reason, this book attempts to require as little background as possible.
The background required will, in a technical sense, be surprisingly minimal — ide-
ally just some commutative ring theory and point-set topology, and some comfort
with things like prime ideals and localization. This is misleading of course — the
more you know, the better. And the less background you have, the harder you will
have to work — this is not a light read. On a related note...
The book is intended to be as self-contained as possible. I have tried to
follow the motto: “if you use it, you must prove it”. I have noticed that most
students are human beings: if you tell them that some algebraic fact is in some late
chapter of a book in commutative algebra, they will not immediately go and read it.
Surprisingly often, what we need can be developed quickly from scratch, and even
if people do not read it, they can see what is involved. The cost is that the book is
much denser, and that significant sophistication and maturity is demanded of the
reader. The benefit is that more people can follow it; they are less likely to reach a
point where they get thrown. On the other hand, people who already have some
familiarity with algebraic geometry, but want to understand the foundations more
completely, should not be bored, and can focus on more subtle issues.
This goal is important because one should not just know what is true, but also
know why things are true, and what is hard, and what is not hard. Also, this helps
the previous goal, by reducing the number of prerequisites.
The book is intended to build intuition for the formidable machinery of
algebraic geometry. The exercises are central for this (see §0.0.1). Informal language
can sometimes be helpful. Many examples are given. (If you do not have pictures
in your head which provide you with insight into why things are true, and how
to prove things, then you cannot really say that you are “thinking geometrically”.)
Learning how to think cleanly (and in particular categorically) is essential. The
advantages of appropriate generality should be made clear by example, and not
through intimidation. The motivation is more local than global. For example, there
is no introductory chapter explaining why one might be interested in algebraic
geometry, and instead there is an introductory chapter explaining why you should
want to think categorically (and how to actually do this).
Balancing the above goals is already impossible. We must thus give up any
hope of achieving any other desiderata. There are no other goals.

0.4.1. Inadequate acknowledgments.

This entire project consists of communal wisdom passed from person to person.
I have tried to collect it and distill it and curate it, but I can in no way begin to
give correct credit for the ideas. I cannot even begin to correctly credit the people I
personally learned from. There are far too many to list, and any list I have tried to
make has had too many painful omissions. Instead, I will try to decribe the broad
classes of people who have influenced this work.
My life began in Toronto, and I had an early glimpse of algebraic geometry at
the University of Toronto, through Bierstone, Milman, Murty, Arthur, and many
others. In my graduate and postdoctoral years at Harvard, Princeton, and MIT, I
fell in love with algebraic geometry and developed into the mathematician that
I am, and if you know who I met during those years — older role models and
teachers, slightly older mentors, many peers, and brilliant younger students — you
 © 2024 Ravi Vakil. Published by Princeton University Press. 23

will see their personalities in these pages. In some important algebro-geometric

sense, Joe Harris gave me a heart, Brendan Hassett gave me a brain, and Johan
de Jong gave me a spine. But the entire ethos of algebraic geometry at the time,
where subfields were rapidly developing, yet people learned from and talked to
each other across the entire field, with generosity and friendship, is what attracted
me into the subject. Perhaps most of the famous algebraic geometers of today have
their ideas in these pages (although they may not realize it), coming from long
conversations at conferences, or just ideas gleaned from talks.
It may surprise the next group how much I learned from them: my students
and postdocs (both official ones and those who merely passed through). My main
contribution as a mathematician has been in thinking about how to develop talent,
and I have have taken great joy in watching extraordinary talent develop.
I (and soon, you) also owe an extreme debt to those have contributed to these
notes over the years. Many people have indivually given hundreds of detailed
useful comments, and the resulting conversations have been particularly rewarding.
At one point I kept a list of those in this group who merited particular thanks, but
the length and arbitrariness of that list led me to decide to leave the evidence
in public on the website where these notes came to life, math216.wordpress.com.
(Many people preferred to give their comments by email, so their contributions
may be less visible.)
The phrase “The Rising Sea” is due to Grothendieck [Gr5, p. 552-3], with this
particular translation by McLarty [Mc, p. 1]. The phrase was popularized as the title
of Daniel Murfet’s excellent blog [Mur], and I want to particularly thank Daniel for
his generosity in sharing this title. Mike Stay is the author of Jokes 1.2.12 and 21.5.2.
Many particulars of the design of the prepublication version of the book are due
to a number of people, but in particular to Sándor Kovács. I am also indebted to
Diana Gilooly for her care and wisdom in shepherding this book to publication.
I am grateful for the financial support this project has indirectly received
over the years, from Stanford, the National Science Foundation, and the Simons
Foundation. The generous support of fundamental research and basic science has
been an important engine of human development over centuries, so it is a pleasure
to acknowledge it here.
Finally, at the most basic level, I am grateful to my family, and for my family.
 Part I

Preliminaries
 CHAPTER 1

Just enough category theory to be dangerous

Was mich nicht umbringt, macht mich stärker.

That which does not kill me, makes me stronger.
— F. Nietzsche [N, aphorism number 8]
Before we get to any interesting geometry, we need to develop a language
to discuss things cleanly and effectively. This is best done in the language of
categories. There is not much to know about categories to get started; it is just a
very useful language. Like all mathematical languages, category theory comes with
an embedded logic, which allows us to abstract intuitions in settings we know well
to far more general situations.
Our motivation is as follows. We will be creating some new mathematical
objects (such as schemes, and certain kinds of sheaves), and we expect them to
act like objects we have seen before. We could try to nail down precisely what
we mean by “act like”, and what minimal set of things we have to check in order
to verify that they act the way we expect. Fortunately, we don’t have to — other
people have done this before us, by defining key notions, such as abelian categories,
which behave like modules over a ring.
Our general approach will be as follows. I will try to tell you what you need to
know, and no more. (This I promise: if I use the word “topoi”, you can shoot me.) I
will begin by telling you things you already know, and describing what is essential
about the examples, in a way that we can abstract a more general definition. We
will then see this definition in less familiar settings, and get comfortable with using
it to solve problems and prove theorems.

1.0.1. Example: product. For example, we will define the notion of product of
schemes. We could just give a definition of product, but then you should want to
know why this precise definition deserves the name of “product”. As a motivation,
we revisit the notion of product in a situation we know well: (the category of)
sets. One way to define the product of sets U and V is as the set of ordered pairs
{(u, v) : u ∈ U, v ∈ V}. But someone from a different mathematical culture might
u
reasonably define it as the set of symbols { v : u ∈ U, v ∈ V}. These notions are
“obviously the same”. Better: there is “an obvious bijection between the two”.
This can be made precise by giving a better definition of product, in terms
of a universal property. Given two sets M and N, a product is a set P, along with
maps µ : P → M and ν : P → N, such that for any set P ′ with maps µ ′ : P ′ → M and

This material is published by Princeton University Press as The Rising Sea: Foundations of Algebraic
Geometry by Ravi Vakil. This version is free to view and download for personal use only. Not for
redistribution, resale, or use in derivative works. © 2024 Ravi Vakil.

27
28 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

ν ′ : P ′ → N, these maps must factor uniquely through P:

(1.0.1.1) P′
ν′
∃!

P /( N
µ′ ν
µ
 
M
(The symbol ∃ means “there exists”, and the symbol ! means “unique”.) Thus a
product is a diagram
ν /
P N
µ

M
and not just a set P, although the maps µ and ν are often left implicit.
This definition agrees with the traditional definition, with one twist: there
isn’t just a single product; but any two products come with a unique isomorphism
between them. In other words, the product is unique up to unique isomorphism.
Here is why: if you have a product

P1
ν1
/N
µ1

M
and I have a product
P2
ν2
/N
µ2

M
then by the universal property of my product (letting (P2 , µ2 , ν2 ) play the role of
(P, µ, ν), and (P1 , µ1 , ν1 ) play the role of (P ′ , µ ′ , ν ′ ) in (1.0.1.1)), there is a unique
map f : P1 → P2 making the appropriate diagram commute (i.e., µ1 = µ2 ◦ f and
ν1 = ν2 ◦ f). Similarly by the universal property of your product, there is a unique
map g : P2 → P1 making the appropriate diagram commute. Now consider the
universal property of my product, this time letting (P2 , µ2 , ν2 ) play the role of both
(P, µ, ν) and (P ′ , µ ′ , ν ′ ) in (1.0.1.1). There is a unique map h : P2 → P2 such that
P2
ν2
h
'/
µ2 P2 ν2
N
µ2

M
commutes. However, I can name two such maps: the identity map idP2 , and f ◦ g.
Thus f ◦ g = idP2 . Similarly, g ◦ f = idP1 . Thus the maps f and g arising from the
 © 2024 Ravi Vakil. Published by Princeton University Press. 29

universal property are bijections. In short, there is a unique bijection between P1

and P2 preserving the “product structure” (the maps to M and N). This gives us the
right to name any such product M × N, since any two such products are uniquely
identified.
This definition has the advantage that it works in many circumstances, and once
we define categories, we will soon see that the above argument applies verbatim
in any category to show that products, if they exist, are unique up to unique
isomorphism. Even if you haven’t seen the definition of category before, you can
verify that this agrees with your notion of product in some category that you have
seen before (such as the category of vector spaces, or the category of manifolds).
This is handy even in cases that you understand. For example, one way of
defining the product of two manifolds M and N is to cut them both up into charts,
then take products of charts, then glue them together. But if I cut up the manifolds
in one way, and you cut them up in another, how do we know our resulting
manifolds are the “same”? We could wave our hands, or make an annoying
argument about refining covers, but instead, we should just show that they are
“categorical products” and hence canonically the “same” (i.e., isomorphic). We will
formalize this argument in §1.2.
Another set of notions we will abstract are categories that “behave like mod-
ules”. We will want to define kernels and cokernels for new notions, and we should
make sure that these notions behave the way we expect them to. This leads us
to the definition of abelian categories, first defined by Grothendieck in his Tôhoku
paper [Gr1].
In this chapter, we will give an informal introduction to these and related
notions, in the hope of giving just enough familiarity to comfortably use them in
practice.

1.1 Categories and functors

The introduction of the digit 0 or the group concept was general nonsense too, and
mathematics was more or less stagnating for thousands of years because nobody was around
to take such childish steps...
— A. Grothendieck, [BP, p. 4–5]
Before functoriality, people lived in caves.
— B. Conrad
We begin with an informal definition of categories and functors.

1.1.1. Categories.
A category consists of a collection of objects, and for each pair of objects, a
set of morphisms (or arrows) between them. (For experts: technically, this is the
definition of a locally small category. In the correct definition, the morphisms need
only form a class, not necessarily a set, but see Caution 0.3.1.) Morphisms are often
informally called maps. The collection of objects of a category C is often denoted
obj(C ), but we will usually denote the collection also by C . If A, B ∈ C , then the
set of morphisms from A to B is denoted Mor(A, B). A morphism is often written
30 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

f : A → B, and A is said to be the source of f, and B the target of f. (Of course,

Mor(A, B) is taken to be disjoint from Mor(A ′ , B ′ ) unless A = A ′ and B = B ′ .)
Morphisms compose as expected: there is a composition Mor(B, C)×Mor(A, B) →
Mor(A, C), and if f ∈ Mor(A, B) and g ∈ Mor(B, C), then their composition is de-
noted g ◦ f. Composition is associative: if f ∈ Mor(A, B), g ∈ Mor(B, C), and
h ∈ Mor(C, D), then h ◦ (g ◦ f) = (h ◦ g) ◦ f. For each object A ∈ C , there is always
an identity morphism idA : A → A, such that when you (left- or right-)compose
a morphism with the identity, you get the same morphism. More precisely, for
any morphisms f : A → B and g : B → C, idB ◦f = f and g ◦ idB = g. (If you wish,
you may check that “identity morphisms are unique”: there is only one morphism
deserving the name idA .) This ends the definition of a category.
We have a notion of isomorphism between two objects of a category (a mor-
phism f : A → B such that there exists some — necessarily unique — morphism
g : B → A, where f ◦ g and g ◦ f are the identity on B and A respectively).

1.1.2. Example. The prototypical example to keep in mind is the category of sets,
denoted Sets. The objects are sets, and the morphisms are maps of sets. (Because
Russell’s paradox shows that there is no set of all sets, we did not say earlier that
there is a set of all objects. But as stated in §0.3, we are deliberately omitting all
set-theoretic issues.)

1.1.3. Example. Another good example is the category Veck of vector spaces over
a given field k. The objects are k-vector spaces, and the morphisms are linear
transformations. (What are the isomorphisms?)

1.1.A. U NIMPORTANT EXERCISE . A category in which each morphism is an

isomorphism is called a groupoid. (This notion is not important in what we will
discuss. The point of this exercise is to give you some practice with categories, by
relating them to an object you know well.)
(a) A perverse definition of a group is: a groupoid with one object. Make sense of
this. (Similarly, in case you care: a perverse definition of a monoid is: a category
with one object.)
(b) Describe a groupoid that is not a group.

1.1.B. E XERCISE . If A is an object in a category C , show that the invertible elements

of Mor(A, A) form a group (called the automorphism group of A, denoted Aut(A)).
What are the automorphism groups of the objects in Examples 1.1.2 and 1.1.3? Show
that two isomorphic objects have isomorphic automorphism groups. (For readers
with a topological background: if X is a topological space, then the fundamental
groupoid is the category where the objects are points of X, and the morphisms
x → y are paths from x to y, up to homotopy. Then the automorphism group of
x0 is the (pointed) fundamental group π1 (X, x0 ). In the case where X is connected,
and π1 (X) is not abelian, this illustrates the fact that for a connected groupoid —
whose definition you can guess — the automorphism groups of the objects are all
isomorphic, but not canonically isomorphic.)

1.1.4. Example: abelian groups. The abelian groups, along with group homomor-
phisms, form a category Ab.
 © 2024 Ravi Vakil. Published by Princeton University Press. 31

1.1.5. Important Example: Modules over a ring. If A is a ring, then the A-modules form
a category ModA . (This category has additional structure; it will be the prototypical
example of an abelian category, see §1.5.) Taking A = k, we obtain Example 1.1.3;
taking A = Z, we obtain Example 1.1.4.

1.1.6. Example: rings. There is a category Rings, where the objects are rings, and
the morphisms are maps of rings in the usual sense (maps of sets which respect
addition and multiplication, and which send 1 to 1 by our conventions, §0.3).

1.1.7. Example: topological spaces. The topological spaces, along with continuous
maps, form a category Top. The isomorphisms are homeomorphisms.
In all of the above examples, the objects of the categories were in obvious ways
sets with additional structure (a concrete category, although we won’t use this
terminology). This needn’t be the case, as the next example shows.

1.1.8. Example: partially ordered sets. A partially ordered set, (or poset), is a set S
along with a binary relation ≥ on S satisfying:
(i) x ≥ x (reflexivity),
(ii) x ≥ y and y ≥ z imply x ≥ z (transitivity), and
(iii) if x ≥ y and y ≥ x then x = y (antisymmetry).
A partially ordered set (S, ≥) can be interpreted as a category whose objects are the
elements of S, and with a single morphism from x to y if and only if x ≥ y (and no
morphism otherwise).
A trivial example is (S, ≥) where x ≥ y if and only if x = y. Another example
is
(1.1.8.1) •

• / •

Here there are three objects. The identity morphisms are omitted for convenience,
and the two non-identity morphisms are depicted. A third example is
(1.1.8.2) • /•

 / •
•
Here the “obvious” morphisms are again omitted: the identity morphisms, and the
morphism from the upper left to the lower right. Similarly,
··· /• /• /•

depicts a partially ordered set, where again, only the “generating morphisms” are
depicted.

1.1.9. Example: the category of subsets of a set, and the category of open subsets of a
topological space. If X is a set, then the subsets form a partially ordered set, where
arrows are given by inclusion. (Be careful: you may be expecting the arrows to
go the other way, because of Example 1.1.8.) Informally, if U ⊂ V, then we have
exactly one morphism U → V in the category (and otherwise none). Similarly, if
32 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

X is a topological space, then the open sets form a partially ordered set, where the
maps are given by inclusions.

1.1.10. Definition. A subcategory A of a category B has as its objects some of the

objects of B, and some of the morphisms of B, such that the objects of A include
the sources and targets of the morphisms of A , and the morphisms of A include
the identity morphisms of the objects of A , and are preserved by composition. (For
example, (1.1.8.1) is in an obvious way a subcategory of (1.1.8.2). Also, we have an
obvious “inclusion” i : A → B, which will soon be an example of a functor.)

1.1.11. Functors.
A covariant functor F from a category A to a category B, denoted F : A → B,
is the following data. It is a map of objects F : obj(A ) → obj(B), and for each A1 ,
A2 ∈ A , and morphism m : A1 → A2 , a morphism F(m) : F(A1 ) → F(A2 ) in B. We
require that F preserves identity morphisms (for A ∈ A , F(idA ) = idF(A) ), and that
F preserves composition (F(m2 ◦ m1 ) = F(m2 ) ◦ F(m1 )). (You may wish to verify
that covariant functors send isomorphisms to isomorphisms.) A trivial example is
the identity functor id : A → A , whose definition you can guess. Here are some
less trivial examples.

1.1.12. Example: a forgetful functor. Consider the functor from the category of
vector spaces (over a field k) Veck to Sets, that associates to each vector space its
underlying set. The functor sends a linear transformation to its underlying map of
sets. This is an example of a forgetful functor, where some additional structure is
forgotten. Another example of a forgetful functor is ModA → Ab from A-modules
to abelian groups, remembering only the abelian group structure of the A-module.

1.1.13. Topological examples. Examples of covariant functors include the funda-

mental group functor π1 , which sends a topological space X with choice of a point
x0 ∈ X to a group π1 (X, x0 ) (what are the objects and morphisms of the source cate-
gory?), and the ith homology functor Top → Ab, which sends a topological space X
to its ith homology group Hi (X, Z). The covariance corresponds to the fact that a
(continuous) morphism of pointed topological spaces ϕ : X → Y with ϕ(x0 ) = y0
induces a map of fundamental groups π1 (X, x0 ) → π1 (Y, y0 ), and similarly for
homology groups.

1.1.14. Example. Suppose A is an object in a category C . Then there is a func-

tor hA : C → Sets sending B ∈ C to Mor(A, B), and sending f : B1 → B2 to
Mor(A, B1 ) → Mor(A, B2 ) described by

[g : A → B1 ]  / [f ◦ g : A → B1 → B2 ].

This seemingly silly functor ends up surprisingly being an important concept.

1.1.15. Definitions. If F : A → B and G : B → C are covariant functors, then we

define a functor G ◦ F : A → C (the composition of G and F) in the obvious way.
Composition of functors is associative in an evident sense.
A covariant functor F : A → B is faithful if for all A, A ′ ∈ A , the map
MorA (A, A ′ ) → MorB (F(A), F(A ′ )) is injective, and full if it is surjective. A functor
that is full and faithful is fully faithful. (For various philosophical reasons, the
notion of “full” functor on its own is unimportant; “fully faithful” is the useful
 © 2024 Ravi Vakil. Published by Princeton University Press. 33

notion.) A subcategory i : A → B is a full subcategory if i is full. (Inclusions are

always faithful, so there is no need for the phrase “faithful subcategory”.) Thus a
subcategory A ′ of A is full if and only if for all A, B ∈ obj(A ′ ), MorA ′ (A, B) =
MorA (A, B). For example, the forgetful functor Veck → Sets is faithful, but not full;
and if A is a ring, the category of finitely generated A-modules is a full subcategory
of the category ModA of A-modules.

1.1.16. Definition. A contravariant functor is defined in the same way as a

covariant functor, except the arrows switch directions: in the above language,
F(A1 → A2 ) is now an arrow from F(A2 ) to F(A1 ). (Thus F(m2 ◦ m1 ) = F(m1 ) ◦
F(m2 ), not F(m2 ) ◦ F(m1 ).)
It is wise to state whether a functor is covariant or contravariant, unless the
context makes it very clear. If it is not stated (and the context does not make it
clear), the functor is often assumed to be covariant.
Sometimes people describe a contravariant functor C → D as a covariant
functor C opp → D, where C opp is the same category as C except that the arrows go
in the opposite direction. Here C opp is said to be the opposite category to C .
One can define fullness, etc. for contravariant functors, and you should do so.

1.1.17. Linear algebra example. If Veck is the category of k-vector spaces (introduced
in Example 1.1.3), then taking duals gives a contravariant functor (·)∨ : Veck → Veck .
Indeed, to each linear transformation f : V → W, we have a dual transformation
f∨ : W ∨ → V ∨ , and (f ◦ g)∨ = g∨ ◦ f∨ .

1.1.18. Topological example (cf. Example 1.1.13) for those who have seen cohomology. The
ith cohomology functor Hi (·, Z) : Top → Ab is a contravariant functor.

1.1.19. Example. There is a contravariant functor Top → Rings taking a topological

space X to the ring of real-valued continuous functions on X. A morphism of
topological spaces X → Y (a continuous map) induces the pullback map from
functions on Y to functions on X.

1.1.20. Example (the functor of points, cf. Example 1.1.14). Suppose A is an object
of a category C . Then there is a contravariant functor hA : C → Sets sending
B ∈ C to Mor(B, A), and sending the morphism f : B1 → B2 to the morphism
Mor(B2 , A) → Mor(B1 , A) via
 / [g ◦ f : B1 → B2 → A].
[g : B2 → A]

This example initially looks weird and different, but Examples 1.1.17 and 1.1.19
may be interpreted as special cases; do you see how? What is A in each case? This
functor might reasonably be called the functor of maps (to A), but is actually known
as the functor of points. We will meet this functor again in §1.2.11 and (in the
category of schemes) in Definition 7.3.10.

1.1.21. ⋆ Natural transformations (and natural isomorphisms) of covariant func-

tors, and equivalences of categories.
(This notion won’t come up in an essential way until at least Chapter 7, so you
shouldn’t read this section until then.) Suppose F and G are two covariant functors
from A to B. A natural transformation of covariant functors F → G is the data of
a morphism mA : F(A) → G(A) for each A ∈ A such that for each f : A → A ′ in A ,
34 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

the diagram

F(f)
F(A) / F(A ′ )

mA mA ′
 
G(A) / G(A ′ )
G(f)

commutes. A natural isomorphism of functors is a natural transformation such

that each mA is an isomorphism. (We make analogous definitions when F and G
are both contravariant.)
The data of functors F : A → B and F ′ : B → A such that F ◦ F ′ is naturally
isomorphic to the identity functor idB on B and F ′ ◦ F is naturally isomorphic
to idA is said to be an equivalence of categories. The right notion of when two
categories are “essentially the same” is not isomorphism (a functor giving bijections
of objects and morphisms) but equivalence. Exercises 1.1.C and 1.1.D might give
you some vague sense of this. Later exercises (for example, that “rings” and “affine
schemes” are essentially the same, once arrows are reversed, Exercise 7.3.E) may
help too.
Two examples might make this strange concept more comprehensible. The
double dual of a finite-dimensional vector space V is not V, but we learn early to
say that it is canonically isomorphic to V. We can make that precise as follows. Let
f.d.Veck be the category of finite-dimensional vector spaces over k. Note that this
category contains oodles of vector spaces of each dimension.

1.1.C. E XERCISE . Let (·)∨∨ : f.d.Veck → f.d.Veck be the double dual functor from
the category of finite-dimensional vector spaces over k to itself. Show that (·)∨∨
is naturally isomorphic to the identity functor on f.d.Veck . (Without the finite-
dimensionality hypothesis, we only get a natural transformation of functors from
id to (·)∨∨ .)
Let V be the category whose objects are the k-vector spaces kn for each n ≥ 0
(there is one vector space for each n), and whose morphisms are linear transfor-
mations. The objects of V can be thought of as vector spaces with bases, and the
morphisms as matrices. There is an obvious functor V → f.d.Veck , as each kn is a
finite-dimensional vector space.

1.1.D. E XERCISE . Show that V → f.d.Veck gives an equivalence of categories,

by describing an “inverse” functor. (Recall that we are being cavalier about set-
theoretic assumptions, see Caution 0.3.1, so feel free to simultaneously choose
bases for each vector space in f.d.Veck . To make this precise, you will need to use
Gödel-Bernays set theory or else replace f.d.Veck with a very similar small category,
but we won’t worry about this.)

1.1.22. ⋆⋆ Aside for experts. Your argument for Exercise 1.1.D will show that
(modulo set-theoretic issues) this definition of equivalence of categories is the same
as another one commonly given: a covariant functor F : A → B is an equivalence
of categories if it is fully faithful and every object of B is isomorphic to an object
of the form F(A) for some A ∈ A (F is essentially surjective, a term we will not
need).
 © 2024 Ravi Vakil. Published by Princeton University Press. 35

1.2 Universal properties determine an object up to unique

isomorphism

Given some category that we come up with, we often will have ways of pro-
ducing new objects from old. In good circumstances, such a definition can be
made using the notion of a universal property. Informally, we wish that there were
an object with some property. We first show that if it exists, then it is essentially
unique, or more precisely, is unique up to unique isomorphism. Then we go about
constructing an example of such an object to show existence.
Explicit constructions are sometimes easier to work with than universal prop-
erties, but with a little practice, universal properties are useful in proving things
quickly and slickly. Indeed, when learning the subject, people often find explicit
constructions more appealing, and use them more often in proofs, but as they
become more experienced, they find universal property arguments more elegant
and insightful.

1.2.1. Products were defined by a universal property. We have seen one impor-
tant example of a universal property argument already in §1.0.1: products. You
should go back and verify that our discussion there gives a notion of product
in any category, and shows that products, if they exist, are unique up to unique
isomorphism.

1.2.2. Initial, final, and zero objects. Here are some simple but useful concepts that
will give you practice with universal property arguments. An object of a category
C is an initial object if it has precisely one map to every object. It is a final object if
it has precisely one map from every object. It is a zero object if it is both an initial
object and a final object.

1.2.A. E XERCISE . Show that any two initial objects are uniquely isomorphic. Show
that any two final objects are uniquely isomorphic.
In other words, if an initial object exists, it is unique up to unique isomorphism,
and similarly for final objects. This (partially) justifies the phrase “the initial object”
rather than “an initial object”, and similarly for “the final object” and “the zero
object”. (Convention: we often say “the”, not “a”, for anything defined up to
unique isomorphism.)

1.2.B. E XERCISE . What are the initial and final objects in Sets, Rings, and Top (if
they exist)? How about in the two examples of §1.1.9?

1.2.3. Localization of rings and modules. Another important example of a defi-

nition by universal property is the notion of localization of a ring. We first review
a constructive definition, and then reinterpret the notion in terms of universal
property. A multiplicative subset S of a ring A is a subset closed under multipli-
cation containing 1. We define a ring S−1 A. The elements of S−1 A are of the form
a/s where a ∈ A and s ∈ S, and where a1 /s1 = a2 /s2 if (and only if) for some
s ∈ S, s(s2 a1 − s1 a2 ) = 0. We define (a1 /s1 ) + (a2 /s2 ) = (s2 a1 + s1 a2 )/(s1 s2 ),
and (a1 /s1 ) × (a2 /s2 ) = (a1 a2 )/(s1 s2 ). (If you wish, you may check that this
equality of fractions really is an equivalence relation and the two binary operations
on fractions are well-defined on equivalence classes and make S−1 A into a ring.)
36 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

We have a canonical ring map

(1.2.3.1) A / S−1 A

given by a 7→ a/1. Note that if 0 ∈ S, S−1 A is the 0-ring.

There are two particularly important flavors of multiplicative subsets. The first
is {1, f, f2 , . . . }, where f ∈ A. This localization is denoted Af . (Can you describe an
∼
isomorphism Af ←→ A[t]/(tf − 1)?) The second is A \ p, where p is a prime ideal.
This localization S−1 A is denoted Ap . (Notational warning: If p is a prime ideal,
then Ap means you’re allowed to divide by elements not in p. However, if f ∈ A,
Af means you’re allowed to divide by f. This can be confusing. For example, if (f)
is a prime ideal, then Af ̸= A(f) .)
Warning: sometimes localization is first introduced in the special case where A
is an integral domain and 0 ∈ / S. In that case, A ,→ S−1 A, but this isn’t always true,
as shown by the following exercise. (But we will see that noninjective localizations
needn’t be pathological, and we can sometimes understand them geometrically, see
Exercise 3.2.L.)

1.2.C. E XERCISE . Show that A → S−1 A is injective if and only if S contains no

zerodivisors. (A zerodivisor of a ring A is an element a such that there is a nonzero
element b with ab = 0. The other elements of A are called non-zerodivisors. For
example, an invertible element is never a zerodivisor. Counter-intuitively, 0 is a
zerodivisor in every ring but the 0-ring. More generally, if M is an A-module, then
a ∈ A is a zerodivisor for M if there is a nonzero m ∈ M with am = 0. The other
elements of A are called non-zerodivisors for M. Equivalently, and very usefully,
a ∈ A is a non-zerodivisor for M if and only if ×a : M → M is an injection, or
equivalently in the language of §1.5, if

0 /M ×a
/M

is exact.)
If A is an integral domain and S = A \ {0}, then S−1 A is called the fraction field
of A, which we denote K(A). The previous exercise shows that A is a subring of its
fraction field K(A). We now return to the case where A is a general (commutative)
ring.

1.2.D. E XERCISE . Verify that A → S−1 A satisfies the following universal property:
S−1 A is initial among A-algebras B where every element of S is sent to an invertible
element in B. (Recall: the data of “an A-algebra B” and “a ring map A → B” are
the same.) Translation: any map A → B where every element of S is sent to an
invertible element must factor uniquely through A → S−1 A. Another translation:
a ring map out of S−1 A is the same thing as a ring map from A that sends every
element of S to an invertible element. Furthermore, an S−1 A-module is the same
thing as an A-module for which s × · : M → M is an A-module isomorphism for all
s ∈ S.
In fact, it is cleaner to define A → S−1 A by the universal property, and to
show that it exists, and to use the universal property to check various properties
S−1 A has. Let’s get some practice with this by defining localizations of modules
by universal property. Suppose M is an A-module. We define the A-module map
 © 2024 Ravi Vakil. Published by Princeton University Press. 37

ϕ : M → S−1 M as being initial among A-module maps M → N such that elements

of S are invertible in N (s × · : N → N is an isomorphism for all s ∈ S). More
precisely, any such map α : M → N factors uniquely through ϕ:

M
ϕ
/ S−1 M

∃!
α
# 
N
(Translation: M → S−1 M is universal (initial) among A-module maps from M to
modules that are actually S−1 A-modules. Can you make this precise by defining
clearly the objects and morphisms in this category?)
Notice: (i) this determines ϕ : M → S−1 M up to unique isomorphism (you
should think through what this means); (ii) we are defining not only S−1 M, but
also the map ϕ at the same time; and (iii) essentially by definition the A-module
structure on S−1 M extends to an S−1 A-module structure.

1.2.E. E XERCISE . Show that ϕ : M → S−1 M exists, by constructing something

satisfying the universal property. Hint: define elements of S−1 M to be of the
form m/s where m ∈ M and s ∈ S, and m1 /s1 = m2 /s2 if and only if for some
s ∈ S, s(s2 m1 − s1 m2 ) = 0. Define the additive structure by (m1 /s1 ) + (m2 /s2 ) =
(s2 m1 + s1 m2 )/(s1 s2 ), and the S−1 A-module structure (and hence the A-module
structure) is given by (a1 /s1 ) · (m2 /s2 ) = (a1 m2 )/(s1 s2 ).

1.2.F. E XERCISE .
(a) Show that localization commutes with finite products, or equivalently, with
finite direct sums. In other words, if M1 , . . . , Mn are A-modules, describe an
isomorphism (of A-modules, and of S−1 A-modules) S−1 (M1 × · · · × Mn ) →
S−1 M1 × · · · × S−1 Mn .
(b) Show that localization commutes with arbitrary direct sums.
(c) Show that “localization
Q does not Q necessarily commute with infinite products”:
the obvious map S−1 ( i Mi ) → i S−1 Mi induced by the universal property of
localization is not always an isomorphism. (Hint: (1, 1/2, 1/3, 1/4, . . . ) ∈ Q × Q ×
· · · .)

1.2.4. Remark. Localization does not always commute with Hom, see Exam-
ple 1.5.10. But Exercise 1.5.H will show that in good situations (if the first argument
of Hom is finitely presented), localization does commute with Hom.

1.2.5. Tensor products. Another important example of a universal property

construction is the notion of a tensor product of A-modules
⊗A : obj(ModA ) × obj(ModA ) / obj(ModA )

(M, N)  / M ⊗A N

The subscript A is often suppressed when it is clear from context. The tensor
product is often defined as follows. Suppose you have two A-modules M and N.
Then elements of the tensor product M ⊗A N are finite A-linear combinations of
symbols m⊗n (m ∈ M, n ∈ N), subject to relations (m1 +m2 )⊗n = m1 ⊗n+m2 ⊗n,
38 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

m ⊗ (n1 + n2 ) = m ⊗ n1 + m ⊗ n2 , a(m ⊗ n) = (am) ⊗ n = m ⊗ (an) (where a ∈ A,

m1 , m2 ∈ M, n1 , n2 ∈ N). More formally, M ⊗A N is the free A-module generated
by M × N, quotiented by the submodule generated by (m1 + m2 , n) − (m1 , n) −
(m2 , n), (m, n1 +n2 )−(m, n1 )−(m, n2 ), a(m, n)−(am, n), and a(m, n)−(m, an)
for a ∈ A, m, m1 , m2 ∈ M, n, n1 , n2 ∈ N. The image of (m, n) in this quotient is
m ⊗ n.
If A is a field k, we recover the tensor product of vector spaces.

1.2.G. E XERCISE ( IF YOU HAVEN ’ T SEEN TENSOR PRODUCTS BEFORE ). Show that
Z/(10)⊗Z Z/(12) = ∼ Z/(2). (This exercise is intended to give some hands-on practice
with tensor products.)

1.2.H. I MPORTANT E XERCISE : RIGHT- EXACTNESS OF (·) ⊗A N. Show that (·) ⊗A N

gives a covariant functor ModA → ModA . Show that (·) ⊗A N is a right-exact
functor, i.e., if
M ′ → M → M ′′ → 0
is an exact sequence of A-modules (which means f : M → M ′′ is surjective, and M ′
surjects onto the kernel of f; see §1.5), then the induced sequence
M ′ ⊗A N → M ⊗A N → M ′′ ⊗A N → 0
is also exact. This exercise is repeated in Exercise 1.5.G, but you may get a lot out of
doing it now. (You will be reminded of the definition of right-exactness in §1.5.6.)
In contrast, you can quickly check that tensor product is not left-exact: tensor
the exact sequence of Z-modules

0 /Z ×2
/Z / Z/(2) /0

with Z/(2).
The constructive definition of ⊗ is a weird definition, and really the “wrong”
definition. To motivate a better one: notice that there is a natural A-bilinear map
M × N → M ⊗A N. (If M, N, P ∈ ModA , a map f : M × N → P is A-bilinear if
f(m1 + m2 , n) = f(m1 , n) + f(m2 , n), f(m, n1 + n2 ) = f(m, n1 ) + f(m, n2 ), and
f(am, n) = f(m, an) = af(m, n).) Any A-bilinear map M × N → P factors through
the tensor product uniquely: M × N → M ⊗A N → P. (Think this through!)
We can take this as the definition of the tensor product as follows. It is an
A-module T along with an A-bilinear map t : M × N → T , such that given any
A-bilinear map t ′ : M × N → T ′ , there is a unique A-linear map f : T → T ′ such
that t ′ = f ◦ t.
M×N
t /T

t′
#  ∃!f
T′

1.2.I. E XERCISE . Show that (T, t : M × N → T ) is unique up to unique isomorphism.

Hint: first figure out what “unique up to unique isomorphism” means for such
pairs, using a category of pairs (T, t). Then follow the analogous argument for the
product.
In short: given M and N, there is an A-bilinear map t : M × N → M ⊗A N,
unique up to unique isomorphism, defined by the following universal property: for
 © 2024 Ravi Vakil. Published by Princeton University Press. 39

any A-bilinear map t ′ : M×N → T ′ there is a unique A-linear map f : M⊗A N → T ′

such that t ′ = f ◦ t.
As with all universal property arguments, this argument shows uniqueness
assuming existence. To show existence, we need an explicit construction.

1.2.J. E XERCISE . Show that the construction of §1.2.5 satisfies the universal property
of tensor product.
The three exercises below are useful facts about tensor products with which
you should be familiar.

1.2.K. I MPORTANT E XERCISE .

(a) If M is an A-module and A → B is a morphism of rings, give B ⊗A M the
structure of a B-module (this is part of the exercise). Show that this describes a
functor ModA → ModB .
(b) (tensor product of rings) If further A → C is another morphism of rings, show
that B ⊗A C has a natural structure of a ring. Hint: multiplication will be given by
(b1 ⊗ c1 )(b2 ⊗ c2 ) = (b1 b2 ) ⊗ (c1 c2 ). (Exercise 1.2.U will interpret this construction
as a fibered coproduct.)

1.2.L. I MPORTANT E XERCISE . If S is a multiplicative subset of A and M is an

∼
A-module, describe a natural isomorphism (S−1 A) ⊗A M −→ S−1 M (as S−1 A-
modules and as A-modules).

1.2.M. E XERCISE (⊗ COMMUTES WITH ⊕). Show that tensor products commute
with arbitrary direct sums: if M and {Ni }i∈I are all A-modules, describe an isomor-
phism
∼ / ⊕i∈I (M ⊗ Ni ) .
M ⊗ (⊕i∈I Ni )

1.2.6. Essential Example: Fibered products. Suppose we have morphisms α : X →

Z and β : Y → Z (in any category). Then the fibered product (or fibred product) is
an object X ×Z Y along with morphisms prX : X ×Z Y → X and prY : X ×Z Y → Y,
where the two compositions α ◦ prX , β ◦ prY : X ×Z Y → Z agree, such that given
any object W with maps to X and Y (whose compositions to Z agree), these maps
factor through some unique W → X ×Z Y:

W
∃!
#
X ×Z Y prY )/ Y

prX β
  
X
α /Z

(Warning: the definition of the fibered product depends on α and β, even though
they are omitted from the notation X ×Z Y.)
By the usual universal property argument, if it exists, it is unique up to unique
isomorphism. (You should think this through until it is clear to you.) Thus the use
of the phrase “the fibered product” (rather than “a fibered product”) is reasonable,
40 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

and we should reasonably be allowed to give it the name X ×Z Y. We know what

maps to it are: they are precisely maps to X and maps to Y that agree as maps to Z.

1.2.7. Definition. As an example, if π : X → Y is a morphism, and the fibered

product X ×Y X exists, then this determines a diagonal morphism δπ : X → X ×Y X.
The diagonal morphism will turn out to be a very useful notion.
Depending on your religion, the diagram

X ×Z Y
prY
/Y
prX β
 
X
α /Z

is called a fibered/pullback/Cartesian diagram/square (six possibilities — and even

more are possible if you prefer “fibred” to “fibered”).
The right way to interpret the notion of fibered product is first to think about
what it means in the category of sets.

1.2.N. E XERCISE ( FIBERED PRODUCTS OF SETS ). Show that in Sets,

X ×Z Y = {(x, y) ∈ X × Y : α(x) = β(y)}.
More precisely, show that the right side, equipped with its evident maps to X and
Y, satisfies the universal property of the fibered product. (This will help you build
intuition for fibered products.)

1.2.O. E XERCISE . If X is a topological space, show that fibered products always

exist in the category of open sets of X, by describing what a fibered product is.
(Hint: it has a one-word description.)

1.2.P. E XERCISE . If Z is the final object in a category C , and X, Y ∈ C , show that

“X ×Z Y = X × Y”: “the” fibered product over Z is uniquely isomorphic to “the”
product. Assume all relevant (fibered) products exist. (This is an exercise about
unwinding the definition.)

1.2.Q. U SEFUL E XERCISE : TOWERS OF C ARTESIAN DIAGRAMS ARE C ARTESIAN

DIAGRAMS . If the two squares in the following commutative diagram are Cartesian
diagrams, show that the “outside rectangle” (involving U, V, Y, and Z) is also a
Cartesian diagram.
U /V

 
W /X

 
Y /Z

1.2.R. E XERCISE . Given morphisms X1 → Y, X2 → Y, and Y → Z, show that there

is a natural morphism X1 ×Y X2 → X1 ×Z X2 , assuming that both fibered products
exist. (This is trivial once you figure out what it is saying. The point of this exercise
is to see why it is trivial.)
 © 2024 Ravi Vakil. Published by Princeton University Press. 41

1.2.S. I MPORTANT EXERCISE : T HE D IAGONAL -B ASE -C HANGE DIAGRAM . Suppose

we are given morphisms X1 , X2 → Y and Y → Z. Show that the following diagram
is a Cartesian square.
X1 ×Y X2 / X1 ×Z X2

 
Y / Y ×Z Y

Assume all relevant (fibered) products exist. (If this exercise is too hard now, you
can try it again at Exercise 1.3.B.) You will appreciate how useful this diagram is
when you repeatedly use the diagonal morphism in proofs and constructions.
If you liked this problem, you may enjoy Exercise 11.1.C.

1.2.8. Coproducts. Define coproduct in a category by reversing all the arrows in the
definition of product. Define fibered coproduct in a category by reversing
` all the
arrows in the definition of fibered product. Coproduct is denoted by .

1.2.T. E XERCISE . `Show that coproduct for Sets is disjoint union. This is why we
use the notation for disjoint union.

1.2.U. E XERCISE . Suppose A → B and A → C are two ring morphisms, so in

particular B and C are A-modules. Recall (Exercise 1.2.K) that B ⊗A C has a ring
structure. Show that there is a natural morphism B → B ⊗A C given by b 7→ b ⊗ 1.
(This is not necessarily an inclusion; see Exercise 1.2.G.) Similarly, there is a natural
morphism C → B ⊗A C. Show that this gives a fibered coproduct on rings, i.e., that

B ⊗OA C o CO

Bo A
satisfies the universal property of fibered coproduct.

1.2.9. Monomorphisms and epimorphisms.

1.2.10. Definition. A morphism π : X → Y is a monomorphism if any two

morphisms µ1 : Z → X and µ2 : Z → X such that π ◦ µ1 = π ◦ µ2 must satisfy
µ1 = µ2 . In other words, there is at most one way of filling in the dotted arrow so
that the diagram
Z
≤1
 
X /Y
π

commutes — for any object Z, the natural map Mor(Z, X) → Mor(Z, Y) is an

injection. Intuitively, it is the categorical version of an injective map, and indeed
this notion generalizes the familiar notion of injective maps of sets. (The reason
we don’t use the word “injective” is that in some contexts, “injective” will have
an intuitive meaning which may not agree with “monomorphism”. One example:
in the category of divisible groups, the map Q → Q/Z is a monomorphism but
42 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

not injective. This is also the case with “epimorphism” (to be defined shortly) vs.
“surjective”.)

1.2.V. E XERCISE . Show that the composition of two monomorphisms is a monomor-

phism.

1.2.W. E XERCISE . Prove that a morphism π : X → Y is a monomorphism if and

only if the fibered product X ×Y X exists, and the induced diagonal morphism
δπ : X → X ×Y X (Definition 1.2.7) is an isomorphism. We may then take this as the
definition of monomorphism. (Monomorphisms aren’t central to future discussions,
although they will come up again. This exercise is just good practice.)

1.2.X. E ASY E XERCISE . We use the notation of Exercise 1.2.R. Show that if Y → Z
is a monomorphism, then the morphism X1 ×Y X2 → X1 ×Z X2 you described in
Exercise 1.2.R is an isomorphism. (Hint: for any object V, give a natural bijection
between maps from V to the first and maps from V to the second. It is also possible
to use the Diagonal-Base-Change diagram, Exercise 1.2.S.)
The notion of an epimorphism is “dual” to the definition of monomorphism,
where all the arrows are reversed. This concept will not be central for us, although
it turns up in the definition of an abelian category. Intuitively, it is the categorical
version of a surjective map. (But be careful when working with categories of objects
that are sets with additional structure, as epimorphisms need not be surjective.
Example: in the category Rings, Z → Q is an epimorphism, but obviously not
surjective.)

1.2.11. Representable functors and Yoneda’s Lemma. Much of our discussion

about universal properties can be cleanly expressed in terms of representable
functors, under the rubric of “Yoneda’s Lemma”. Yoneda’s lemma is an easy fact
stated in a complicated way. Informally speaking, you can essentially recover an
object in a category by knowing the maps into it. For example, we have seen that
the data of maps to X × Y are naturally (canonically) the data of maps to X and to Y.
Indeed, we have now taken this as the definition of X × Y.
Recall Example 1.1.20. Suppose A is an object of category C . For any object
C ∈ C , we have a set of morphisms Mor(C, A). If we have a morphism f : B → C,
we get a map of sets

(1.2.11.1) Mor(C, A) / Mor(B, A),

by composition: given a map from C to A, we get a map from B to A by precom-

posing with f : B → C. Hence this gives a contravariant functor hA : C → Sets.
Yoneda’s Lemma states that the functor hA determines A up to unique isomor-
phism. More precisely:

1.2.Y. I MPORTANT E XERCISE THAT YOU SHOULD DO ONCE IN YOUR LIFE (Y ONEDA’ S
L EMMA ).
(a) Suppose you have two objects A and A ′ in a category C , and maps

(1.2.11.2) iC : Mor(C, A) / Mor(C, A ′ )

that commute with the maps (1.2.11.1). Show that the iC (as C ranges over the
objects of C ) are induced from a unique morphism g : A → A ′ . More precisely,
 © 2024 Ravi Vakil. Published by Princeton University Press. 43

show that there is a unique morphism g : A → A ′ such that for all C ∈ C , iC is

u 7→ g ◦ u.
(b) If furthermore the iC are all bijections, show that the resulting g is an isomor-
phism. (Hint for both: This is much easier than it looks. This statement is so general
that there are really only a couple of things that you could possibly try. For example,
if you’re hoping to find a morphism A → A ′ , where will you find it? Well, you
are looking for an element Mor(A, A ′ ). So just plug in C = A to (1.2.11.2), and see
where the identity goes.)
There is an analogous statement with the arrows reversed, where instead of
maps into A, you think of maps from A. The role of the contravariant functor hA of
Example 1.1.20 is played by the covariant functor hA of Example 1.1.14. Because
the proof is the same (with the arrows reversed), you needn’t think it through.
The phrase “Yoneda’s Lemma” properly refers to a more general statement.
Although it looks more complicated, it is no harder to prove.

1.2.Z. ⋆ E XERCISE .
(a) Suppose A and B are objects in a category C . Give a bijection between the natu-
ral transformations hA → hB of covariant functors C → Sets (see Example 1.1.14
for the definition) and the morphisms B → A.
(b) State and prove the corresponding fact for contravariant functors hA (see Ex-
ample 1.1.20). Remark: A contravariant functor F from C to Sets is said to be
representable if there is a natural isomorphism
∼ / hA .
ξ: F

Thus the representing object A is determined up to unique isomorphism by the pair

(F, ξ). There is a similar definition for covariant functors. (We will revisit this in §7.6,
and this problem will appear again as Exercise 7.6.C. The element ξ−1 (idA ) ∈ F(A)
is often called the “universal object”; do you see why?)
(c) Yoneda’s Lemma. Suppose F is a covariant functor C → Sets, and A ∈ C .
Give a bijection between the natural transformations hA → F and F(A). (The
corresponding fact for contravariant functors is essentially Exercise 10.1.B.)
In fancy terms, Yoneda’s lemma states the following. Given a category C , we
can produce a new category, called the functor category of C , where the objects are
contravariant functors C → Sets, and the morphisms are natural transformations
of such functors. We have a functor (which we can usefully call h) from C to its
functor category, which sends A to hA . Yoneda’s Lemma states that this is a fully
faithful functor, called the Yoneda embedding. (Fully faithful functors were defined
in §1.1.15.)

1.2.12. Joke. The Yoda embedding, contravariant it is.

1.3 Limits and colimits

Limits and colimits are two important definitions determined by universal

properties. They generalize a number of familiar constructions. I will give the
definition first, and then show you why it is familiar. For example, fractions will
44 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

be motivating examples of colimits (Exercise 1.3.D(a)), and the p-adic integers

(Example 1.3.4) will be motivating examples of limits.

1.3.1. Limits. We say that a category is a small category if the objects form a set
and the morphisms form a set. (This is a technical condition intended only for
experts.) Suppose I is any small category, and C is any category. Then a functor
F : I → C (i.e., with an object Ai ∈ C for each element i ∈ I , and appropriate
commuting morphisms dictated by I ) is said to be a diagram indexed by I . We
call I an index category. Our index categories will usually be partially ordered
sets (Example 1.1.8), in which in particular there is at most one morphism between
any two objects. (But other examples are sometimes useful.) For example, if □ is
the category
• /•

 / •
•
and A is a category, then a functor □ → A is precisely the data of a commuting
square in A .
Then the limit of the diagram is an object lim Ai (or lim Ai ) of C along with
I ←−
I
morphisms fj : lim Ai → Aj for each j ∈ I , such that if m : j → k is a morphism in
I
I , then

(1.3.1.1) lim Ai
I
fk
fj
 F(m) "
Aj / Ak

commutes, and this object and maps to each Ai are universal (final) with respect to
this property. More precisely, given any other object W along with maps gi : W →
Ai commuting with the F(m) (if m : j → k is a morphism in I , then gk = F(m) ◦ gj ),
then there is a unique map
g : W → lim Ai
I

so that gi = fi ◦ g for all i. (In some cases, the limit is sometimes called the inverse
limit or projective limit. We won’t use this language.) By the usual universal
property argument, if the limit exists, it is unique up to unique isomorphism.

1.3.2. Examples: products. For example, if I is the partially ordered set

•

• / •

we obtain the fibered product.

If I is
• •
we obtain the product.
 © 2024 Ravi Vakil. Published by Princeton University Press. 45

If I is a set (i.e., the only morphisms areQthe identity maps), then the limit is
called the product of the Ai , and is denoted i Ai . The special case where I has
two elements is the example of the previous paragraph.

1.3.A. E XERCISE ( REALITY CHECK ). Suppose that the partially ordered set I has
an initial object e. Show that the limit of any diagram indexed by I exists.

1.3.B. E XERCISE : THE D IAGONAL -B ASE -C HANGE DIAGRAM , AGAIN . Solve 1.2.S
again by identifying both X1 ×Y X2 and Y ×(Y×Z Y) (X1 ×Z X2 ) as the limit of the
diagram
X1


/Z
?Y

X2

1.3.3. Example: formal power series. For a ring A, the (formal) power series, A[[x]],
are often described informally (and somewhat unnaturally) as being the ring
A[[x]] = {a0 + a1 x + a2 x2 + · · · }
(where ai ∈ A, and the ring operations are the “obvious” ones). It is an example of
a limit in the category of rings:

A[[x]]

% * ,/
··· / A[x]/(x3 ) / A[x]/(x2 ) A[x]/(x).

The universal property of limits yields a natural ring morphism A[x] → A[[x]]. If
A = R or C, this map factors through the ring of convergent power series.

1.3.4. Example: the p-adic integers. For a prime number p, the p-adic integers
(or more informally, p-adics), Zp , are often described informally (and somewhat
unnaturally) as being of the form
a0 + a1 p + a2 p2 + · · ·
(where 0 ≤ ai < p). They are an example of a limit in the category of rings:

Zp

" ) +/
··· / Z/(p3 ) / Z/(p2 ) Z/(p).

(Warning: Zp is sometimes used to denote the integers modulo p, but Z/(p) or

Z/pZ is better to use for this, to avoid confusion. Worse: by §1.2.3, Zp also denotes
those rationals whose denominators are a power of p. Hopefully the meaning of
Zp will be clear from the context.)
46 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

The similarity of Examples 1.3.3 and 1.3.4 is no coincidence. Formal power

series and the p-adic integers are examples of completions, the topic of Chapter 28.
Limits do not always exist for any index category I . However, you can often
easily check that limits exist if the objects of your category can be interpreted as
sets with additional structure, and arbitrary products exist (respecting the set-like
structure).

1.3.C. I MPORTANT E XERCISE . Show that in the category Sets,

 
Y
(ai )i∈I ∈ Ai : F(m)(aj ) = ak for all m ∈ MorI (j, k) ⊂ Mor(I ) ,
i

along with the obvious projection maps to each Ai , is the limit lim Ai .
I

This clearly also works in the category ModA of A-modules (in particular Veck
and Ab), as well as Rings.
From this point of view, 2 + 3p + 2p2 + · · · ∈ Zp can be understood as the
sequence (2, 2 + 3p, 2 + 3p + 2p2 , . . . ).

1.3.5. Colimits. More immediately relevant for us will be the dual (arrow-reversed
version) of the notion of limit (or inverse limit). We just flip the arrows fi in (1.3.1.1),
and get the notion of a colimit, which is denoted colimI Ai (or limI Ai ). (You
−→
should draw the corresponding diagram.) Again, if it exists, it is unique up to
unique isomorphism. (In some cases, the colimit is sometimes called the direct
limit, inductive limit, or injective limit. We won’t use this language. I prefer
using limit/colimit in analogy with kernel/cokernel and product/coproduct. This
is more than analogy, as kernels and products may be interpreted as limits, and
similarly with cokernels and coproducts. Also, I remember that kernels “map to”,
and cokernels are “mapped to”, which reminds me that a limit maps to all the
objects in the big commutative diagram indexed by I ; and a colimit has a map
from all the objects.)

1.3.6. Joke. A comathematician is a device for turning cotheorems into ffee.

Even though we have just flipped the arrows, colimits behave quite differently
from limits.

1.3.7. Example. The abelian group 5−∞ Z of rational numbers whose denominators
are powers of 5 is a colimit colimi∈Z+ 5−i Z. More precisely, 5−∞ Z is the colimit of
the diagram
Z / 5−1 Z / 5−2 Z / ···
in the category of abelian groups. `
The colimit over an index set I is called the coproduct, denoted i Ai , and is
the dual (arrow-reversed) notion to the product.

1.3.D. E XERCISE .
1
(a) Interpret the statement “Q = colim n Z”.
(b) Interpret the union of some subsets of a given set as a colimit. (Dually, the
intersection can be interpreted as a limit.) The objects of the category in question
are the subsets of the given set.
 © 2024 Ravi Vakil. Published by Princeton University Press. 47

Colimits do not always exist, but there are two useful large classes of examples
for which they do.

1.3.8. Definition. A nonempty partially ordered set (S, ≥) is filtered (or is said to
be a filtered set) if for each x, y ∈ S, there is a z such that x ≥ z and y ≥ z. More
generally (see Figure 1.1, a nonempty category I is filtered if:
(i) for each x, y ∈ I , there is a z ∈ I and arrows x → z and y → z, and
(ii) for every two arrows u : x → y and v : x → y, there is an arrow w : y → z
such that w ◦ u = w ◦ v.
(Other terminologies are also commonly used, such as “directed partially ordered
set” and “filtered index category”, respectively.)

F IGURE 1.1. A filtered category (pictorial definition)

1.3.E. E XERCISE . Suppose I is filtered. (We `will almost exclusively use the case
where I is a filtered set.) Recall the symbol for disjoint union of sets. Show that
any diagram in Sets indexed by I has the following, with the obvious maps to it,
as a colimit:
a 
(ai , i) ∼ (aj , j) if and only if there are f : Ai → Ak and

(ai , i) ∈ Ai
g : Aj → Ak in the diagram for which f(ai ) = g(aj ) in Ak
i∈I

(You will see that the “filtered” hypothesis is there is to ensure that ∼ is an equiva-
lence relation.)
For example, in Example 1.3.7, each element of the colimit is an element of
something upstairs, but you can’t say in advance what it is an element of. For
instance, 17/125 is an element of the 5−3 Z (or 5−4 Z, or later ones), but not 5−2 Z.
This idea applies to many categories whose objects can be interpreted as sets
with additional structure (such as abelian groups, A-modules, groups, etc.). For
example, the colimit colim Mi in the category of A-modules ModA can be described
as follows. The set underlying colim Mi is defined as in Exercise 1.3.E. To add
the elements mi ∈ Mi and mj ∈ Mj , choose an ℓ ∈ I with arrows u : i → ℓ and
v : j → ℓ, and then define the sum of mi and mj to be F(u)(mi ) + F(v)(mj ) ∈ Mℓ .
48 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

The element mi ∈ Mi is 0 if and only if there is some arrow u : i → k for which

F(u)(mi ) = 0, i.e., if it becomes 0 “later in the diagram”. Last, multiplication by an
element of A is defined in the obvious way.

1.3.F. E XERCISE . Verify that the A-module described above is indeed the colimit.
(Make sure you verify that addition is well-defined, i.e., is independent of the
choice of representatives mi and mj , the choice of ℓ, and the choice of arrows u and
v. Similarly, make sure that scalar multiplication is well-defined.)

1.3.G. U SEFUL E XERCISE ( LOCALIZATION AS A COLIMIT ). Generalize Exer-

cise 1.3.D(a) to interpret localization of an integral domain as a colimit over a filtered
set: suppose S is a multiplicative set of A, and interpret S−1 A = colim 1s A where
the colimit is over s ∈ S, and in the category of A-modules. (Aside: Can you make
some version of this work even if A isn’t an integral domain, e.g., S−1 A = colim As ?
This will work in the category of A-algebras.)
A variant of this construction works without the filtered condition, if you have
another means of “connecting elements in different objects of your diagram”. For
example:

1.3.H. E XERCISE : COLIMITS OF A- MODULES WITHOUT THE FILTERED CONDITION .

Suppose you are given a diagram of A-modules indexed by I : F : I → ModA ,
where we let Mi := F(i). Show that the colimit is ⊕i∈I Mi modulo the relations
mi − F(n)(mi ) for every n : i → j in I (i.e., for every arrow in the diagram). (Some-
what more precisely: “modulo” means “quotiented by the submodule generated
by”.)

1.3.9. Summary. One useful thing to informally keep in mind is the following. In a
category where the objects are “set-like”, an element of a limit can be thought of
as a family of elements of each object in the diagram, that are “compatible” (Exer-
cise 1.3.C). And an element of a colimit can be thought of as (“has a representative
that is”) an element of a single object in the diagram (Exercise 1.3.E). Even though
the definitions of limit and colimit are the same, just with arrows reversed, these
interpretations are quite different.

1.3.10. Small remark. In fact, colimits exist in the category of sets for all reasonable
(“small”) index categories (see for example [E, Thm. A6.1]), but that won’t matter
to us.

1.3.11. Joke. What do you call someone who reads a paper on category theory?
Answer: A coauthor.

1.4 Adjoints

We next come to a very useful notion closely related to universal properties.

Just as a universal property “essentially” (up to unique isomorphism) determines
an object in a category (assuming such an object exists), “adjoints” essentially
determine a functor (again, assuming it exists). Two covariant functors F : A → B
 © 2024 Ravi Vakil. Published by Princeton University Press. 49

and G : B → A are adjoint if there is a natural bijection for all A ∈ A and B ∈ B

(1.4.0.1) τAB : MorB (F(A), B) → MorA (A, G(B)).
We say that (F, G) form an adjoint pair, and that F is left-adjoint to G (and G is
right-adjoint to F). We say F is a left adjoint (and G is a right adjoint). By “natural”
we mean the following. For all f : A → A ′ in A , we require
Ff∗ / MorB (F(A), B)
(1.4.0.2) MorB (F(A ′ ), B)
τA ′ B τAB
 ∗ 
MorA (A ′ , G(B))
f / MorA (A, G(B))

to commute, and for all g : B → B ′ in B we want a similar commutative diagram to

commute. (Here f∗ is the map induced by f : A → A ′ , and Ff∗ is the map induced
by Ff : F(A) → F(A ′ ).)

1.4.A. E XERCISE . Write down what this diagram should be.

1.4.B. E XERCISE . Show that the map τAB (1.4.0.1) has the following properties.
For each A there is a map ηA : A → GF(A) so that for any g : F(A) → B, the
corresponding τAB (g) : A → G(B) is given by the composition

A
ηA
/ GF(A) Gg
/ G(B).

Similarly, there is a map ϵB : FG(B) → B for each B so that for any f : A → G(B),
the corresponding map τ−1AB (f) : F(A) → B is given by the composition

F(A)
Ff / FG(B) ϵB
/ B.

Here is a key example of an adjoint pair.

1.4.C. E XERCISE . Suppose M, N, and P are A-modules (where A is a ring). Describe

a bijection HomA (M ⊗A N, P) ↔ HomA (M, HomA (N, P)). (Hint: try to use the
universal property of ⊗.)

1.4.D. E XERCISE ( TENSOR -H OM ADJUNCTION ). Show that (·)⊗A N and HomA (N, ·)
are adjoint functors.

1.4.E. E XERCISE . Suppose B → A is a morphism of rings. If M is an A-module, you

can create a B-module MB by considering it as a B-module. This gives a functor
·B : ModA → ModB . Show that this functor is right-adjoint to · ⊗B A. In other words,
describe a bijection
HomA (N ⊗B A, M) = ∼ HomB (N, MB )
functorial in both arguments. (This adjoint pair is very important.)

1.4.1. ⋆ Fancier remarks we won’t use. You can check that the left adjoint determines
the right adjoint up to natural isomorphism, and vice versa. The maps ηA and ϵB of
Exercise 1.4.B are called the unit and counit of the adjunction. This leads to a different
characterization of adjunction. Suppose functors F : A → B and G : B → A are
given, along with natural transformations η : idA → GF and ϵ : FG → idB with the
property that Gϵ ◦ ηG = idG (for each B ∈ B, the composition of ηG(B) : G(B) →
50 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

GFG(B) and G(ϵB ) : GFG(B) → G(B) is the identity) and ϵF ◦ Fη = idF . Then you
can check that F is left-adjoint to G. These facts aren’t hard to check, so if you want
to use them, you should verify everything for yourself.

1.4.2. Examples from other fields. For those familiar with representation theory:
Frobenius reciprocity may be understood in terms of adjoints. Suppose V is a
finite-dimensional representation of a finite group G, and W is a representation of
a subgroup H < G. Then induction and restriction are an adjoint pair (IndG G
H , ResH )
between the category of G-modules and the category of H-modules.
Topologists’ favorite adjoint pair may be the suspension functor and the loop
space functor.

1.4.3. Example: groupification of abelian semigroups. Here is another motivating

example: getting an abelian group from an abelian semigroup. (An abelian semi-
group is just like an abelian group, except we don’t require an identity or an inverse.
Morphisms of abelian semigroups are maps of sets preserving the binary operation.
One example is the non-negative integers Z≥0 = {0, 1, 2, . . . } under addition. An-
other is the positive integers Z+ = {1, 2, 3, . . . } under addition. Yet another is the
positive integers Z+ under multiplication. You may enjoy groupifying all three.)
From an abelian semigroup, you can create an abelian group. In our examples, from
the nonnegative integers under addition (Z≥0 , +), we create the integers (Z, +),
and from the positive integers under multiplication (Z+ , ×), we create the positive
rationals (Q+ , ×). Here is a formalization of that notion. A groupification of an
abelian semigroup S is a map of abelian semigroups π : S → G such that G is an
abelian group, and any map of abelian semigroups from S to an abelian group G ′
factors uniquely through G:

S
π /G

∃!

G′

(Perhaps “abelian groupification” would be more precise than “groupification”.)

1.4.F. E XERCISE ( AN ABELIAN GROUP IS GROUPIFIED BY ITSELF ). Show that if an

abelian semigroup is already a group then the identity morphism is the groupifica-
tion. (More correct: the identity morphism is a groupification.) Note that you don’t
need to construct groupification (or even know that it exists in general) to solve this
exercise.

1.4.G. E XERCISE . Construct the “groupification functor” H from the category of

nonempty abelian semigroups to the category of abelian groups. (One possible
construction: given an abelian semigroup S, the elements of its groupification
H(S) are ordered pairs (a, b) ∈ S × S, which you may think of as a − b, with the
equivalence that (a, b) ∼ (c, d) if a + d + e = b + c + e for some e ∈ S. Describe
addition in this group, and show that it satisfies the properties of an abelian group.
Describe the abelian semigroup map S → H(S).) Let F be the forgetful functor from
the category of abelian groups Ab to the category of abelian semigroups. Show that
H is left-adjoint to F.
 © 2024 Ravi Vakil. Published by Princeton University Press. 51

(Here is the general idea for experts: We have a full subcategory of a category.
We want to “project” from the category to the subcategory. We have
Morcategory (S, H) = Morsubcategory (G, H)
automatically; thus we are describing the left adjoint to the forgetful functor. How
the argument worked: we constructed something which was in the smaller category,
which automatically satisfies the universal property.)

1.4.H. E XERCISE ( CF. E XERCISE 1.4.E). The purpose of this exercise is to give
you more practice with “left adjoints of forgetful functors”, the means by which
we get abelian groups from abelian semigroups, and sheaves from presheaves.
Suppose A is a ring, and S is a multiplicative subset. Then S−1 A-modules are a
full subcategory (§1.1.15) of the category of A-modules (via the obvious inclusion
ModS−1 A ,→ ModA ). Then ModA → ModS−1 A can be interpreted as an adjoint to
the forgetful functor ModS−1 A → ModA . State and prove the correct statements.
(Here is the larger story. Every S−1 A-module is an A-module, and this is an
injective map, so we have a forgetful functor F : ModS−1 A → ModA . In fact this
is a fully faithful functor: it is injective on objects, and the morphisms between
any two S−1 A-modules as A-modules are just the same when they are considered
as S−1 A-modules. Then there is a functor G : ModA → ModS−1 A , which might
reasonably be called “localization with respect to S”, which is left-adjoint to the
forgetful functor. Translation: If M is an A-module, and N is an S−1 A-module, then
Mor(GM, N) (morphisms as S−1 A-modules, which are the same as morphisms as
A-modules) are in natural bijection with Mor(M, FN) (morphisms as A-modules).)
Table 1 gives most of the adjoints that will come up for us. Other examples
will also come up, such as the adjoint pair (∼, Γ• ) between graded modules over a
graded ring, and quasicoherent sheaves on the corresponding projective scheme
(§15.7).

1.4.4. Last comments only for people who have seen adjoints before. If (F, G) is an adjoint
pair of functors, then F commutes with colimits, and G commutes with limits. Also,
limits commute with limits and colimits commute with colimits. We will prove
these facts (and a little more) in §1.5.14.

1.5 An introduction to abelian categories

Ton papier sur l’Algèbre homologique a été lu soigneusement, et a converti tout le
monde (même Dieudonné, qui semble complètement fonctorisé!) à ton point de vue.
Your paper on homological algebra was read carefully and converted everyone (even
Dieudonné, who seems to be completely functorised!) to your point of view.
— J.-P. Serre, letter to A. Grothendieck, Jul 13, 1955 [GrS, p. 17-18]
Since learning linear algebra, you have been familiar with the notions and
behaviors of kernels, cokernels, etc. Later in your life you saw them in the category
of abelian groups, and later still in the category of A-modules.
We will soon define some new categories (certain sheaves) that will have
familiar-looking behavior, reminiscent of that of modules over a ring. The notions
52 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

situation category category left adjoint right adjoint

A B F: A → B G: B → A
A-modules (Ex. 1.4.D) ModA ModA (·) ⊗A N HomA (N, ·)
ring maps (·) ⊗B A M 7→ MB
B → A (Ex. 1.4.E) ModB ModA (extension (restriction
of scalars) of scalars)
(pre)sheaves on a presheaves sheaves
topological space on X on X sheafification forgetful
X (Ex. 2.4.K)
(semi)groups (§1.4.3) semigroups groups groupification forgetful
sheaves, sheaves sheaves π−1 π∗
π : X → Y (Ex. 2.7.B) on Y on X
sheaves of abelian
groups or O-modules, sheaves sheaves π! π−1
open embeddings on U on Y
π : U ,→ Y (Ex. 23.4.G)
quasicoherent sheaves, QCohY QCohX π∗ π∗
π : X → Y (Prop. 14.5.7)
ring maps M 7→ MB N 7→
B → A (Ex. 17.1.J) ModA ModB (restriction HomB (A, N)
of scalars)
quasicoherent sheaves,
affine π : X → Y QCohX QCohY π∗ π!?
(Ex. 17.1.K(b))
TABLE 1. Some important adjoint pairs

of kernels, cokernels, images, and more will make sense, and they will behave
“the way we expect” from our experience with modules. This can be made precise
through the notion of an abelian category. Abelian categories are the right general
setting in which one can do “homological algebra”, in which notions of kernel,
cokernel, and so on are used, and one can work with complexes and exact sequences.

We will see enough to motivate the definitions that we will see in general:
monomorphism (and subobject), epimorphism, kernel, cokernel, and image. But in
this book we will avoid having to show that they behave “the way we expect” in a
general abelian category because the examples we will see are directly interpretable
in terms of modules over rings. In particular, it is not worth memorizing the
definition of abelian category.
Two central examples of an abelian category are the category Ab of abelian
groups, and the category ModA of A-modules. The first is a special case of the
second (just take A = Z). As we give the definitions, you should verify that ModA
is an abelian category.
We first define the notion of additive category. We will use it only as a stepping
stone to the notion of an abelian category. Two examples you can keep in mind
while reading the definition: the category of free A-modules (where A is a ring),
and real (or complex) Banach spaces.
 © 2024 Ravi Vakil. Published by Princeton University Press. 53

1.5.1. Definition. A category C is said to be additive if it satisfies the following

properties.
Ad1. For each A, B ∈ C , Mor(A, B) is an abelian group, such that composition
of morphisms distributes over addition. (You should think about what
this means — it translates to two distinct statements.)
Ad2. C has a zero object, denoted 0. (This is an object that is simultaneously an
initial object and a final object, Definition 1.2.2.)
Ad3. It has products of two objects (a product A × B for any pair of objects),
and hence by induction, products of any finite number of objects.
In an additive category, the morphisms are often called homomorphisms, and
Mor is denoted by Hom. In fact, this notation Hom is a good indication that you’re
working in an additive category. A functor between additive categories preserving
the additive structure of Hom, is called an additive functor.

1.5.2. Remarks. It is a consequence of the definition of additive category that finite

products are also finite coproducts (i.e., sums) — the details don’t matter to us. The
symbol ⊕ is used for this notion. Also, it is quick to show that additive functors
send zero objects to zero objects (show that Z is a 0-object if and only if idZ = 0Z ;
additive functors preserve both id and 0), and preserve products.
One motivation for the name 0-object is that the 0-morphism in the abelian
group Hom(A, B) is the composition A → 0 → B. (We also remark that the notion
of 0-morphism thus makes sense in any category with a 0-object.)
(A cleaner axiomatization of additive categories that makes clear that the
abelian group structure of Mor(A, B) is intrinsic to the category itself is the fol-
lowing, [Lur, p. 21–22]. A0. C has a zero object. A1. C has products of any two
objects, and coproducts of any two objects. By the universal
` property of product and
coproduct, we have natural morphisms ϕAB : A B → A × B. A2. ϕAB is an
isomorphism. This allows us to define a binary operation on Mor(A, B), with f + g
(for f, g ∈ Mor(A, B)) defined by the composition

(f,g)
/ B×B
ϕ−1 `
A
BB
/B B /B

where the last map is the “codiagonal” defined by universal property of coproduct.
A little work shows that this endows Mor(A, B) with the structure of a commutative
monoid, i.e., an abelian semigroup with identity. The identity is the composition
A → 0 → B. A3. This commutative monoid Mor(A, B) is an abelian group.)

1.5.3. The category of A-modules ModA is clearly an additive category, but it has
even more structure, which we now formalize as an example of an abelian category.

1.5.4. Definition. Let C be a category with a 0-object (and thus 0-morphisms). A

kernel of a morphism f : B → C is defined to be a map i : A → B such that f ◦ i = 0,
and that is universal with respect to this property. Diagramatically:

Z
0
∃!
  '
A
i /B f
8/ C
0
54 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

(Note that the kernel is not just an object; it is a morphism of an object to B. In

practice the term is often applied to just the object, and the intended interpretation
is clear from the context.) Hence it is unique up to unique isomorphism by universal
property nonsense. The kernel is written ker f → B. A cokernel (denoted coker f) is
defined dually by reversing the arrows — do this yourself. The kernel of f : B → C
is the limit (§1.3) of the diagram

(1.5.4.1) 0


B
f /C

and similarly the cokernel is a colimit (see (2.6.0.1)).

If i : A → B is a monomorphism, then we say that A is a subobject of B, where
the map i is implicit. There is also the notion of quotient object, defined dually to
subobject.
An abelian category is an additive category satisfying three additional proper-
ties.
(1) Every map has a kernel and cokernel.
(2) Every monomorphism is the kernel of its cokernel.
(3) Every epimorphism is the cokernel of its kernel.
It is a nonobvious (and imprecisely stated) fact that every property you want
to be true about kernels, cokernels, etc. follows from these three. (Warning: in part
of the literature, additional hypotheses are imposed as part of the definition.)
The image of a morphism f : A → B is defined as im(f) = ker(coker f) when-
ever it exists (e.g., in every abelian category). The morphism f : A → B factors
uniquely through im f → B whenever im f exists, and A → im f is an epimorphism
and a cokernel of ker f → A in every abelian category. The reader may want to
verify this as a (hard!) exercise.
The cokernel of a monomorphism is called the quotient. The quotient of a
monomorphism A → B is often denoted B/A (with the map from B implicit).
We will leave the foundations of abelian categories untouched. The key thing
to remember is that if you understand kernels, cokernels, images and so on in
the category of modules over a given ring, you can manipulate objects in any
abelian category. This is made precise by the Freyd-Mitchell Embedding Theorem
(Remark 1.5.5).
However, the abelian categories we will come across will obviously be related
to modules, and our intuition will clearly carry over, so we needn’t invoke a
theorem whose proof we haven’t read. For example, we will show that sheaves
of abelian groups on a topological space X form an abelian category (§2.6), and
the interpretation in terms of “compatible germs” will connect notions of kernels,
cokernels etc. of sheaves of abelian groups to the corresponding notions of abelian
groups.

1.5.5. Small remark on chasing diagrams. It is useful to prove facts (and solve
exercises) about abelian categories by chasing elements. Unfortunately, some
commonly used abelian categories, such as the category of complexes (to be defined
in Exercise 1.5.C), do not have “elements” — they are not naturally “sets with
additional structure” in any obvious way. Nonetheless, proof by element-chasing
 © 2024 Ravi Vakil. Published by Princeton University Press. 55

can be justified by the Freyd-Mitchell Embedding Theorem: If C is an abelian

category whose objects form a set, then there is a ring A and an exact, fully faithful
functor from C into ModA , which embeds C as a full subcategory. (Unfortunately,
the ring A need not be commutative.) A proof is sketched in [Weib, §1.6], and
references to a complete proof are given there. A proof is also given in [KS1, §9.7].
The upshot is that to prove something about a diagram in some abelian category,
we may assume that it is a diagram of modules over some ring, and we may then
“diagram-chase” elements. Moreover, any fact about kernels, cokernels, and so on
that holds in ModA holds in any abelian category.
If invoking a theorem whose proof you haven’t read bothers you, a short
alternative is Mac Lane’s “elementary rules for chasing diagrams”, [Mac, Thm. 3,
p. 204]; [Mac, Lem. 4, p. 205] gives a proof of the Five Lemma (Exercise 1.6.6) as an
example.
But in any case, do what you need to do to put your mind at ease, so you can
move forward. Do as little as your conscience will allow.

1.5.6. Complexes, exactness, and homology.

(In this entire discussion, we assume we are working in an abelian category.)
We say a sequence

(1.5.6.1) ··· /A f /B g
/C / ···
is a complex at B if g ◦ f = 0, and is exact at B if ker g = im f. (More specifically, g
has a kernel that is an image of f. Exactness at B implies being a complex at B — do
you see why?) A sequence is a complex (resp., exact) if it is a complex (resp., exact)
at each (internal) term. A short exact sequence is an exact sequence with five terms,
the first and last of which are zeros — in other words, an exact sequence of the form
0 /A /B /C / 0.

For example, 0 /A / 0 is exact if and only if A = 0;

0 /A f /B

is exact if and only if f is a monomorphism (with a similar statement for A

f /B / 0 );

0 /A f /B /0
is exact if and only if f is an isomorphism; and

0 /A f /B g
/C
f / g
/C / 0 ).
is exact if and only if f is a kernel of g (with a similar statement for A B
To show some of these facts it may be helpful to prove that (1.5.6.1) is exact at B if
and only if the cokernel of f is a cokernel of the kernel of g.
If you would like practice in playing with these notions before thinking about
homology, you can prove the Snake Lemma (stated in Example 1.6.5, with a stronger
version in Exercise 1.6.B), or the Five Lemma (stated in Example 1.6.6, with a
stronger version in Exercise 1.6.C). (I would do this in the category of A-modules,
but see [KS1, Lem. 12.1.1, Lem. 8.3.13] for proofs in general.)
If (1.5.6.1) is a complex at B, then its homology at B (often denoted by H) is
ker g / im f. (More precisely, there is some monomorphism im f ,→ ker g, and
56 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

that H is the cokernel of this monomorphism.) Therefore, (1.5.6.1) is exact at B if

and only if its homology at B is 0. We say that elements of ker g (assuming the
objects of the category are sets with some additional structure) are the cycles, and
elements of im f are the boundaries (so homology is “cycles mod boundaries”).
If the complex is indexed in decreasing order, the indices are often written as
subscripts, and Hi is the homology at Ai+1 → Ai → Ai−1 . If the complex is
indexed in increasing order, the indices are often written as superscripts, and the
homology Hi at Ai−1 → Ai → Ai+1 is often called cohomology.
An exact sequence

(1.5.6.2) A• : ··· / Ai−1 fi−1 / Ai fi / Ai+1 fi+1 / ···

can be “factored” into short exact sequences

0 / ker fi / Ai / ker fi+1 /0

which is helpful in proving facts about long exact sequences by reducing them to
facts about short exact sequences.
More generally, if (1.5.6.2) is assumed only to be a complex, then it can be
“factored” into short exact sequences.

(1.5.6.3) 0 / ker fi / Ai / im fi /0

0 / im fi−1 / ker fi / Hi (A• ) /0

1.5.A. E XERCISE . Describe exact sequences

(1.5.6.4) 0 / im fi / Ai+1 / coker fi /0

0 / Hi (A• ) / coker fi−1 / im fi /0

(These are somehow dual to (1.5.6.3). In fact in some mirror universe this might
have been given as the standard definition of homology.) Assume the category is
that of modules over a fixed ring for convenience, but be aware that the result is
true for any abelian category.

1.5.B. E XERCISE AND IMPORTANT DEFINITION . Suppose

d0 / A1 d1 / ··· dn−1 / dn /0
0 An
is a complex of finite-dimensional k-vector
P spaces (often
P called A• for short). Define
h (A ) := dim H (AP). Show that (−1) dim A = (−1) h (A• ). In particular,
i • i • i i i i

if A• is exact, then (−1)i dim Ai = 0. (If you haven’t dealt much with cohomol-
ogy, this will give you some practice.)

1.5.C. I MPORTANT E XERCISE . Suppose C is an abelian category. Define the

category ComC of complexes) as follows. The objects are infinite complexes

A• : ··· / Ai−1 fi−1 / Ai fi / Ai+1 fi+1 / ···

 © 2024 Ravi Vakil. Published by Princeton University Press. 57

in C , and the morphisms A• → B• are commuting diagrams

(1.5.6.5) ··· / Ai−1 fi−1 / Ai fi / Ai+1 fi+1 / ···

 gi−1
 gi
 gi+1
··· / Bi−1 / Bi / Bi+1 / ···
Show that ComC is an abelian category.
Feel free to deal with the special case of modules over a fixed ring. (Remark
for experts: Essentially the same argument shows that C I is an abelian category
for any small category I and any abelian category C . This immediately implies
that the category of presheaves on a topological space X with values in an abelian
category C is automatically an abelian category, cf. §2.3.5.)

1.5.D. I MPORTANT E XERCISE . Show that (1.5.6.5) induces a map of homology

Hi (A• ) → Hi (B• ). Show furthermore that Hi is a covariant functor ComC → C .
(Again, feel free to deal with the special case ModA .)

1.5.7. Homotopic maps induce the same maps on homology. We say two maps of
complexes f : C• → D• and g : C• → D• are homotopic if there is a sequence of
maps w : Ci → Di−1 such that f − g = dw + wd.

1.5.E. E XERCISE . Show that two homotopic maps give the same map on homology.

1.5.8. Theorem (Long exact sequences). — A short exact sequence of complexes

0• : ··· /0 /0 /0 / ···

   
A• : ··· / Ai−1 fi−1 / Ai fi / Ai+1 fi+1 / ···

  gi−1
 gi
 gi+1
B• : ··· / Bi−1 / Bi / Bi+1 / ···

   
C• : ··· / Ci−1 hi−1 / Ci hi / Ci+1 hi+1 / ···

   
0• : ··· /0 /0 /0 / ···
induces a long exact sequence in cohomology

··· / Hi−1 (C• ) /

Hi (A• ) / Hi (B• ) / Hi (C• ) /

Hi+1 (A• ) / ···

58 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

(This requires a definition of the connecting homomorphism Hi−1 (C• ) →

H (A• ), which is “natural” in an appropriate sense.) In the category of modules
i

over a ring, Theorem 1.5.8 will come out of our discussion of spectral sequences,
see Exercise 1.6.F, but this is a somewhat perverse way of proving it. For a proof
in general, see [KS1, Theorem 12.3.3]. You may want to prove it yourself, by
first proving a weaker version of the Snake Lemma (Example 1.6.5), where in the
hypotheses (1.6.5.1), the 0’s in the bottom left and top right are removed, and in the
conclusion (1.6.5.2), the first and last 0’s are removed.

1.5.9. Exactness of functors. If F : A → B is an additive covariant functor from one

abelian category to another, we say that F is right-exact if the exactness of
A′ /A / A ′′ / 0,
in A implies that
F(A ′ ) / F(A) / F(A ′′ ) /0

is also exact. Dually, we say that F is left-exact if the exactness of

0 / A′ /A / A ′′ implies

0 / F(A ′ ) / F(A) / F(A ′′ ) is exact.

An additive contravariant functor is left-exact if the exactness of
A′ /A / A ′′ /0 implies

0 / F(A ′′ ) / F(A) / F(A ′ ) is exact.

The reader should be able to deduce what it means for a contravariant functor to
be right-exact.
An additive covariant or contravariant functor is exact if it is both left-exact
and right-exact.

1.5.F. E XERCISE . Suppose F is an exact functor. Show that applying F to an exact

sequence preserves exactness. For example, if F is covariant, and A ′ → A → A ′′ is
exact, then FA ′ → FA → FA ′′ is exact. (This will be generalized in Exercise 1.5.I(c).)

1.5.G. E XERCISE . Suppose A is a ring, S ⊂ A is a multiplicative subset, and M is

an A-module.
(a) Show that localization of A-modules ModA → ModS−1 A is an exact covariant
functor.
(b) Show that (·) ⊗A M is a right-exact covariant functor ModA → ModA . (This is a
repeat of Exercise 1.2.H.)
(c) Show that Hom(M, ·) is a left-exact covariant functor ModA → ModA . If C is
any abelian category, and C ∈ C , show that Hom(C, ·) is a left-exact covariant
functor C → Ab.
(d) Show that Hom(·, M) is a left-exact contravariant functor ModA → ModA . If C is
any abelian category, and C ∈ C , show that Hom(·, C) is a left-exact contravariant
functor C → Ab.
 © 2024 Ravi Vakil. Published by Princeton University Press. 59

1.5.H. E XERCISE . Suppose M is a finitely presented A-module: M has a finite

number of generators, and with these generators it has a finite number of relations;
or usefully equivalently, fits in an exact sequence

(1.5.9.1) A⊕q / A⊕p /M /0

Use (1.5.9.1) and the left-exactness of Hom to describe an isomorphism
∼
S−1 HomA (M, N) o / HomS−1 A (S−1 M, S−1 N).

(You might be able to interpret this in light of a variant of Exercise 1.5.I below, for
left-exact contravariant functors rather than right-exact covariant functors.)

1.5.10. Example: Hom doesn’t always commute with localization. In the language of
Exercise 1.5.H, take A = N = Z, M = Q, and S = Z \ {0}.

1.5.11. ⋆ Two useful facts in homological algebra.

We now come to two (sets of) facts I wish I had learned as a child, as they
would have saved me lots of grief. They encapsulate what is best and worst of
abstract nonsense. The statements are so general as to be nonintuitive. The proofs
are very short. They generalize some specific behavior that is easy to prove on an
ad hoc basis. Once they are second nature to you, many subtle facts will become
obvious to you as special cases. And you will see that they will get used (implicitly
or explicitly) repeatedly.

1.5.12. ⋆ Interaction of homology and (right/left-)exact functors.

You might wait to prove this until you learn about cohomology in Chapter 18,
when it will first be used in a serious way.

1.5.I. I MPORTANT E XERCISE ( THE FHHF T HEOREM ). This result can take you
far, and perhaps for that reason it has sometimes been called the Fernbahnhof
(FernbaHnHoF) Theorem, notably in [Vak1, Exer. 1.5.I]. Suppose F : A → B is a
covariant functor of abelian categories, and C• is a complex in A .
(a) (F right-exact yields FH• / H• F ) If F is right-exact, describe a natural
morphism FH → H F. (More precisely, for each i, the left side is F applied
• •

to the cohomology at piece i of C• , while the right side is the cohomology

at piece i of FC• .)
(b) (F left-exact yields FH• o H• F ) If F is left-exact, describe a natural
morphism H F → FH .
• •
∼ /
(c) (F exact yields FH• o H• F ) If F is exact, show that the morphisms of
(a) and (b) are inverses and thus isomorphisms.
d / i+1
i
/ coker di / 0 to give an isomorphism
Hint for (a): use Ci C
∼
F coker d ←→ coker Fd . Then use the first line of (1.5.6.4) to give an epimorphism
i i

F im di ↠ im Fdi . Then use the second line of (1.5.6.4) to give the desired map
FHi C• → Hi FC• . While you are at it, you may as well describe a map for the fourth
member of the quartet {coker, im, H, ker}: F ker di → ker Fdi .

1.5.13. If this makes your head spin, you may prefer to think of it in the following
specific case, where both A and B are the category of A-modules, and F is (·) ⊗A N
for some fixed A-module N. Your argument in this case will translate without
60 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

change to yield a solution to Exercise 1.5.I(a) and (c) in general. If ⊗N is exact,

then N is called a flat A-module. (The notion of flatness will turn out to be very
important, and is discussed in detail in Chapter 24.)
For example, localization is exact (Exercise 1.5.G(a)), so S−1 A is a flat A-algebra
for all multiplicative sets S. Thus taking cohomology of a complex of A-modules
commutes with localization — something you could verify directly.

1.5.14. Interaction of adjoints, (co)limits, and (left- and right-) exactness.

A surprising number of arguments boil down to the statement:
Limits commute with limits and right adjoints. In particular, in an abelian category,
because kernels are limits, both limits and right adjoints are left-exact.
as well as its dual:
Colimits commute with colimits and left adjoints. In particular, because cokernels are
colimits, both colimits and left adjoints are right-exact.
These statements were promised in §1.4.4, and will be proved below. The latter
has a useful extension:
In ModA , colimits over filtered index categories are exact. “Filtered” was defined in
§1.3.8.

1.5.15. ⋆⋆ Caution. It is not true that in abelian categories in general, colimits

over filtered index categories are exact. (Grothendieck realized the desirability of
such colimits being exact, and formalized this as his “AB5” axiom, see for example
[Stacks, tag 079A].) Here is a counterexample. Because the axioms of abelian
categories are self-dual, it suffices to give an example in which a cofiltered limit fails
to be exact (where cofiltered has the obvious dual definition to filtered), and we do
this. Fix a prime p. In the category Ab of abelian groups, for each positive integer
n, we have an exact sequence Z → Z/(pn ) → 0. Taking the limit over all n in the
obvious way, we obtain Z → Zp → 0, which is certainly not exact.)
Unimportant Remark 1.5.18 will dash another hope you may have.

1.5.16. If you want to use these statements (for example, later in this book), you will
have to prove them. Let’s now make them precise.

1.5.J. E XERCISE ( KERNELS COMMUTE WITH LIMITS ). Suppose C is an abelian

category, and a : I → C and b : I → C are two diagrams in C indexed by I .
For convenience, let Ai = a(i) and Bi = b(i) be the objects in those two diagrams.
Let hi : Ai → Bi be maps commuting with the maps in the diagram. (Translation:
h is a natural transformation of functors a → b, see §1.1.21.) Then the ker hi
form another diagram in C indexed by I . Describe a canonical isomorphism
∼
lim ker hi ←→ ker(lim Ai → lim Bi ), assuming the limits exist.
Implicit in the previous exercise is the idea that limits should somehow be
understood as functors.

1.5.K. E XERCISE . Make sense of the statement that “limits commute with limits” in
a general category, and prove it. (Hint: recall that kernels are limits. The previous
exercise should be a corollary of this one.)

1.5.17. Proposition (right adjoints commute with limits). — Suppose (F : C →

D, G : D → C ) is a pair of adjoint functors. If A = lim Ai is a limit in D of a diagram
 © 2024 Ravi Vakil. Published by Princeton University Press. 61

indexed by I , then GA = lim GAi (with the corresponding maps GA → GAi ) is a limit
in C .

Proof. We must show that GA → GAi satisfies the universal property of limits.
Suppose we have maps W → GAi commuting with the maps of I . We wish to
show that there exists a unique W → GA extending the W → GAi . By adjointness
of F and G, we can restate this as: Suppose we have maps FW → Ai commuting
with the maps of I . We wish to show that there exists a unique FW → A extending
the FW → Ai . But this is precisely the universal property of the limit. □
Of course, the dual statements to Exercise 1.5.K and Proposition 1.5.17 hold by
the dual arguments.
If F and G are additive functors between abelian categories, and (F, G) is an
adjoint pair, then (as kernels are limits and cokernels are colimits) G is left-exact
and F is right-exact.

1.5.L. E XERCISE . Show that in ModA , colimits over filtered index categories are
exact. (Your argument will apply without change to any abelian category whose
objects can be interpreted as “sets with additional structure”.) Right-exactness
follows from the above discussion, so the issue is left-exactness. (Possible hint:
After you show that localization is exact, Exercise 1.5.G(a), or stalkification is exact,
Exercise 2.6.E, in a hands-on way, you will be easily able to prove this. Conversely,
if you do this exercise, those two will be easy.)

1.5.M. E XERCISE . Show that filtered colimits commute with homology in ModA .
Hint: use the FHHF Theorem (Exercise 1.5.I), and the previous Exercise.
In light of Exercise 1.5.M, you may want to think about how limits (and colimits)
commute with homology in general, and which way maps go. The statement of
the FHHF Theorem should suggest the answer. (Are limits analogous to left-exact
functors, or right-exact functors?) We won’t directly use this insight, but see §18.1
(vii) for an example.
Just as colimits are exact (not just right-exact) in especially good circumstances,
limits are exact (not just left-exact) too. The following will be used twice in Chap-
ter 28.
62 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

1.5.N. E XERCISE . Suppose

.. .. ..
. . .

  
0 / An+1 / Bn+1 / Cn+1 /0

  
0 / An / Bn / Cn /0

  
.. .. ..
. . .

  
0 / A0 / B0 / C0 /0

is an inverse system of exact sequences of modules over a ring, such that the
maps An+1 → An are surjective. (We say: “transition maps of the left term are
surjective”.) Show that the limit

(1.5.17.1) 0 / lim An / lim Bn / lim Cn /0

is also exact. (You will need to define the maps in (1.5.17.1).)

1.5.18. Unimportant Remark. Based on these ideas, you may suspect that right-exact
functors always commute with colimits. The fact that tensor product commutes
with infinite direct sums (Exercise 1.2.M) may reinforce this idea. Unfortunately,
it is not true — “double dual” ·∨∨ : Veck → Veck is covariant and right exact (in
fact, exact), but does not commute with infinite direct sums, as ⊕∞i=1 (k
∨∨
) is not
∞ ∨∨
isomorphic to (⊕i=1 k) .

1.5.19. ⋆ Dreaming of derived functors. When you see a left-exact functor, you
should always dream that you are seeing the end of a long exact sequence. If

0 / M′ /M / M ′′ /0

is an exact sequence in abelian category A , and F : A → B is a left-exact functor,

then
0 / FM ′ / FM / FM ′′
is exact, and you should always dream that it should continue in some natural way.
For example, the next term should depend only on M ′ , call it R1 FM ′ , and if it is
zero, then FM → FM ′′ is an epimorphism. This remark holds true for left-exact
and contravariant functors too. In good cases, such a continuation exists, and is
incredibly useful. We will discuss this in Chapter 23.

1.6 ⋆ Spectral sequences

 © 2024 Ravi Vakil. Published by Princeton University Press. 63

Je suis quelque peu affolé par ce déluge de cohomologie, mais j’ai courageusement tenu
le coup. Ta suite spectrale me paraı̂t raisonnable (je croyais, sur un cas particulier, l’avoir
mise en défaut, mais je m’étais trompé, et cela marche au contraire admirablement bien).
I am a bit panic-stricken by this flood of cohomology, but have borne up courageously.
Your spectral sequence seems reasonable to me (I thought I had shown that it was wrong in
a special case, but I was mistaken, on the contrary it works remarkably well).
— J.-P. Serre, letter to A. Grothendieck, March 14, 1956 [GrS, p. 38]
Spectral sequences are a powerful book-keeping tool for proving things involv-
ing complicated commutative diagrams. They were introduced by Leray in the
1940’s at the same time as he introduced sheaves. They have a reputation for being
abstruse and difficult. It has been suggested that the name ‘spectral’ was given
because, like specters, spectral sequences are terrifying, evil, and dangerous. I have
heard no one disagree with this interpretation, which is perhaps not surprising
since I just made it up.
Nonetheless, the goal of this section is to tell you enough that you can use
spectral sequences without hesitation or fear, and why you shouldn’t be frightened
when they come up in a seminar. What is perhaps different in this presentation is
that we will use spectral sequences to prove things that you may have already seen,
and that you can prove easily in other ways. This will allow you to get some hands-
on experience for how to use them. We will also see them only in the special case of
double complexes (the version by far the most often used in algebraic geometry),
and not in the general form usually presented (filtered complexes, exact couples,
etc.). See [Weib, Ch. 5] for more detailed information if you wish.
You should not read this section when you are reading the rest of Chapter 1.
Instead, you should read it just before you need it for the first time. When you
finally do read this section, you must do the exercises up to Exercise 1.6.F.
For concreteness, we work in the category ModA of module over a ring A.
However, everything we say will apply in any abelian category. (And if it helps
you feel secure, work instead in the category Veck of vector spaces over a field k.)

1.6.1. Double complexes.

A double complex is a collection of A-modules Ep,q (p, q ∈ Z), and “rightward”
morphisms dp,q → : E
p,q
→ Ep+1,q and “upward” morphisms dp,q ↑ : Ep,q → Ep,q+1 .
In the superscript, the first entry denotes the column number (the “x-coordinate”),
and the second entry denotes the row number (the “y-coordinate”). (Warning: this
is opposite to the convention for matrices.) The subscript is meant to suggest the
direction of the arrows. We will always write these as d→ and d↑ and ignore the
superscripts. We require that d→ and d↑ satisfy (a) d2→ = 0, (b) d2↑ = 0, and one more
condition: (c) either d→ d↑ = d↑ d→ (all the squares commute) or d→ d↑ + d↑ d→ = 0
(they all anticommute). Both come up in nature, and you can switch from one to
the other by replacing dp,q↑ with (−1)p dp,q
↑ . So I will assume that all the squares
anticommute, but that you know how to turn the commuting case into this one.
(You will see that there is no difference in the recipe, basically because the image
64 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

and kernel of a homomorphism f equal the image and kernel respectively of −f.)

dp,q+1
Ep,q+1
→
/ Ep+1,q+1
O O

dp,q
↑ anticommutes dp+1,q
↑

dp,q
Ep,q
→
/ Ep+1,q

There are variations on this definition, where for example the vertical arrows
go downwards, or some subset of the Ep,q is required to be zero.
From the double complex we construct a corresponding (single) complex E•
with Ek = ⊕i Ei,k−i , with d = d→ + d↑ . In other words, when there is a single
superscript k, we mean a sum of the kth antidiagonal of the double complex. The
single complex is sometimes called the total complex. Note that d2 = (d→ + d↑ )2 =
d2→ + (d→ d↑ + d↑ d→ ) + d2↑ = 0, so E• is indeed a complex.
The cohomology of the single complex is sometimes called the hypercohomol-
ogy of the double complex. We will instead use the phrase cohomology of the
double complex.
Our initial goal will be to find the cohomology of the double complex. You will
see later that we secretly also have other goals.
A spectral sequence is a recipe for computing some information about the
cohomology of the double complex. I won’t yet give the full recipe. Surprisingly,
this fragmentary bit of information is sufficient to prove lots of things.

1.6.2. Approximate Definition. A spectral sequence with rightward orientation is

a sequence of tables or pages → Ep,q p,q p,q p,q
0 , → E1 , → E2 , . . . (p, q ∈ Z), where → E0 =
p,q
E , along with a differential

p,q
→ dr : → Ep,q
r
/ p−r+1,q+r
→ Er

(r ∈ Z≥0 ) with → dp,q

r ◦ → dp+r−1,q−r
r = 0, and with an isomorphism of the coho-
mology of → dr at → Er (i.e., ker → dp,q
p,q
r / im → dr
p+r−1,q−r
) with → Ep,q
r+1 .
The orientation indicates that our 0th differential is the rightward one: d0 = d→ .
The left subscript “→” is usually omitted.
The order of the morphisms is best understood visually:

(1.6.2.1) •Z

•W

d3
d2 •O
d1

• /•
d0
 © 2024 Ravi Vakil. Published by Princeton University Press. 65

(the morphisms each apply to different pages). Notice that the map always is
“degree 1” in terms of the grading of the single complex E• . (You should figure out
what this informal statement really means.)
The actual definition describes what E•,• •,•
r and dr really are, in terms of E
•,•
.
We will describe d0 , d1 , and d2 below, and you should for now take on faith that
this sequence continues in some natural way.
Note that Ep,q
r is always a subquotient of the corresponding term on the ith
page Ep,q
i for all i < r. In particular, if Ep,q = 0, then Ep,q
r = 0 for all r.
Suppose now that E•,• is a first quadrant double complex, i.e., Ep,q = 0 for
p < 0 or q < 0 (so Ep,q r = 0 for all r unless p, q ∈ Z≥0 ). Then for any fixed p, q,
once r is sufficiently large, Ep,q •,•
r+1 is computed from (Er , dr ) using the complex

0Y

dp,q
r

Ep,q
r Y

p+r−1,q−r
dr

and thus we have canonical isomorphisms

Ep,q ∼ Ep,q =
= ∼ Ep,q =
∼ ··· .
r r+1 r+2

We denote this module Ep,q ∞ . The same idea works in other circumstances, for
example if the double complex is only nonzero in a finite number of rows — Ep,q =
0 unless q0 < q < q1 . This will come up for example in the mapping cone and long
exact sequence discussion (Exercises 1.6.F and 1.6.E below).
We now describe the first few pages of the spectral sequence explicitly. As
stated above, the differential d0 on E•,•
0 = E
•,•
is defined to be d→ . The rows are
complexes:
• /• /•

The 0th page E0 : • /• /•

• /• /•

and so E1 is just the table of cohomologies of the rows. You should check that
there are now vertical maps dp,q
1 : E1
p,q
→ Ep,q+1
1 of the row cohomology groups,
induced by d↑ , and that these make the columns into complexes. (This is essentially
the fact that a map of complexes induces a map on homology.) We have “used up
66 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

the horizontal morphisms”, but “the vertical differentials live on”.

•O •O •O

The 1st page E1 : •O •O •O

• • •

We take cohomology of d1 on E1 , giving us a new table, Ep,q

2 . It turns out that
there are natural morphisms from each entry to the entry two above and one to the
left, and that the composition of these two is 0. (It is a very worthwhile exercise
to work out how this natural morphism d2 should be defined. Your argument
may be reminiscent of the connecting homomorphism in the Snake Lemma 1.6.5 or
in the long exact sequence in cohomology arising from a short exact sequence of
complexes, Theorem 1.5.8. This is no coincidence.)

•W •W •

The 2nd page E2 : • • •

• • •

This is the beginning of a pattern.

Then it is a theorem that there is a filtration of Hk (E• ) by Ep,q
∞ where p + q = k.
(We can’t yet state it as an official Theorem because we haven’t precisely defined
the pages and differentials in the spectral sequence.) More precisely, there is a
filtration

E ∞   E∞ 
1,k−1 2,k−2
Ek,0
(1.6.2.2) E0,k
∞
/ ? / ···
∞
/ Hk (E• )

where the quotients are displayed above each inclusion. (Here is a tip for remember
which way the quotients are supposed to go. The differentials on later and later
pages point deeper and deeper into the filtration. Thus the entries in the direction
of the later arrowheads are the subobjects, and the entries in the direction of the
later “arrowtails” are quotients. This tip has the advantage of being independent of
the details of the spectral sequence, e.g., the “quadrant” or the orientation.)
We say that the spectral sequence → E•,• • •
• converges to H (E ). We often say
•,• • •
that → E2 (or any other page) abuts to H (E ).
Although the filtration gives only partial information about H• (E• ), sometimes
one can find H• (E• ) precisely. One example is if all Ei,k−i
∞ are zero, or if all but one
of them are zero (e.g., if E•,•
r has precisely one nonzero row or column, in which
case one says that the spectral sequence collapses at the rth step, although we will
not use this term). Another example is in the category of vector spaces over a field,
in which case we can find the dimension of Hk (E• ). Also, in lucky circumstances,
E2 (or some other small page) already equals E∞ .
 © 2024 Ravi Vakil. Published by Princeton University Press. 67

1.6.A. E XERCISE : INFORMATION FROM THE SECOND PAGE . Show that H0 (E• ) =
E0,0 0,0
∞ = E2 and

d1,0
0 / E0,1 / H1 (E• ) / E1,0 2
/ E0,2 / H2 (E• )
2 2 2

is exact.

1.6.3. The other orientation.

You may have observed that we could as well have done everything in the
opposite direction, i.e., reversing the roles of horizontal and vertical morphisms.
Then the sequences of arrows giving the spectral sequence would look like this
(compare to (1.6.2.1)).
(1.6.3.1) •O

• /•

'
•

%
•
•,•
This spectral sequence is denoted (“with the upward orientation”). Then
↑ E•
we would again get pieces of a filtration of H• (E• ) (where we have to be a bit
careful with the order with which ↑ Ep,q∞ corresponds to the subquotients — it is
the opposite order to that of (1.6.2.2) for → Ep,q
∞ ). Warning: in general there is no
isomorphism between → Ep,q ∞ and Ep,q
↑ ∞ .
In fact, this observation that we can start with either the horizontal or vertical
maps was our secret goal all along. Both algorithms compute information about the
same thing (H• (E• )), and usually we don’t care about the final answer — we often
care about the answer we get in one way, and we get at it by doing the spectral
sequence in the other way.

1.6.4. Examples.
We are now ready to see how this is useful. The moral of these examples is
the following. In the past, you may have proved various facts involving various
sorts of diagrams, by chasing elements around. Now, you will just plug them into
a spectral sequence, and let the spectral sequence machinery do your chasing for
you.

1.6.5. Example: Proving the Snake Lemma. Consider the diagram

(1.6.5.1) 0 /D /E /F /0
O O O
α β γ

0 /A /B /C /0

where the rows are exact in the middle (at A, B, C, D, E, F) and the squares
commute. (Normally the Snake Lemma is described with the vertical arrows
pointing downwards, but I want to fit this into my spectral sequence conventions.)
68 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

We wish to show that there is an exact sequence

(1.6.5.2) 0 → ker α → ker β → ker γ → coker α → coker β → coker γ → 0.

We plug this into our spectral sequence machinery. We first compute the coho-
mology using the rightward orientation, i.e., using the order (1.6.2.1). Then because
the rows are exact, Ep,q
1 = 0, so the spectral sequence has already converged:
Ep,q
∞ = 0.
We next compute this “0” in another way, by computing the spectral sequence
using the upward orientation. Then ↑ E•,•
1 (with its differentials) is:

0 / coker α / coker β / coker γ /0

0 / ker α / ker β / ker γ / 0.

Then ↑ E•,•
2 is of the form:

0 0

' '?
0 ?? ? 0

'? ' '

0 ? ?? 0

' '
0 0

We see that after ↑ E2 , all the terms will stabilize except for the double question marks
— all maps to and from the single question marks are to and from 0-entries. And
after ↑ E3 , even these two double-question-mark terms will stabilize. But in the end
our complex must be the 0 complex. This means that in ↑ E2 , all the entries must be
zero, except for the two double question marks, and these two must be isomorphic.
This means that 0 → ker α → ker β → ker γ and coker α → coker β → coker γ → 0
are both exact (that comes from the vanishing of the single question marks), and

∼ ker(coker α → coker β)
coker(ker β → ker γ) =

is an isomorphism (that comes from the equality of the double question marks).
Taken together, we have proved the exactness of (1.6.5.2), and hence the Snake
Lemma! (Notice: in the end we didn’t really care about the double complex. We
just used it as a prop to prove the Snake Lemma.)
Spectral sequences make it easy to see how to generalize results further. For
example, if A → B is no longer assumed to be injective, how would the conclusion
change?

1.6.B. U NIMPORTANT E XERCISE ( GRAFTING EXACT SEQUENCES , A VARIANT OF

THE S NAKE L EMMA ). Extend the Snake Lemma as follows. Suppose we have a
 © 2024 Ravi Vakil. Published by Princeton University Press. 69

commuting diagram
0O / X′ / Y′ / Z′ / A′ / ···
O O O O
a b c

··· /W /X /Y /Z / 0.
where the top and bottom rows are exact. Show that the top and bottom rows can
be “grafted together” to an exact sequence
··· /W / ker a / ker b / ker c

/ coker a / coker b / coker c / A′ / ··· .

1.6.6. Example: the Five Lemma. Suppose

(1.6.6.1) FO /G /H /I /J
O O O O
α β γ δ ϵ

A /B /C /D /E
where the rows are exact and the squares commute.
Suppose α, β, δ, ϵ are isomorphisms. We will show that γ is an isomorphism.
We first compute the cohomology of the total complex using the rightward
orientation (1.6.2.1). We choose this because we see that we will get lots of zeros.
Then → E•,•
1 looks like this:

?O 0O 0O 0O ?O

? 0 0 0 ?
Then → E2 looks similar, and the sequence will converge by E2 , as we will never get
any arrows between two nonzero entries in a table thereafter. We can’t conclude
that the cohomology of the total complex vanishes, but we can note that it vanishes
in all but four degrees — and most important, it vanishes in the two degrees
corresponding to the entries C and H (the source and target of γ).
We next compute this using the upward orientation (1.6.3.1). Then ↑ E1 looks
like this:
0 /0 /? /0 /0

0 /0 /? /0 /0
and the spectral sequence converges at this step. We wish to show that those two
question marks are zero. But they are precisely the cohomology groups of the total
complex that we just showed were zero — so we are done!
The best way to become comfortable with this sort of argument is to try it
out yourself several times, and realize that it really is easy. So you should do the
following exercises! Many can readily be done directly, but you should deliberately
try to use this spectral sequence machinery in order to get practice and develop
confidence.
70 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

1.6.C. E XERCISE : A SUBTLER F IVE L EMMA . By looking at the spectral sequence

proof of the Five Lemma above, prove a subtler version of the Five Lemma, where
one of the isomorphisms can instead just be required to be an injection, and another
can instead just be required to be a surjection. (I am deliberately not telling you
which ones, so you can see how the spectral sequence is telling you how to improve
the result.)

1.6.D. E XERCISE : ANOTHER SUBTLE VERSION OF THE FIVE LEMMA . If β and δ (in
(1.6.6.1)) are injective, and α is surjective, show that γ is injective. Give the dual
statement (whose proof is of course essentially the same).
The next two exercises no longer involve first quadrant double complexes. You
will have to think a little to realize why there is no reason for confusion or alarm.

1.6.E. I MPORTANT E XERCISE ( THE MAPPING CONE ). Suppose µ : A• → B• is a

morphism of complexes. Suppose C• is the single complex associated to the double
complex A• → B• . (C• is called the mapping cone of µ.) Show that there is a long
exact sequence of complexes:
· · · → Hi−1 (C• ) → Hi (A• ) → Hi (B• ) → Hi (C• ) → Hi+1 (A• ) → · · · .
(There is a slight notational ambiguity here; depending on how you index your
double complex, your long exact sequence might look slightly different.) In particu-
lar, we will use the fact that µ induces an isomorphism on cohomology if and only
if the mapping cone is exact. (We won’t use it until the proof of Theorem 18.2.4.)

1.6.F. E XERCISE . Use spectral sequences to show that a short exact sequence of
complexes gives a long exact sequence in cohomology (Theorem 1.5.8). (This is a
generalization of Exercise 1.6.E.)
The Grothendieck composition-of-functors spectral sequence (Theorem 23.3.5)
will be an important example of a spectral sequence that specializes in a number of
useful ways.
You are now ready to go out into the world and use spectral sequences to your
heart’s content!

1.6.7. Complete definition of spectral sequences, and proof.

You should most definitely not read the precise definition of a spectral se-
quence, and the proof that they work as advertised, any time soon after reading the
introduction to spectral sequences above. But after a suitable interval, you should
at least flip through a construction and proof to convince yourself that nothing
fancy is involved. The idea is not as bad as you might think, see [Vak2].
It is useful to notice that the proof implies that spectral sequences are functorial
in the 0th page: the spectral sequence formalism has good functorial properties in
the double complex. Unfortunately, Grothendieck’s terminology “spectral functor”
[Gr1, §2.4] did not catch on.
 CHAPTER 2

Sheaves

It is perhaps surprising that geometric spaces are often best understood in

terms of (nice) functions on them. For example, a differentiable manifold that
is a subset of Rn can be studied in terms of its smooth (C∞ ) functions. Because
“geometric spaces” can have few (everywhere-defined) functions, a more precise
version of this insight is that the structure of the space can be well understood
by considering all functions on all open subsets of the space. This information is
encoded and organized in something called a sheaf. Sheaves were introduced by
Leray in the 1940’s, and Serre introduced them to algebraic geometry. (The reason
for the name will be somewhat explained in Remark 2.4.3.) We will define sheaves
and describe useful facts about them. We will begin with a motivating example to
convince you that the notion is not so foreign.
One reason sheaves are slippery to work with is that they keep track of a huge
amount of information, and there are some subtle local-to-global issues. There are
also three different ways of getting a hold of them:
• in terms of open sets (the definition §2.2) — intuitive but in some ways
the least helpful;
• in terms of stalks (see §2.4.1); and
• in terms of a base of a topology (§2.5).
(Some people strongly prefer the espace étalé interpretation, §2.2.11, as well.)
Knowing which to use requires experience, so it is essential to do a number of
exercises on different aspects of sheaves in order to truly understand the concept.

2.1 Motivating example: The sheaf of smooth functions

Consider smooth (C∞ ) functions on the topological space X = Rn (or more
generally on a manifold X). The sheaf of smooth functions on X is the data of all
smooth functions on all open subsets on X. We will see how to manage these data,
and observe some of their properties. On each open set U ⊂ X, we have a ring of
smooth functions. We denote this ring of functions O(U).
Given a smooth function on an open set, you can restrict it to a smaller open
set, obtaining a smooth function there. In other words, if U ⊂ V is an inclusion of
open sets, we have a “restriction map” resV,U : O(V) → O(U).
Take a smooth function on a big open set, and restrict it to a medium open set,
and then restrict that to a small open set. The result is the same as if you restrict the

This material is published by Princeton University Press as The Rising Sea: Foundations of Algebraic
Geometry by Ravi Vakil. This version is free to view and download for personal use only. Not for
redistribution, resale, or use in derivative works. © 2024 Ravi Vakil.

71
72 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

smooth function on the big open set directly to the small open set. In other words,
if U ,→ V ,→ W, then the following diagram commutes:

O(W)
resW,V
/ O(V)

resW,U resV,U
# {
O(U)

Next take two smooth functions f1 and f2 on a big open set U, and an open
cover of U by some collection of open subsets {Ui }. (We say {Ui } covers U, or is
an open cover of U, if U = ∪Ui .) Suppose that f1 and f2 agree on each of these
Ui . Then they must have been the same function to begin with. In other words, if
{Ui }i∈I is a cover of U, and f1 , f2 ∈ O(U), and resU,Ui f1 = resU,Ui f2 , then f1 = f2 .
Thus we can identify functions on an open set by looking at them on a covering by
small open sets.
Finally, suppose you are given the same U and cover {Ui }, take a smooth
function on each of the Ui — a function f1 on U1 , a function f2 on U2 , and so on —
and assume they agree on the pairwise overlaps. Then they can be “glued together”
to make one smooth function on all of U. In other words, given fi ∈ O(Ui ) for all i,
such that resUi ,Ui ∩Uj fi = resUj ,Ui ∩Uj fj for all i and j, then there is some f ∈ O(U)
such that resU,Ui f = fi for all i.
The entire example above would have worked just as well with continuous
functions, or real-analytic functions, or just plain real-valued functions. Thus all
of these classes of “nice” functions share some common properties. We will soon
formalize these properties in the notion of a sheaf.

2.1.1. The germ of a smooth function. Before we do, we first give another
definition, that of the germ of a smooth function at a point p ∈ X. Intuitively, it is a
“shred” of a smooth function at p. Germs are objects of the form

(f, open set U) such that p ∈ U, f ∈ O(U)

modulo the relation that (f, U) ∼ (g, V) if there is some open set W ⊂ U, V contain-
ing p where f|W = g|W (i.e., resU,W f = resV,W g). In other words, two functions
that are the same in an open neighborhood of p (but may differ elsewhere) have
the same germ. We call this set of germs the stalk at p, and denote it Op . Notice
that the stalk is a ring: you can add two germs, and get another germ: if you have
a function f defined on U, and a function g defined on V, then f + g is defined on
U ∩ V. Moreover, f + g is well-defined: if f̃ has the same germ as f, meaning that
there is some open set W containing p on which they agree, and g̃ has the same
germ as g, meaning they agree on some open W ′ containing p, then f̃ + g̃ is the
same function as f + g on U ∩ V ∩ W ∩ W ′ .
Notice also that if p ∈ U, you get a map O(U) → Op . Experts may already see
that we are talking about germs as colimits.
We can see that Op is a local ring as follows. Consider those germs vanishing
at p, which we denote mp ⊂ Op . They certainly form an ideal: mp is closed under
addition, and when you multiply something vanishing at p by any function, the
result also vanishes at p. We check that this ideal is maximal by showing that the
 © 2024 Ravi Vakil. Published by Princeton University Press. 73

quotient ring is a field:

(2.1.1.1) 0 / mp := ideal of germs vanishing at p / Op f7→f(p)/ R /0

2.1.A. E XERCISE . Show that this is the only maximal ideal of Op . (Hint: show that
every element of Op \ mp is invertible.)
Note that we can interpret the value of a function at a point, or the value of
a germ at a point, as an element of the local ring modulo the maximal ideal. (We
will see that this doesn’t work for more general sheaves, but does work for things
behaving like sheaves of functions. This will be formalized in the notion of a locally
ringed space, which we will see, briefly, in §7.3.)

2.1.2. Aside. Notice that mp /m2p is a module over Op /mp = ∼ R, i.e., it is a real vector
space. It turns out to be naturally (whatever that means) the cotangent space to the
differentiable or analytic manifold at p. This insight will prove handy later, when
we define tangent and cotangent spaces of schemes.

2.1.B. ⋆ E XERCISE FOR THOSE WITH DIFFERENTIAL GEOMETRIC BACKGROUND .

Prove this. (Rhetorical question for experts: what goes wrong if the sheaf of
continuous functions is substituted for the sheaf of smooth functions? What goes
wrong if you use the sheaf of C1 functions?)

2.2 Definition of sheaf and presheaf

We now formalize these notions, by defining presheaves and sheaves. Presheaves
are simpler to define, and notions such as kernel and cokernel are straightforward.
Sheaves are more complicated to define, and some notions such as cokernel require
more thought. But sheaves are more useful because they are in some vague sense
more geometric; you can get information about a sheaf locally.

2.2.1. Definition of sheaf and presheaf on a topological space X.

To be concrete, we will define sheaves of sets. However, in the definition the
category Sets can be replaced by any reasonable category, and other important
examples are abelian groups Ab, k-vector spaces Veck , rings Rings, modules over a
ring ModA , and more. (You may have to think more when dealing with a category
of objects that aren’t “sets with additional structure”, but there aren’t any surprising
complications. In any case, this won’t be relevant for us, although people who
want to do this should start by solving Exercise 2.2.C.) Sheaves (and presheaves)
are often written in calligraphic font. The fact that F is a sheaf on a topological
space X is often written as
F

2.2.2. Definition: Presheaf. A presheaf of sets F on a topological space X is the

following data.
74 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

• To each open set U ⊂ X, we have a set F (U) (e.g., the set of differentiable
functions in our motivating example). (Notational warning: Several notations are
in use, for various good reasons: F (U) = Γ (U, F ) = H0 (U, F ). We will use them
all.) The elements of F (U) are called sections of F over U. (§2.2.11 combined
with Exercise 2.2.G gives a motivation for this terminology, although this isn’t so
important for us.)
By convention, if the “U” is omitted, it is implicitly taken to be X: “sections of
F ” means “sections of F over X”. These are also called global sections.
• For each inclusion U ,→ V of open sets, we have a restriction map
resV,U : F (V) → F (U)
(just as we did for differentiable functions). If f ∈ F (V), we often write
f|U
for resV,U (f).
The data is required to satisfy the following two conditions.
• The map resU,U is the identity: resU,U = idF (U) .
• If U ,→ V ,→ W are inclusions of open sets, then the restriction maps compose
as you would expect, i.e., the diagram

F (W)
resW,V
/ F (V)

resW,U resV,U
$ {
F (U)
commutes.

2.2.A. E XERCISE FOR CATEGORY- LOVERS : “A PRESHEAF IS THE SAME AS A CON -

TRAVARIANT FUNCTOR ”. Given any topological space X, we have a “category
of open sets” (Example 1.1.9), where the objects are the open sets and the mor-
phisms are inclusions. Verify that the data of a presheaf is precisely the data of a
contravariant functor from the category of open sets of X to the category of sets.
(This interpretation is surprisingly useful.)

2.2.3. Definition: Stalks and germs. We define the stalk of a presheaf at a point in
two equivalent ways. One will be hands-on, and the other will be as a colimit.

2.2.4. Define the stalk of a presheaf F at a point p to be the set of germs of F at p,

denoted Fp , as in the example of §2.1.1. Germs correspond to sections over some
open set containing p, and two of these sections are considered the same if they
agree on some smaller open set containing p. More precisely: the stalk is
{(f, open U) : p ∈ U, f ∈ F (U)}
modulo the relation that (f, U) ∼ (g, V) if there is some open set W ⊂ U, V where
p ∈ W and resU,W f = resV,W g. (To explain the agricultural terminology: the
French name “germe” is meant to suggest a tiny shoot sprouting from a seed, cf.
“germinate”.)

2.2.5. A useful equivalent definition of a stalk is as a colimit of all F (U) over all
open sets U containing p:
Fp = colim F (U).
 © 2024 Ravi Vakil. Published by Princeton University Press. 75

The index category is a filtered set (given any two such open sets, there is a third
such set contained in both), so these two definitions are the same by Exercise 1.3.E.
Hence we can define stalks for sheaves of sets, groups, rings, and other things for
which colimits exist for directed sets. It is very helpful to keep both definitions of
stalk in mind at the same time.
If p ∈ U, and f ∈ F (U), then the image of f in Fp is called the germ of f at p.
(Warning: unlike the example of §2.1.1, in general, the value of a section at a point
doesn’t make sense.)

2.2.6. Definition: Sheaf. A presheaf is a sheaf if it satisfies two more axioms.

Notice that these axioms use the additional information of when some open sets
cover another.
Identity axiom. For any open set U, if {Ui }i∈I is an open cover of U, and
f1 , f2 ∈ F (U), and f1 |Ui = f2 |Ui for all i, then f1 = f2 .
(A presheaf satisfying the identity axiom is called a separated presheaf, but
we will not use that notation in any essential way.)
Gluability axiom. If {Ui }i∈I is an open cover of U, then given fi ∈ F (Ui ) for
all i, such that fi |Ui ∩Uj = fj |Ui ∩Uj for all i, j, then there is some f ∈ F (U) such that
resU,Ui f = fi for all i.
In mathematics, definitions often come paired: “at most one” and “at least one”.
In this case, identity means there is at most one way to glue, and gluability means
that there is at least one way to glue.
(For experts and scholars of the empty set only: an additional axiom some-
times included is that F (∅) is a one-element set, and in general, for a sheaf with
values in a category, F (∅) is required to be the final object in the category. This
actually follows from the above definitions, assuming that the empty product is
appropriately defined as the final object.)
Example. If U and V are disjoint, then F (U ∪ V) = F (U) × F (V). Here we use
the fact that F (∅) is the final object.
The stalk of a sheaf at a point is just its stalk as a presheaf — the same definition
applies — and similarly for the germs of a section of a sheaf.

2.2.B. U NIMPORTANT EXERCISE : PRESHEAVES THAT ARE NOT SHEAVES . Show that
the following are presheaves on C (with the classical topology), but not sheaves:
(a) bounded functions, (b) holomorphic functions admitting a holomorphic square
root.
Both of the presheaves in the previous Exercise satisfy the identity axiom. A
“natural” example failing even the identity axiom is implicit in Remark 2.5.5.
We now make a couple of points intended only for category-lovers.

2.2.7. Interpretation in terms of the equalizer exact sequence. The two axioms for a
presheaf to be a sheaf can be interpreted as “exactness” of the “equalizer exact
Q Q
sequence”: · / F (U) / F (Ui ) // F (Ui ∩ Uj ). Identity is exact-
Q
ness at F (U), and gluability is exactness at F (Ui ). I won’t make this precise, or
even explain what the double right arrow means. (What is an exact sequence of
sets?!) But you may be able to figure it out from the context.

2.2.C. E XERCISE . The identity and gluability axioms may be interpreted as saying
that F (∪i∈I Ui ) is a certain limit. What is that limit?
76 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Here are a number of examples of sheaves.

2.2.D. E XERCISE .
(a) Verify that the examples of §2.1 are indeed sheaves (of smooth functions, or
continuous functions, or real-analytic functions, or plain real-valued functions, on
a manifold or Rn ).
(b) Show that real-valued continuous functions on (open sets of) a topological space
X form a sheaf.

2.2.8. Important Example: Restriction of a sheaf. Suppose F is a sheaf on X, and U

is an open subset of X. Define the restriction of F to U, denoted F |U , to be the
collection F |U (V) = F (V) for all open subsets V ⊂ U. Clearly this is a sheaf on
U. (Unimportant but fun fact: §2.7 will tell us how to restrict sheaves to arbitrary
subsets.)

2.2.9. Important Example: the skyscraper sheaf. Suppose X is a topological space, with
p ∈ X, and S is a set. Let ip : p → X be the inclusion. Then ip,∗ S defined by

S if p ∈ U, and
ip,∗ S(U) =
{e} if p ∈
/U

forms a sheaf. Here {e} is any one-element set. (Check this if it isn’t clear to you —
what are the restriction maps?) This is called a skyscraper sheaf supported at p,
because the informal picture of it looks like a skyscraper at p. (Mild caution: this
informal picture suggests that the only nontrivial stalk of a skyscraper sheaf is at p,
which isn’t the case. Exercise 6.1.B(b) gives an example, although it certainly isn’t
the simplest one.) There is an analogous definition for sheaves of abelian groups,
except ip,∗ S(U) = {0} if p ∈
/ U; and for sheaves with values in a category more
generally, ip,∗ S(U) should be a final object.
(This notation is admittedly hideous, and the alternative (ip )∗ S is equally bad.
In §2.2.12 we explain this notation.)

2.2.10. Constant presheaves and constant sheaves. Let X be a topological space, and S
a set. Define Spre (U) = S for all open sets U. You will readily verify that Spre forms
a presheaf (with restriction maps the identity). This is called the constant presheaf
associated to S. This isn’t (in general) a sheaf. (It may be distracting to say why.
Lovers of the empty set will insist that the sheaf axioms force the sections over the
empty set to be the final object in the category, i.e., a one-element set. But even if
we patch the definition by setting Spre (∅) = {e}, if S has more than one element,
and X is the two-point space with the discrete topology, i.e., where every subset is
open, you can check that Spre fails gluability.)

2.2.E. E XERCISE ( CONSTANT SHEAVES ). Now let F (U) be the set of maps U → S
that are locally constant, i.e., for any point p in U, there is an open neighborhood
of p where the function is constant. Show that this is a sheaf. (A better description
is this: endow S with the discrete topology, and let F (U) be the continuous maps
U → S.) This is called the constant sheaf (with values in S); do not confuse it with
the constant presheaf. (I would prefer the name “locally constant sheaf”, but it is
too late in history for this change.) We denote this sheaf S.
 © 2024 Ravi Vakil. Published by Princeton University Press. 77

2.2.F. E XERCISE (“ MORPHISMS GLUE ”). Suppose Y is a topological space. Show

that “continuous maps to Y” form a sheaf of sets on X. More precisely, to each open
set U of X, we associate the set of continuous maps of U to Y. Show that this forms
a sheaf. (Exercise 2.2.D(b), with Y = R, and Exercise 2.2.E, with Y = S with the
discrete topology, are both special cases.)

2.2.G. E XERCISE . This is a fancier version of the previous exercise.

(a) (sheaf of sections of a map) Suppose we are given a continuous map µ : Y → X.
Show that “sections of µ” form a sheaf. More precisely, to each open set U of X,
associate the set of continuous maps s : U → Y such that µ ◦ s = id|U . Show that
this forms a sheaf. (For those who have heard of vector bundles, these are a good
example.) This is motivation for the phrase “section of a sheaf”.
(b) (This exercise is for those who know what a topological group is. If you don’t
know what a topological group is, you might be able to guess.) Suppose that Y is a
topological group. Show that continuous maps to Y form a sheaf of groups.

2.2.11. ⋆ The space of sections (espace étalé) of a (pre)sheaf. Depending on your back-
ground, you may prefer the following perspective on sheaves. Suppose F is a
presheaf of sets (e.g., a sheaf) on a topological space X. Construct a topological
space F along with a continuous map π : F → X as follows: as a set, F is the disjoint
union of all the stalks of F . This naturally gives a map of sets π : F → X. Topologize
F as follows. Each s in F (U) determines a subset {(x, sx ) : x ∈ U} of F. The
topology on F is the weakest topology such that these subsets are open. (These
subsets form a base of the topology. For each y ∈ F, there is an open neighborhood
V of y and an open neighborhood U of π(y) such that π|V is a homeomorphism
from V to U. Do you see why these facts are true?) The topological space F could
be thought of as the space of sections of F (and in French is called the espace étalé
of F ). We will not discuss this construction at any length, but it can have some
advantages: (a) It is always better to know as many ways as possible of thinking
about a concept. (b) “Inverse image” (informally, “pullback”) has a natural inter-
pretation in this language (mentioned briefly in Exercise 2.7.C). (c) Sheafification
has a natural interpretation in this language (see Remark 2.4.7).

2.2.H. I MPORTANT E XERCISE /D EFINITION : THE PUSHFORWARD SHEAF OR DIRECT

IMAGE SHEAF. Suppose π : X → Y is a continuous map, and F is a presheaf on
X. Then define a presheaf π∗ F on Y by π∗ F (V) = F (π−1 (V)), where V is an
open subset of Y. Show that π∗ F is a presheaf on Y, and is a sheaf if F is. This is
called the pushforward (or direct image) of F . More precisely, π∗ F is called the
pushforward of F by π.

2.2.12. As the notation suggests, the skyscraper sheaf (Example 2.2.9) can be
interpreted as the pushforward of the constant sheaf S on a one-point space p,
under the inclusion morphism ip : {p} → X.
Once we endow sheaves with the structure of a category, we will see that the
pushforward is a functor from sheaves on X to sheaves on Y (Exercise 2.3.B).

2.2.I. E XERCISE ( PUSHFORWARD INDUCES MAPS OF STALKS ). Suppose π : X → Y

is a continuous map, and F is a sheaf of sets (or rings or A-modules) on X. If
π(p) = q, describe the natural morphism of stalks (π∗ F )q → Fp . (You can use the
78 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

explicit definition of stalk using representatives, §2.2.4, or the universal property,

§2.2.5. If you prefer one way, you should try the other.)

2.2.13. Important Example: Ringed spaces, and OX -modules. Suppose OX is a

sheaf of rings on a topological space X (i.e., a sheaf on X with values in the category
of Rings). Then (X, OX ) is called a ringed space. The sheaf of rings is often denoted
by OX , pronounced “oh-X”. This sheaf is called the structure sheaf of the ringed
space. Sections of the structure sheaf OX over an open subset U are called functions
on U. Functions on X are called global functions, or just functions. (Caution:
what we call “functions”, others sometimes call “regular functions”. Furthermore, we will
later define “rational functions” on schemes in §5.2.I and §6.6.36, which are not precisely
functions in this sense; they are a particular type of “partially-defined function”.)
The symbol OX will always refer to the structure sheaf of a ringed space X.
The restriction OX |U of OX to an open subset U ⊂ X is denoted OU . (We will later
call (U, OU ) → (X, OX ) an open embedding of ringed spaces, see Definition 7.2.1.)
The stalk of OX at a point p is written “OX,p ”, because this looks less hideous than
“OXp ”.
Just as we have modules over a ring, we have OX -modules over a sheaf of rings
OX . There is only one possible definition that could go with the name OX -module
(or often O-module) — a sheaf of abelian groups F with the following additional
structure. For each U, F (U) is an OX (U)-module. Furthermore, this structure
should behave well with respect to restriction maps: if U ⊂ V, then

OX (V) × F (V)
action / F (V)

(2.2.13.1) resV,U × resV,U resV,U

 action /

OX (U) × F (U) F (U)

commutes. (You should convince yourself that I haven’t forgotten anything.)

Recall that the notion of A-module generalizes the notion of abelian group,
because an abelian group is the same thing as a Z-module. Similarly, the notion
of OX -module generalizes the notion of sheaf of abelian groups, because the latter
is the same thing as a Z-module. Hence when we are proving things about OX -
modules, we are also proving things about sheaves of abelian groups.

2.2.J. E XERCISE . If (X, OX ) is a ringed space, and F is an OX -module, describe

how for each p ∈ X, Fp is an OX,p -module.

2.2.14. For those who know about vector bundles. The motivating example of OX -
modules is the sheaf of sections of a vector bundle. If (X, OX ) is a differentiable
manifold (so OX is the sheaf of smooth functions), and π : V → X is a vector bundle
over X, then the sheaf of smooth sections σ : X → V is an OX -module. Indeed, given
a section s of π over an open subset U ⊂ X, and a function f on U, we can multiply
s by f to get a new section fs of π over U. Moreover, if U ′ is a smaller subset, then
we could multiply f by s and then restrict to U ′ , or we could restrict both f and s
to U ′ and then multiply, and we would get the same answer. That is precisely the
commutativity of (2.2.13.1).
 © 2024 Ravi Vakil. Published by Princeton University Press. 79

2.3 Morphisms of presheaves and sheaves

2.3.1. Definitions. Whenever one defines a new mathematical object, category theory
teaches to try to understand maps between them. We now define morphisms of
presheaves, and similarly for sheaves. In other words, we will describe the category
of presheaves (of sets, abelian groups, etc.) and the category of sheaves.
A morphism of presheaves of sets (or indeed of presheaves with values in
any category) on X, ϕ : F → G , is the data of maps ϕ(U) : F (U) → G (U) for all U
behaving well with respect to restriction: if U ,→ V then

ϕ(V)
F (V) / G (V)

resV,U resV,U
 ϕ(U) 
F (U) / G (U)

commutes. (Notice: the underlying space of both F and G is X.)

Morphisms of sheaves are defined identically: the morphisms from a sheaf F
to a sheaf G are precisely the morphisms from F to G as presheaves. (Translation:
The category of sheaves on X is a full subcategory of the category of presheaves on
X.) If (X, OX ) is a ringed space, then morphisms of OX -modules have the obvious
definition. (Can you write it down?)
An example of a morphism of sheaves is the map from the sheaf of smooth
functions on R to the sheaf of continuous functions. This is a “forgetful map”: we
are forgetting that these functions are differentiable, and remembering only that
they are continuous.

2.3.2. Notation. We may as well set some notation: let SetsX , AbX , etc. denote
the category of sheaves of sets, abelian groups, etc. on a topological space X. Let
pre
ModOX denote the category of OX -modules on a ringed space (X, OX ). Let SetsX ,
etc. denote the category of presheaves of sets, etc. on X.

2.3.3. Aside for category-lovers. If you interpret a presheaf on X as a contravariant

functor (from the category of open sets), a morphism of presheaves on X is a natural
transformation of functors (§1.1.21).

2.3.A. E XERCISE : MORPHISMS OF ( PRE ) SHEAVES INDUCE MORPHISMS OF STALKS .

If ϕ : F → G is a morphism of presheaves on X, and p ∈ X, describe an induced
morphism of stalks ϕp : Fp → Gp . Translation: taking the stalk at p induces a
stalkification functor SetsX → Sets. (Your proof will extend in obvious ways. For
example, if ϕ is a morphism of OX -modules, then ϕp is a map of OX,p -modules.)

2.3.B. E XERCISE . Suppose π : X → Y is a continuous map of topological spaces (i.e.,

a morphism in the category of topological spaces). Show that pushforward gives a
functor π∗ : SetsX → SetsY . Here Sets can be replaced by other categories. (Watch
out for some possible confusion: a presheaf is a functor, and presheaves form a
category. It may be best to forget that presheaves are functors for now.)

2.3.C. I MPORTANT E XERCISE AND D EFINITION : “S HEAF Hom”. Suppose F and

G are two sheaves of sets on X. (In fact, it will suffice that F is a presheaf.) Let
80 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Hom (F , G ) be the collection of data

Hom (F , G )(U) := Mor(F |U , G |U ).
(Recall the notation F |U , the restriction of the sheaf to the open set U, Exam-
ple 2.2.8.) Show that this is a sheaf of sets on X. (To avoid a common confusion:
the right side does not say Mor(F (U), G (U)).) This sheaf is called “sheaf Hom ”.
(Strictly speaking, we should reserve Hom for when we are in an additive category,
so this should possibly be called “sheaf Mor”. But the terminology “sheaf Hom” is
too established to uproot.) It will be clear from your construction that, like Hom,
Hom is a contravariant functor in its first argument and a covariant functor in its
second argument.
Warning: Hom does not commute with taking stalks. More precisely: it is
not true that Hom (F , G )p is isomorphic to Hom(Fp , Gp ). (Can you think of a
counterexample? There is at least a map from one of these to the other — in which
direction?)

2.3.4. We will use many variants of the definition of Hom . For example, if F and
G are sheaves of abelian groups on X, then Hom AbX (F , G ) is defined by taking
Hom AbX (F , G )(U) to be the maps as sheaves of abelian groups F |U → G |U . (Note
that Hom AbX (F , G ) has the structure of a sheaf of abelian groups in a natural way.)
Similarly, if F and G are OX -modules, we define Hom ModOX (F , G ) in the analogous
way (and it is an OX -module). Obnoxiously, the subscripts AbX and ModOX are
often dropped (here and in the literature), so be careful which category you are
working in! We call Hom ModOX (F , OX ) the dual of the OX -module F , and denote it
F ∨.

2.3.D. U NIMPORTANT E XERCISE ( REALITY CHECK ).

∼ F , where {p} is
(a) If F is a sheaf of sets on X, then show that Hom ({p}, F ) =
the constant sheaf “with values in the one element set {p}”.
(b) If F is a sheaf of abelian groups on X, then show that Hom AbX (Z, F ) =∼F
(an isomorphism of sheaves of abelian groups).
(c) If F is an OX -module, then show that Hom ModOX (OX , F ) = ∼ F (an isomor-
phism of OX -modules).
A key idea in (b) and (c) is that 1 “generates” (in some sense) Z (in (b)) and OX (in
(c)).

2.3.5. Presheaves of abelian groups (and even “presheaf OX -modules”) form an

abelian category.
We can make module-like constructions using presheaves of abelian groups
on a topological space X. (Throughout this section, all (pre)sheaves are of abelian
groups.) For example, we can clearly add maps of presheaves and get another map
of presheaves: if ϕ, ψ : F → G , then we define the map ϕ + ψ by (ϕ + ψ)(V) =
ϕ(V) + ψ(V). (There is something small to check here: that the result is indeed a
map of presheaves.) In this way, presheaves of abelian groups form an additive
category (Definition 1.5.1: the morphisms between any two presheaves of abelian
groups form an abelian group; there is a 0-object; and one can take finite products).
For exactly the same reasons, sheaves of abelian groups also form an additive
category.
 © 2024 Ravi Vakil. Published by Princeton University Press. 81

If ϕ : F → G is a morphism of presheaves, define the presheaf kernel kerpre ϕ

by (kerpre ϕ)(U) := ker ϕ(U).

2.3.E. E XERCISE . Show that kerpre ϕ is a presheaf. (Hint: if U ,→ V, define the

restriction map by chasing the following diagram:

0 / kerpre ϕ(V) / F (V) / G (V)

∃! resV,U resV,U
  
0 / kerpre ϕ(U) / F (U) / G (U)

You should check that the restriction maps compose as desired.)

Define the presheaf cokernel cokerpre ϕ similarly. It is a presheaf by essentially
the same (dual) argument.

2.3.F. E XERCISE : THE COKERNEL DESERVES ITS NAME . Show that the presheaf cok-
ernel satisfies the universal property of cokernels (Definition 1.5.4) in the category
of presheaves.
Similarly, kerpre ϕ → F satisfies the universal property for kernels in the
category of presheaves.
It is not too tedious to verify that presheaves of abelian groups form an abelian
category, and the reader is free to do so. The key idea is that all abelian-categorical
notions may be defined and verified “open set by open set”. We needn’t worry
about restriction maps — they “come along for the ride”. Hence we can define terms
such as subpresheaf, image presheaf (or presheaf image), and quotient presheaf
(or presheaf quotient), and they behave as you would expect. You construct kernels,
quotients, cokernels, and images open set by open set. Homological algebra (exact
sequences and so forth) works, and also “works open set by open set”. In particular:

2.3.G. E ASY E XERCISE . Show (or observe) that for a topological space X with open
pre
set U, F 7→ F (U) gives a functor from presheaves of abelian groups on X, AbX , to
abelian groups, Ab. Then show that this functor is exact.

2.3.H. E XERCISE . Show that a sequence of presheaves 0 → F1 → F2 → · · · →

Fn → 0 is exact if and only if 0 → F1 (U) → F2 (U) → · · · → Fn (U) → 0 is exact
for all U.
The above discussion essentially carries over without change to presheaves
with values in any abelian category. (Think this through if you wish.)
However, we are interested in more geometric objects, sheaves, where things
can be understood in terms of their local behavior, thanks to the identity and gluing
axioms. We will soon see that sheaves of abelian groups also form an abelian
category, but a complication will arise that will force the notion of sheafification on
us. Sheafification will be the answer to many of our prayers. We just haven’t yet
realized what we should be praying for.
To begin with, sheaves AbX form an additive category, as described in the first
paragraph of §2.3.5.
Kernels work just as with presheaves:
82 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

2.3.I. I MPORTANT E XERCISE . Suppose ϕ : F → G is a morphism of sheaves.

Show that the presheaf kernel kerpre ϕ is in fact a sheaf. Show that it satisfies
the universal property of kernels (Definition 1.5.4). (Hint: the second question
follows immediately from the fact that kerpre ϕ satisfies the universal property in
the category of presheaves.)
Thus if ϕ is a morphism of sheaves, we define
ker ϕ := kerpre ϕ.
The problem arises with the cokernel.

2.3.J. I MPORTANT E XERCISE . Let X be C with the classical topology, let OX be the
sheaf of holomorphic functions, and let F be the presheaf of functions admitting a
holomorphic logarithm. Describe an exact sequence of presheaves on X:
0 /Z / OX /F /0

where Z → OX is the natural inclusion and OX → F is given by f 7→ exp(2πif).

(Be sure to verify exactness.) Show that F is not a sheaf. (Hint: F does not satisfy
the gluability axiom. The problem is that there are functions that don’t have a
logarithm but locally have a logarithm.) This will come up again in Example 2.4.9.
We will have to put our hopes for understanding cokernels of sheaves on hold
for a while. We will first learn to understand sheaves using stalks.

2.4 Properties determined at the level of stalks, and sheafification

2.4.1. Properties determined by stalks. We now come to the second way of

understanding sheaves mentioned at the start of the chapter. In this section, we
will see that lots of facts about sheaves can be checked “at the level of stalks”. We
call any property determined at the level of stalks stalk-local. This isn’t true for
presheaves, and reflects the local nature of sheaves. We will see that sections and
morphisms are determined “by their stalks”, and the property of a morphism being
an isomorphism may be checked at stalks. (The last one is the trickiest.)

2.4.A. I MPORTANT EASY EXERCISE (sections are determined by germs). Prove that
a section of a sheaf of sets is determined by its germs, i.e., the natural map
F (U) /Q
(2.4.1.1) p∈U Fp

is injective. Hint 1: you won’t use the gluability axiom, so this is true for separated
presheaves. Hint 2: it is false for presheaves in general, see Exercise 2.4.E, so you
will use the identity axiom. (Your proof will also apply to sheaves of groups, rings,
etc. — to categories of “sets with additional structure”. The same is true of many
exercises in this section.)
Exercise 2.4.A suggests a question: which elements of the right side of (2.4.1.1)
are in the image of the left side?

2.4.2.
Q Important definition. We say that an element (sp )p∈U of the right side
p∈U Fp of (2.4.1.1) consists of compatible germs if for all p ∈ U, there is some
 © 2024 Ravi Vakil. Published by Princeton University Press. 83

representative
(s̃p ∈ F (Up ), Up open in U)
for sp (where p ∈ Up ⊂ U) such that the germ of s̃p at all q ∈ Up is sq . Equivalently,
there is an open cover {Ui } of U, and sections fi ∈ F (Ui ), such that if p ∈ Ui , then
sp is the germ of fi at p. Clearly any section s of F over U gives a choice of
compatible germs for U.

2.4.B. I MPORTANT E XERCISE . Prove that any choice of compatible germs for a
sheaf of sets F over U is the image of a section of F over U. (Hint: you will use
gluability.)
We have thus completely described the image of (2.4.1.1), in a way that will
prove useful.

2.4.3. Remark. This perspective motivates the agricultural terminology “sheaf”: a

sheaf is (the data of) a bunch of stalks, bundled together appropriately.
Now we throw morphisms into the mix. Recall Exercise 2.3.A: morphisms of
(pre)sheaves induce morphisms of stalks.

2.4.C. E XERCISE (morphisms are determined by stalks). If ϕ1 and ϕ2 are mor-

phisms from a presheaf of sets F to a sheaf of sets G that induce the same maps on
each stalk, show that ϕ1 = ϕ2 . Hint: consider the following diagram.

(2.4.3.1) F (U) / G (U)

_

Q  
Fp /Q Gp
p∈U p∈U

2.4.D. T RICKY E XERCISE (isomorphisms are determined by stalks). Show that

a morphism of sheaves of sets is an isomorphism if and only if it induces an
isomorphism of all stalks. Hint: Use (2.4.3.1). Once you have injectivity, show
surjectivity, perhaps using Exercise 2.4.B, or gluability in some other way; this
is more subtle. Warning: This exercise does not say that if two sheaves have
isomorphic stalks, then they are isomorphic.

2.4.E. E XERCISE .
(a) Show that Exercise 2.4.A is false for general presheaves.
(b) Show that Exercise 2.4.C is false for general presheaves.
(c) Show that Exercise 2.4.D is false for general presheaves.
(General hint for finding counterexamples of this sort: consider a 2-point space
with the discrete topology.)

2.4.4. Sheafification.
Every sheaf is a presheaf (and indeed by definition sheaves on X form a full
subcategory of the category of presheaves on X). Just as groupification (§1.4.3)
gives an abelian group that best approximates an abelian semigroup, sheafification
gives the sheaf that best approximates a presheaf, with an analogous universal
property. (One possible example to keep in mind is the sheafification of the presheaf
84 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

of holomorphic functions admitting a square root on C with the classical topology.

See also the exponential exact sequence, Example 2.4.9.)

2.4.5. Definition. If F is a presheaf on X, then a morphism of presheaves sh : F →

F sh on X is a sheafification of F if F sh is a sheaf, and for any sheaf G , and
any presheaf morphism g : F → G , there exists a unique morphism of sheaves
f : F sh → G making the diagram

F
sh / F sh

f
g
! 
G

commute.
We still have to show that it exists. The following two exercises require existence
(which we will show shortly), but not the details of the construction.

2.4.F. E XERCISE . Show that sheafification is unique up to unique isomorphism, as-

suming it exists. Show that if F is a sheaf, then the sheafification is id : F /F .
(This should be second nature by now.)

2.4.G. E ASY E XERCISE ( SHEAFIFICATION IS A FUNCTOR ). Assume for now that

sheafification exists. Use the universal property to show that for any morphism of
presheaves ϕ : F → G , we get a natural induced morphism of sheaves ϕsh : F sh →
G sh . Show that sheafification is a functor from presheaves on X to sheaves on X.

2.4.6. Construction. We next show that any presheaf of sets (or groups, rings, etc.)
has a sheafification. Suppose F is a presheaf. Define F sh by defining F sh (U) as the
set of “compatible (families of) germs” of the presheaf F over U. Explicitly:

F sh (U) := {(fp ∈ Fp )p∈U : for all p ∈ U, there exists an open neighborhood V

of p, contained in U, and s ∈ F (V) with sq = fq for all q ∈ V}.

Here sq means the germ of s at q — the image of s in the stalk Fq .

2.4.H. E ASY E XERCISE . Show that F sh (using the tautological restriction maps)
forms a sheaf.

2.4.I. E ASY E XERCISE . Describe a natural map of presheaves sh : F → F sh .

2.4.J. E XERCISE . Show that the map sh satisfies the universal property of sheafifi-
cation (Definition 2.4.5). (This is easier than you might fear.)

2.4.K. U SEFUL EXERCISE , NOT JUST FOR CATEGORY- LOVERS . Show that the
sheafification functor is left-adjoint to the forgetful functor from sheaves on X to
presheaves on X. This is not difficult — it is largely a restatement of the universal
property. But it lets you use results from §1.5.14, and can “explain” why you
don’t need to sheafify when taking kernel (why the presheaf kernel is already the
sheaf kernel), and why you need to sheafify when taking cokernel and (soon, in
Exercise 2.6.K) ⊗.
 © 2024 Ravi Vakil. Published by Princeton University Press. 85

2.4.L. E XERCISE . Show F → F sh induces an isomorphism of stalks. (Possible

hint: Use the concrete description of the stalks. Another possibility once you read
Remark 2.7.3: judicious use of adjoints.)
As a reality check, you may want to verify that “the sheafification of a constant
presheaf is the corresponding constant sheaf” (see §2.2.10): if X is a topological
space and S is a set, then (Spre )sh may be naturally identified with S.

2.4.7. ⋆ Remark. The “space of sections” (or “espace étalé”) construction (§2.2.11)
yields a different-sounding description of sheafification which may be preferred
by some readers. The main idea is identical: if F is a presheaf, let F be the space
of sections (or espace étalé) of F . You may wish to show that if F is a presheaf,
the sheaf of sections of F → X (defined in Exercise 2.2.G(a)) is the sheafification
of F . Exercise 2.2.E may be interpreted as an example of this construction. The
“space of sections” construction of the sheafification is essentially the same as
Construction 2.4.6.

2.4.8. Subsheaves and quotient sheaves.

We now discuss subsheaves and quotient sheaves from the perspective of
stalks.

2.4.M. E XERCISE . Suppose ϕ : F → G is a morphism of sheaves of sets on a

topological space X. Show that the following are equivalent.
(a) ϕ is a monomorphism in the category of sheaves.
(b) ϕ is injective on the level of stalks: ϕp : Fp → Gp is injective for all p ∈ X.
(c) ϕ is injective on the level of open sets: ϕ(U) : F (U) → G (U) is injective
for all open U ⊂ X.
(Possible hints: for (b) implies (a), recall that morphisms are determined by stalks,
Exercise 2.4.C. For (a) implies (c), use the “indicator sheaf” with one section over
every open set contained in U, and no section over any other open set.) If these
conditions hold, we say that F is a subsheaf of G (where the “inclusion” ϕ is
sometimes left implicit).
(You may later wish to extend your solution to Exercise 2.4.M to show that
for any morphism of presheaves, if all maps of sections are injective, then all stalk
maps are injective. And furthermore, if ϕ : F → G is a morphism from a separated
presheaf to an arbitrary presheaf, then injectivity on the level of stalks implies
that ϕ is a monomorphism in the category of presheaves. This is useful in some
approaches to Exercise 2.6.C.)

2.4.N. E XERCISE . Continuing the notation of the previous exercise, show that the
following are equivalent.
(a) ϕ is an epimorphism in the category of sheaves.
(b) ϕ is surjective on the level of stalks: ϕp : Fp → Gp is surjective for all
p ∈ X.
(Possible hint: use a skyscraper sheaf.)
If these conditions hold, we say that G is a quotient sheaf of F .
Thus monomorphisms and epimorphisms — subsheafiness and quotient sheafiness —
can be checked at the level of stalks.
86 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Both exercises generalize readily to sheaves with values in any reasonable

category, where “injective” is replaced by “monomorphism” and “surjective” is
replaced by “epimorphism”.
Notice that there was no part (c) to Exercise 2.4.N, and Example 2.4.9 shows
why. (But there is a version of (c) that implies (a) and (b): surjectivity on all open
sets in any base of a topology implies that the corresponding map of sheaves is an
epimorphism, Exercise 2.5.D.)

2.4.9. Example (cf. Exercise 2.3.J). Let X = C with the classical topology, and define
OX to be the sheaf of holomorphic functions, and OX∗ to be the sheaf of invertible
(nowhere zero) holomorphic functions. This is a sheaf of abelian groups under
multiplication. We have maps of sheaves
exp
(2.4.9.1) 0 /Z ×2πi
/ OX / O∗ / 1.
X

(You can figure out what the sheaves 0 and 1 mean; they are isomorphic, and are
written in this way for reasons that may be clear.) We will soon interpret this as an
exact sequence of sheaves of abelian groups (the exponential exact sequence, see
Exercise 2.6.F), although we don’t yet have the language to do so.

2.4.O. E NLIGHTENING E XERCISE . Show that exp : OX / O ∗ describes O ∗ as

X X
a quotient sheaf of OX . Find an open set on which this map is not surjective.
This is a great example to get a sense of what “surjectivity” means for sheaves:
nowhere vanishing holomorphic functions (such as the function x away from the
origin) have logarithms locally, but they need not have logarithms globally.

2.5 Recovering sheaves from a “sheaf on a base”

Sheaves are natural things to want to think about, but hard to get our hands on.
We like the identity and gluability axioms, but they make proving things trickier
than for presheaves. We have discussed how we can understand sheaves using
stalks (using “compatible germs”). We now introduce a second way of getting a
hold of sheaves, by introducing the notion of a sheaf on a base. Warning: this way
of understanding an entire sheaf from limited information is confusing. It may
help to keep sight of the central insight that this partial information is enough
to understand germs, and the notion of when they are compatible (with nearby
germs).
First, we define the notion of a base of a topology. Suppose we have a topo-
logical space X, i.e., we know which subsets Ui of X are open. Then a base of a
topology is a subcollection of the open sets {Bj } ⊂ {Ui }, such that each Ui is a union
of the Bj . Here is one example that you have seen early in your mathematical life.
Suppose X = Rn . Then the way the classical topology is often first defined is by
defining open balls Br (x) = {y ∈ Rn : |y − x| < r}, and declaring that any union
of open balls is open. So the balls form a base of the classical topology — we say
they generate the classical topology. As an application of how we use them, to check
continuity of some map π : X → Rn , you need only think about the pullback of
balls on Rn — part of the traditional δ-ϵ definition of continuity.
 © 2024 Ravi Vakil. Published by Princeton University Press. 87

Now suppose we have a sheaf F on a topological space X, and a base {Bi } of

open sets on X. Then consider the information
({F (Bi )}, {resBi ,Bj : F (Bi ) → F (Bj )}),
which is a subset of the information contained in the sheaf — we are only paying
attention to the information involving elements of the base, not all open sets.
We can recover the entire sheaf from this information. This is because we can
determine the stalks from this information, and we can determine when germs are
compatible.

2.5.A. I MPORTANT E XERCISE . Make this precise. How can you recover a sheaf F
from this partial information?
This suggests a notion, called a sheaf on a base. A sheaf of sets (or abelian
groups, rings, . . . ) on a base {Bi } is the following. For each Bi in the base, we have a
set F(Bi ). If Bi ⊂ Bj , we have maps resBj ,Bi : F(Bj ) → F(Bi ), with resBi ,Bi = idF(Bi ) .
(Things called “B” are always assumed to be in the base.) If Bi ⊂ Bj ⊂ Bk , then
resBk ,Bi = resBj ,Bi ◦ resBk ,Bj . So far we have defined a presheaf on a base {Bi }.
We also require the base identity axiom: If B = ∪Bi , then if f, g ∈ F(B) are such
that resB,Bi f = resB,Bi g for all i, then f = g.
We require the base gluability axiom too: If B = ∪Bi , and we have fi ∈
F(Bi ) such that fi agrees with fj on any basic open set contained in Bi ∩ Bj (i.e.,
resBi ,Bk fi = resBj ,Bk fj for all Bk ⊂ Bi ∩ Bj ) then there exists f ∈ F(B) such that
resB,Bi f = fi for all i.

2.5.1. Theorem. — Suppose {Bi } is a base on X, and F is a sheaf of sets on this base. Then
∼
there is a sheaf F extending F (with isomorphisms F (Bi ) ←→ F(Bi ) agreeing with the
restriction maps). This sheaf F is unique up to unique isomorphism.

Proof. We will define F as the sheaf of (families of) compatible germs of F.

Define the stalk of a presheaf F on a base at p ∈ X by
Fp = colim F(Bi )
where the colimit is over all Bi (in the base) containing p.
We will say a family of germs in an open set U is compatible near p if there is a
section s of F over some Bi containing p such that the germs over Bi are precisely
the germs of s. More formally, define
F (U) := {(fp ∈ Fp )p∈U : for all p ∈ U, there exists B with p ∈ B ⊂ U, s ∈ F(B),
with sq = fq for all q ∈ B}
where each B is in our base.
This is a sheaf (for the same reasons that the sheaf of compatible germs was, cf.
Exercise 2.4.H).
I next claim that if B is in our base, the natural map F(B) → F (B) is an
isomorphism.

2.5.B. E XERCISE . Verify that F(B) → F (B) is an isomorphism, likely by showing

that it is injective and surjective (or else by describing the inverse map and verifying
that it is indeed inverse). Possible hint: elements of F (B) are determined by stalks,
as are elements of F(B).
88 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

It will be clear from your solution to Exercise 2.5.B that the restriction maps for
F are the same as the restriction maps of F (for elements of the base).
Finally, you should verify to your satisfaction that F is indeed unique up to
unique isomorphism. (First be sure that you understand what this means!) □
Theorem 2.5.1 shows that sheaves on X can be recovered from their “restriction
to a base”. It is clear from the argument (and in particular the solution to the
Exercise 2.5.B) that if F is a sheaf and F is the corresponding sheaf on the base B,
then for any p ∈ X, Fp is naturally isomorphic to Fp .
Theorem 2.5.1 is a statement about objects in a category, so we should hope for
a similar statement about morphisms.

2.5.C. I MPORTANT E XERCISE : M ORPHISMS OF SHEAVES CORRESPOND TO MOR -

PHISMS OF SHEAVES ON A BASE . Suppose {Bi } is a base for the topology of X. A
morphism F → G of sheaves on the base is a collection of maps F(Bk ) → G(Bk )
such that the diagram
F(Bi ) / G(Bi )
resBi ,Bj resBi ,Bj
 
F(Bj ) / G(Bj )

commutes for all Bj ,→ Bi .

(a) Verify that a morphism of sheaves is determined by the induced morphism of
sheaves on the base.
(b) Show that a morphism of sheaves on the base gives a morphism of the induced
sheaves. (Possible hint: compatible stalks.)

2.5.2. Remark. The above constructions and arguments describe an equivalence of

categories (§1.1.21) between sheaves on X and sheaves on a given base of X. There
is no new content to this statement, but you may wish to think through what it
means. What are the functors in each direction? Why aren’t their compositions the
identity?

2.5.3. Remark. It will be useful to extend these notions to OX -modules (see for
example Exercise 6.2.C). You will readily be able to verify that there is a correspon-
dence (really, equivalence of categories) between OX -modules and “OX -modules
on a base”. Rather than working out the details, you should just informally think
through the main points: what is an “OX -module on a base”? Given an OX -module
on a base, why is the corresponding sheaf naturally an OX -module? Later, if you
are forced at gunpoint to fill in details, you will be able to.

2.5.D. U NIMPORTANT EXERCISE . Suppose a morphism of sheaves of sets F → G on

a base {Bi } is surjective for all Bi (i.e., F(Bi ) → G(Bi ) is surjective for all i). Show
that the corresponding morphism of sheaves (not on the base) is surjective (or more
precisely: an epimorphism). The converse is not true, unlike the case for injectivity.
This gives a useful sufficient criterion for “surjectivity”: a morphism of sheaves
is an epimorphism (“surjective”) if it is surjective for sections on a base. You may
enjoy trying this out with Example 2.4.9 (dealing with holomorphic functions in
the classical topology on X = C), showing that the exponential map exp : OX → OX∗
is surjective, using the base of contractible open sets.
 © 2024 Ravi Vakil. Published by Princeton University Press. 89

2.5.4. Gluing sheaves.

We will repeatedly see the theme of constructing some object by gluing, in
many different contexts. Keep an eye out for it! In each case, we carefully consider
what information we need in order to glue.

2.5.E. I MPORTANT E XERCISE . Suppose X = ∪Ui is an open cover of X, and we

have sheaves Fi on Ui along with isomorphisms
∼
ϕij : Fi |Ui ∩Uj −→ Fj |Ui ∩Uj
(with ϕii the identity) that agree on triple overlaps, i.e.,
(2.5.4.1) ϕjk ◦ ϕij = ϕik
on Ui ∩ Uj ∩ Uk (this is called the cocycle condition, for reasons we ignore). Show
that these sheaves can be glued together into a sheaf F on X (unique up to unique
∼
isomorphism), with isomorphisms F |Ui −→ Fi , and the isomorphisms over
Ui ∩ Uj are the obvious ones. Warning: we are not assuming this is a finite cover, so
you cannot use induction. For this reason this exercise can be perplexing. (You can
use the ideas of this section to solve this problem, but you don’t necessarily need to.
Hint: As the base, take those open sets contained in some Ui .)
Thus we can “glue sheaves together”, using limited patching information.
Small observation: the hypothesis ϕii is the identity is extraneous, as it follows
from the cocycle condition.

2.5.5. Remark for experts. Exercise 2.5.E almost says that the “set” of sheaves forms
a sheaf itself, but not quite. Making this precise leads one to the notion of a stack.

2.6 Sheaves of abelian groups, and OX -modules, form abelian

categories
We are now ready to see that sheaves of abelian groups, and their cousins,
OX -modules, form abelian categories. In other words, we may treat them similarly
to vector spaces, and modules over a ring. In the process of doing this, we will see
that this is much stronger than an analogy; kernels, cokernels, exactness, and so
forth can be understood at the level of stalks (which are just abelian groups), and
the compatibility of the germs will come for free.
The category of sheaves of abelian groups on a topological space X is clearly an
additive category (Definition 1.5.1). In order to show that it is an abelian category,
we must begin by showing that any morphism ϕ : F → G has a kernel and a
cokernel. We have already seen that ϕ has a kernel (Exercise 2.3.I): the presheaf
kernel is a sheaf, and is a kernel in the category of sheaves.

2.6.A. E XERCISE . Show that the stalk of the kernel is the kernel of the stalks: for all
p ∈ X, there is a natural isomorphism
∼
(ker(F → G ))p o / ker(Fp → Gp ).

We next address the issue of the cokernel. Now ϕ : F → G has a cokernel in

the category of presheaves; call it Hpre (where the subscript is meant to remind
90 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

us that this is a presheaf). Let sh : Hpre → H be its sheafification. Recall that the
cokernel is defined using a universal property: it is the colimit of the diagram

(2.6.0.1) F
ϕ
/G


0
in the category of presheaves (cf. (1.5.4.1) and the comment thereafter).

2.6.1. Proposition. — The composition G → H is the cokernel of ϕ in the category of

sheaves.

Proof. We show that it satisfies the universal property. Given any sheaf E and a
commutative diagram

F
ϕ
/G

 
0 /E
We construct
F
ϕ
/G

 
0 / Hpre sh /H


*E

We show that there is a unique morphism H → E making the diagram commute.

As Hpre is the cokernel in the category of presheaves, there is a unique morphism
of presheaves Hpre → E making the diagram commute. But then by the universal
property of sheafification (Definition 2.4.5), there is a unique morphism of sheaves
H → E making the diagram commute. □

2.6.B. E XERCISE . Show that the stalk of the cokernel is naturally isomorphic to the
cokernel of the stalk.
We have now defined the notions of kernel and cokernel, and verified that they
may be checked at the level of stalks. We have also verified that the properties of
a morphism being a monomorphism or epimorphism are also determined at the
level of stalks (Exercises 2.4.M and 2.4.N). Hence we have proved the following:

2.6.2. Theorem. — Sheaves of abelian groups on a topological space X form an abelian

category.
That’s all there is to it — what needs to be proved has been shifted to the stalks,
where everything works because stalks are abelian groups!
And we see more: all structures coming from the abelian nature of this category
may be checked at the level of stalks. For example:
 © 2024 Ravi Vakil. Published by Princeton University Press. 91

2.6.C. E XERCISE . Suppose ϕ : F → G is a morphism of sheaves of abelian groups.

Show that the image sheaf im ϕ is the sheafification of the image presheaf. (You
must use the definition of image in an abelian category. In fact, this gives the
accepted definition of image sheaf for a morphism of sheaves of sets.) Show that
the stalk of the image is the image of the stalk.
As a consequence, exactness of a sequence of sheaves may be checked at the
level of stalks. If you are not sure about this, you should do the following exercise.

2.6.D. E XERCISE . Suppose α : F → G and β : G → H are two morphisms of

sheaves of abelian groups on X. Show that

F
α /G β
/H

is exact (at G ) if and only if for all p ∈ X,

αp βp
Fp / Gp / Hp

is exact.
In particular:

2.6.E. I MPORTANT E XERCISE ( CF. E XERCISE 2.3.A). Show that taking the stalk of
a sheaf of abelian groups is an exact functor. More precisely, if X is a topological
space and p ∈ X is a point, show that taking the stalk at p defines an exact functor
AbX → Ab.

2.6.F. E XERCISE . Check that the exponential exact sequence (2.4.9.1) is indeed an
exact sequence of sheaves of abelian groups.

2.6.G. E XERCISE : L EFT- EXACTNESS OF THE FUNCTOR OF “ SECTIONS OVER U”.

Suppose U ⊂ X is an open set, and 0 → F → G → H is an exact sequence of
sheaves of abelian groups. Show that

0 / F (U) / G (U) / H (U)

is exact. (You should do this “by hand”, even if you realize there is a very fast proof
using the left-exactness of the “forgetful” right adjoint to the sheafification functor.)
Show that the section functor need not be exact: show that if 0 → F → G → H →
0 is an exact sequence of sheaves of abelian groups, then

0 / F (U) / G (U) / H (U) /0

need not be exact. (Hint: the exponential exact sequence (2.4.9.1). But feel free to
make up a different example.)

2.6.H. E XERCISE : L EFT- EXACTNESS OF PUSHFORWARD . Suppose 0 → F →

G → H is an exact sequence of sheaves of abelian groups on X. If π : X → Y is a
continuous map, show that

0 / π∗ F / π∗ G / π∗ H

is exact. (The previous exercise, dealing with the left-exactness of the global section
functor can be interpreted as a special case of this, in the case where Y is a point.)
92 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

2.6.I. E XERCISE : L EFT- EXACTNESS OF Hom ( CF. E XERCISE 1.5.G( C ) AND ( D )). Sup-
pose F is a sheaf of abelian groups on a topological space X. Show that Hom (F , ·)
is a left-exact covariant functor AbX → AbX . Show that Hom (·, F ) is a left-exact
contravariant functor AbX → AbX .

2.6.3. OX -modules.

2.6.J. E XERCISE . Show that if (X, OX ) is a ringed space, then OX -modules form an
abelian category. (There is a fair bit to check, but there aren’t many new ideas.)

2.6.4. Many facts about sheaves of abelian groups carry over to OX -modules without
change, because a sequence of OX -modules is exact if and only if the underlying
sequence of sheaves of abelian groups is exact. You should be able to easily check
that all of the statements of the earlier exercises in §2.6 also hold for OX -modules,
when modified appropriately. For example (Exercise 2.6.I), Hom OX (·, ·) is a left-exact
contravariant functor in its first argument and a left-exact covariant functor in its
second argument.
We end with a useful construction using some of the ideas in this section.

2.6.K. I MPORTANT E XERCISE : T ENSOR PRODUCTS OF OX - MODULES .

(a) Suppose OX is a sheaf of rings on X. Define (categorically) what we should
mean by tensor product of two OX -modules. Give an explicit construction, and
show that it satisfies your categorical definition. Hint: take the “presheaf tensor
product” — which needs to be defined — and sheafify. Note: ⊗OX is often written
⊗ when the subscript is clear from the context. (An example showing sheafification
is necessary will arise in Example 15.1.1.)
(b) Show that the tensor product of stalks is the stalk of the tensor product. (If you
can show this, you may be able to make sense of the phrase “colimits commute
with tensor products”.)

2.6.5. Conclusion. Just as presheaves of abelian groups on a topological space

form an abelian category because all abelian-categorical notions make sense open
set by open set, sheaves of abelian groups on a topological space form an abelian
category because all abelian-categorical notions make sense stalk by stalk.

2.7 The inverse image sheaf

We next describe a notion that is fundamental, but rather intricate. We will

not need it for some time, so this may be best left for a second reading. Suppose
we have a continuous map π : X → Y. If F is a sheaf on X, we have defined the
pushforward or direct image sheaf π∗ F , which is a sheaf on Y. There is also a
notion of inverse image sheaf. (We will not call it the pullback sheaf, reserving that
name for a later construction for quasicoherent sheaves, §14.5.) This is a covariant
functor π−1 from sheaves on Y to sheaves on X. If the sheaves on Y have some
additional structure (e.g., group or ring), then this structure is respected by π−1 .

2.7.1. Definition by adjoint: elegant but abstract. We define the inverse image π−1 as
the left adjoint to π∗ .
 © 2024 Ravi Vakil. Published by Princeton University Press. 93

This isn’t really a definition; we need a construction to show that the adjoint
exists. Note that we then get canonical maps π−1 π∗ F → F (associated to the
identity in MorY (π∗ F , π∗ F )) and G → π∗ π−1 G (associated to the identity in
MorX (π−1 G , π−1 G )).
π−1 G /F

X G / π∗ F

π

Y

2.7.2. Construction: concrete but ugly. Define the temporary notation

pre G (U) = colimV⊃π(U) G (V).

π−1

(Recall the explicit description of colimit: sections of π−1

pre over U are sections on
open sets containing π(U), with an equivalence relation. Note that π(U) won’t be
an open set in general.)

2.7.A. E XERCISE . Show that this defines a presheaf on X. Show that it needn’t form
a sheaf. (Hint: map 2 points to 1 point.)
Now define the inverse image of G by π−1 G := (π−1 pre G ) . Note that π
sh −1
is a
functor from sheaves on Y to sheaves on X. The next exercise shows that π−1 is
indeed left-adjoint to π∗ . But you may wish to try the later exercises first, and come
back to Exercise 2.7.B at another time. (For the later exercises, try to give two proofs,
one using the universal property, and the other using the explicit description.)

2.7.B. I MPORTANT T RICKY E XERCISE ((π−1 , π∗ ) ARE ADJOINT ). If π : X → Y is a

continuous map, and F is a sheaf on X and G is a sheaf on Y, describe a bijection
∼
MorX (π−1 G , F ) o / MorY (G , π∗ F ).

Observe that your bijection is “natural” in the sense of the definition of adjoints
(i.e., functorial in both F and G ). Thus Construction 2.7.2 satisfies the universal
property of Definition 2.7.1. Possible hint: Show that both sides agree with the
following third construction, which we denote MorYX (G , F ). A collection of maps
ϕVU : G (V) → F (U) (as U runs through all open sets of X, and V runs through all
open sets of Y containing π(U)) is said to be compatible if for all open U ′ ⊂ U ⊂ X
and all open V ′ ⊂ V ⊂ Y with π(U) ⊂ V, π(U ′ ) ⊂ V ′ , the diagram

(2.7.2.1) G (V)
ϕVU
/ F (U)

resV,V ′ resU,U ′
 ϕV ′ U ′ 
G (V ′ ) / F (U ′ )

commutes. Define MorYX (G , F ) to be the set of all compatible collections ϕ =

{ϕVU }.
94 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

2.7.3. Remark (“stalk and skyscraper are an adjoint pair”). As a special case, if X is a
point p ∈ Y, we see that π−1 G is the stalk Gp of G , and maps from the stalk Gp to a
set S are the same as maps of sheaves on Y from G to the skyscraper sheaf with set
S supported at p. You may prefer to prove this special case by hand directly before
solving Exercise 2.7.B, as it is enlightening. (It can also be useful — can you use it
to solve Exercises 2.4.L and 2.4.N?)

2.7.C. E XERCISE . Show that the stalks of π−1 G are the same as the stalks of G . More
∼
precisely, if π(p) = q, describe a natural isomorphism Gq −→ (π−1 G )p . (Possible
hint: use the concrete description of the stalk, as a colimit. Recall that stalks are
preserved by sheafification, Exercise 2.4.L. Alternatively, use adjointness.)
Exercise 2.7.C, along with the notion of compatible germs, may give you a
simple way of thinking about (and perhaps visualizing) inverse image sheaves.
Closely related: you can think of sections of the inverse image sheaf as, locally,
inverse images of sections on the target. (Those preferring the “espace étalé” or
“space of sections” perspective, §2.2.11, can check that the “inverse image of the
space of sections” is the “space of sections” of the inverse image.)

2.7.D. E XERCISE ( EASY BUT USEFUL ). If U is an open subset of Y, i : U → Y is the

inclusion, and G is a sheaf on Y, show that i−1 G is naturally isomorphic to G |U (the
restriction of G to U, §2.2.8).

2.7.E. E XERCISE . Show that π−1 is an exact functor from sheaves of abelian groups
on Y to sheaves of abelian groups on X (cf. Exercise 2.6.E). (Hint: exactness can be
checked on stalks, and by Exercise 2.7.C, the stalks are the same.) Essentially the
same argument will show that π−1 is an exact functor from OY -modules (on Y) to
(π−1 OY )-modules (on X), but don’t bother writing that down. (Remark for experts:
π−1 is a left adjoint, hence right-exact by abstract nonsense, as discussed in §1.5.14.
Left-exactness holds because colimits of abelian groups over filtered index sets are
exact, Exercise 1.5.L.)

2.7.4. Definition: The push-pull map.

Suppose

β′
(2.7.4.1) W /X

α′ α
 
Y
β
/Z

is a commutative (not necessarily Cartesian!) diagram, and F is a sheaf on X.

Define the push-pull map

(2.7.4.2) β−1 α∗ F / α∗′ (β ′ )−1 F

∼
of sheaves on Y as follows. Start with the identity (β ′ )−1 F −→ (β ′ )−1 F on
W. By adjointness of ((β ′ )−1 , β∗′ ), this is the same as the data of a morphism
F → (β∗′ )(β ′ )−1 F on X. Apply α∗ to get a map α∗ F → α∗ (β∗′ )(β ′ )−1 F on Z. By
the commutativity of (2.7.4.1), this is the map α∗ F → β∗ (α∗′ )(β ′ )−1 F on Z. By
adjointness of (β−1 , β∗ ), this yields a map (2.7.4.2).
 © 2024 Ravi Vakil. Published by Princeton University Press. 95

We observe that this entire construction is functorial in F (i.e., given a map

F → G of sheaves on X, we get a certain commutative diagram of sheaves on Y —
what is it?). (We will later extend this to O-modules, quasicoherent sheaves, and
cohomology, see Exercises 7.2.D(f), 14.5.K, and 18.7.B.)

2.7.F. ⋆⋆ S URPRISINGLY H ARD E XERCISE . We could have defined the push-pull

map in a “dual way” starting with the identity α∗ F → α∗ F on Z, then using
adjointness of (α−1 , α∗ ), and continuing from there. Why does this give the same
definition of the push-pull map?

2.7.5. The support of a sheaf, and the support of a section of a sheaf.

Exercise 2.7.H below gives us an excuse to introduce the notion of support,
which we use repeatedly later.

2.7.6. Definition. Suppose F is a sheaf (or indeed separated presheaf) of abelian

groups on X, and s is a global section of F . Define the support of the section s,
denoted Supp s, to be the set of points p of X where s has a nonzero germ:
Supp s := {p ∈ X : sp ̸= 0 in Fp }.
We think of this as the subset of X where “the section s lives” — the complement
is the locus where s is the 0-section. (Unimportant: We Q could define this even if
F is a presheaf, but without the inclusion F (U) ,→ p∈U Fp of Exercise 2.4.A,
we could have the strange situation where we have a nonzero section that “lives
nowhere”, because it is 0 “near every point”, i.e., is 0 in every stalk.)

2.7.G. E XERCISE ( THE SUPPORT OF A SECTION IS CLOSED ). Show that Supp s is a

closed subset of X.

2.7.7. Caution: the locus where a continuous function is nonzero is open; the locus where
the germ of a function is nonzero is closed. Basically by the definition of continuity, the
locus where the value of a continuous function is nonzero is open. (More generally,
the locus where the value of a function on a locally ringed space is nonzero is open,
see Exercise 4.3.F(a).) In contrast, Exercise 2.7.G shows that the locus where the
germ of a function is nonzero is closed. We will try to avoid misunderstanding by
using phrases like “f is 0 at p” (the value of f is zero, i.e., f(p) = 0) and “f is 0
near p” (the germ of f is zero, i.e., f = 0 in OX,p , or equivalently, f is zero in some
neighborhood of p).

2.7.8. Definition. Define the support of a sheaf of groups G , denoted Supp G , as

the locus where the stalks are nontrivial:
Supp G := {p ∈ X : |Gp | ̸= 1}.
Equivalently, Supp G is the union of supports of sections over all open sets. Clearly
support is a “stalk-local notion”, and hence “commutes” with restriction to open
sets. (Irrelevant for us: more generally, if the sheaf has values in some category,
such as the category of sets, the support can be defined as the points where the
stalk is not the terminal object.)

2.7.H. E XERCISE .
(a) Suppose Z ⊂ Y is a closed subset, and i : Z ,→ Y is the inclusion. If F is a sheaf
of groups on Z, then show that the stalk (i∗ F )q is the one-element group if q ∈ / Z,
96 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

and Fq if q ∈ Z.
(b) Suppose Supp G ⊂ Z where Z is closed. Show that the natural map G → i∗ i−1 G
is an isomorphism. Thus a sheaf supported on a closed subset can be considered a
sheaf on that closed subset.

2.7.9. Extension by zero, an occasional left adjoint to the inverse image functor.
In addition to always being a left adjoint, π−1 can sometimes be a right adjoint,
when π is an inclusion of an open subset. We discuss this when we need it, in
§23.4.7.
 Part II

Schemes
 L’idée même de schéma est d’une simplicité enfantine — si simple, si humble, que
personne avant moi n’avait songé à se pencher si bas. Si “bébête” même, pour tout dire,
que pendant des années encore et en dépit de l’évidence, pour beaucoup de mes savants
collègues, ça faisait vraiment “pas sérieux”!
The very idea of scheme is of infantile simplicity — so simple, so humble, that no one
before me thought of stooping so low. So childish, in short, that for years, despite all the
evidence, for many of my erudite colleagues, it was really “not serious”!
— A. Grothendieck [Gr5, p. P32], translated by C. McLarty [Mc, p. 313]
 CHAPTER 3

Toward affine schemes: the underlying set, and

topological space

There is no serious historical question of how Grothendieck found his definition of

schemes. It was in the air. Serre has well said that no one invented schemes... The question
is, what made Grothendieck believe he should use this definition to simplify an 80 page
paper by Serre into some 1000 pages of Élements de Géométrie Algébrique?
— C. McLarty [Mc, p. 313]
We are now ready to consider the notion of a scheme, which is the type of
geometric space central to algebraic geometry. We should first think through what
we mean by “geometric space”. You have likely seen the notion of a manifold,
and we wish to abstract this notion so that it can be generalized to other settings,
notably so that we can deal with nonsmooth and arithmetic objects.

3.1 Toward schemes

The key insight behind this generalization from the notion of something like
a manifold to a more versatile notion of a “geometric space” is the following: we
can understand a geometric space (such as a manifold) well by understanding
the functions on this space. More precisely, we will understand it through the
sheaf of functions on the space. If we are interested in differentiable manifolds, we
will consider smooth functions; if we are interested in analytic manifolds, we will
consider real analytic functions; and so on.
Thus we will define a scheme to be the following data
• The set: the points of the scheme
• The topology: the open sets of the scheme
• The structure sheaf: the sheaf of “algebraic functions” (a sheaf of rings) on
the scheme.
Recall that a topological space with a sheaf of rings is called a ringed space (§2.2.13).
We will try to draw pictures throughout. Pictures can help develop geometric
intuition, which can guide the algebraic development (and, eventually, vice versa).
Some people find pictures very helpful, while others are repulsed or confused.
We will try to make all three notions as intuitive as possible. For the set, in
the key example of complex (affine) varieties (roughly, things cut out in Cn by
polynomials), we will see that the points are the “traditional points” (n-tuples of

This material is published by Princeton University Press as The Rising Sea: Foundations of Algebraic
Geometry by Ravi Vakil. This version is free to view and download for personal use only. Not for
redistribution, resale, or use in derivative works. © 2024 Ravi Vakil.

99
100 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

complex numbers), plus some extra points that will be handy to have around. For
the topology, we will require that “the subset where an algebraic function vanishes
must be closed”, and require nothing else. For the sheaf of algebraic functions (the
structure sheaf), we will expect that in the complex plane C2 , (3x2 + y2 )/(2x +
4xy + 1) should be an algebraic function on the open set consisting of points where
the denominator doesn’t vanish, and this will largely motivate our definition.

3.1.1. Example: Differentiable manifolds. As motivation, we return to our

example of differentiable manifolds, reinterpreting them in this light. We will be
quite informal in this discussion. Suppose X is a differentiable manifold. It is a
topological space, and has a sheaf of smooth (C∞ ) functions OX (see §2.1). This gives
X the structure of a ringed space. We have observed that evaluation at a point p ∈ X
gives a surjective map from the stalk to R

OX,p / / R,

so the kernel, the (germs of) functions vanishing at p, is a maximal ideal mX,p (see
§2.1.1).
We could define a differentiable real manifold as a topological space X with
a sheaf of rings (see Definition 4.3.9). We would require that there is a cover of
X by open sets such that on each open set the ringed space is isomorphic to a
ball around the origin in Rn (with the sheaf of smooth functions on that ball).
With this definition, the ball is the basic patch, and a general manifold is obtained
by gluing these patches together. (Admittedly, a great deal of geometry comes
from how one chooses to patch the balls together!) In the algebraic setting, the
basic patch is the notion of an affine scheme, which we will discuss soon. (In the
definition of manifold, there is an additional requirement that the topological
space be Hausdorff and second-countable, to avoid pathologies. Schemes are often
required to be “separated” to avoid essentially the same pathologies. Separatedness
will be discussed in Chapter 11.)
Functions are determined by their values at points. This is an obvious statement, but
won’t be true for schemes in general. We will see an example in Exercise 3.2.A(a),
and discuss this behavior further in §3.2.12.
Morphisms of manifolds. How can we describe maps of differentiable manifolds
π : X → Y? They are certainly continuous maps — but which ones? We can pull
back functions along continuous maps. Smooth functions pull back to smooth
functions. More formally, we have a map π−1 OY → OX . (The inverse image sheaf
π−1 was defined in §2.7.) Inverse image is left-adjoint to pushforward, so we also
get a map π♯ : OY → π∗ OX .
Certainly given a map of differentiable manifolds, smooth functions pull back
to smooth functions. It is less obvious that this is a sufficient condition for a continuous
map to be smooth.

3.1.A. I MPORTANT E XERCISE FOR THOSE WITH A LITTLE EXPERIENCE WITH MANI -
FOLDS . Suppose that π : X → Y is a continuous map of differentiable manifolds (as
topological spaces — not a priori smooth). Show that π is smooth if smooth func-
tions pull back to smooth functions, i.e., if pullback by π gives a map OY → π∗ OX .
(Hint: check this on small patches. Once you figure out what you are trying to
show, you will realize that the result is immediate.)
 © 2024 Ravi Vakil. Published by Princeton University Press. 101

3.1.B. E XERCISE . Show that a morphism of differentiable manifolds π : X → Y with

π(p) = q induces a morphism of stalks π♯ : OY,q → OX,p . Show that π♯ (mY,q ) ⊂
mX,p . In other words, if you pull back a function that vanishes at q, you get a
function that vanishes at p — not a huge surprise. (In §7.3, we formalize this by
saying that maps of manifolds are maps of locally ringed spaces.)

3.1.2. Aside. Here is a little more for experts: Notice that π induces a map on
tangent spaces (see Aside 2.1.2)

(mX,p /m2X,p )∨ / (mY,q /m2 )∨ .

Y,q

This is the tangent map you would geometrically expect. Again, it is interesting
that the cotangent map mY,q /m2Y,q → mX,p /m2X,p is algebraically more natural than
the tangent map (there are no “duals”).
Experts are now free to try to interpret other differential-geometric information
using only the map of topological spaces and map of sheaves. For example: how
can one check if π is a submersion of manifolds? How can one check if f is an im-
mersion? (We will see that the algebro-geometric version of these notions are smooth
morphism and unramified morphism; see Definition 13.6.2 and §21.7, respectively.)

3.1.3. Side Remark. Manifolds are covered by disks that are all isomorphic. This
isn’t true for schemes (even for “smooth complex varieties”). There are examples
of two “smooth complex curves” (the algebraic version of Riemann surfaces) X and
Y so that no nonempty open subset of X is isomorphic to a nonempty open subset
of Y (see Exercise 7.5.L). And there is a Riemann surface X such that no two open
subsets of X are isomorphic (see Exercise 19.7.E). Informally, this is because in the
Zariski topology on schemes, all nonempty open sets are “huge” and have more
“structure”.

3.1.4. Other examples. If you are interested in differential geometry, you might
be interested in differentiable manifolds, on which the functions under considera-
tion are smooth functions. Similarly, if you are interested in topology, you will be
interested in topological spaces, on which you will consider the continuous func-
tions. If you are interested in complex geometry, you will be interested in complex
manifolds (or possibly “complex analytic varieties”), on which the functions are
holomorphic functions. In each of these cases of interesting “geometric spaces”, the
topological space and sheaf of functions is clear. The notion of scheme fits naturally
into this family.

3.2 The underlying set of an affine scheme

For any ring A, we are going to define something called Spec A, the spectrum
of A. In this section, we will define it as a set, but we will soon endow it with a
topology, and later we will define a sheaf of rings on it (the structure sheaf). Such
an object is called an affine scheme. Later Spec A will denote the set along with
the topology, and a sheaf of functions. But for now, as there is no possibility of
confusion, Spec A will just be the set.
102 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

3.2.1. The set Spec A is the set of prime ideals of A. The prime ideal p of A when
considered as an element of Spec A will be denoted [p], to avoid confusion. Elements
a ∈ A will be called functions on Spec A, and their value at the point [p] will be a
(mod p). This is weird: a function can take values in different rings at different points —
the function 5 on Spec Z takes the value 1 (mod 2) at [(2)] and 2 (mod 3) at [(3)].
“An element a of the ring lying in a prime ideal p” translates to “a function a that is
0 at the point [p]” or “a function a vanishing at the point [p]”, and we will use these
phrases interchangeably. Notice that if you add or multiply two functions, you add
or multiply their values at all points; this is a translation of the fact that A → A/p
is a ring morphism. These translations are important — make sure you are very
comfortable with them! They should become second nature.
If A is generated over a base field (or base ring) by elements x1 , . . . , xr , the
elements x1 , . . . , xr are often called coordinates, because we will later be able to
reinterpret them as restrictions of “coordinates on r-space”, via the idea of §3.2.10,
made precise in Exercise 7.2.F.

3.2.2. Beginning of a grand dictionary between algebra and geometry. We are building
up the beginning of a grand dictionary; see the first part of §3.7.2. At some point
you will get the sense that we are slowly decoding some timeless Rosetta Stone
whose etchings we struggle to make out.

3.2.3. Glimpses of the future. In §4.1.3: we will interpret functions on Spec A as

global sections of the “structure sheaf”, i.e., as a function on a ringed space, in the
sense of §2.2.13. We repeat a caution from §2.2.13: what we will call “functions”,
others may call “regular functions”. And we will later define “rational functions”
(§6.6.36), which are not precisely functions in this sense; they are a particular type
of “partially-defined function”.
The notion of “value of a function” will be later interpreted as a value of a
function on a particular locally ringed space, see Definition 4.3.7.

3.2.4. We now give some examples.

Example 1 (the complex affine line): A1C := Spec C[x]. Let’s find the prime
ideals of C[x]. As C[x] is an integral domain, 0 is prime. Also, (x − a) is prime, for
any a ∈ C: it is even a maximal ideal, as the quotient by this ideal is a field:

f7→f(a)
0 / (x − a) / C[x] /C /0

(This exact sequence may remind you of (2.1.1.1) in our motivating example of
manifolds.)
We now show that there are no other prime ideals. We use the fact that C[x] has
a division algorithm, and is a unique factorization domain. Suppose p is a prime
ideal. If p ̸= (0), then suppose f(x) ∈ p is a nonzero element of smallest degree.
It is not constant, as prime ideals can’t contain 1. If f(x) is not linear, then factor
f(x) = g(x)h(x), where g(x) and h(x) have positive degree. (Here we use that C is
algebraically closed.) Then g(x) ∈ p or h(x) ∈ p, contradicting the minimality of the
degree of f. Hence there is a linear element x − a of p. Then I claim that p = (x − a).
Suppose f(x) ∈ p. Then the division algorithm would give f(x) = g(x)(x − a) + m
where m ∈ C. Then m = f(x) − g(x)(x − a) ∈ p. If m ̸= 0, then 1 ∈ p, giving a
contradiction.
 © 2024 Ravi Vakil. Published by Princeton University Press. 103

Thus we can and should (and must!) make a picture of A1C = Spec C[x] (see
Figure 3.1). This is just the first illustration of a point of view of Sophie Germain
[Ge]: “L’algèbre n’est qu’une géométrie écrite; la géométrie n’est qu’une algèbre
figurée.” (Algebra is but written geometry; geometry is but drawn algebra.)
There is one “traditional” point for each complex number, plus one extra
(“bonus”) point [(0)]. We can mostly picture A1C as C: the point [(x − a)] we will
reasonably associate to a ∈ C. Where should we picture the point [(0)]? The best
way of thinking about it is somewhat zen. It is somewhere on the complex line,
but nowhere in particular. Because (0) is contained in all of these prime ideals,
we will somehow associate it with this line passing through all the other points.
This new point [(0)] is called the “generic point” of the line. (We will formally
define “generic point” in §3.6.) It is “generically on the line” but you can’t pin it
down any further than that. It is not at any particular place on the line. (This is
misleading too — we will see in Easy Exercise 3.6.N that it is “near” every point.
So it is near everything, but located nowhere precisely.) We will place it far to the
right for lack of anywhere better to put it. You will notice that we sketch A1C as
one-(real-)dimensional (even though we picture it as an enhanced version of C);
this is to later remind ourselves that this will be a one-dimensional space, where
dimensions are defined in an algebraic (or complex-geometric) sense. (Dimension
will be defined in Chapter 12.)

F IGURE 3.1. A picture of A1C = Spec C[x]

To give you some feeling for this space, we make some statements that are
currently undefined, but suggestive. The functions on A1C are the polynomials. So
f(x) = x2 − 3x + 1 is a function. What is its value at [(x − 1)], which we think of as
the point 1 ∈ C? Answer: f(1)! Or equivalently, we can evaluate f(x) modulo x − 1
— this is the same thing by the division algorithm. (What is its value at [(0)]? It is
f(x) (mod 0), which is just f(x).)
104 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Here is a more complicated example: g(x) = (x − 3)3 /(x − 2) is a “rational

function”. It is defined everywhere but x = 2. (When we know what the structure
sheaf is, we will be able to say that it is an element of the structure sheaf on the
open set A1C \ {2}.) We want to say that g(x) has a triple zero at 3, and a single pole
at 2, and we will be able to after §13.5.
Example 2 (the affine line over k = k): A1k := Spec k[x] where k is an alge-
braically closed field. This is called the affine line over k. All of our discussion in
the previous example carries over without change. We will use the same picture,
which is after all intended to just be a metaphor.
Example 3: Spec Z. An amazing fact is that from our perspective, this will
look a lot like the affine line A1k . The integers, like k[x], form a unique factorization
domain, with a division algorithm. The prime ideals are: (0), and (p) where p is
prime. Thus everything from Example 1 carries over without change, even the
picture. Our picture of Spec Z is shown in Figure 3.2.

F IGURE 3.2. A “picture” of Spec Z, which looks suspiciously like Figure 3.1

Let’s blithely carry over our discussion of functions to this space. 100 is a
function on Spec Z. Its value at (3) is “1 (mod 3)”. Its value at (2) is “0 (mod 2)”,
and in fact it has a double zero. 27/4 is a “rational function” on Spec Z, defined
away from (2). We want to say that it has a double pole at (2), and a triple zero at
(3). Its value at (5) is

27 × 4−1 ≡ 2 × (−1) ≡ 3 (mod 5).

(We will gradually make this discussion precise over time.)
Example 4: silly but important examples, and the German word for bacon.
The set Spec k where k is any field is boring: one point. Spec 0, where 0 is the
zero-ring, is the empty set, as 0 has no prime ideals.

3.2.A. A SMALL EXERCISE ABOUT SMALL SCHEMES .

(a) Describe the set Spec k[ϵ]/(ϵ2 ). The ring k[ϵ]/(ϵ2 ) is called the ring of dual
numbers, and will turn out to be quite useful. You should think of ϵ as a very
small number, so small that its square is 0 (although it itself is not 0). It is a nonzero
function whose value at all points is zero, thus giving our first example of functions
not being determined by their values at points. We will discuss this phenomenon
further in §3.2.12.
(b) Describe the set Spec k[x](x) (see §1.2.3 for a discussion of localization). We will
see this scheme again repeatedly, starting with §3.2.9 and Exercise 3.4.K. You might
later think of it as a shred of a particularly nice “smooth curve”.
In Example 2, we restricted to the case of algebraically closed fields for a reason:
things are more subtle if the field is not algebraically closed.
 © 2024 Ravi Vakil. Published by Princeton University Press. 105

Example 5 (the affine line over R): A1R = Spec R[x]. Using the fact that R[x]
is a Euclidean domain, similar arguments to those of Examples 1–3 show that the
prime ideals are (0), (x − a) where a ∈ R, and (x2 + ax + b) where x2 + ax + b is
an irreducible quadratic. The latter two are maximal ideals, i.e., their quotients are
fields. For example: R[x]/(x − 3) =∼ R, R[x]/(x2 + 1) =
∼ C.

3.2.B. U NIMPORTANT E XERCISE . Show that for the last type of prime, of the form
(x2 + ax + b), the quotient is always isomorphic to C.
So we have the points that we would normally expect to see on the real line,
corresponding to real numbers; the generic point 0; and new points which we may
interpret as conjugate pairs of complex numbers (the roots of the quadratic). This last
type of point should be seen as more akin to the real numbers than to the generic
point. You can picture A1R as the complex plane, folded along the real axis. But the
key point is that Galois-conjugate points (such as i and −i) are considered glued.
Let’s explore functions on this space. Consider the function f(x) = x3 − 1. Its
value at the point [(x − 2)] is 7, or perhaps better, “7 (mod x − 2)”. How about at
(x2 + 1)? We get
x3 − 1 ≡ −x − 1 (mod x2 + 1),
which may be profitably interpreted as −i − 1.
One moral of this example is that we can work over a non-algebraically closed
field if we wish. It is more complicated, but we can recover much of the information
we care about.

3.2.C. I MPORTANT E XERCISE . Describe the set A1Q . (This is harder to picture in a
way analogous to A1R . But the rough cartoon of points on a line, as in Figure 3.1,
remains a reasonable sketch.)
Example 6 (the affine line over Fp ): A1Fp = Spec Fp [x]. As in the previous
examples, Fp [x] is a Euclidean domain, so the prime ideals are of the form (0) or
(f(x)) where f(x) ∈ Fp [x] is an irreducible polynomial, which can be of any degree.
Irreducible polynomials correspond to sets of Galois conjugates in Fp .
Note that Spec Fp [x] has p points corresponding to the elements of Fp , but also
many more (infinitely more, see Exercise 3.2.D). This makes this space much richer
than simply p points. For example, a polynomial f(x) is not determined by its
values at the p elements of Fp , but it is determined by its values at the points of
Spec Fp [x]. (As we have mentioned before, this is not true for all schemes.)
You should think about this, even if you are a geometric person — this intuition
will later turn up in geometric situations. Even if you think you are interested
only in working over an algebraically closed field (such as C), you will have non-
algebraically closed fields (such as C(x)) forced upon you.

3.2.D. E XERCISE . If k is a field, show that Spec k[x] has infinitely many points.
(Hint: Euclid’s proof of the infinitude of primes of Z.)
Example 7 (the complex affine plane): A2C = Spec C[x, y]. (As with Examples 1
and 2, our discussion will apply with C replaced by any algebraically closed field.)
Sadly, C[x, y] is not a principal ideal domain: (x, y) is not a principal ideal. We
can quickly name some prime ideals. One is (0), which has the same flavor as the
(0) ideals in the previous examples. (x − 2, y − 3) is prime, and indeed maximal,
106 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

∼ C, where this isomorphism is via f(x, y) 7→ f(2, 3).

because C[x, y]/(x − 2, y − 3) =
More generally, (x − a, y − b) is prime for any (a, b) ∈ C2 . Also, if f(x, y) is an
irreducible polynomial (e.g., y − x2 or y2 − x3 ) then (f(x, y)) is prime.

3.2.E. E XERCISE . Show that we have identified all the prime ideals of C[x, y]. Hint:
Suppose p is a prime ideal that is not principal. Show you can find f(x, y), g(x, y) ∈
p with no common factor. By considering the Euclidean algorithm in the Euclidean
domain C(x)[y], show that you can find a nonzero h(x) ∈ (f(x, y), g(x, y)) ⊂ p.
Using primality, show that one of the linear factors of h(x), say (x − a), is in p.
Similarly show there is some (y − b) ∈ p.

F IGURE 3.3. Picturing A2C = Spec C[x, y]

We now attempt to draw a picture of A2C (see Figure 3.3). The maximal prime
ideals of C[x, y] correspond to the traditional points in C2 : [(x − a, y − b)] corre-
sponds to (a, b) ∈ C2 . We now have to visualize the “bonus points”. [(0)] somehow
lives behind all of the traditional points; it is somewhere on the affine plane, but
nowhere in particular. So for example, it does not lie on the parabola y = x2 . The
point [(y − x2 )] lies on the parabola y = x2 , but nowhere in particular on it. (Fig-
ure 3.3 is a bit misleading. For example, the point [(0)] isn’t in the fourth quadrant;
it is somehow near every other point, which is why it is depicted as a somewhat
diffuse large dot.) You can see from this picture that we already are implicitly think-
ing about “dimension”. The prime ideals (x − a, y − b) are somehow of dimension
0, the prime ideals (f(x, y)) are of dimension 1, and (0) is of dimension 2. (All of
our dimensions here are complex or algebraic dimensions. The complex plane C2 has
real dimension 4, but complex dimension 2. Complex dimensions are in general
half of real dimensions.) We won’t define dimension precisely until Chapter 12, but
you should feel free to keep it in mind before then.
Note too that maximal ideals correspond to the “smallest” points. Smaller
ideals correspond to “bigger” points. “One prime ideal contains another” means
that the points “have the opposite containment.” All of this will be made precise
once we have a topology. This order-reversal is a little confusing, and will remain
so even once we have made the notions precise.
We now come to the obvious generalization of Example 7.
Example 8 (complex affine n-space — important!): Let An C := Spec C[x1 , . . . , xn ].
(Important definition: More generally, An
A is defined to be Spec A[x1 , . . . , xn ], where
 © 2024 Ravi Vakil. Published by Princeton University Press. 107

A is an arbitrary ring. When the base ring is clear from context, the subscript A is
often omitted. For pedants: the notation An A implicitly includes the data of the n
coordinate functions x1 , . . . , xn .) For concreteness, let’s consider n = 3. We now
have an interesting question in what at first appears to be pure algebra: What are
the prime ideals of C[x, y, z]?
Analogously to before, (x − a, y − b, z − c) is a prime ideal. This is a maximal
ideal, because its residue ring is a field (C); we think of these as “0-dimensional
points”. We will often write (a, b, c) for [(x−a, y−b, z−c)] because of our geometric
interpretation of these ideals. There are no more maximal ideals, by Hilbert’s Weak
Nullstellensatz.

3.2.5. Hilbert’s Weak Nullstellensatz. — If k is an algebraically closed field, then the

maximal ideals of k[x1 , . . . , xn ] are precisely those ideals of the form (x1 −a1 , . . . , xn −an ),
where ai ∈ k.
We may as well state a slightly stronger version now.

3.2.6. Hilbert’s Nullstellensatz. — If k is any field, every maximal ideal of k[x1 , . . . , xn ]

has residue field a finite extension of k. Translation: any field extension of k that is finitely
generated as a k-algebra is necessarily also finitely generated as a k-vector space (i.e., is a
finite extension of fields).
This statement is also often called Zariski’s Lemma.

3.2.F. E XERCISE . Show that the Nullstellensatz 3.2.6 implies the Weak Nullstellen-
satz 3.2.5.
We will prove the Nullstellensatz in §8.4.3, and again in Exercise 12.2.B.
The following fact is a useful accompaniment to the Nullstellensatz.

3.2.G. E XERCISE ( NOT REQUIRING THE N ULLSTELLENSATZ ). Any integral domain

A which is a finite k-algebra (i.e., a k-algebra that is a finite-dimensional vector
space over k) must be a field. Hint: for any nonzero x ∈ A, show ×x : A → A is an
isomorphism. (Thus, in combination the Nullstellensatz 3.2.6, we see that prime
ideals of k[x1 , . . . , xn ] with finite-dimensional residue ring are the same as maximal ideals
of k[x1 , . . . , xn ]. This is worth remembering.)
There are other prime ideals of C[x, y, z] too. We have (0), which corresponds
to a “3-dimensional point”. We have (f(x, y, z)), where f is irreducible. To this we
associate the “hypersurface” f = 0, so this is “2-dimensional” in nature. But we
have not found them all! One clue: we have prime ideals of “dimension” 0, 2, and
3 — we are missing “dimension 1”. Here is one such prime ideal: (x, y). We picture
this as the locus where x = y = 0, which is the z-axis. This is a prime ideal, as the
corresponding quotient C[x, y, z]/(x, y) = ∼ C[z] is an integral domain (and should
be interpreted as the functions on the z-axis). There are lots of “1-dimensional
prime ideals”, and it is not possible to classify them in a reasonable way. It will
turn out that they correspond to things that we think of as irreducible curves. Thus
remarkably the answer to the purely algebraic question (“what are the prime ideals
of C[x, y, z]”) is fundamentally geometric!
The fact that the points of A1Q corresponding to maximal ideals of the ring Q[x]
(what we will soon call “closed points”, see Definition 3.6.8) can be interpreted as
108 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

points of Q where Galois-conjugates

√ √ are glued together
√ √(Exercise 3.2.C) √extends
√
to An
Q . For example, in A 2
Q , ( 2, 2) is glued to (− 2, − 2) but not to ( 2, − 2).
The following exercise will give you some idea of how this works.
√ √
3.2.H. E√XERCISE
√ . Describe the maximal ideal of Q[x, y] corresponding to√( 2, √2)
and (− √2, −√ 2). Describe the maximal ideal of Q[x, y] corresponding to ( 2, − 2)
and (− 2, 2). What are the residue fields in each case?
The description of “closed points” of A2Q (those points corresponding to maxi-
2
mal ideals of the ring Q[x, y]) as Galois-orbits of points in Q can even be extended
to other “nonclosed” points, as follows.

3.2.I. U NIMPORTANT AND TRICKY BUT FUN EXERCISE . Consider the map of
sets ϕ : C2 → A2Q defined as follows. (z1 , z2 ) is sent to the prime ideal of Q[x, y]
consisting of polynomials vanishing at (z1 , z2 ).
(a) What is the image of (π, π2 )?
⋆ (b) Show that ϕ is surjective. (Warning: You will need some ideas we haven’t
discussed in order to solve this. Once we define the Zariski topology on A2Q , you
will be able to check that ϕ is continuous, where we give C2 the classical topology.
This example generalizes. For example, you may later be able to generalize this to
arbitrary dimension.)

3.2.7. Quotients and localizations. Two natural ways of getting new rings from
old — quotients and localizations — have interpretations in terms of spectra.

3.2.8. Quotients: Spec A/I as a subset of Spec A. It is an important fact that the prime
ideals of A/I are in bijection with the prime ideals of A containing I.

3.2.J. E SSENTIAL ALGEBRA EXERCISE ( MANDATORY IF YOU HAVEN ’ T SEEN IT BE -

FORE ). Suppose A is a ring, and I an ideal of A. Let ϕ : A → A/I. Show that
ϕ−1 gives an inclusion-preserving bijection between prime ideals of A/I and prime
ideals of A containing I. Thus we can picture Spec A/I as a subset of Spec A.
As an important motivational special case, you now have a picture of affine
complex varieties. Suppose A is a finitely generated C-algebra, generated by x1 , . . . ,
xn , with relations f1 (x1 , . . . , xn ) = · · · = fr (x1 , . . . , xn ) = 0. Then this description
in terms of generators and relations naturally gives us an interpretation of Spec A
as a subset of AnC , which we think of as “traditional points” (n-tuples of complex
numbers) along with some “bonus” points we haven’t yet fully described. To
see which of the traditional points are in Spec A, we simply solve the equations
f1 = · · · = fr = 0. For example, Spec C[x, y, z]/(x2 + y2 − z2 ) may be pictured
as shown in Figure 3.4. (Admittedly this is just a “sketch of the R-points”, but
we will still find it helpful later.) This entire picture carries over (along with the
Nullstellensatz) with C replaced by any algebraically closed field. Indeed, the
picture of Figure 3.4 can be said to depict Spec k[x, y, z]/(x2 + y2 − z2 ) for most
algebraically closed fields k (although it is misleading in characteristic 2, because
of the coincidence x2 + y2 − z2 = (x + y + z)2 ).

3.2.9. Localizations: Spec S−1 A as a subset of Spec A. The following exercise shows
how prime ideals behave under localization.
 © 2024 Ravi Vakil. Published by Princeton University Press. 109

F IGURE 3.4. A “picture” of Spec C[x, y, z]/(x2 + y2 − z2 )

3.2.K. E SSENTIAL ALGEBRA EXERCISE ( MANDATORY IF YOU HAVEN ’ T SEEN IT

BEFORE ). Suppose S is a multiplicative subset of A. Describe an order-preserving
bijection of the prime ideals of S−1 A with the prime ideals of A that don’t meet the
multiplicative set S.
Recall from §1.2.3 that there are two important flavors of localization. The
first is Af = {1, f, f2 , . . . }−1 A where f ∈ A. A motivating example is A = C[x, y],
f = y − x2 . The second is Ap = (A \ p)−1 A, where p is a prime ideal. A motivating
example is A = C[x, y], S = A \ (x, y).
If S = {1, f, f2 , . . . }, the prime ideals of S−1 A are just those prime ideals not
containing f — the points where “f doesn’t vanish”. (In §3.5, we will call this a dis-
tinguished open set, once we know what open sets are.) So to picture Spec C[x, y]y−x2 ,
we picture the affine plane, and throw out those points on the parabola y − x2 —
the points (a, a2 ) for a ∈ C (by which we mean [(x − a, y − a2 )]), as well as the
“new kind of point” [(y − x2 )].
It can be initially confusing to think about localization in the case where zerodi-
visors are inverted, because localization A → S−1 A is not injective (Exercise 1.2.C).
Geometric intuition can help. Consider the case A = C[x, y]/(xy) and f = x. What
is the localization Af ? The space Spec C[x, y]/(xy) “is” the union of the two axes in
the affine plane. Localizing means throwing out the locus where x vanishes. So we
are left with the x-axis, minus the origin, so we expect Spec C[x]x . So there should
be some natural isomorphism
∼ / C[x]x .
(C[x, y]/(xy))x

3.2.L. E XERCISE . Show that these two rings are isomorphic. (You will see that y on
the left goes to 0 on the right.)
If S = A \ p, the prime ideals of S−1 A are just the prime ideals of A contained in
p. In our example A = C[x, y], p = (x, y), we keep all those points corresponding to
“things through the origin”, i.e., the 0-dimensional point (x, y), the 2-dimensional
point (0), and those 1-dimensional points (f(x, y)) where f(0, 0) = 0, i.e., those
“irreducible curves through the origin”. You can think of this being a shred of the
plane near the origin; anything not actually “visible” at the origin is discarded (see
Figure 3.5).
Another example is when A = k[x], and p = (x) (or more generally when p is
any maximal ideal). Then Ap has only two prime ideals (Exercise 3.2.A(b)). You
should see this as the germ of a “smooth curve”, where one point is the “classical
110 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

F IGURE 3.5. Picturing Spec C[x, y](x,y) as a “shred of A2C ”. Only

those points near the origin remain.

point”, and the other is the “generic point of the curve”. This is an example of a
discrete valuation ring, and indeed all discrete valuation rings should be visualized
in such a way. We will discuss discrete valuation rings in §13.5. By then we will
have justified the use of the words “smooth” and “curve”. (Reality check: try to
picture Spec of Z localized at (2) and at (0). How do the two pictures differ?)

3.2.10. Important fact: Maps of rings induce maps of spectra (as sets). We now
make an observation that will later grow up to be the notion of morphisms of
schemes.

3.2.M. I MPORTANT EASY EXERCISE . If ϕ : B → A is a map of rings, and p is a prime

ideal of A, show that ϕ−1 (p) is a prime ideal of B.
Hence a map of rings ϕ : B → A induces a map of sets Spec A → Spec B “in the
opposite direction”. This gives a contravariant functor from the category of rings to
the category of sets: the composition of two maps of rings induces the composition
of the corresponding maps of spectra.

3.2.N. E ASY E XERCISE ( REALITY CHECK ). Let B be a ring.

(a) Suppose I ⊂ B is an ideal. Show that the map Spec B/I → Spec B is the inclusion
of §3.2.8.
(b) Suppose S ⊂ B is a multiplicative set. Show that the map Spec S−1 B → Spec B
is the inclusion of §3.2.9.

3.2.11. An explicit example. In the case of “affine complex varieties” (or indeed
affine varieties over any algebraically closed field), the translation between maps
given by explicit formulas and maps of rings is quite direct. For example, consider
a map from the parabola in C2 (with coordinates a and b) given by b = a2 , to the
“curve” in C3 (with coordinates x, y, and z) cut out by the equations y = x2 and
z = y2 . Suppose the map sends the point (a, b) ∈ C2 to the point (a, b, b2 ) ∈ C3 .
 © 2024 Ravi Vakil. Published by Princeton University Press. 111

In our new language, we have a map

Spec C[a, b]/(b − a2 ) / Spec C[x, y, z]/(y − x2 , z − y2 )

given by
C[a, b]/(b − a2 ) o C[x, y, z]/(y − x2 , z − y2 )

(a, b, b2 ) o  (x, y, z),

i.e., x 7→ a, y 7→ b, and z 7→ b2 . If the idea is not yet clear, the following two
exercises are very much worth doing — they can be very confusing the first time
you see them, and very enlightening (and finally, trivial) when you finally figure
them out.

F IGURE 3.6. The map C → C given by x 7→ y = x2

3.2.O. I MPORTANT E XERCISE ( SPECIAL CASE ). Consider the map of complex

manifolds sending C → C via x 7→ y = x2 . We interpret the “source” C as the
“x-line”, and the “target” C the “y-line”. You can picture it as the projection of
the parabola y = x2 in the xy-plane to the y-axis (see Figure 3.6). Interpret the
corresponding map of rings as given by C[y] → C[x] by y 7→√x2 . Verify that the
preimage (the fiber) above the point a ∈ C is the point(s) √ ± a ∈ C, using
√ the
definition given above (identifying a with [(y − a)], and a with [(x − a)]). (A
more sophisticated version of this example appears in Example 10.3.4. Warning:
the roles of x and y are swapped there, in order to picture double covers in a certain
way.)

3.2.P. I MPORTANT E XERCISE ( GENERALIZING E XAMPLE 3.2.11). Suppose k is a

field, and f1 , . . . , fn ∈ k[x1 , . . . , xm ] are given. Let ϕ : k[y1 , . . . , yn ] → k[x1 , . . . , xm ]
be the morphism of k-algebras defined by yi 7→ fi .
(a) Show that ϕ induces a map of sets Spec k[x1 , . . . , xm ]/I → Spec k[y1 , . . . , yn ]/J
for any ideals I ⊂ k[x1 , . . . , xm ] and J ⊂ k[y1 , . . . , yn ] such that ϕ(J) ⊂ I. (You may
wish to consider the case I = 0 and J = 0 first. In fact, part (a) has nothing to do
with k-algebras; you may wish to prove the statement when the rings k[x1 , . . . , xm ]
and k[y1 , . . . , yn ] are replaced by general rings A and B.)
112 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

(b) Show that the map of part (a) sends the point (a1 , . . . , am ) ∈ km (or more
precisely, [(x1 − a1 , . . . , xm − am )] ∈ Spec k[x1 , . . . , xm ]) to
(f1 (a1 , . . . , am ), . . . , fn (a1 , . . . , am )) ∈ kn .

3.2.Q. E XERCISE : P ICTURING An Z . Consider the map of sets π : AZ → Spec Z, given

n

by the ring map Z → Z[x1 , . . . , xn ]. If p is prime, describe a bijection between the

fiber π−1 ([(p)]) and AnFp . (You won’t need to describe either set! Which is good
because you can’t.) This exercise may give you a sense of how to picture maps (see
Figure 3.7), and in particular why you can think of An n
Z as an “A -bundle” over
n
Spec Z. (Can you interpret the fiber over [(0)] as Ak for some field k?)

F IGURE 3.7. A picture of An Z → Spec Z as a “family of A ’s”, or

n

an “An -bundle over Spec Z”. What is the field k? How should you
“geometrically” think of the three points indicated?

3.2.12. Functions are not determined by their values at points: the fault of
nilpotents. We conclude this section by describing some strange behavior. We
are developing machinery that will let us bring our geometric intuition to algebra.
There is one serious serious point where your intuition will be false, so you should
know now, and adjust your intuition appropriately. As noted by Mumford ([Mu2,
p. 12]), “it is this aspect of schemes which was most scandalous when Grothendieck
defined them.”
Suppose we have a function (ring element) vanishing at all points. Then it
is not necessarily the zero function! The translation of this question is: is the
intersection of all prime ideals necessarily just 0? The answer is no, as is shown by
the example of the ring of dual numbers k[ϵ]/(ϵ2 ): ϵ ̸= 0, but ϵ2 = 0. (We saw this
ring in Exercise 3.2.A(a).) Any function whose power is zero certainly lies in the
intersection of all prime ideals.

3.2.R. E XERCISE . Ring elements that have a power that is 0 are called nilpotents.
(a) Show that if I is an ideal of nilpotents, then the inclusion Spec B/I → Spec B of
Exercise 3.2.J is a bijection. Thus nilpotents don’t affect the underlying set. (We will
soon see in §3.4.7 that they won’t affect the topology either — the difference will be
 © 2024 Ravi Vakil. Published by Princeton University Press. 113

in the structure sheaf.)

(b) Show that the nilpotents of a ring B form an ideal. This ideal is called the
nilradical, and is denoted N = N(B).
Thus the nilradical is contained in the intersection of all the prime ideals. The
converse is also true:

3.2.13. Theorem. — The nilradical N(A) is the intersection of all the prime ideals of A.
Geometrically: a function on Spec A vanishes at every point if and only if it is nilpotent.

3.2.S. E XERCISE . If you don’t know this theorem, then look it up, or better yet,
prove it yourself. (Hint: Use the fact that any proper ideal of A is contained in
a maximal ideal, which requires Zorn’s Lemma. Possible further hint: Suppose
x∈/ N(A). We wish to show that there is a prime ideal not containing x. Show that
Ax is not the 0-ring, by showing that 1 ̸= 0.)

3.2.14. In particular, although it is upsetting that functions are not determined by

their values at points, we have precisely specified what the failure of this intuition
is: two functions have the same values at points if and only if they differ by a
nilpotent. You should think of this geometrically: a function vanishes at every
point of the spectrum of a ring if and only if it has a power that is zero. And if there
are no nonzero nilpotents — if N = (0) — then functions are determined by their
values at points. If a ring has no nonzero nilpotents, we say that it is reduced.

3.2.T. F UN UNIMPORTANT EXERCISE : DERIVATIVES WITHOUT DELTAS AND EP -

SILONS ( OR AT LEAST WITHOUT DELTAS ). Suppose we have a polynomial
f(x) ∈ k[x]. Instead, we work in k[x, ϵ]/(ϵ2 ). What then is f(x + ϵ)? (Do a couple of
examples, then prove the pattern you observe.) This is a hint that nilpotents will be
important in defining differential information (Chapter 21).

3.3 Visualizing schemes: Generic points

A heavy warning used to be given that pictures are not rigorous; this has never had
its bluff called and has permanently frightened its victims into playing for safety. Some
pictures, of course, are not rigorous, but I should say most are (and I use them whenever
possible myself).
— J. E. Littlewood, [Lit, p. 54]
We all know that Art is not truth. Art is a lie that makes us realize truth, at least the
truth that is given us to understand. The artist must know the manner whereby to convince
others of the truthfulness of his lies.
— P. Picasso [Pi, p. 315]
For years, you have been able to picture x2 + y2 = 1 in the plane, and you
now have an idea of how to picture Spec Z. If we are claiming to understand rings
as geometric objects (through the Spec functor), then we should wish to develop
geometric insight into them. To develop geometric intuition about schemes, it is
helpful to have pictures in your mind, extending your intuition about geometric
spaces you are already familiar with. As we go along, we will empirically develop
114 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

some idea of what schemes should look like. This section summarizes what we
have gleaned so far.
Some mathematicians prefer to think completely algebraically, and never think
in terms of pictures. Others will be disturbed by the fact that this is an art, not a
science. And finally, this hand-waving will necessarily never be used in the rigorous
development of the theory. For these reasons, you may wish to skip these sections.
However, having the right picture in your mind can greatly help understanding
what facts should be true, and how to prove them. Fitgerald’s exhortation on the
importance of stretching one’s mind is essential advice to a mathematician.
The test of a first-rate intelligence is the ability to hold two opposed ideas in the mind
at the same time, and still retain the ability to function.
— F. Scott Fitzgerald, The Crack-Up [Fi, p. 41]
Our starting point is the example of “affine complex varieties” (things cut out
by equations involving a finite number of variables over C), and more generally
similar examples over arbitrary algebraically closed fields. We begin with notions
that are intuitive (“traditional” points behaving the way you expect them to), and
then add in the two features which are new and disturbing, generic points and
nonreduced behavior. You can then extend this notion to seemingly different spaces,
such as Spec Z.
Hilbert’s Weak Nullstellensatz 3.2.5 shows that the “traditional points” are
present as points of the scheme, and this carries over to any algebraically closed
field. If the field is not algebraically closed, the traditional points are glued together
into clumps by Galois conjugation, as in Examples 5 (the real affine line) and 6
(the affine line over Fp ) in §3.2. This is a geometric interpretation of Hilbert’s
Nullstellensatz 3.2.6.
But we have some additional points to add to the picture. You should remember
that they “correspond” to “irreducible” “closed” (algebraic) subsets. As motivation,
consider the case of the complex affine plane (Example 7): we had one for each
irreducible polynomial, plus one corresponding to the entire plane. We will make
“closed” precise when we define the Zariski topology (in the next section). You may
already have an idea of what “irreducible” should mean; we make that precise at
the start of §3.6. By “correspond” we mean that each closed irreducible subset has
a corresponding point sitting on it, called its generic point (defined in §3.6). It is a
new point, distinct from all the other points in the subset. (The correspondence is
described in Exercise 3.7.F for Spec A, and in Exercise 5.1.B for schemes in general.)
We don’t know precisely where to draw the generic point, so we may stick it
arbitrarily anywhere, but you should think of it as being “almost everywhere”, and
in particular, near every other point in the subset.
In §3.2.8, we saw how the points of Spec A/I should be interpreted as a subset
of Spec A. So for example, when you see Spec C[x, y]/(x + y), you should picture
this not just as a line, but as a line in the xy-plane; the choice of generators x and y
of the algebra C[x, y] implies an inclusion into affine space.
In §3.2.9, we saw how the points of Spec S−1 A should be interpreted as subsets
of Spec A. The two most important cases were discussed. The points of Spec Af
correspond to the points of Spec A where f doesn’t vanish; we will later (§3.5)
interpret this as a distinguished open set.
 © 2024 Ravi Vakil. Published by Princeton University Press. 115

If p is a prime ideal, then Spec Ap should be seen as a “shred of the space Spec A
near the subset corresponding to p”. The simplest nontrivial case of this is A = k[x]
and p = (x) ⊂ A (see Exercise 3.2.A, which we discuss again in Exercise 3.4.K).
“If any one of them can explain it,” said Alice, (she had grown so large in the last few
minutes that she wasn’t a bit afraid of interrupting him), “I’ll give him sixpence. I don’t
believe there’s an atom of meaning in it.” ...
“If there’s no meaning in it,” said the King, “that saves a world of trouble, you know,
as we needn’t try to find any.”
— Lewis Carroll [Carr, Ch. XII]

3.4 The underlying topological space of an affine scheme

We next introduce the Zariski topology on the spectrum of a ring. When you first
hear the definition, it seems odd, but with a little experience it becomes reasonable.
As motivation, consider A2C = Spec C[x, y], the complex (affine) plane (with a few
extra points). In algebraic geometry, we will only be allowed to consider algebraic
functions, i.e., polynomials in x and y. The locus where a polynomial vanishes
should reasonably be a closed set, and the Zariski topology is defined by saying
that the only sets we should consider closed should be these sets, and other sets
forced to be closed by these. In other words, it is the coarsest topology where these
sets are closed.

3.4.1. In particular, although topologies are often described using open subsets, it
will be more convenient for us to define this topology in terms of closed subsets. If
S is a subset of a ring A, define the vanishing set of S by

V(S) := {[p] ∈ Spec A : S ⊂ p}.

It is the set of points on which all elements of S are zero. (It should now be second
nature to equate “vanishing at a point” with “contained in a prime”.) We declare
that these — and no others — are the closed subsets.
For example, consider V(xy, yz) ⊂ A3C = Spec C[x, y, z]. Which points are
contained in this locus? We think of this as solving xy = yz = 0. Of the “traditional”
points (interpreted as ordered triples of complex numbers, thanks to the Hilbert’s
Nullstellensatz 3.2.5), we have the points where y = 0 or x = z = 0: the xz-plane
and the y-axis respectively. Of the “new” points, we have the generic point of the
xz-plane (also known as the point [(y)]), and the generic point of the y-axis (also
known as the point [(x, z)]). You might imagine that we also have a number of
“one-dimensional” points contained in the xz-plane.

3.4.A. E ASY E XERCISE . Check that the x-axis is contained in V(xy, yz). (The x-axis
is defined by y = z = 0, and the y-axis and z-axis are defined analogously.)
Let’s return to the general situation. The following exercise lets us restrict
attention to vanishing sets of ideals.

3.4.B. E ASY E XERCISE . Show that if (S) is the ideal generated by S, then V(S) =
V((S)).
116 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

3.4.2. Definition. We define the Zariski topology by declaring that V(S) is closed for
all S. (We may as well state here that Zariski closure means closure in the Zariski
topology.) Let’s check that the Zariski topology is indeed a topology:

3.4.C. E XERCISE .
(a) Show that ∅ and Spec A are both open subsets of Spec A.
(b)PIf Ii is a collection of ideals (as i runs over some index set), show that ∩i V(Ii ) =
V( i Ii ). Hence the union of any collection of open sets is open.
(c) Show that V(I1 ) ∪ V(I2 ) = V(I1 I2 ). (The product of two ideals I1 and I2 of A
are finite A-linear
Pn combinations of products of elements of I1 and I2 , i.e., elements
of the form j=1 i1,j i2,j , where ik,j ∈ Ik . Equivalently, it is the ideal generated by
products of elements of I1 and I2 . You should quickly check that this is an ideal,
and that products are associative, i.e., (I1 I2 )I3 = I1 (I2 I3 ).) Hence the intersection
of any finite number of open sets is open.

3.4.3. Properties of the “vanishing set” function V(·). The function V(·) is
obviously inclusion-reversing: If S1 ⊂ S2 , then V(S2 ) ⊂ V(S1 ). Warning: We could
have equality in the second inclusion without equality in the first, as the next
exercise shows.

3.4.D. E XERCISE /D EFINITION . If I ⊂ A is an ideal, then define its radical by

√
I := {r ∈ A : rn ∈ I for some n ∈ Z>0 }.
p √
For example, the nilradical N
√ (§3.2.R) is (0). Show that I is an ideal (cf. Exer-
cise 3.2.R(b)). Show thatpV( I) = V(I). We say an ideal is radical if it equals its
√ √
own radical. Show that I = I, and that prime ideals are radical.
Here are two useful consequences. As (I ∩ J)2 ⊂ IJ ⊂ I ∩ J (products of ideals
were defined in Exercise 3.4.C), we have that V(IJ) = V(I ∩ J) (= V(I) ∪ V(J) by
Exercise
p 3.4.C(c)). Also, combining this with Exercise 3.4.B, we see V(S) = V((S)) =
V( (S)).

3.4.E. E XERCISE ( RADICALS COMMUTE p WITH FINITE √INTERSECTIONS ). If I1 , . . . ,

n
In are ideals of a ring A, show that ∩n I
i=1 i = ∩i=1 Ii . We will use this property
repeatedly without referring back to this exercise.
√
3.4.F. E XERCISE FOR LATER USE . Show that I is the intersection of all the prime
ideals containing I. (Hint: Use Theorem 3.2.13 on an appropriate ring.)

3.4.4. Examples. Let’s see how this meshes with our examples from the previous
section.
Recall that A1C , as a set, was just the “traditional” points (corresponding to
maximal ideals, in bijection with a ∈ C), and one “new” point [(0)]. The Zariski
topology on A1C is not that exciting: the open sets are the empty set, and A1C minus a
finite number of maximal ideals. (It “almost” has the cofinite topology. Notice that
the open sets are determined by their intersections with the “traditional points”.
The “new” point [(0)] comes along for the ride, which is a good sign that it is
harmless. Ignoring the “new” point, observe that the topology on A1C is a coarser
topology than the classical topology on C.)
 © 2024 Ravi Vakil. Published by Princeton University Press. 117

3.4.G. E XERCISE . Describe the topological space A1k (cf. Exercise 3.2.D). (Notice that
the strange new point [(0)] is “near every other point” — every neighborhood of ev-
ery point contains [(0)]. This is typical of these new points, see Easy Exercise 3.6.N.)

The case of Spec Z is similar. The topology is “almost” the cofinite topology in
the same way. The open sets are the empty set, and Spec Z minus a finite number
of “ordinary” ((p) where p is prime) primes.

3.4.5. Closed subsets of A2C . The case A2C is more interesting. You should think
through where the “one-dimensional prime ideals” fit into the picture. In Exer-
cise 3.2.E, we identified all the prime ideals of C[x, y] (i.e., the points of A2C ) as
the maximal ideals [(x − a, y − b)] (where a, b ∈ C — “zero-dimensional points”),
the “one-dimensional points” [(f(x, y))] (where f(x, y) is irreducible), and the “two-
dimensional point” [(0)].
Then the closed subsets are of the following form:
(a) the entire space (the closure of the “two-dimensional point” [(0)]),
(b) a finite number (possibly none) of “curves” (each the closure of a “one-
dimensional point” — the “one-dimensional point” along with the “zero-dimensional
points” “lying on it”) and a finite number (possibly none) of “zero-dimensional”
points (what we will soon call “closed points”, see Definition 3.6.8).
We will soon know enough to verify this using general theory, but you can
prove it yourself now, using ideas in Exercise 3.2.E. (The key idea: if f(x, y) and
g(x, y) are irreducible polynomials that are not multiples of each other, why do
their zero sets intersect in a finite number of points?)

3.4.6. Important fact: Maps of rings induce continuous maps of topological

spaces. We saw in §3.2.10 that a map of rings ϕ : B → A induces a map of sets
π : Spec A → Spec B.

3.4.H. I MPORTANT EASY EXERCISE . By showing that closed sets pull back to closed
sets, show that π is a continuous map. Interpret Spec as a contravariant functor
Rings → Top.
Not all continuous maps arise in this way. Consider for example the continuous
map on A1C that is the identity except 0 and 1 (i.e., [(x)] and [(x − 1)]) are swapped;
no polynomial can manage this marvelous feat.
In §3.2.10, we saw that Spec B/I and Spec S−1 B are naturally subsets of Spec B. It
is natural to ask if the Zariski topology behaves well with respect to these inclusions,
and indeed it does.

3.4.I. I MPORTANT E XERCISE ( CF. E XERCISE 3.2.N). Suppose that I, S ⊂ B are an

ideal and multiplicative subset respectively.
(a) Show that Spec B/I is naturally a closed subset of Spec B. If S = {1, f, f2 , . . . }
(f ∈ B), show that Spec S−1 B is naturally an open subset of Spec B. Show that for
arbitrary S, Spec S−1 B need not be open or closed. (Hint: Spec Q ⊂ Spec Z, or
possibly Figure 3.5.)
(b) Show that the Zariski topology on Spec B/I (resp., Spec S−1 B) is the subspace
topology induced by inclusion in Spec B. (Hint: compare closed subsets.)
118 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

3.4.7. In particular, if I ⊂ N is an ideal of nilpotents, the bijection Spec B/I → Spec B

(Exercise 3.2.R) is a homeomorphism. Thus nilpotents don’t affect the topological
space. (The difference will be in the structure sheaf.)

√LATER . Suppose I ⊂ B is an ideal. Show that f vanishes

3.4.J. U SEFUL EXERCISE FOR
on V(I) if and only if f ∈ I (i.e., fn ∈ I for some n ≥ 1). (Hint: Exercise 3.4.F. If
you are stuck, you will get another hint when you see Exercise 3.5.E.)

3.4.K. E ASY E XERCISE ( CF. E XERCISE 3.2.A). Describe the topological space
Spec k[x](x) .

3.5 A base of the Zariski topology on Spec A: Distinguished open

sets

If f ∈ A, define the distinguished open set

D(f) := {[p] ∈ Spec A : f ∈

/ p}
= {[p] ∈ Spec A : f([p]) ̸= 0}.

It is the locus where f doesn’t vanish. (I often privately write this as D(f ̸= 0) to
remind myself of this. I also privately call this the “Doesn’t-vanish set” of f in
analogy with V(f) being the Vanishing set of f.) We have already seen this set when
discussing Spec Af as a subset of Spec A. For example, we have observed that the
Zariski-topology on the distinguished open set D(f) ⊂ Spec A coincides with the
Zariski topology on Spec Af (Exercise 3.4.I).
The reason these sets are important is that they form a particularly nice base
for the (Zariski) topology:

3.5.A. E ASY E XERCISE . Show that the distinguished open sets form a base for the
(Zariski) topology. (Hint: Given a subset S ⊂ A, show that the complement of V(S)
is ∪f∈S D(f).)
Here are some important but not difficult exercises to give you a feel for this
concept.

3.5.B. N IFTY E XERCISE . Suppose fi ∈ A as i runs over some index set J. Show that
∪i∈J D(fi ) = Spec A if and only if ({fi }i∈J ) = A, or equivalently
P and very usefully,
if there are ai (i ∈ J), all but finitely many 0, such that i∈J ai fi = 1. (One of the
directions will use the fact that any proper ideal of A is contained in some maximal
ideal.)

3.5.C. E XERCISE . Show that if Spec A is an infinite union of distinguished open

sets ∪j∈J D(fj ), then in fact it is a union of a finite number of these, i.e., there is a
finite subset J ′ so that Spec A = ∪j∈J ′ D(fj ). (Hint: Exercise 3.5.B.)

3.5.D. E ASY E XERCISE . Show that D(f) ∩ D(g) = D(fg).

3.5.E. I MPORTANT E XERCISE ( CF. E XERCISE 3.4.J). Show that D(f) ⊂ D(g) if and
only if fn ∈ (g) for some n ≥ 1, if and only if g is an invertible element of Af .
 © 2024 Ravi Vakil. Published by Princeton University Press. 119

We will use Exercise 3.5.E often. You can solve it thinking purely algebraically,
but the following geometric interpretation may be helpful. (You should try to draw
your own picture to go with this discussion.) Inside Spec A, we have the closed
subset V(g) = Spec A/(g), where g vanishes, and its complement D(g), where g
doesn’t vanish. Then f is a function on this closed subset V(g) (or more precisely, on
Spec A/(g)), and by assumption it vanishes at all points of the closed subset. Now
any function vanishing at every point of the spectrum of a ring must be nilpotent
(Theorem 3.2.13). In other words, there is some n such that fn = 0 in A/(g), i.e.,
fn ≡ 0 (mod g) in A, i.e., fn ∈ (g).

3.5.F. E ASY E XERCISE . Show that D(f) = ∅ if and only if f ∈ N.

3.5.G. U NIMPORTANT E XERCISE ( INJECTIVE MAPS OF RINGS INDUCE MAPS OF

SPECTRA WITH DENSE IMAGE ). Show that if B ,→ A, then the induced map of
topological spaces Spec A → Spec B has dense image.

3.5.1. Remark/caution about notation. There will be two variations on the notation
D(f): the projective distinguished open set D+ (f) (the locus on a projective scheme
where a “form f of positive degree” doesn’t vanish, §4.5.8), and Xf (the locus on a
scheme X where a function f doesn’t vanish, Definition 6.2.8). These are almost the
same thing as D(f), but not quite, so we keep the notation separate for reasons of
hygiene.

3.6 Topological (and Noetherian) properties

Many topological notions are useful when applied to the topological space
Spec A, and later, to schemes.

3.6.1. Possible topological attributes of Spec A: connectedness, irreducibility,

quasicompactness.

3.6.2. Connectedness.
A topological space X is connected if it cannot be written as the disjoint union of
two nonempty open sets. Exercise 3.6.A below gives an example of a nonconnected
Spec A, and the subsequent remark explains that all examples are of this form.

3.6.A. E`XERCISE .
`If A`= A1 × A2 × · · · × An , describe a homeomorphism
Spec A1 Spec A2 · · · Spec An → Spec A for which each QnSpec Ai `is mapped
n
onto a distinguished open subset D(fi ) of Spec A. Thus Spec i=1 Ai = i=1 Spec Ai
as topological spaces. (Hint: reduce to n = 2 for convenience. Let f1 = (1, 0) and
f2 = (0, 1).)

3.6.3. Remark: the idempotent-connectedness package. An extension of Exercise 3.6.A is

that Spec A is not connected if and only if A is isomorphic to the product of nonzero
rings A1 and A2 . The key idea is to show that both conditions are equivalent to
there existing nonzero a1 , a2 ∈ A for which a21 = a1 , a22 = a2 , a1 + a2 = 1, and
hence a1 a2 = 0. An element a ∈ A satisfying a2 = a is called an idempotent. This
will appear as Exercise 10.5.L.
120 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

3.6.4. Irreducibility.
A topological space is said to be irreducible if it is nonempty, and it is not
the union of two proper closed subsets. In other words, a nonempty topological
space X is irreducible if whenever X = Y ∪ Z with Y and Z closed in X, we have
Y = X or Z = X. Equivalently (and helpfully): any two nonempty open subsets of
X intersect. This is a less useful notion in classical geometry — C2 is reducible (i.e.,
not irreducible), but we will see that A2C is irreducible (Exercise 3.6.C).

3.6.B. E ASY E XERCISE .

(a) Show that in an irreducible topological space, any nonempty open set is dense.
(For this reason, you will see that unlike in the classical topology, in the Zariski
topology, nonempty open sets are all “huge”.)
(b) If X is a topological space, and Z is a subset (with the subspace topology), show
that Z is irreducible if and only if the closure Z in X is irreducible.

3.6.C. E ASY E XERCISE . If A is an integral domain, show that Spec A is irreducible.

(Hint: pay attention to the generic point [(0)].) We will generalize this in Exer-
cise 3.7.G.

3.6.D. E XERCISE . Show that an irreducible topological space is connected.

3.6.E. E XERCISE . Give (with proof!) an example of a ring A where Spec A is

connected but reducible. (Possible hint: a picture may help. The symbol “×” has
two “pieces” yet is connected.)

3.6.F. T RICKY E XERCISE .

(a) Suppose I = (wz−xy, wy−x2 , xz−y2 ) ⊂ k[w, x, y, z]. Show that Spec k[w, x, y, z]/I
is irreducible, by showing that k[w, x, y, z]/I is an integral domain. (This is hard, so
here is one of several possible hints: Show that k[w, x, y, z]/I is isomorphic to the
k-subalgebra of k[a, b] generated by monomials of degree divisible by 3. There are
other approaches as well, some of which we will see later. This is an example of a
hard question: how do you tell if an ideal is prime?) We will later see this as the
cone over the twisted cubic curve (the twisted cubic curve is defined in Exercise 9.3.A,
and is a special case of a Veronese embedding, §9.3.6).
⋆ (b) Note that the generators of the ideal of part (a) may be rewritten as the
equations ensuring that
 
w x y
rank ≤ 1,
x y z
i.e., as the determinants of the 2 × 2 submatrices. Generalize part (a) to the ideal of
rank one 2 × n matrices. This notion will correspond to the cone (§9.3.11) over the
degree n rational normal curve (Exercise 9.3.I).

3.6.5. Quasicompactness.
A topological space X is quasicompact if given any cover X = ∪i∈I Ui by open
sets, there is a finite subset S of the index set I such that X = ∪i∈S Ui . Informally:
every open cover has a finite subcover. We will like this condition, because we
are afraid of infinity. Depending on your definition of “compactness”, this is the
definition of compactness, minus possibly a Hausdorff condition. However, this
isn’t really the algebro-geometric analog of “compact” (we certainly wouldn’t want
A1C to be compact) — the right analog is “properness” (§11.4).
 © 2024 Ravi Vakil. Published by Princeton University Press. 121

3.6.G. E XERCISE .
(a) Show that Spec A is quasicompact. (Hint: Exercise 3.5.C.)
⋆ (b) (less important) Show that in general Spec A can have nonquasicompact
open sets. Possible hint: let A = k[x1 , x2 , x3 , . . . ] and m = (x1 , x2 , . . . ) ⊂ A, and
consider the complement of V(m). This example will be useful to construct other
“counterexamples” later, e.g., Exercises 8.1.F and 5.1.K. In Exercise 3.6.U, we will see
that such weird behavior doesn’t happen for “suitably nice” (Noetherian) rings.

3.6.H. E XERCISE .
(a) If X is a topological space that is a finite union of quasicompact spaces, show
that X is quasicompact.
(b) Show that every closed subset of a quasicompact topological space is quasicom-
pact.
`n ∼
3.6.6.Q⋆⋆ Fun but irrelevant remark. Exercise 3.6.A shows that i=1 Spec Ai =
n
Spec i=1 Ai , but this never holds if “n is infinite” and all Ai are nonzero, as
Spec of any ring is quasicompact (Exercise
Q∞ 3.6.G(a)). This leads to an`∞ interesting
phenomenon. We show that Spec i=1 Ai is “strictly bigger” than i=1 Spec Ai
where
`∞ each Ai is isomorphic Q∞ to the field k. First, we have Q an inclusion of sets
i=1 Spec A i ,→ Spec A
i=1 i , as there is a maximal ideal of Ai corresponding
to each i (precisely,
Q those elements 0 in the ith component.) But there are other
maximal ideals of Ai . Hint: describe a proper Q ideal not contained in any of these
maximal ideals. (One idea: consider elements ai that are “eventually zero”, i.e.,
ai = 0 for i ≫ 0.) This leads to the notion of ultrafilters, which are very useful, but
irrelevant to our current discussion.

3.6.7. Possible topological properties of points of Spec A.

3.6.8. Definition. A point of a topological space p ∈ X is said to be a closed point

if {p} is a closed subset. In the classical topology on Cn , all points are closed. In
Spec Z and Spec k[t], all the points are closed except for [(0)].

3.6.I. E XERCISE . Show that the closed points of Spec A correspond to the maximal
ideals. (In particular, nonempty affine schemes have closed points, as nonzero rings
have maximal ideals, §0.3.)

3.6.9. Connection to the classical theory of varieties. Hilbert’s Nullstellensatz lets us

interpret the closed points of An C as the n-tuples of complex numbers. More gener-
ally, the closed points of Spec k[x1 , . . . , xn ]/(f1 , . . . , fr ) (where k is an algebraically
n
closed field) are naturally interpreted as those points in k satisfying the equations
f1 = · · · = fr = 0 (see Exercises 3.2.J and 3.2.N(a) for example). Hence from now on
we will say “closed point” instead of “traditional point” and “non-closed point” in-
stead of “bonus” point when discussing subsets of An k
. The following exercise is the
reason that on algebraic varieties, the set of closed points is dense (cf. Exercise 5.3.F
and Definition 11.2.9).

3.6.J. E XERCISE .
(a) Suppose that k is a field, and A is a finitely generated k-algebra. Show that the set
of closed points of Spec A is dense, by showing that if f ∈ A, and D(f) is a nonempty
(distinguished) open subset of Spec A, then D(f) contains a closed point of Spec A.
122 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Hint: note that Af is also a finitely generated k-algebra. Use the Nullstellensatz 3.2.6
to recognize closed points of Spec of a finitely generated k-algebra B as those for
which the residue field (the quotient of A by the corresponding maximal ideal) is a
finite extension of k. Apply this to both B = A and B = Af .
(b) Show that if A is a k-algebra that is not finitely generated, the set of closed
points need not be dense. (Hint: Exercise 3.4.K.)

3.6.K. E XERCISE . Suppose k is an algebraically closed field, and A = k[x1 , . . . , xn ]/I

is a finitely generated k-algebra with N(A) = {0} (so the discussion of §3.2.14
applies). Consider the set X = Spec A as a subset of An n
k . The space Ak contains the
n
“classical” points k . Show that functions on X are determined by their values on the
closed points (by the weak Nullstellensatz 3.2.5, the “classical” points kn ∩ Spec A
of Spec A). Hint: if f and g are different functions on X, then f − g is nowhere zero
on an open subset of X. Use Exercise 3.6.J(a).
Once we know what a variety is (Definition 11.2.9), this will immediately imply
that a function on a variety over an algebraically closed field is determined by its values on
the “classical points”. (Before the advent of scheme theory, functions on varieties —
over algebraically closed fields — were thought of as functions on “classical” points,
and Exercise 3.6.K basically shows that there is no harm in thinking of “traditional”
varieties as a particular flavor of schemes.)

3.6.10. Specialization and generization. Given two points x, y of a topological space

X, we say that x is a specialization of y, and y is a generization of x, if x ∈ {y}.
This (and Exercise 3.6.L) now makes precise our hand-waving about “one point
containing another”. It is of course nonsense for a point to contain another. But it is
not nonsense to say that the closure of a point contains another. For example, in
A2C = Spec C[x, y], [(y − x2 )] is a generization of [(x − 2, y − 4)] = (2, 4) ∈ C2 , and
(2, 4) is a specialization of [(y − x2 )] (see Figure 3.8).

F IGURE 3.8. (2, 4) = [(x − 2, y − 4)] is a specialization of [(y − x2 )].

[(y − x2 )] is a generization of (2, 4).

3.6.L. E XERCISE . If X = Spec A, show that [q] is a specialization of [p] if and only if
p ⊂ q. Hence show that V(p) = {[p]}.

3.6.11. Definition. We say that a point p ∈ X is a generic point for a closed subset K
if {p} = K.
This important notion predates Grothendieck. The early twentieth-century
Italian algebraic geometers had a notion of “generic points” of a variety, by which
 © 2024 Ravi Vakil. Published by Princeton University Press. 123

they meant points with no special properties, so that anything proved of “a generic
point” was true of “almost all” the points on that variety. The modern “generic
point” has the same intuitive meaning. If something is “generically” or “mostly”
true for the points of an irreducible subset, in the sense of being true for a dense
open subset (for “almost all points”), then it is true for the generic point, and
vice versa. (This is a statement of principle, not of fact. An interesting case is
“reducedness”, for which this principle does not hold in general, but does hold for
“reasonable” schemes such as varieties, see Remark 5.2.2.) For example, a function
has value zero at the generic point of an irreducible scheme if and only if it has the
value zero at all points. You should keep an eye out for other examples of this.
The phrase general point does not have the same meaning. The phrase “the
general point of X satisfies such-and-such a property” means “every point of some
dense open subset of X satisfies such-and-such a property”. Be careful not to confuse
“general” and “generic”. But be warned that accepted terminology does not always
follow this convention; witness “generic smoothness” (§21.6.3, for example) and
“generic flatness” (Theorem 24.5.13).

3.6.M. E XERCISE . Verify that [(y − x2 )] ∈ A2C is a generic point for V(y − x2 ).
As more motivation for the terminology “generic”: we think of [(y − x2 )] as
being some nonspecific point on the parabola (with the closed points (a, a2 ) ∈
C2 , i.e., (x − a, y − a2 ) for a ∈ C, being “specific points”); it is “generic” in the
conventional sense of the word. We might “specialize it” to a specific point of
the parabola; hence for example (2, 4) is a specialization of [(y − x2 )]. (Again, see
Figure 3.8.) To make this somewhat more precise:

3.6.N. E ASY E XERCISE . Suppose p is a generic point for the closed subset K. Show
that it is “near every point q of K” (every neighborhood of q contains p), and “not
near any point r not in K” (there is a neighborhood of r not containing p). (This
idea was mentioned in §3.2.4 Example 1 and in Exercise 3.4.G.)
We will soon see (Exercise 3.7.F) that there is a natural bijection between points
of Spec A and irreducible closed subsets of Spec A, sending each point to its closure,
and each irreducible closed subset to its (unique) generic point. You can prove this
now, but we will wait until we have developed some convenient terminology.

3.6.12. Irreducible and connected components, and Noetherian conditions.

An irreducible component of a topological space is a maximal irreducible
subset (an irreducible subset not contained in any larger irreducible subset). Irre-
ducible components are closed (as the closure of irreducible subsets are irreducible,
Exercise 3.6.B(b)), and it can be helpful to think of irreducible components of a
topological space X as maximal among the irreducible closed subsets of X. We think
of these as the “pieces of X” (see Figure 3.9).
Similarly, a subset Y of a topological space X is a connected component if it
is a maximal connected subset (a connected subset not contained in any larger
connected subset).

3.6.O. E XERCISE ( EVERY TOPOLOGICAL SPACE IS THE UNION OF IRREDUCIBLE

COMPONENTS ). Show that every point p of a topological space X is contained in
an irreducible component of X. Hint: Zorn’s Lemma. More precisely, consider the
124 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

F IGURE 3.9. This closed subset of A2C has six irreducible components

partially ordered set S of irreducible subsets of X containing p. Show that there

exists a maximal totally ordered subset {Zα } of S . Show that ∪Zα is irreducible.

3.6.13. Remark. Every point is contained in a connected component, and con-

nected components are always closed. You can prove this now, but we deliberately
postpone this until we need it, in an optional starred section (Exercise 10.5.J). On
the other hand, connected components need not be open, see [Stacks, tag 004T].
An example
Q∞ of an affine scheme with connected components that are not open is
Spec( 1 F2 ).

3.6.14. In the examples we have considered, the spaces have naturally broken up
into a finite number of irreducible components. For example, the locus xy = 0
in A2C we think of as having two “pieces” — the two axes. The reason for this is
that their underlying topological spaces (as we shall soon establish) are Noetherian.
A topological space X is called Noetherian if it satisfies the descending chain
condition for closed subsets: any sequence Z1 ⊃ Z2 ⊃ · · · ⊃ Zn ⊃ · · · of closed
subsets eventually stabilizes: there is an r such that Zr = Zr+1 = · · · . Here is a first
example (which you should work out explicitly, not using Noetherian rings).

3.6.P. E XERCISE . Show that A2C is a Noetherian topological space: any decreasing
sequence of closed subsets of A2C = Spec C[x, y] must eventually stabilize. Note that
it can take arbitrarily long to stabilize. (The closed subsets of A2C were described in
§3.4.5.) Show that C2 with the classical topology is not a Noetherian topological
space.

3.6.15. Proposition. — Suppose X is a Noetherian topological space. Then every

nonempty closed subset Z can be expressed uniquely as a finite union Z = Z1 ∪ · · · ∪ Zn
of irreducible closed subsets, none contained in any other.
Translation: any closed subset Z has a finite number of “pieces”.

Proof. The following technique is called Noetherian induction, for reasons that will
be clear. We will use it again, many times.
Consider the collection of nonempty closed subsets of X that cannot be ex-
pressed as a finite union of irreducible closed subsets. We will show that this
collection is empty. Otherwise, let Y1 be one such. If Y1 properly contains another
such, then choose one, and call it Y2 . If Y2 properly contains another such, then
choose one, and call it Y3 , and so on. By the descending chain condition, this must
eventually stop, and we must have some Yr that cannot be written as a finite union
 © 2024 Ravi Vakil. Published by Princeton University Press. 125

of irreducible closed subsets, but every closed subset properly contained in it can
be so written. But then Yr is not itself irreducible, so we can write Yr = Y ′ ∪ Y ′′
where Y ′ and Y ′′ are both proper closed subsets. Both of these by hypothesis can be
written as the union of a finite number of irreducible closed subsets, and hence so
can Yr , yielding a contradiction. Thus each closed subset can be written as a finite
union of irreducible closed subsets. We can assume that none of these irreducible
closed subsets contain any others, by discarding some of them.
We now show uniqueness. Suppose

Z = Z1 ∪ Z2 ∪ · · · ∪ Zr = Z1′ ∪ Z2′ ∪ · · · ∪ Zs′

are two such representations. Then Z1′ ⊂ Z1 ∪ Z2 ∪ · · · ∪ Zr , so Z1′ = (Z1 ∩ Z1′ ) ∪

· · · ∪ (Zr ∩ Z1′ ). Now Z1′ is irreducible, so one of these is Z1′ itself, say (without
loss of generality) Z1 ∩ Z1′ . Thus Z1′ ⊂ Z1 . Similarly, Z1 ⊂ Za′ for some a; but
because Z1′ ⊂ Z1 ⊂ Za′ , and Z1′ is contained in no other Zi′ , we must have a = 1,
and Z1′ = Z1 . Thus each element of the list of Z’s is in the list of Z ′ ’s, and vice versa,
so they must be the same list. □

3.6.Q. E XERCISE . Show that every connected component of a topological space

X is the union of irreducible components of X. Show that any subset of X that
is simultaneously open and closed must be the union of some of the connected
components of X. If X is a Noetherian topological space, show that the union of
any subset of the connected components of X is always open and closed in X. (In
particular, connected components of Noetherian topological spaces are always
open, which is not true for more general topological spaces, see Remark 3.6.13.)

3.6.16. Noetherian rings. It turns out that all of the spectra we have considered
(except in starred Exercise 3.6.G(b)) are Noetherian topological spaces, but that
isn’t true of the spectra of all rings. The key characteristic all of our examples have
had in common is that the rings were Noetherian. A ring is Noetherian if every
ascending sequence I1 ⊂ I2 ⊂ · · · of ideals eventually stabilizes: there is an r such
that Ir = Ir+1 = · · · . (This is called the ascending chain condition on ideals.)
Here are some quick facts about Noetherian rings. You should be able to prove
them all.

• Fields are Noetherian. Z is Noetherian.

• (Noetherianness is preserved by quotients.) If A is Noetherian, and I is any
ideal of A, then A/I is Noetherian.
• (Noetherianness is preserved by localization.) If A is Noetherian, and S is any
multiplicative set, then S−1 A is Noetherian.

3.6.R. E XERCISE . Show that principal ideal domains are Noetherian rings.

3.6.S. I MPORTANT E XERCISE . Show that a ring A is Noetherian if and only if every
ideal of A is finitely generated.
The next fact is nontrivial.

3.6.17. The Hilbert Basis Theorem. — If A is Noetherian, then so is A[x].

126 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Hilbert proved this in the epochal paper [Hil] where he also proved the Hilbert
Syzygy Theorem (§16.1.2), and defined Hilbert functions and showed that they are
eventually polynomial (§18.6).

3.6.18. By the results described above, any polynomial ring in finitely many
variables over any field, or over the integers, is Noetherian — and also any
quotient or localization thereof. Hence for example any finitely generated alge-
bra over k or Z, or any localization thereof, is Noetherian. Most “nice” rings
are Noetherian, but not all rings are Noetherian: k[x1 , x2 , . . . ] is not, because
(x1 ) ⊂ (x1 , x2 ) ⊂ (x1 , x2 , x3 ) ⊂ · · · is a strictly ascending chain of ideals (cf.
Exercise 3.6.G(b)).

3.6.19. Proof of the Hilbert Basis Theorem 3.6.17. We show that any ideal I ⊂ A[x]
is finitely generated. We inductively produce a set of generators f1 , . . . as follows.
For n ≥ 0, if I ̸= (f1 , . . . , fn ), let fn+1 be any nonzero element of I − (f1 , . . . , fn )
of lowest degree. Thus f1 is any element of I of lowest degree, assuming I ̸= (0).
If this procedure terminates, we are done. Otherwise, let an ∈ A be the initial
coefficient of fn for all n > 0. As A is Noetherian, (a1 , a2 , . . . ) = (a1 , . . . , aN ) for
PN
some N. Say aN+1 = i=1 bi ai . Then

X
N
fN+1 − bi fi xdeg fN+1 −deg fi
i=1

is an element of I that is not in (f1 , . . . , fN ) (as fN+1 ∈

/ (f1 , . . . , fN )), and of lower
degree than fN+1 , contradicting how we were supposed to have chosen fN+1 . □
We now connect Noetherian rings and Noetherian topological spaces.

3.6.T. E XERCISE . If A is Noetherian, show that Spec A is a Noetherian topological

space. Describe a ring A such that Spec A is not a Noetherian topological space.
(Aside: if Spec A is a Noetherian topological space, A need not be Noetherian. One
example is A = k[x1 , x2 , x3 , . . . ]/(x1 , x22 , x33 , . . . ). Then Spec A has one point, so is
Noetherian. But A is not Noetherian as (x1 ) ⊊ (x1 , x2 ) ⊊ (x1 , x2 , x3 ) ⊊ · · · in A.)

3.6.U. E XERCISE ( PROMISED IN E XERCISE 3.6.G( B )). Show that every open subset
of a Noetherian topological space is quasicompact. Hence if A is Noetherian, every
open subset of Spec A is quasicompact. (If you prefer, show the result for any subset,
with the induced topology.)

3.6.20. For future use: Noetherian conditions for modules. An important related
notion is that of a Noetherian module. Although we won’t use this notion for some
time (§10.7.3), we will develop their most important properties, while Noetherian
ideas are still fresh in your mind. If A is any ring, not necessarily Noetherian,
we say an A-module is Noetherian if it satisfies the ascending chain condition
for submodules. Thus for example a ring A is Noetherian if and only if it is a
Noetherian A-module.

3.6.V. E XERCISE ( CF. I MPORTANT E XERCISE 3.6.S). Show that if M is a Noetherian

A-module, then any submodule of M is a finitely generated A-module. (In fact the
converse is true, and you are welcome to prove that too.)
 © 2024 Ravi Vakil. Published by Princeton University Press. 127

3.6.W. E XERCISE . If 0 → M ′ → M → M ′′ → 0 is exact, show that M ′ and M ′′ are

Noetherian if and only if M is Noetherian. (Hint: Given an ascending chain in M,
we get two simultaneous ascending chains in M ′ and M ′′ . Possible further hint:
prove that if

M′ /M ϕ
/ M ′′

is exact, and N ⊂ N ′ ⊂ M, and N ∩ M ′ = N ′ ∩ M ′ and ϕ(N) = ϕ(N ′ ), then

N = N ′ .)

3.6.X. E XERCISE . Show that if A is a Noetherian ring, then A⊕n is a Noetherian

A-module.

3.6.Y. E XERCISE . Show that if A is a Noetherian ring and M is a finitely generated A-

module, then M is a Noetherian module. Hence by Exercise 3.6.V, any submodule
of a finitely generated module over a Noetherian ring is finitely generated.

3.6.21. Why you should not worry about Noetherian hypotheses. Should you
work hard to eliminate Noetherian hypotheses? Should you worry about Noe-
therian hypotheses? Should you stay up at night thinking about non-Noetherian
rings? For the most part, the answer to these questions is “no”. Most people
will never need to worry about non-Noetherian rings, but there are reasons to be
open to them. First, they can actually come up. For example, fibered products of
Noetherian schemes over Noetherian schemes (and even fibered products of Noe-
therian points over Noetherian points!) can be non-Noetherian (Warning 10.1.5),
and the normalization of Noetherian rings can be non-Noetherian (Warning 10.7.4).
You can either work hard to show that the rings or schemes you care about don’t
have this pathology, or you can just relax and not worry about it. Second, there is
often no harm in working with schemes in general. Knowing when Noetherian
conditions are needed will help you remember why results are true, because you
will have some sense of where Noetherian conditions enter into arguments. Finally,
for some people, non-Noetherian rings naturally come up. For example, adeles are
not Noetherian. And many valuation rings that naturally arise in arithmetic and
tropical geometry are not Noetherian.

3.7 The function I(·), taking subsets of Spec A to ideals of A

We now introduce a notion that is in some sense “inverse” to the vanishing set
function V(·). Given a subset
T S ⊂ Spec A, I(S) is the set of functions vanishing on S.
In other words, I(S) = [p]∈S p ⊂ A (at least when S is nonempty).
We make three quick observations. (Do you see why they are true?)
• I(S) is clearly an ideal of A.
• I(·) is inclusion-reversing: if S1 ⊂ S2 , then I(S2 ) ⊂ I(S1 ).
• I(S) = I(S).

3.7.A. E XERCISE . Let A = k[x, y]. If S = {[(y)], [(x, y − 1)]} (see Figure 3.10), then
I(S) consists of those polynomials vanishing on the x-axis, and at the point (0, 1).
Give generators for this ideal.
128 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

F IGURE 3.10. The set S of Exercise 3.7.A, pictured as a subset of A2

3.7.B. E XERCISE . Suppose S ⊂ A3C is the union of the three axes. Give generators
for the ideal I(S). Be sure to prove it! We will see in Exercise 13.1.F that this ideal is
not generated by less than three elements.

3.7.C. E XERCISE . Show that V(I(S)) = S. Hence V(I(S)) = S for a closed set S.
p
Note that I(S) is always a radical ideal — if f ∈ I(S), then fn vanishes on S
for some n > 0, so then f vanishes on S, so f ∈ I(S).
√
3.7.D. E ASY E XERCISE . Prove that if J ⊂ A is an ideal, then I(V(J)) = J. (Huge
hint: Exercise 3.4.J.)
Exercises 3.7.C and 3.7.D show that V and I are “almost” inverse. More pre-
cisely:

3.7.1. Theorem. — V(·) and I(·) give an inclusion-reversing bijection between closed
subsets of Spec A and radical ideals of A (where a closed subset gives a radical ideal by I(·),
and a radical ideal gives a closed subset by V(·)).
Theorem 3.7.1 is sometimes called Hilbert’s Nullstellensatz, but we reserve that
name for Theorem 3.2.6.

3.7.E. E XERCISE . Let J = (x2 + y2 − 1, y − 1). Find, with proof, an element of

I(V(J)) \ J.

3.7.F. I MPORTANT E XERCISE ( CF. E XERCISE 3.7.G). Show that V(·) and I(·) give a
bijection between irreducible closed subsets of Spec A and prime ideals of A. From this
conclude that in Spec A there is a bijection between points of Spec A and irreducible
closed subsets of Spec A (where a point determines an irreducible closed subset
by taking the closure). Hence each irreducible closed subset of Spec A has precisely one
generic point — any irreducible closed subset Z can be written uniquely as {z}.

3.7.G. E XERCISE / DEFINITION . A prime ideal of a ring A is a minimal prime ideal

(or more simply, minimal prime) if it is minimal with respect to inclusion. (For
example, the only minimal prime ideal of k[x, y] is (0).) If A is any ring, show that
the irreducible components of Spec A are in bijection with the minimal prime ideals
of A. In particular, Spec A is irreducible if and only if A has only one minimal
prime ideal; this generalizes Exercise 3.6.C.
Proposition 3.6.15, Exercise 3.6.T, and Exercise 3.7.G imply that every Noether-
ian ring has a finite number of minimal prime ideals. An algebraic fact is now
revealed to be really a “geometric” fact!
 © 2024 Ravi Vakil. Published by Princeton University Press. 129

3.7.H. E XERCISE ( REALITY CHECK ). In An k = Spec k[x1 , . . . , xn ], the subset cut

out by f(x1 , . . . , xn ) ∈ k[x1 , . . . , xn ] should certainly have irreducible components
corresponding to the distinct irreducible factors of f. Prove this. (What property of
the ring k[x1 , . . . , xn ] makes this work?)

3.7.I. E XERCISE . What are the minimal prime ideals of k[x, y]/(xy) (where k is a
field)?

3.7.2. Beginning of a grand dictionary between algebra and geometry. We

are now well on our away to building a grand dictionary between algebra and
geometry. You should add more entries to the table as your understanding deepens.

algebra geometry reference

ring A affine scheme Spec A §3.2.1
p = I(p) prime ideal of A p = [p] point of Spec A §3.2.1, §3.7
element f ∈ A function f on Spec A §3.2.1
f (mod p) f(p); value of the function f at p §3.2.1
f∈p p the function f vanishes (is zero) at p §3.2.1
nilradical ideal N = (0) = ∩p of A functions vanishing at every point Thm. 3.2.13
maximal ideal of A irreducible closed subset of Spec A
that is a point
(= closed point of Spec A) Ex. 3.6.I
p = I(Z) prime ideal of A irreducible closed subset Z = V(p)
(a repeat!) of Spec A Ex. 3.7.F
minimal prime ideal
√ of A irreducible component of Spec A Ex. 3.7.G
radical ideal I = I = I(S) of A closed subset S = V(I) of Spec A Thm. 3.7.1
 CHAPTER 4

The structure sheaf, and the definition of schemes in

general

The final ingredient in the definition of an affine scheme is the structure sheaf
OSpec A , which we think of as the “sheaf of algebraic functions”. You should keep
in your mind the example of “algebraic functions” on Cn , which you understand
well. For example, for the plane A2 , we expect that on the open set D(xy) (away
from the two axes), (3x4 + y + 4)/x7 y3 should be an algebraic function.
These functions will have values at points, but won’t be determined by their
values at points. But like all sections of sheaves, they will be determined by their
germs (see §4.3.6).

4.1 The structure sheaf of an affine scheme

We want now to give a ring of functions on any open set, but we know that
it suffices to describe the functions on a base of the topology. We will see that the
base of distinguished open sets will make things much more tractable than you
might fear. So we now describe the structure sheaf as a sheaf (of rings) on the base
of distinguished open sets (Theorem 2.5.1 and Exercise 3.5.A).

4.1.1. Definition. Define OSpec A (D(f)) to be the localization of A at the multiplica-

tive set of all functions that do not vanish outside of V(f) (i.e., those g ∈ A such
that V(g) ⊂ V(f), or equivalently D(f) ⊂ D(g), cf. Exercise 3.5.E). This depends
only on D(f), and not on f itself. (Scholars of the empty set might notice that by
Exercise 3.5.F, we have that OSpec A (∅) = {0}.)

4.1.A. G REAT E XERCISE . Show that the natural map Af → OSpec A (D(f)) is an
isomorphism. (Possible hint: Exercise 3.5.E.)
If D(f ′ ) ⊂ D(f), define the restriction map
resD(f),D(f ′ ) : OSpec A (D(f)) / OSpec A (D(f ′ ))

in the obvious way: the latter ring is a further localization of the former ring. The
restriction maps obviously compose: this is a “presheaf on the distinguished base”.

4.1.2. Theorem. — The data just described give a sheaf on the distinguished base, and
hence determine a sheaf on the topological space Spec A.

This material is published by Princeton University Press as The Rising Sea: Foundations of Algebraic
Geometry by Ravi Vakil. This version is free to view and download for personal use only. Not for
redistribution, resale, or use in derivative works. © 2024 Ravi Vakil.

131
132 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

4.1.3. This sheaf is called the structure sheaf, and will be denoted OSpec A , or some-
times O if the subscript is clear from the context. The notation Spec A will hereafter
denote the data of a topological space with a structure sheaf. Such a topological
space, with structure sheaf, will be called an affine scheme (Definition 4.3.1). We
continue to use the language of ringed spaces (§2.2.13): functions on open sets,
global functions, and so forth. An important lesson of Theorem 4.1.2 is not just
that OSpec A is a sheaf, but also that the distinguished base provides a good way of
working with OSpec A . Notice also that we have justified interpreting elements of A
as functions on Spec A.

Proof. We must show the base identity and base gluability axioms hold (§2.5). We
show that they both hold for the open set that is the entire space Spec A, and leave
to you the trick which extends them to arbitrary distinguished open sets (Exer-
cises 4.1.B and 4.1.C). Suppose Spec A = ∪i∈I D(fi ), or equivalently (Exercise 3.5.B)
the ideal generated by the fi is the entire ring A.
(Aside: Experts familiar with the equalizer exact sequence of §2.2.7 will realize
that we are showing exactness of
(4.1.3.1) 0 /A / Q Af /Q
i∈I i i̸=j∈I Afi fj

where {fi }i∈I is a set of functions with (fi )i∈I = A. Signs are involved in the right-
hand map: the map Afi → Afi fj is the localization map, and the map Afj → Afi fj
is the negative of the localization map. BaseQ identity corresponds to injectivity at A,
and gluability corresponds to exactness at i Afi .)
We check identity on the base. Suppose that Spec A = ∪i∈I D(fi ) where i
runs over some index set I. Then there is some finite subset of I, which we name
{1, . . . , n}, such that Spec A = ∪n i=1 D(fi ), i.e., (f1 , . . . , fn ) = A (quasicompactness
of Spec A, Exercise 3.5.C). Suppose we are given s ∈ A such that resSpec A,D(fi ) s = 0
in Afi for all i. We wish to show that s = 0. The fact that resSpec A,D(fi ) s = 0 in
Afi implies that there is some m such that for each i ∈ {1, . . . , n}, fm i s = 0. Now
(fm
1 , . . . , fm
)
n P = A (for example, from Spec A = ∪D(f i ) = ∪D(fm
i )), so there are
n
ri ∈ A with i=1 ri fm i = 1 in A, from which
X  X
s= ri fm
i s= ri (fmi s) = 0.

Thus we have checked the “base identity” axiom for Spec A.

4.1.B. E XERCISE . Make tiny changes to the above argument to show base identity
for any distinguished open D(f). (Hint: judiciously replace A by Af in the above
argument.)
We next show base gluability. (Serre has described this as a “partition of unity”
argument, and if you look at it in the right way, his insight is very enlightening.)
Suppose again ∪i∈I D(fi ) = Spec A, where I is an index set (possibly horribly
infinite). Suppose we are given elements in each Afi that agree on the overlaps
Afi fj . Note that intersections of distinguished open sets are also distinguished
open sets.
Assume first that I is finite, say I = {1, . . . , n}. We have elements ai /flii ∈ Afi
agreeing on overlaps Afi fj (see Figure 4.1(a)). Letting gi = flii , using D(fi ) = D(gi ),
we can simplify notation by considering our elements as of the form ai /gi ∈ Agi
(Figure 4.1(b)).
 © 2024 Ravi Vakil. Published by Princeton University Press. 133

F IGURE 4.1. Base gluability of the structure sheaf

The fact that ai /gi and aj /gj “agree on the overlap” (i.e., in Agi gj ) means that
for some mij ,
(gi gj )mij (gj ai − gi aj ) = 0
in A. By taking m = max mij (here we use the finiteness of I), we can simplify
notation:
(gi gj )m (gj ai − gi aj ) = 0
m+1
for all i, j (Figure 4.1(c)). Let bi = ai gm
i for all i, and hi = gi (so D(hi ) = D(gi )).
Then we can simplify notation even more (Figure 4.1(d)): on each D(hi ), we have a
function bi /hi , and the overlap condition is
(4.1.3.2) h j bi = h i bj .
Pn
Now ∪i D(hi ) = Spec A, implying that 1 = i=1 ri hi for some ri ∈ A. Define
X
(4.1.3.3) r= r i bi .

This will be the element of A that restricts to each bj /hj . Indeed, from the overlap
condition (4.1.3.2),
X X
rhj = ri b i h j = r i h i bj = bj .
i i
We next deal with the case where I is infinite. Choose a finite subset {1, . . . , n} ⊂
I with (f1 , . . . , fn ) = A (or equivalently, use quasicompactness of Spec A to choose
a finite subcover by D(fi )). Construct r as above, using (4.1.3.3). We will show that
for any z ∈ I − {1, . . . , n}, r restricts to the desired element az /flzz of Afz . Repeat the
entire process above with {1, . . . , n, z} in place of {1, . . . , n}, to obtain r ′ ∈ A which
restricts to ai /flii for i ∈ {1, . . . , n, z}. Then by the base identity axiom, r ′ = r. (Note
that we use base identity to prove base gluability. This is an example of how the
identity axiom is somehow “prior” to the gluability axiom.) Hence r restricts to the
desired element az /flzz of Afz .
134 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

4.1.C. E XERCISE . Alter this argument appropriately to show base gluability for any
distinguished open D(f).
We have now completed the proof of Theorem 4.1.2. □

4.1.4. ⋆⋆ Remark. Definition 4.1.1 and Theorem 4.1.2 suggests a potentially slick
way of describing sections of OSpec A over any open subset: perhaps OSpec A (U) is
the localization of A at the multiplicative set of all functions that do not vanish at
any point of U. This is not true. A counterexample (that you will later be able to
make precise): let Spec A be two copies of A2k glued together at their origins (see
(24.3.8.1) for explicit equations) and let U be the complement of the origin(s). Then
the function which is 1 on the first copy of A2k \ {(0, 0)} and 0 on the second copy of
A2k \ {(0, 0)} is not of this form. (Follow-up question: why would this discussion
not work for two copies of A1k glued at a point?)

4.1.5. Important notion for future use: OSpec A -modules coming from A-modules.
The following generalization of Theorem 4.1.2 will be essential in the definition of
a quasicoherent sheaf in Chapter 6. You can leave these exercises for then, but they
may be easiest to solve now, while the ideas are fresh in your mind.

4.1.D. I MPORTANT E XERCISE /D EFINITION . Suppose M is an A-module. Show that

the following construction describes a sheaf M f on the distinguished base. Define
M(D(f))
f to be the localization of M at the multiplicative set of all functions that do
not vanish outside of V(f). Define restriction maps resD(f),D(g) in the analogous
way to OSpec A . Show that this defines a sheaf on the distinguished base, and hence
a sheaf on Spec A. Then show that this is an OSpec A -module.

4.1.6. Important Remark. In the course of answering the previous exercise, you will
show that if (fi )i∈I = A,

(4.1.6.1) 0 /M /Q Mf i /Q Mfi fj
i∈I i̸=j∈I

(cf. (4.1.3.1)) is exact. In particular, M can be identified with a specific submodule

of Mf1 × · · · × Mfr . Even though M → Mfi may not be an inclusion for any fi ,
M → Mf1 × · · · × Mfr is an inclusion. This will be useful later: we will want to
show that if M has some nice property, then Mf does too, which will be easy. We
will also want to show that if (f1 , . . . , fn ) = A, and the Mfi have this property,
then M does too. This idea will be made precise in the Affine Communication
Lemma 5.3.2.

4.1.E. E XERCISE . Suppose p is a prime ideal of A. Describe a canonical isomorphism

f[p] −→ ∼
M Mp between the stalk of M f and the localization of M. In particular, the
stalk of OSpec A at the point [p] is the local ring Ap .
The following exercise shows how an important result can be understood quite
differently from algebraic and geometric perspectives.

4.1.F. E XERCISE . Suppose M is an A-module. Show that the natural map

 /Q
M p∈Spec A Mp
 © 2024 Ravi Vakil. Published by Princeton University Press. 135

is an injection in two ways: (a) by considering the kernel of the map, and show that
any element of it must be 0, and (b) by applying Exercise 2.4.A (sections of a sheaf
are determined by germs) to the sheaf M. f (Thus an A-module is zero if and only if
all its localizations at prime ideals are zero, and we see this as a simultaneously (a)
algebraic and (b) geometric fact.)

4.1.G. E XERCISE . Describe a bijection between maps of A-modules M → N, and

maps of OSpec A -modules Mf → N. e (Fancy translation: you will have described
ModA as a full subcategory of ModOSpec A .)

4.1.7. Definition: Support of a module. Motivated by Exercise 4.1.E, and the notion of
support of a section of a sheaf (Definition 2.7.6), define the support of m ∈ M by
Supp m := {[p] ∈ Spec A : mp ̸= 0} ⊂ Spec A,
which is the support of m considered as a section of M.f By Exercise 2.7.G, Supp m
is a closed subset of Spec A.
Similarly (following Definition 2.7.8), define the support of M by
f = {[p] ∈ Spec A : Mp ̸= 0} ⊂ Spec A.
Supp M := Supp M
These notions will come up repeatedly. We will discuss support in more detail in
§6.6.3.

4.1.8. Recurring (counter)examples. The example of two planes meeting at a

point (§4.1.4) will appear many times as an example or a counterexample. It is
important to have such examples (or counterexamples) at the back of your mind, as
“boundary markers” reminding you of what is reasonable, and what to watch out
for when you hear a new statement. The appearances of the following are listed in
the index (under “recurring (counter)examples”), so you can see how often they
come up, and where. You do not have the language to understand many of these
yet, but you can come back to this list later.
• two planes meeting at a point
• affine space minus the origin, and its inclusion in affine space
• affine space with doubled origin
• (the cone over) the quadric surface, A = k[w, x, y, z]/(wz − xy)
• an embedded point on a line, k[x, y]/(y2 , xy)
• x2 = 0 in the projective plane `
• infinite disjoint unions of schemes (especially Spec k[x]/(xn ))
• the morphism Spec Q → Spec Q
• k(x)
Q∞ ⊗k k(y)
• F2

4.2 Visualizing schemes: Nilpotents

The price of metaphor is eternal vigilance.
— A. Rosenbluth and N. Wiener (attribution by [Lew, p. 4])
In §3.3, we discussed how to visualize the underlying set of schemes, adding in
generic points to our previous intuition of “classical” (or closed) points. Our later
136 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

discussion of the Zariski topology fit well with that picture. In our definition of the
“affine scheme” (Spec A, OSpec A ), we have the additional information of nilpotents,
which are invisible on the level of points (§3.2.12), so now we figure out how to
picture them. We will then readily be able to glue them together to picture schemes
in general, once we have made the appropriate definitions. As we are building
intuition, we cannot be rigorous or precise.
As motivation, note that we have inclusion-reversing bijections

maximal ideals of A o / closed points of Spec A (Exercise 3.6.I)

prime ideals of A o / irreducible closed subsets of Spec A (Exercise 3.7.F)

radical ideals of A o / closed subsets of Spec A (Theorem 3.7.1)

If we take the things on the right as “pictures”, our goal is to figure out how to
picture ideals that are not radical:

ideals of A o / ???

(We will later fill this in rigorously in a different way with the notion of a closed
subscheme, the scheme-theoretic version of closed subsets, §9.1. But our goal now is
to create a picture.)
As motivation, when we see the expression, Spec C[x]/(x(x − 1)(x − 2)), we
immediately interpret it as a closed subset of A1C , namely {0, 1, 2}. In particular, the
map C[x] → C[x]/(x(x − 1)(x − 2)) can be interpreted (via the Chinese Remainder
Theorem) as: take a function on A1 , and restrict it to the three points 0, 1, and 2.
This will guide us in how to visualize a nonradical ideal. The simplest example
to consider is Spec C[x]/(x2 ) (Exercise 3.2.A(a)). As a subset of A1 , it is just the origin
0 = [(x)], which we are used to thinking of as Spec C[x]/(x) (i.e., corresponding to
the ideal (x), not (x2 )). We want to enrich this picture in some way. We should
picture C[x]/(x2 ) in terms of the information the quotient remembers. The image
of a polynomial f(x) is the information of its value at 0, and its derivative (cf.
Exercise 3.2.T). We thus picture this as being the point, plus a little bit more —
a little bit of infinitesimal “fuzz” on the point (see Figure 4.2). The sequence of
restrictions C[x] → C[x]/(x2 ) → C[x]/(x) should be interpreted as nested pictures.

C[x] / / C[x]/(x2 ) / / C[x]/(x)

f(x)  / f(0),

Similarly, C[x]/(x3 ) remembers even more information — the second derivative as

well. Thus we picture this as the point 0 with even more fuzz.
More subtleties arise in two dimensions (see Figure 4.3). Consider

Spec C[x, y]/(x, y)2 ,

 © 2024 Ravi Vakil. Published by Princeton University Press. 137

F IGURE 4.2. Picturing quotients of C[x]

which is sandwiched between two rings we know well:

C[x, y] / / C[x, y]/(x, y)2 / / C[x, y]/(x, y)

f(x, y)  / f(0, 0).

Again, taking the quotient by (x, y)2 remembers the first derivative, “in all direc-
tions”. We picture this as fuzz around the point, in the shape of a circle (no direction
is privileged). Similarly, (x, y)3 remembers the second derivative “in all directions”
— bigger circular fuzz.

F IGURE 4.3. Picturing quotients of C[x, y]

Consider instead the ideal (x2 , y). What it remembers is the derivative only
in the x direction — given a polynomial, we remember its value at 0, and the
coefficient of x. We remember this by picturing the fuzz only in the x direction.
This gives us some handle on picturing more things of this sort, but now it
becomes more an art than a science. For example, Spec C[x, y]/(x2 , y2 ) we might
picture as a fuzzy square around the origin. (Could you believe that this square is
circumscribed by the circular fuzz Spec C[x, y]/(x, y)3 , and inscribed by the circular
fuzz Spec C[x, y]/(x, y)2 ?) One feature of this example is that given two ideals I
and J of a ring A (such as C[x, y]), your fuzzy picture of Spec A/(I, J) should be the
“intersection” of your picture of Spec A/I and Spec A/J in Spec A. (You will make
this precise in Exercise 9.1.I(a).) For example, Spec C[x, y]/(x2 , y2 ) should be the
intersection of two thickened lines. (How would you picture Spec C[x, y]/(x5 , y3 )?
Spec C[x, y, z]/(x3 , y4 , z5 , (x + y + z)2 )? Spec C[x, y]/((x, y)5 , y3 )?)
138 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

One final example that will motivate us in §6.6 is Spec C[x, y]/(y2 , xy). Know-
ing what a polynomial in C[x, y] is modulo (y2 , xy) is the same as knowing its value
on the x-axis, as well as first-order differential information around the origin. This
is worth thinking through carefully: do you see how this information is captured
(however imperfectly) in Figure 4.4?

F IGURE 4.4. A picture of the scheme Spec k[x, y]/(y2 , xy). The
fuzz at the origin indicates where “there is nonreducedness”.

Our pictures capture useful information that you already have some intuition
for. For example, consider the intersection of the parabola y = x2 and the x-axis
(in the xy-plane), see Figure 4.5. You already have a sense that the intersection has
multiplicity two. In terms of this visualization, we interpret this as intersecting (in
Spec C[x, y]):
Spec C[x, y]/(y − x2 ) ∩ Spec C[x, y]/(y) = Spec C[x, y]/(y − x2 , y)
= Spec C[x, y]/(y, x2 )
which we interpret as the fact that the parabola and line not just meet with multi-
plicity two, but that the “multiplicity 2” part is in the direction of the x-axis. You
will make this example precise in Exercise 9.1.I(b).

F IGURE 4.5. The “scheme-theoretic” intersection of the parabola

y = x2 and the x-axis is a nonreduced scheme (with fuzz in the
x-direction)

4.2.1. We will later make the location of the fuzz more precise when we discuss asso-
ciated points. We will see that in reasonable circumstances, the fuzz is concentrated
on closed subsets (Exercise 6.6.N).
On a bien souvent répété que la Géométrie est l’art de bien raisonner sur des figures
mal faites; encore ces figures, pour ne pas nous tromper, doivent-elles satisfaire à certaines
conditions; les proportions peuvent être grossièrement altérées, mais les positions relatives
des diverses parties ne doivent pas être bouleversées.
It is often said that geometry is the art of reasoning well from badly made figures;
however, these figures, if they are not to deceive us, must satisfy certain conditions; the
proportions may be grossly altered, but the relative positions of the different parts must not
be upset.
— H. Poincaré, [Po1, p. 2] (see J. Stillwell’s translation [Po2, p. ix])
 © 2024 Ravi Vakil. Published by Princeton University Press. 139

4.3 Definition of schemes

4.3.1. Definitions. We can now define scheme in general. First, define an isomor-
phism of ringed spaces (X, OX ) → (Y, OY ) as (i) a homeomorphism π : X → Y, and
(ii) an isomorphism of sheaves OX and OY , considered to be on the same space
∼
via π. (Part (ii), more precisely, is an isomorphism OY −→ π∗ OX of sheaves on
∼
Y, or equivalently, an isomorphism π−1 OY −→ OX of sheaves on X.) In other
words, we have a “correspondence” of sets, topologies, and structure sheaves.
Every isomorphism α : (X, OX ) → (Y, OY ) clearly has an “inverse isomorphism”
α−1 : (Y, OY ) → (X, OX ).
An affine scheme is a ringed space that is isomorphic to (Spec A, OSpec A ) for
some A. A scheme (X, OX ) is a ringed space such that any point of X has an open
neighborhood U such that (U, OX |U ) is an affine scheme. The topology on a scheme
is called the Zariski topology. The scheme can be denoted (X, OX ), although it is
often denoted X, with the structure sheaf OX left implicit.
An isomorphism of schemes (X, OX ) → (Y, OY ) is an isomorphism as ringed
spaces. Recall the definition of Γ (·, ·) in §2.2.2. If U ⊂ X is an open subset, then
the elements of Γ (U, OX ) are said to be the functions on U; this generalizes the
definition of functions on an affine scheme, §3.2.1.

4.3.2. Remark. From the definition of the structure sheaf on an affine scheme,
several things are clear. First of all, if we are told that (X, OX ) is an affine scheme,
we may recover its ring (i.e., find the ring A such that Spec A = X) by taking the
ring of global sections, as X = D(1), so:

Γ (X, OX ) = Γ (D(1), OSpec A ) as D(1) = Spec A

= A.

(You can verify that we get more, and can “recognize X as the scheme Spec A”: we
∼
get an isomorphism π : (Spec Γ (X, OX ), OSpec Γ (X,OX ) ) −→ (X, OX ). For example, if
m is a maximal ideal of Γ (X, OX ), then {π([m])} = V(m).) The following exercise will
make these ideas rigorous — they are subtler than they first appear.

4.3.3. Caution. It is not part of the definition that the overlap of two affine open
subsets is also affine (see Exercise 4.4.C for an example), although this is true in
most cases of interest (see Proposition 11.2.13).

4.3.A. E NLIGHTENING E XERCISE ( WHICH CAN BE STRANGELY CONFUSING ). De-

∼
scribe a bijection between the isomorphisms Spec A −→ Spec A ′ (of ringed spaces)
∼
and the ring isomorphisms A ′ −→ A. Hint: the hardest part is to show that if an
∼ ∼
isomorphism π : Spec A −→ Spec A ′ induces an isomorphism π♯ : A ′ −→ A, which
∼
in turn induces an isomorphism ρ : Spec A −→ Spec A ′ , then π = ρ. First show this
on the level of points; this is (surprisingly) the trickiest part. Then show π = ρ as
maps of topological spaces. Finally, to show π = ρ on the level of structure sheaves,
use the distinguished base. Feel free to use insights from later in this section, but be
careful to avoid circular arguments. Even struggling with this exercise and failing
(until reading later sections) will be helpful.
140 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

More generally, given f ∈ A, Γ (D(f), OSpec A ) = ∼ Af . Thus under the natural

inclusion of sets Spec Af ,→ Spec A, the Zariski topology on Spec A restricts to give
the Zariski topology on Spec Af (Exercise 3.4.I), and the structure sheaf of Spec A
restricts to the structure sheaf of Spec Af , as the next exercise shows.

4.3.B. I MPORTANT BUT EASY EXERCISE . Suppose f ∈ A. Show that under the
identification of D(f) in Spec A with Spec Af (§3.5), there is a natural isomorphism
∼
of ringed spaces (D(f), OSpec A |D(f) ) ←→ (Spec Af , OSpec Af ). Hint: notice that
distinguished open sets of Spec Af are already distinguished open sets in Spec A.

4.3.C. E ASY E XERCISE . If X is a scheme, and U is any open subset, prove that
(U, OX |U ) is also a scheme.

4.3.4. Definitions. We say (U, OX |U ) is an open subscheme of X. If U is also an affine

scheme, we often say U is an affine open subset, or an affine open subscheme,
or sometimes informally just an affine open. For example, D(f) is an affine open
subscheme of Spec A. (Fun fact: it is not true that every affine open subscheme of
Spec A is of the form D(f); see §19.11.10 for an example.)

4.3.D. E ASY E XERCISE . Show that if X is a scheme, then the affine open sets form a
base for the Zariski topology.

4.3.E. E ASY E XERCISE . The disjoint union of schemes is defined as you would
expect: it is the disjoint union of sets, with the expected topology (thus it is the
disjoint
` union of topological spaces), with the expected sheaf. We use the symbol
X Y for the disjoint union of X and Y, because once we know what morphisms of
schemes are, this construction will turn out to be the coproduct in the category of
schemes (Exercise 7.3.K).
(a) Show that the disjoint union of a finite number of affine schemes is also an affine
scheme. (Hint: Exercise 3.6.A.)
(b) (a first example of a non-affine scheme) Show that an infinite disjoint union of
(nonempty) affine schemes is not an affine scheme. (Hint: affine schemes are
quasicompact, Exercise 3.6.G(a). This is basically answered in Remark 3.6.6.)

4.3.5. Remark: A first glimpse of closed subschemes. Open subsets of a scheme come
with a natural scheme structure (Definition 4.3.4). For comparison, closed subsets
can have many “natural” scheme structures. We will discuss this later (in §9.1), but
for now, it suffices for you to know that a closed subscheme of X is, informally, a
particular kind of scheme structure on a closed subset of X. As an example: if I ⊂ A
is an ideal, then Spec A/I endows the closed subset V(I) ⊂ Spec A with a scheme
structure; but note that there can be different ideals with the same vanishing set
(for example (x) and (x2 ) in k[x]).

4.3.6. Stalks of the structure sheaf: germs, values at a point, and the residue field
of a point. Like every sheaf, the structure sheaf has stalks, and unsurprisingly,
they are interesting from an algebraic point of view. In fact, we have seen them
before.

4.3.7. Definition. We say a ringed space is a locally ringed space if its stalks are
local rings. By Exercise 4.1.E, the stalk of OSpec A at the point [p] is the local ring Ap .
 © 2024 Ravi Vakil. Published by Princeton University Press. 141

Hence schemes are locally ringed spaces. Manifolds are another example of locally
ringed spaces, see §2.1.1. In both cases, taking the quotient by the maximal ideal
may be interpreted as evaluating at the point. The maximal ideal of the local ring
OX,p is denoted mX,p or mp , and the residue field OX,p /mp is denoted κ(p). Each
function f on an open subset U of a locally ringed space X has a value at each point
p of U: the value of f at p (denoted f(p)) is the image of the germ of f at p under
the canonical map OX,p → κ(p). We say that a function vanishes at a point p if its
value at p is 0. (This generalizes our notion of the value of a function on Spec A,
defined in §3.2.1.) Notice that we can’t even make sense of the phrase “function
vanishing” on ringed spaces in general.

4.3.F. U SEFUL E XERCISE .

(a) If f is a function on a locally ringed space X, show that the subset of X where f
vanishes is closed. Hint: if f is a function on a ringed space X, show that the subset of
X where the germ of f is invertible is open. (Thus the locus on a locally ringed space
where a function has nonzero value is open. In contrast, the locus it has nonzero
germ is closed, as mentioned in Caution 2.7.7.)
(b) Show that if f is a function on a locally ringed space that vanishes nowhere, then
f is invertible.
Consider a point [p] of an affine scheme Spec A. (Of course, any point of a
scheme can be interpreted in this way, as each point has an affine open neigh-
borhood.) The residue field at [p] is Ap /pAp , which is isomorphic to K(A/p), the
fraction field of the quotient. It is useful to note that localization at p and taking the
quotient by p “commute”, i.e., the following diagram commutes.

(4.3.7.1) Ap
>
localize quotient

$
A Ap /pAp K(A/p)
:
quotient localize, i.e., K(·)

A/p

For example, consider the scheme A2k = Spec k[x, y], where k is a field of
characteristic not 2. Then (x2 + y2 )/x(y2 − x5 ) is a function away from the y-axis
and the curve y2 − x5 . Its value at (2, 4) (by which we mean [(x − 2, y − 4)]) is
(22 + 42 )/(2(42 − 25 )), as
x 2 + y2 22 + 42
2 5
≡
x(y − x ) 2(42 − 25 )
in the residue field — check this if it seems mysterious. And its value at [(y)], the
x2 4
generic point of the x-axis, is −x6 = −1/x , which we see by setting y to 0. This is

indeed an element of the fraction field of k[x, y]/(y), i.e., k(x). (If you think you care
only about algebraically closed fields, let this example be a first warning: Ap /pAp
won’t be algebraically closed in general, even if A is a finitely generated C-algebra!)
If anything makes you nervous, you should make up an example to make you
feel better. Here is one: 27/4 is a function on Spec Z \ {[(2)], [(7)]} or indeed on an
even bigger open set. What is its value at [(5)]? Answer: 2/(−1) ≡ −2 (mod 5).
142 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

What is its value at the generic point [(0)]? Answer: 27/4. Where does it vanish? At
[(3)].

4.3.8. Definition: the fiber and rank of an OX -module at a point. If F is an OX -module

on a scheme X (or more generally, a locally ringed space), define the fiber (or fibre)
of F at a point p ∈ X by
F |p := Fp ⊗OX,p κ(p).
The fiber is a vector space over κ(p). The dimension of this vector space dimκ(p) F |p
is called the rank of F at p, and is denoted rankp F . For example, OX |p is κ(p),
and rankp OX = 1.

4.3.9. ⋆ Definition: Manifolds of various sorts. In the same way, we can cleanly
and concisely define (various notions of) manifolds. A topological (respectively,
differentiable, C∞ = smooth) (real) manifold is a ringed space (X, OX ), whose
underlying topological space X is Hausdorff and second countable, such that any
point of X has an open neighborhood U such that (U, OX |U ) is isomorphic to an
open ball in Rn with the sheaf of continuous (respectively, differentiable, C∞ =
smooth) real-valued functions on it. The phrase manifold (without adjective) often
means smooth manifold. A complex manifold is a ringed space (X, OX ), whose
underlying topological space X is Hausdorff and second countable, such that any
point of X has an open neighborhood U such that (U, OX |U ) is isomorphic to an
open ball in Cn with the sheaf of holomorphic functions on it.

4.4 Three examples

We now give three extended examples. Our short-term goal is to see that we
can really work with the structure sheaf, and can compute the ring of sections of
interesting open sets that aren’t just distinguished open sets of affine schemes. Our
long-term goal is to meet interesting examples that will come up repeatedly in the
future.

4.4.1. First example: The (affine) plane minus the origin. This example will
show you that the distinguished base is something that you can work with. Let
A = k[x, y], so Spec A = A2k . Let’s work out the space of functions on the open set
U = A2 \ {(0, 0)} = A2 \ {[(x, y)]}.
It is not immediately obvious whether this is a distinguished open set. (In fact
it is not — you may be able to figure out why within a few paragraphs, if you can’t
right now. It is not enough to show that (x, y) is not a principal ideal.) But in any
case, we can describe it as the union of two things which are distinguished open
sets: U = D(x) ∪ D(y). We will find the functions on U by gluing together functions
on D(x) and D(y).
The functions on D(x) are, by Definition 4.1.1, Ax = k[x, y, 1/x]. The functions
on D(y) are Ay = k[x, y, 1/y]. Note that A injects into its localizations (if 0 is not
inverted), as it is an integral domain (Exercise 1.2.C), so A injects into both Ax
and Ay , and both inject into Axy (and indeed into k(x, y) = K(A)). So we are
looking for functions on D(x) and D(y) that agree on D(x) ∩ D(y) = D(xy), i.e.,
we are interpreting Ax ∩ Ay in Axy (or in k(x, y)). Clearly those rational functions
with only powers of x in the denominator, and also with only powers of y in the
 © 2024 Ravi Vakil. Published by Princeton University Press. 143

denominator, are the polynomials. Translation: Ax ∩ Ay = A. Thus we conclude:

(4.4.1.1) Γ (U, OA2 ) ≡ k[x, y].
In other words, we get no extra functions by removing the origin. Notice how easy
that was to calculate!

4.4.2. Aside. Notice that any function on A2 \ {(0, 0)} extends over all of A2 . This is
an analog of Hartogs’s Lemma in complex geometry: you can extend a holomorphic
function defined on the complement of a set of codimension at least two on a
complex manifold over the missing set. This will work more generally in the
algebraic setting: you can extend over points in codimension at least 2 not only if
they are “smooth”, but also if they are mildly singular — what we will call normal.
We will make this precise in §13.5.19. This fact will be very useful for us.

4.4.3. We now show an interesting fact: (U, OA2 |U ) is a scheme, but it is not an
affine scheme. (This is confusing, so you will have to pay attention.) Here’s why:
∼ (Spec A ′ , OSpec A ′ ), then we can recover A ′ by taking
otherwise, if (U, OA2 |U ) =
global sections:
A ′ = Γ (U, OA2 |U ),
which we have already identified in (4.4.1.1) as k[x, y]. So if U is affine, then we
have the isomorphism
∼
Uo / Spec A ′ = Spec k[x, y] = A2 .
k

But this bijection between prime ideals in a ring and points of the spectrum is more
constructive than that: given the prime ideal I, you can recover the point as the generic
point of the closed subset cut out by I, i.e., V(I), and given the point p, you can recover the
ideal as those functions vanishing at p, i.e., I(p). In particular, the prime ideal (x, y) of
A ′ should cut out a point of Spec A ′ . But on U, V(x) ∩ V(y) = ∅. Conclusion: U is
not an affine scheme. (If you are ever looking for a counterexample to something,
and you are expecting one involving a non-affine scheme, keep this example in
mind!)

4.4.4. Gluing two copies of A1 together in two different ways. We have now
seen two examples of non-affine schemes: an infinite disjoint union of nonempty
schemes (Exercise 4.3.E) and A2 \ {(0, 0)}. I want to give you two more examples.
They are important because they are the first examples of fundamental behavior,
the first pathological, and the second central.
First, I need to tell you how to glue two schemes together. Before that, you
should review how to glue topological spaces together along isomorphic open sets.
Given two topological spaces X and Y, and open subsets U ⊂ X and V ⊂ Y along
∼ /
with a homeomorphism U o V , we can create a new topological space W, that
∼ /
we think of as gluing
` X and Y together along U o V . It is the quotient of the
disjoint union X Y by the equivalence relation U ∼ V, where the quotient is given
the quotient topology. Then X and Y are naturally (identified with) open subsets
of W, and indeed cover W. Can you restate this cleanly with an arbitrary (not
necessarily finite) number of topological spaces?
Now that we have discussed gluing topological spaces, let’s glue schemes
together. (This applies without change more generally to ringed spaces.) Suppose
you have two schemes (X, OX ) and (Y, OY ), and open subsets U ⊂ X and V ⊂ Y,
144 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

∼ / V , and an isomorphism of structure

along with a homeomorphism π : U
∼ / ∼
sheaves OV π∗ OU (i.e., an isomorphism of schemes (U, OX |U ) ←→ (V, OY |V )).
Then we can glue these together to get a single scheme. Reason: let W be X and Y
∼
glued together using the isomorphism U ←→ V. Then Exercise 2.5.E shows that
the structure sheaves can be glued together to get a sheaf of rings. Note that this
is indeed a scheme: any point has an open neighborhood that is an affine scheme.
(Do you see why?)

4.4.A. E SSENTIAL E XERCISE ( CF. E XERCISE 2.5.E). Show that you can glue an
arbitrary collection of schemes together. Suppose we are given:
• schemes Xi (as i runs over some index set I, not necessarily finite),
• open subschemes Xij ⊂ Xi with Xii = Xi ,
∼
• isomorphisms fij : Xij −→ Xji with fii the identity
such that
• (the cocycle condition) the isomorphisms “agree on triple intersections”,
i.e., fik |Xij ∩Xik = fjk |Xji ∩Xjk ◦ fij |Xij ∩Xik (so implicitly, to make sense of
the right side, fij (Xik ∩ Xij ) ⊂ Xjk ).
(The cocycle condition ensures that fij and fji are inverses. In fact, the hypothesis
that fii is the identity also follows from the cocycle condition.) Show that there
is a unique scheme X (up to unique isomorphism) along with open subschemes
isomorphic to the Xi respecting this gluing data in the obvious sense. (Hint: what
is X as a set? What is the topology on this set? In terms of your description of the
open sets of X, what are the sections of this sheaf over each open set?) You may
wish to show uniqueness (up to unique isomorphism) by describing a universal
property satisfied by X.
I will now give you two non-affine schemes. Both are handy to know. In both
cases, I will glue together two copies of the affine line A1k . Let X = Spec k[t], and
Y = Spec k[u]. Let U = D(t) = Spec k[t, 1/t] ⊂ X and V = D(u) = Spec k[u, 1/u] ⊂
Y. We will get both examples by gluing X and Y together along U and V. The
difference will be in how we glue.

4.4.5. Second example: the affine line with the doubled origin. Consider
∼ ∼
the isomorphism U ←→ V via the isomorphism k[t, 1/t] ←→ k[u, 1/u] given by
t ←→ u (cf. Exercise 4.3.A). The resulting scheme is called the affine line with
doubled origin. Figure 4.6 is a picture of it.

F IGURE 4.6. The affine line with doubled origin

As the picture suggests, intuitively this is an analog of a failure of Haus-

dorffness. Now A1 itself is not Hausdorff, so we can’t say that it is a failure of
Hausdorffness. We see this as weird and bad, so we will want to make a definition
that will prevent this from happening. This will be the notion of separatedness (to be
discussed in Chapter 11). (In fact, in some sense, separatedness is the right defini-
tion of Hausdorfness, see Exercise 11.2.B!) This will answer other of our prayers as
 © 2024 Ravi Vakil. Published by Princeton University Press. 145

well. For example, on a separated scheme, the “affine base of the Zariski topology”
is nice — the intersection of two affine open sets will be affine (Proposition 11.2.13).

4.4.B. E XERCISE . Show that the affine line with doubled origin is not affine. Hint:
calculate the ring of global sections, and look back at the argument for A2 \ {(0, 0)}.

4.4.C. E ASY E XERCISE . Do the same construction with A1 replaced by A2 . You

will have defined the affine plane with doubled origin. Describe two affine open
subsets of this scheme whose intersection is not an affine open subset. (This example
was promised in Caution 4.3.3. Also, an “infinite-dimensional” version comes up
in Exercise 5.1.K.)

4.4.6. Third example: the projective line. Consider the isomorphism U = ∼ V via
the isomorphism k[t, 1/t] = ∼ k[u, 1/u] given by t ↔ 1/u. Figure 4.7 is a suggestive
picture of this gluing. The resulting scheme is called the projective line over the
field k, and is denoted P1k .

F IGURE 4.7. Gluing two affine lines together to get P1

Notice how the points glue. Let me assume that k is algebraically closed for
convenience. (You can think about how this changes otherwise.) On the first affine
line, we have the closed (“traditional”) points [(t − a)], which we think of as “a
on the t-line”, and we have the generic point [(0)]. On the second affine line, we
have closed points that are “b on the u-line”, and the generic point. Then a on
the t-line is glued to 1/a on the u-line (if a ̸= 0 of course), and the generic point
is glued to the generic point (the ideal (0) of k[t] becomes the ideal (0) of k[t, 1/t]
upon localization, and the ideal (0) of k[u] becomes the ideal (0) of k[u, 1/u]. And
∼
(0) in k[t, 1/t] is (0) in k[u, 1/u] under the isomorphism t ←→ 1/u).

4.4.7. If k is algebraically closed, we can interpret the closed points of P1k in the
following way, which may make this sound closer to the way you have seen
projective space defined earlier. The points are of the form [a, b], where a and b are
not both zero, and [a, b] is identified with [ac, bc] where c ∈ k× . Then if b ̸= 0, this
is identified with a/b on the t-line, and if a ̸= 0, this is identified with b/a on the
u-line.

4.4.8. Proposition. — P1k is not affine.

Proof. We do this by calculating the ring of global sections. The global sections
correspond to sections over X and sections over Y that agree on the overlap. A
section on X is a polynomial f(t). A section on Y is a polynomial g(u). If we restrict
f(t) to the overlap, we get something we can still call f(t); and similarly for g(u).
Now we want them to be equal: f(t) = g(1/t). But the only polynomials in t that
146 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

are at the same time polynomials in 1/t are the constants k. Thus Γ (P1 , OP1 ) = k.
If P1 were affine, then it would be Spec Γ (P1 , OP1 ) = Spec k, i.e., one point. But it
isn’t — it has lots of points. □
We have proved an analog of an important theorem: the only holomorphic
functions on CP1 are the constants! (See §11.4.7 for a serious yet easy generaliza-
tion.)

4.4.9. Important Example: Projective space. We now make a preliminary definition

of projective n-space over a field k, denoted Pn k , by gluing together n + 1 open sets
each isomorphic to An k . Judicious choice of notation for these open sets will make
our life easier. Our motivation is as follows. In the construction of P1 above, we
thought of points of projective space as [x0 , x1 ], where (x0 , x1 ) are only determined
up to scalars, i.e., (x0 , x1 ) is considered the same as (λx0 , λx1 ). Then the first patch
can be interpreted by taking the locus where x0 ̸= 0, and then we consider the
points [1, t], and we think of t as x1 /x0 ; even though x0 and x1 are not well-defined,
x1 /x0 is. The second corresponds to where x1 ̸= 0, and we consider the points
[u, 1], and we think of u as x0 /x1 . It will be useful to instead use the notation x1/0
for t and x0/1 for u.
For Pn , we glue together n + 1 open sets, one for each of i = 0, . . . , n. The ith
open set Ui will have coordinates x0/i , . . . , x(i−1)/i , x(i+1)/i , . . . , xn/i . It will be
convenient to write this as
(4.4.9.1) Spec k[x0/i , x1/i , . . . , xn/i ]/(xi/i − 1)
(so we have introduced a “dummy variable” xi/i which we immediately set to
1). We glue the distinguished open set D(xj/i ) of Ui to the distinguished open set
D(xi/j ) of Uj , by identifying these two schemes by describing the identification of
rings
∼
k[x0/i , x1/i , . . . , xn/i , 1/xj/i ]/(xi/i − 1) =
k[x0/j , x1/j , . . . , xn/j , 1/xi/j ]/(xj/j − 1)
via xk/i = xk/j /xi/j and xk/j = xk/i /xj/i (which implies xi/j xj/i = 1). We need to
check that this gluing information agrees over triple overlaps.

4.4.D. E XERCISE . Check this, as painlessly as possible. (Possible hint: the triple
intersection is affine; describe the corresponding ring.)

4.4.10. Definition: Pn
A . Note that our definition does not use the fact that k is a field.
Hence we may as well define Pn A for any ring A. This will be useful later.

∼
4.4.E. E XERCISE . Show that the only functions on Pn k are constants (Γ (Pk , O) = k),
n
n
and hence that Pk is not affine if n > 0. Hint: you might fear that you will need
some delicate interplay among all of your affine open sets, but you will only need
two of your open sets to see this. There is even some geometric intuition behind this:
the complement of the union of two open sets has codimension 2. But “Algebraic
Hartogs’s Lemma” (discussed informally in §4.4.2, and to be stated rigorously
in Theorem 13.5.19) says that any function defined on this union extends to be a
function on all of projective space. Because we are expecting to see only constants
as functions on all of projective space, we should already see this for this union of
our two affine open sets.
 © 2024 Ravi Vakil. Published by Princeton University Press. 147

4.4.F. E XERCISE ( GENERALIZING §4.4.7). Show that if k is algebraically closed, the

closed points of Pn k may be interpreted in the traditional way: the points are of the
form [a0 , . . . , an ], where the ai are not all zero, and [a0 , . . . , an ] is identified with
[λa0 , . . . , λan ] where λ ∈ k× .
Helpful translation: we think of the closed points of Pn k (where k = k) as
[x0 , x1 , . . . , xn ], with xi ∈ k (not all xi zero), and we identify [x0 , x1 , . . . , xn ] with
[λx0 , λx1 , . . . , λxn ] for all λ ∈ k× . Then [x0 , x1 , . . . , xn ] corresponds to a line through
the origin, and (x0/i , x1/i , . . . , xn/i ) is where this line meets the hyperplane xi = 1.
We will later give other definitions of projective space (Definition 4.5.9, §15.2.3),
and the xi will be called projective coordinates there. Our first definition here will
often be handy for computing things. But there is something unnatural about it —
projective space is highly symmetric, and that isn’t clear from our current definition.
Furthermore, as noted by Herman Weyl [Wey, p. 90], “The introduction of numbers
as coordinates is an act of violence.”

4.4.11. Fun aside: The Chinese Remainder Theorem is a geometric fact. The
Chinese Remainder Theorem is embedded in what we have done. We will see this
in a single example, but you should then figure out the general statement. The
Chinese Remainder Theorem says that knowing an integer modulo 60 is the same
as knowing an integer modulo 3, 4, and 5. Here is how to see this in the language
of schemes. What is Spec Z/(60)? What are the prime ideals of this ring? Answer:
those prime ideals of Z containing (60), i.e., those primes dividing 60, i.e., (2), (3),
and (5). Figure 4.8 is a sketch of Spec Z/(60). They are all closed points, as these
are all maximal ideals, so the topology is the discrete topology. What are the stalks?
You can check that they are Z/(4), Z/(3), and Z/(5). The nilpotents “at (2)” are
indicated by the “fuzz” on that point. (We discussed visualizing nilpotents with
“infinitesimal fuzz” in §4.2.) So what are global sections on this scheme? They are
sections on this open set (2), this other open set (3), and this third open set (5). In
other words, we have a natural isomorphism of rings
∼ / Z/(22 ) × Z/(3) × Z/(5).
Z/(60)

F IGURE 4.8. A picture of the scheme Spec Z/(60)

4.4.12. ⋆⋆ Example. Here is an example of a function on an open subset of a scheme

with some surprising behavior. On X = Spec k[w, x, y, z]/(wz − xy), consider the
open subset D(y) ∪ D(w). Clearly the function z/y on D(y) agrees with x/w on
D(w) on the overlap D(y) ∩ D(w). Hence they glue together to give a section. You
may have seen this before when thinking about analytic continuation in complex
geometry — we have a “holomorphic” function which has the description z/y
on an open set, and this description breaks down elsewhere, but you can still
“analytically continue” it by giving the function a different definition on different
parts of the space.
148 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Follow-up for curious experts. This function has no “single description” as a well-
defined expression in terms of w, x, y, z! There is a lot of interesting geometry here,
and this scheme will be a constant source of (counter)examples for us (look in the
index under “recurring (counter)examples”). Here is a glimpse, in terms of words
we have not yet defined. The space Spec k[w, x, y, z] is A4 , and is, not surprisingly,
4-dimensional. We are working with the subset X, which is a hypersurface, and is
3-dimensional. It is a cone over a smooth quadric surface in P3 (flip to Figure 9.2).
The open subset D(y) ⊂ X is X minus some hypersurface, so we are throwing away
a codimension 1 locus. The open subset D(w) involves throwing away another
codimension 1 locus. You might think that the intersection of these two discarded
loci is then codimension 2, and that failure of extending this weird function to
a global polynomial comes because of a failure of our Hartogs’s Lemma-type
theorem. But that’s not true — V(y) ∩ V(w) is in fact codimension 1. Here is
what is actually going on. The space V(y) is obtained by throwing away the (cone
over the) union of two lines ℓ and m1 , one in each “ruling” of the surface, and
V(w) also involves throwing away the (cone over the) union of two lines ℓ and
m2 . The intersection is the (cone over the) line ℓ, which is a codimension 1 set.
Remarkably, despite being “pure codimension 1” the cone over ℓ is not cut out even
set-theoretically by a single equation (see Exercise 15.5.N). This means that any
expression f(w, x, y, z)/g(w, x, y, z) for our function cannot correctly describe our
function on D(y) ∪ D(w) — at some point of D(y) ∪ D(w) it must be 0/0. Here’s
why. Our function can’t be defined on V(y) ∩ V(w), so g must vanish here. But g
can’t vanish just on the cone over ℓ — it must vanish elsewhere too.

4.5 Projective schemes, and the Proj construction

Projective schemes are important for a number of reasons. Here are a few.
Schemes that were of “classical interest” in geometry — and those that you would
have cared about before knowing about schemes — are all projective or an open
subset thereof (basically, quasiprojective, see §4.5.10). Moreover, most “schemes of
interest” tend to be projective or quasiprojective. In fact, it is very hard to even give
an example of a scheme satisfying basic properties — for example, finite type and
“Hausdorff” (“separated”) over a field — that is provably not quasiprojective. For
complex geometers: it is hard to find a compact complex variety that is provably
not projective (see Remark 11.4.6), and it is quite hard to come up with a complex
variety that is provably not an open subset of a projective variety. So projective
schemes are really ubiquitous. Also, the notion of “projective k-scheme” is a good
approximation of the algebro-geometric version of compactness (“properness”, see
§11.4).
Finally, although projective schemes may be obtained by gluing together affine
schemes, and we know that keeping track of gluing can be annoying, there is a
simple means of dealing with them without worrying about gluing. Just as there
is a rough dictionary between rings and affine schemes, we will have a (slightly
looser) analogous dictionary between graded rings and projective schemes. Just
as one can work with affine schemes by instead working with rings, one can work
with projective schemes by instead working with graded rings.
 © 2024 Ravi Vakil. Published by Princeton University Press. 149

4.5.1. Motivation from classical geometry.

For geometric intuition, we recall how one thinks of projective space “classi-
cally” (in the classical topology, over the real numbers). Pn can be interpreted as
the lines through the origin in Rn+1 . Thus subsets of Pn correspond to unions of
lines through the origin of Rn+1 , and closed subsets correspond to such unions
which are closed. (The same is not true with “closed” replaced by “open”!)
One often pictures Pn as being the “points at infinite distance” in Rn+1 , where
the points infinitely far in one direction are associated with the points infinitely far
in the opposite direction. We can make this more precise using the decomposition
a
Pn+1 = Rn+1 Pn

by which we mean that there is an open subset in Pn+1 identified with Rn+1 (the
points with last “projective coordinate” nonzero), and the complementary closed
subset identified with Pn (the points with last “projective coordinate” zero). (The
phrase “projective coordinate” will be formally defined in §4.5.9, but we will use it
even before then, in Exercises 4.5.B and 4.5.E.)
Then for example any equation cutting out some set V of points in Pn will also
cut out some set of points in Rn+1 that will be a closed union of lines. We call this
the affine cone of V. These equations will cut out some union of P1 ’s in Pn+1 , and
we call this the projective cone of V. The projective cone is the disjoint union of the
affine cone and V. For example, the affine cone over x2 + y2 = z2 in P2 is just the
“classical” picture of a cone in R3 , see Figure 4.9. We will make this analogy precise
in our algebraic setting in §9.3.11.

F IGURE 4.9. The affine and projective cone of x2 + y2 = z2 in

classical geometry

4.5.2. Projective schemes, a first description.

We now describe a construction of projective schemes, which will help motivate
the Proj construction. We begin by giving an algebraic interpretation of the cone just
described, and more generally, getting some practice with transforming between
projective coordinates, and coordinates on the standard affine open subsets. We
switch coordinates from x, y, z to x0 , x1 , x2 in order to use the notation of §4.4.9. For
the next questions, in P2k , we take projective coordinates x0 , x1 , x2 . (The terminology
“projective coordinate” will not be formally defined until §4.5.9, but you should
150 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

be able to solve these exercises anyway.) The big open set (or “coordinate chart”)
U0 = {[x0 , x1 , x2 ] : x0 ̸= 0} has coordinates x1/0 and x2/0 , which we interpret as
x1 /x0 and x2 /x0 . We have similar definitions for U1 and U2 . It will be convenient
to define x0/0 as 1.

4.5.A. E XERCISE . Describe (x0/1 , x2/1 ) in terms of (x1/0 , x2/0 ).

4.5.B. E XERCISE . Explain how to define a scheme that should be interpreted as

x20 + x21 − x22 = 0 “in P2k ”. Hint: in the affine open subset corresponding to x2 ̸= 0, it
should (in the language of 4.4.9) be cut out by x20/2 + x21/2 − 1 = 0, i.e., it should
“be” the scheme Spec k[x0/2 , x1/2 ]/(x20/2 + x21/2 − 1). You can similarly guess what
it should be on the other two standard open sets, and show that the three schemes
glue together.

4.5.C. E XERCISE . (a) Describe all homogeneous polynomials in x0 , x1 , x2 that can

be “dehomogenized” to x21/0 + x52/0 = 1. Explain which one is “best”. (b) In what
points do the vanishing sets of these polynomials meet the “line at infinity” x0 = 0
in P2k ?

4.5.D. E XERCISE . Consider the parabola x2/0 = x21/0 (or, if you prefer, y = x2 ).
How does it meet the line at infinity? (What does this mean?) What is its description
in the different coordinate patches U1 and U2 ?

4.5.3. Remark: Degree d hypersurfaces in Pn . The degree d homogeneous polynomials

in n + 1 variables over a field form a vector space of dimension n+d


d . (This is
essentially the content of Exercise 9.3.J and Exercise 15.1.C.) It is almost true that
two polynomials cut out the same subset of Pn k if one is a nonzero scalar multiple
of the other. Unfortunately, the examples of x2 y = 0 and xy2 = 0 show that this
isn’t quite right. You will later be able to check (for example, using Exercise 9.5.C)
that two polynomials cut out the same closed subscheme (whatever that means) if
and only if one is a nonzero scalar multiple of the other. (This is some evidence
that the notion of a “closed subscheme” is better than that of a “closed subset”.)
The zero polynomial doesn’t really cut out a hypersurface in any reasonable sense
of the word. Thus we informally imagine that “degree d hypersurfaces in Pn are
n+d
parametrized by P( d )−1 ”. This intuition will come up repeatedly (in special
cases), and we will give it precise meaning in §25.3.5. (We will properly define
hypersurfaces in §9.3.2, once we have the language of closed subschemes. At that
time we will also define line, hyperplane, quadric hypersurfaces, conic curves, and
other wondrous notions.)

4.5.E. E XERCISE ( GENERALIZING E XERCISE 4.5.B). Consider Pn A , with projective

coordinates x0 , . . . , xn . Given a collection of homogeneous polynomials fi ∈
A[x0 , . . . , xn ], make sense of the scheme “cut out in Pn
A by the fi .” (This will later
be made precise as an example of a “vanishing scheme”, see Exercise 4.5.S.)
This could be taken as the definition of a projective A-scheme, but we will wait
until §4.5.10 to state it a little better.

4.5.4. Preliminaries on graded rings.

 © 2024 Ravi Vakil. Published by Princeton University Press. 151

The Proj construction produces a scheme out of a graded ring. We now give
some background on graded rings.

4.5.5. Definition: Z-graded rings. A Z-graded ring is a ring S• = ⊕n∈Z Sn (the

subscript is called the grading), where multiplication respects the grading, i.e.,
sends Sm × Sn to Sm+n . Clearly S0 is a subring, each Sn is an S0 -module, and S• is
a S0 -algebra. Suppose for the remainder of §4.5.5 that S• is a Z-graded ring. Those
elements of some Sn are called homogeneous elements of S• . An ideal I of S• is a
homogeneous ideal (or a graded ideal, although we won’t use this terminology) if
it is generated by homogeneous elements. Nonzero homogeneous elements have
an obvious degree. An element of Sd is called a form of degree d.

4.5.F. E XERCISE .
(a) Show that an ideal I is homogeneous if and only if it contains the degree n piece
of each of its elements for each n. (Hence I can be decomposed into homogeneous
pieces, I = ⊕In , and S• /I has a natural Z-grading. This is the reason for the name
homogeneous ideal.)
(b) Show that the set of homogeneous ideals of a given Z-graded ring S• is closed
under sum, product, intersection, and radical.
(c) Show that a homogeneous ideal I ⊂ S• is prime if I ̸= S• , and if for any
homogeneous a, b ∈ S• , if ab ∈ I, then a ∈ I or b ∈ I.
If T is a multiplicative subset of S• containing only homogeneous elements,
then T −1 S• has a natural structure as a Z-graded ring.

4.5.6. Z≥0 -graded rings, graded ring over A, and finitely generated graded rings. A
Z≥0 -graded ring is a Z-graded ring with no elements of negative degree.
For the remainder of the book, graded ring will refer to a Z≥0 -graded ring.
Warning: this convention is nonstandard (for good reason).
From now on, unless otherwise stated, S• is assumed to be a graded ring. Fix
a ring A, which we call the base ring. If S0 = A, we say that S• is a graded ring
over A. A key example is A[x0 , . . . , xn ], or more generally A[x0 , . . . , xn ]/I where
I is a homogeneous ideal with I ∩ S0 = 0 (cf. Exercise 4.5.E). Here we take the
conventional grading on A[x0 , . . . , xn ], where each xi has weight 1.
The subset S+ := ⊕i>0 Si ⊂ S• is an ideal, called the irrelevant ideal. The
reason for the name “irrelevant” will be clearer in a few paragraphs. If the irrelevant
ideal S+ is a finitely generated ideal, we say that S• is a finitely generated graded
ring over A. If S• is generated by S1 as an A-algebra, we say that S• is generated
in degree 1. (We will later find it useful to interpret “S• is generated in degree 1” as
“the natural map Sym• S1 → S• is a surjection”. The symmetric algebra construction
will be briefly discussed in §14.2.6.)

4.5.G. E XERCISE .
(a) Show that a graded ring S• over A is a finitely generated graded ring (over
A) if and only if S• is a finitely generated graded A-algebra, i.e., generated over
A = S0 by a finite number of homogeneous elements of positive degree. (Hint
for the forward implication: show that the generators of S+ as an ideal are also
generators of S• as an algebra.)
152 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

(b) Show that a graded ring S• over A is Noetherian if and only if A = S0 is

Noetherian and S• is a finitely generated graded ring.

4.5.7. The Proj construction.

We now define a scheme Proj S• , where S• is a (Z≥0 -)graded ring. Here are two
examples, to provide a light at the end of the tunnel. If S• = A[x0 , . . . , xn ], we will
recover PnA ; and if S• = A[x0 , . . . , xn ]/(f(x0 , . . . , xn )) where f is homogeneous, we
will construct something “cut out in Pn A by the equation f = 0” (cf. Exercise 4.5.E).
As we did with Spec of a ring, we will build Proj S• first as a set, then as a
topological space, and finally as a ringed space. In our preliminary definition of Pn A,
we glued together n + 1 well-chosen affine pieces, but we don’t want to make any
choices, so we do this by simultaneously considering “all possible” affine open sets.
Our affine building blocks will be as follows. For each homogeneous f ∈ S+ , note
that the localization (S• )f is naturally a Z-graded ring, where deg(1/f) = − deg f.
Consider

(4.5.7.1) Spec((S• )f )0 .

where ((S• )f )0 means the 0-graded piece of the graded ring (S• )f . (These will be
our affine building blocks, as f varies over the homogeneous elements of S+ .) The
notation ((S• )f )0 is admittedly horrible — the first and third subscripts refer to the
grading, and the second refers to localization. As motivation for considering this
construction: applying this to S• = k[x0 , . . . , xn ], with f = xi , we obtain the ring
appearing in (4.4.9.1):

k[x0/i , x1/i , . . . , xn/i ]/(xi/i − 1).

(Before we begin the construction: another possible way of defining Proj S• is

by gluing together affines of this form, by jumping straight to Exercises 4.5.N and
4.5.O. If you prefer that, by all means do so.)
The points of Proj S• are the homogeneous prime ideals of S• not containing the
irrelevant ideal S+ (the “relevant homogeneous prime ideals”).

4.5.H. I MPORTANT AND TRICKY EXERCISE . Suppose f ∈ S+ is homogeneous.

(a) Give a bijection between the prime ideals of ((S• )f )0 and the homogeneous
prime ideals of (S• )f . Hint: Avoid notational confusion by proving instead that if
A is a Z-graded ring with a homogeneous invertible element f in positive degree,
then there is a bijection between prime ideals of A0 and homogeneous prime ideals
of A. Using the ring map A0 → A, from each homogeneous prime ideal of A we
find a prime ideal of A0 . The reverse direction is the harder one. Given a prime
ideal P0 ⊂ A0 , define P ⊂ A (a priori only a subset) as ⊕Qi , where Qi ⊂ Ai , and
a ∈ Qi if and only if adeg f /fi ∈ P0 . Note that Q0 = P0 . Show that a ∈ Qi if and
only if a2 ∈ Q2i ; show that if a1 , a2 ∈ Qi then a21 + 2a1 a2 + a22 ∈ Q2i and hence
a1 + a2 ∈ Qi ; then show that P is a homogeneous ideal of A; then show that P is
prime.
(b) Interpret the set of prime ideals of ((S• )f )0 as a subset of Proj S• .
The correspondence of the points of Proj S• with homogeneous prime ideals
helps us picture Proj S• . For example, if S• = k[x, y, z] with the usual grading, then
we picture the homogeneous prime ideal (z2 − x2 − y2 ) first as a subset of Spec S• ;
it is a cone (see Figure 4.9). As in §4.5.1, we picture P2k as the “plane at infinity”.
 © 2024 Ravi Vakil. Published by Princeton University Press. 153

Thus we picture this equation as cutting out a conic “at infinity” (in Proj S• ). We
will make this intuition somewhat more precise in §9.3.11.

4.5.8. Motivated by the affine case, if T is a set of homogeneous elements of S• ,

define the (projective) vanishing set of T , V(T ) ⊂ Proj S• , to be those homogeneous
prime ideals containing T but not S+ . Define V(f) if f is a homogeneous element
of positive degree in the obvious way. Let D+ (f) := Proj S• \ V(f) (the projective
distinguished open set) be the complement of V(f). (The subscript is intended to
distinguish this notation from the similar D(f).) Once we define a scheme structure
on Proj S• , we will (without comment) use D+ (f) to refer to the open subscheme,
not just the open subset. (Although the definition of D+ (f) makes sense even if f
has degree 0, we deliberately allow only f of positive degree. For example, we will
want the D+ (f) to form an affine cover, and if f has degree 0, then D+ (f) needn’t
be affine.)

4.5.I. E XERCISE . Show that D+ (f) “is” (or more precisely, “corresponds to”) the
subset Spec((S• )f )0 you described in Exercise 4.5.H(b). For example, in §4.4.9, the
D+ (xi ) are the standard open sets covering projective space.
As in the affine case, the V(I)’s satisfy the axioms of the closed sets of a topology,
and we call this the Zariski topology on Proj S• . (Other definitions given in the
literature may look superficially different, but can be easily shown to be the same.)
Many statements about the Zariski topology on Spec of a ring carry over to this
situation with little extra work. Clearly D+ (f) ∩ D+ (g) = D+ (fg), by the same
immediate argument as in the affine case (Exercise 3.5.D).

4.5.J. E ASY E XERCISE . Verify that the projective distinguished open sets D+ (f)
(as f runs through the homogeneous elements of S+ ) form a base of the Zariski
topology.

4.5.K. E XERCISE . Fix a graded ring S• .

(a) Suppose I is any homogeneous ideal of S• contained in S+ , and f is a homoge-
neous element of positive degree. Show that f vanishes on V(I) (i.e., V(I) ⊂ V(f)) if
and only if fn ∈ I for some n. (Hint: Mimic the affine case; see Exercise 3.4.J.) In
particular, as in the affine case (Exercise 3.5.E), if D+ (f) ⊂ D+ (g), then fn ∈ (g) for
some n, and vice versa. (Here g is also homogeneous of positive degree.)
(b) If Z ⊂ Proj S• , define I(Z) ⊂ S+ . Show that it is a homogeneous ideal of S• . For
any two subsets, show that I(Z1 ∪ Z2 ) = I(Z1 ) ∩ I(Z2 ).
(c) For any subset Z ⊂ Proj S• , show that V(I(Z)) = Z.

4.5.L. E XERCISE ( CF. E XERCISE 3.5.B). Fix a graded ring S• , and a homogeneous
ideal I ⊆ S+ . Show that the following are equivalent.
(a) V(I) = ∅.
(b) For any homogeneous fi (as i runs through some index set) generating I,
∪D
√ + (fi ) = Proj S• .
(c) I ⊃ S+ .
This is more motivation for the ideal S+ being “irrelevant”: any ideal whose radical
contains it is “geometrically irrelevant”.
We now construct Proj S• as a scheme.
154 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

4.5.M. E XERCISE . Suppose some homogeneous f ∈ S+ is given. Via the inclusion

 / Proj S•
D+ (f) = Spec((S• )f )0
of Exercise 4.5.I, show that the Zariski topology on Proj S• restricts to the Zariski
topology on Spec((S• )f )0 .
Now that we have defined Proj S• as a topological space, we are ready to define
the structure sheaf. On D+ (f), we wish it to be the structure sheaf of Spec((S• )f )0 .
We will glue these sheaves together using Exercise 2.5.E on gluing sheaves.

4.5.N. E XERCISE . If f, g ∈ S+ are homogeneous and nonzero, describe an isomor-

phism between Spec((S• )fg )0 and the distinguished open subset D(gdeg f /fdeg g ) of
Spec((S• )f )0 .
Similarly, Spec((S• )fg )0 is identified with a distinguished open subset of Spec((S• )g )0 .
We then glue the various Spec((S• )f )0 (as f varies) altogether, using these pairwise
gluings.

4.5.O. E XERCISE . By checking that these gluings behave well on triple overlaps
(see Exercise 2.5.E), finish the definition of the scheme Proj S• .

4.5.P. E XERCISE . (Some will find this essential, while others will prefer to ignore
it.) (Re)interpret the structure sheaf of Proj S• in terms of compatible germs.

4.5.9. Definition. We (re)define projective space (over a ring A) by

Pn
A := Proj A[x0 , . . . , xn ].

This definition involves no messy gluing, or special choice of patches. Note that
the variables x0 , . . . , xn , which we call the projective coordinates on Pn
A , are part
of the definition. (They may have other names than x’s, depending on the context.)

4.5.Q. E XERCISE . Check that this agrees with our earlier construction of Pn
A (§4.4.9).
(How do you know that the D+ (xi ) cover Proj A[x0 , . . . , xn ]?)
Notice that with our old definition of projective space, it would have been a non-
trivial exercise to show that D+ (x2 +y2 −z2 ) ⊂ P2k (the complement of a plane conic)
is affine; with our new perspective, it is immediate — it is Spec(k[x, y, z](x2 +y2 −z2 ) )0 .

4.5.R. E XERCISE . Suppose that k is an algebraically closed field. We know from

Exercise 4.4.F that the closed points of Pn k , as defined in §4.4.9, are in bijection
with the points of “classical” projective space. By Exercise 4.5.Q, the scheme
Pnk as defined in §4.4.9 is isomorphic to Proj k[x0 , . . . , xn ]. Therefore, each point
[a0 , . . . , an ] of classical projective space corresponds to a homogeneous prime ideal
of k[x0 , . . . , xn ]. Which homogeneous prime ideal is it?
We now figure out the “right definition” of the vanishing scheme, in anal-
ogy with the vanishing set V(·) (§3.4.1). You will be defining a closed subscheme
(mentioned in Remark 4.3.5, and to be properly defined in §9.1).

4.5.S. E XERCISE . If S• is generated in degree 1, and f ∈ S+ is homogeneous,

explain how to define V(f) “in” Proj S• , the vanishing scheme of f. This is the same
notation as the “vanishing set” V(·), but now means something richer than a mere
set. (Exercise 9.1.K is similar. Warning: in general, f isn’t a function on Proj S• . We
 © 2024 Ravi Vakil. Published by Princeton University Press. 155

will later interpret it as something close: a section of a line bundle, see for example
§15.1.2.) Hence define V(I) for any homogeneous ideal I of S+ . (Another solution
in more general circumstances will be given in Exercise 14.2.D.)

4.5.10. Projective and quasiprojective schemes.

We call a scheme of the form (i.e., isomorphic to) Proj S• , where S• is a finitely
generated graded ring over A, a projective scheme over A, or a projective A-scheme.
A quasiprojective A-scheme is a quasicompact open subscheme of a projective
A-scheme. The “A” is omitted if it is clear from the context; often A is a field.

4.5.11. Unimportant remarks. (i) Note that Proj S• makes sense even when S• is not
finitely generated. This can be useful. For example, you will later be able to do
Exercise 7.4.D without worrying about Exercise 7.4.H.
(ii) The quasicompact requirement in the definition of quasiprojectivity is of
course redundant in the Noetherian case (cf. Exercise 3.6.U), which is all that matters
to most.
∼ Spec A. Thus “Spec A is a
4.5.12. Silly example. Note that P0A = Proj A[T ] =
projective A-scheme”.

4.5.13. Example: projectivization of a vector space PV. We can make this

definition of projective space even more choice-free as follows. Let V be an (n + 1)-
dimensional vector space over k. (Here k can be replaced by any ring A and V by a
free module as usual.) Define

Sym• V ∨ = k ⊕ V ∨ ⊕ Sym2 V ∨ ⊕ · · · .

(The reason for the dual is explained by the next exercise.) If for example V is
the dual of the vector space with basis associated to x0 , . . . , xn , we would have
Sym• V ∨ = k[x0 , . . . , xn ]. Then we can define PV := Proj(Sym• V ∨ ). In this
language, we have an interpretation for x0 , . . . , xn : they are linear functionals
on the underlying vector space V. (Warning: some authors use the definition
PV = Proj(Sym• V), so be cautious.)

4.5.T. U NIMPORTANT EXERCISE . Suppose k is algebraically closed. Describe a

natural bijection between one-dimensional subspaces of V and the closed points
of PV. Thus this construction canonically (in a basis-free manner) describes the
one-dimensional subspaces of the vector space V.
Unimportant remark: you may be surprised at the appearance of the dual in the
definition of PV. This is partially explained by the previous exercise. Most normal
(traditional) people define the projectivization of a vector space V to be the space of
one-dimensional subspaces of V. Grothendieck considered the projectivization to
be the space of one-dimensional quotients. One motivation for this is that it gets rid
of the annoying dual in the definition above. There are better reasons, that we won’t
go into here. In a nutshell, quotients tend to be better-behaved than subobjects for
coherent sheaves, which generalize the notion of vector bundle. (Coherent sheaves
are discussed in Chapter 14.)
On another note related to Exercise 4.5.T: you can also describe a natural bijec-
tion between points of V and the closed points of Spec(Sym• V ∨ ). This construction
respects the affine/projective cone picture of §9.3.11.
156 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

4.5.14. The Grassmannian. At this point, we could describe the fundamental

geometric object known as the Grassmannian, and give the “wrong” (but correct)
definition of it. We will instead wait until §7.7 to give the wrong definition, when
we will know enough to sense that something is amiss. The right definition will be
given in §16.4.
 CHAPTER 5

Some properties of schemes

Now that we have defined the notion of a scheme, we can define some use-
ful properties of schemes. As you see each definition, you should try it out
in specific examples of your choice, such as your favorite schemes of the form
Spec C[x1 . . . , xn ]/(f1 , . . . , fr ).

5.1 Topological properties

The definitions of connected, connected component, (ir)reducible, irreducible com-
ponent, quasicompact, generization, specialization, generic point, Noetherian topological
space, and closed point, were given in §3.6. You should have pictures in your mind
of each of these notions.
Exercise 3.6.C shows that An k is irreducible (it was easy). This argument “be-
haves well under gluing”, yielding:

5.1.A. E ASY E XERCISE . Show that Pn

k is irreducible.

5.1.B. E XERCISE . Exercise 3.7.F showed that there is a bijection between irreducible
closed subsets and points for affine schemes (the map sending a point p to the
closed subset {p} is a bijection). Show that this is true of schemes in general.

5.1.C. E ASY E XERCISE . Prove that if X is a scheme that has a finite cover X =
∪ni=1 Spec Ai where Ai is Noetherian, then X is a Noetherian topological space
(§3.6.14). (We will soon call a scheme with such a cover a Noetherian scheme, §5.3.4.)
Hint: show that a topological space that is a finite union of Noetherian subspaces is
itself Noetherian.
Thus Pn n
k and PZ are Noetherian topological spaces: we built them by gluing
together a finite number of spectra of Noetherian rings.

5.1.D. E ASY E XERCISE . Show that a scheme X is quasicompact if and only if it can
be written as a finite union of affine open subschemes. (Hence Pn
A is quasicompact
for any ring A.)

5.1.E. E XERCISE : Q UASICOMPACT SCHEMES HAVE CLOSED POINTS . Show that if X

is a quasicompact scheme, then every point has a closed point in its closure. Show
that every nonempty closed subset of X contains a closed point of X. In particular,

This material is published by Princeton University Press as The Rising Sea: Foundations of Algebraic
Geometry by Ravi Vakil. This version is free to view and download for personal use only. Not for
redistribution, resale, or use in derivative works. © 2024 Ravi Vakil.

157
158 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

every nonempty quasicompact scheme has a closed point. (Warning: there exist
nonempty schemes with no closed points, such as Exercise 15.2.P, so your argument
had better use the quasicompactness hypothesis!)
This exercise will often be used in the following way. If there is some property
P of points of a scheme that is “open” (if a point p has P, then there is some open
neighborhood U of p such that all the points in U have P), then to check if all points
of a quasicompact scheme have P, it suffices to check only the closed points. (A first
example of this philosophy is Exercise 5.2.E.) This provides a connection between
schemes and the classical theory of varieties — the points of traditional varieties are
the closed points of the corresponding schemes (essentially by the Nullstellensatz,
see §3.6.9 and Exercise 5.3.F). In many good situations, the closed points are dense
(such as for varieties, see §3.6.9 and Exercise 5.3.F again), but this is not true in
some fundamental cases (see Exercise 3.6.J(b)).

5.1.1. Quasiseparated schemes. Quasiseparatedness is a weird notion that comes

in handy for certain people. (Warning: we will later realize that this is really a
property of morphisms, not of schemes, §8.3.1.) Most people, however, can ignore
this notion, as the schemes they will encounter in real life will all have this property.
A topological space is quasiseparated if the intersection of any two quasicompact
open subsets is quasicompact. (The motivation for the “separatedness” in the name
is that it is a weakened version of separated, for which the intersection of any two
affine open sets is affine, see Proposition 11.2.13.)

5.1.F. S HORT EXERCISE . Show that a scheme is quasiseparated if and only if the
intersection of any two affine open subsets is a finite union of affine open subsets.

5.1.G. E XERCISE . Show that affine schemes are quasiseparated. (Possible hint:
Exercise 5.1.F.)

5.1.H. E XERCISE . Show that a scheme X is quasiseparated if and only if there exists
a cover of X by affine open subsets, any two of which have intersection also covered
by a finite number of affine open subsets.
We will see that quasiseparatedness will be a useful hypothesis in theorems
(in conjunction with quasicompactness), and that various interesting kinds of
schemes (affine, locally Noetherian, separated, see Exercises 5.1.G, 5.3.A, and 11.2.D
respectively) are quasiseparated, and this will allow us to state theorems more
succinctly (e.g., “if X is quasicompact and quasiseparated” rather than “if X is
quasicompact, and either this or that or the other thing hold”).
“Quasicompact and quasiseparated” means something concrete:

5.1.I. E XERCISE . Show that a scheme X is quasicompact and quasiseparated if and

only if X can be covered by a finite number of affine open subsets, any two of which
have intersection also covered by a finite number of affine open subsets.

5.1.2. Qcqs. When you see “quasicompact and quasiseparated” as hypotheses in a

theorem, you should take this as a clue that you will use this interpretation, and
that finiteness will be used in an essential way. This combination of hypotheses is
so common that people often describe it as qcqs for short (see for example the Qcqs
Lemma 6.2.9).
 © 2024 Ravi Vakil. Published by Princeton University Press. 159

5.1.J. E ASY E XERCISE . Show that all projective A-schemes are quasicompact and
quasiseparated. (Hint: use the fact that the graded ring in the definition is finitely
generated — those finite number of generators will lead you to a covering set.)

5.1.K. E XERCISE ( A NONQUASISEPARATED SCHEME ). Let X = Spec k[x1 , x2 , . . . ],

and let U be X \ [m] where m is the maximal ideal (x1 , x2 , . . . ). Take two copies
of X, glued along U (“affine ∞-space with a doubled origin”, see Example 4.4.5
and Exercise 4.4.C for “finite-dimensional” versions). Show that the result is not
quasiseparated. Hint: This “open embedding” U ⊂ X came up earlier in Exer-
cise 3.6.G(b) as an example of a nonquasicompact open subset of an affine scheme.

5.1.3. Dimension. One very important topological notion is dimension. (It is

amazing that this is a topological idea.) But despite being intuitively fundamental, it
is more difficult, so we postpone it until Chapter 12.

5.2 Reducedness and integrality

Recall that one of the alarming things about schemes is that functions are not
determined by their values at points, and that was because of the presence of
nilpotents (§3.2.12).

5.2.1. Definition. We say that a scheme X is reduced if its stalks OX,p have no
nonzero nilpotents (for all p).

5.2.A. E XERCISE . Show that if f and g are two functions (global sections of OX )
Q a reduced scheme that agree at all points, then f = g. (Two hints: OX (U) ,→
on
p∈U OX,p from Exercise 2.4.A, and the nilradical is the intersection of all prime
ideals from Theorem 3.2.13.)

5.2.B. E XERCISE . Recall that a ring is said to be reduced if it has no nonzero

nilpotents (§3.2.14). Show that A is a reduced ring if and only if Spec A is reduced.
Show that a scheme X is reduced if OX (U) is reduced for every open set U of X.

5.2.C. E ASY E XERCISE . Show that An n

k and Pk are reduced.

The scheme Spec k[x, y]/(y2 , xy) is nonreduced. When we sketched it in Fig-
ure 4.4, we indicated that the fuzz represented nonreducedness at the origin. The
following exercise is a first stab at making this precise.


5.2.D. E XERCISE . Show that k[x, y]/(y2 , xy) x has no nonzero nilpotent elements.
(Possible hint: show that it is isomorphic to another ring, by considering the
geometric picture. Exercise 3.2.L may give another hint.) Show that the only point
of Spec k[x, y]/(y2 , xy) with a nonreduced stalk is the origin.

5.2.E. U NIMPORTANT E XERCISE . If X is a quasicompact scheme, show that it

suffices to check reducedness at closed points. Hint: Do not try to show that
reducedness is an open condition (see Remark 5.2.2). Instead show that any nonre-
duced point has a nonreduced closed point in its closure, using Exercise 5.1.E. (This
result is interesting, but we won’t use it.)
160 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

5.2.2. Remark. We will see in Exercise 6.6.N that reducedness is an open condi-
tion for locally Noetherian schemes. But in general, reducedness is not an open
condition. You may be able to identify the underlying topological space of

X = Spec C[x, y1 , y2 , . . . ]/(y21 , y22 , y23 , . . . , (x − 1)y1 , (x − 2)y2 , (x − 3)y3 , · · · )

with that of Spec C[x], and then to show that the nonreduced points of X are
precisely the closed points corresponding to the positive integers. The complement
of this set is not Zariski open.

5.2.3. ⋆ Warning for experts. If a scheme X is reduced, then (from Exercise 5.2.B)
its ring of global sections is reduced. However, the converse is not true. You
already know enough to verify that if X is the scheme cut out by x2 = 0 in P2k ,
∼ k, and that X is nonreduced. (This example will come up again in
then Γ (X, OX ) =
§11.4.7 and §18.6.U.)

5.2.F. E XERCISE . Suppose X is quasicompact, and f is a function that vanishes

at all points of X. Show that there is some n such that fn = 0. Show that this
may fail if X is not quasicompact. (This exercise is less important, but shows
why we like quasicompactness, and gives a standard pathology when quasicom-
pactness doesn’t hold.) Hint: take an infinite disjoint union of Spec An with
An := k[ϵ]/(ϵn ). This scheme arises again in §9.4.2 (see Figure 9.5 for a picture)
and in Caution/Example 9.4.11.

5.2.4. Definition. A scheme X is integral if it is nonempty, and OX (U) is an integral

domain for every nonempty open subset U of X.

5.2.G. I MPORTANT E XERCISE . Show that a scheme X is integral if and only if it

is irreducible and reduced. (Thus we picture integral schemes as: “one piece, no
fuzz”. Possible hint: Exercise 4.3.F.)

5.2.H. E XERCISE . Show that an affine scheme Spec A is integral if and only if A is
an integral domain.

5.2.I. E XERCISE . Suppose X is an integral scheme. Then X (being irreducible) has a

generic point η. Suppose Spec A is any nonempty affine open subset of X. Show
that OX,η (the stalk of OX at η) is naturally identified with K(A), the fraction field of
A. This is called the function field K(X) of X. It can be computed on any nonempty
open set of X, as any such open set contains the generic point. The elements of K(X)
are called rational functions on X (to be generalized further in Definition 6.6.36).

5.2.J. E XERCISE . Suppose X is an integral scheme. Show that the restriction maps
resU,V : OX (U) → OX (V) are inclusions so long as V ̸= ∅. Suppose Spec A is any
nonempty affine open subset of X (so A is an integral domain). Show that the
natural map OX (U) → OX,η = K(A) (where U is any nonempty open subset and η
is the generic point) is an inclusion.
Thus irreducible varieties (an important example of integral schemes defined
later) have the convenient property that sections over different open sets can be
considered subsets of the same ring. In particular, restriction maps (except to the
empty set) are always inclusions, and gluing is easy: functions fi on a cover Ui
of U (as i runs over an index set) glue if and only if they are the same element of
 © 2024 Ravi Vakil. Published by Princeton University Press. 161

K(X). This is one reason why (irreducible) varieties are usually introduced before
schemes.

5.2.5. Caution. Integrality is not`stalk-local (the disjoint union of two integral

schemes is not integral, as Spec A Spec B = Spec(A × B) by Exercise 3.6.A), but
it almost is, see Exercise 5.3.C.

5.3 The Affine Communication Lemma, and properties of schemes

that can be checked “affine-locally”
This section is intended to address something tricky in the definition of schemes.
We have defined a scheme as a topological space with a sheaf of rings, that can be
covered by affine schemes. Hence we have all of the affine open sets in the cover,
but we don’t know how to communicate between any two of them. Somewhat
more explicitly, if I have an affine cover, and you have an affine cover, and we want
to compare them, and I calculate something on my cover, there should be some
way of us getting together, and figuring out how to translate my calculation over
to your cover. The Affine Communication Lemma 5.3.2 will provide a convenient
machine for doing this.
Thanks to this lemma, we can define a host of important properties of schemes.
All of these are “affine-local”, in that they can be checked on any affine cover, i.e.,
a covering by open affine sets. (We state this more formally after the statement of
the Affine Communication Lemma.) We like such properties because we can check
them using any affine cover we like. If the scheme in question is quasicompact,
then we need only check a finite number of affine open sets.

5.3.1. Proposition. — Suppose Spec A and Spec B are affine open subschemes of a scheme
X. Then Spec A ∩ Spec B is the union of open sets that are simultaneously distinguished
open subschemes of Spec A and Spec B.

F IGURE 5.1. A trick to show that the intersection of two affine

open sets may be covered by open sets that are simultaneously
distinguished in both affine open sets

Proof. (See Figure 5.1.) Given any point p ∈ Spec A ∩ Spec B, we produce an open
neighborhood of p in Spec A ∩ Spec B that is simultaneously distinguished in both
Spec A and Spec B. Let Spec Af be a distinguished open subset of Spec A contained
162 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

in Spec A ∩ Spec B and containing p. Let Spec Bg be a distinguished open subset

of Spec B contained in Spec Af and containing p. Then g ∈ Γ (Spec B, OX ) restricts
to an element g ′ ∈ Γ (Spec Af , OX ) = Af . The points where g and g ′ vanish on
Spec Af are the same, so
Spec Bg = Spec Af \ {[p] : g ′ ∈ p}
= Spec(Af )g ′ .
If g ′ = g ′′ /fn (g ′′ ∈ A) then Spec(Af )g ′ = Spec Afg ′′ , and we are done. □
The following easy result will be crucial for us.

5.3.2. Affine Communication Lemma. — Let P be some property enjoyed by some

affine open subsets of a scheme X, such that for any affine open subset Spec A ,→ X,
(i) if Spec A ,→ X has property P then for any f ∈ A, Spec Af ,→ X does too.
(ii) if (f1 , . . . , fn ) = A (i.e., the Spec Afi cover Spec A, see Exercise 3.5.B), and
Spec Afi ,→ X has P for all i, then so does Spec A ,→ X.
Suppose that X = ∪i∈I Spec Ai where Spec Ai has property P. Then every affine open
subset of X has P too.
We say such a property is affine-local: a property is affine-local if we can check
it on any affine cover. (For example, reducedness is an affine-local property. Do
you understand why?) Note that if U is an open subscheme of X, then U inherits
any affine-local property of X. Note also that any property that is stalk-local (a
scheme has property P if and only if all its stalks have property Q) is necessarily
affine-local (a scheme has property P if and only if all of its affine open sets have
property R, where an affine scheme has property R if and only if all its stalks have
property Q). But it is sometimes not so obvious what the right definition of Q is;
see for example the discussion of normality in the next section.

Proof. Let Spec A be an affine subscheme of X. Cover Spec A with a finite num-
ber of distinguished open sets Spec Agj , each of which is distinguished in some
Spec Ai . This is possible by Proposition 5.3.1 and the quasicompactness of Spec A
(Exercise 3.6.G(a)). By (i), each Spec Agj has P. By (ii), Spec A has P. □
By choosing the property P appropriately, we define some important properties
of schemes.

5.3.3. Proposition. — Suppose A is a ring, and (f1 , . . . , fn ) = A.

(a) If A is a Noetherian ring, then so is Afi . If each Afi is Noetherian, then so is A.
(b) Suppose B is a ring, and A is a B-algebra. (Hence Ag is a B-algebra for all
g ∈ A.) If A is a finitely generated B-algebra, then so is Afi . If each Afi is a
finitely generated B-algebra, then so is A.
We will prove these shortly (§5.3.9). But let’s first motivate you to read the
proof by giving some interesting definitions and results assuming Proposition 5.3.3
is true.

5.3.4. Important Definition. Suppose X is a scheme. If X can be covered by affine

open sets Spec A where A is Noetherian, we say that X is a locally Noetherian
scheme. If in addition X is quasicompact, or equivalently can be covered by finitely
many such affine open sets, we say that X is a Noetherian scheme. (We will see a
 © 2024 Ravi Vakil. Published by Princeton University Press. 163

number of definitions of the form “if X has this property, we say that it is locally Q;
if further X is quasicompact, we say that it is Q.”) By Exercise 5.1.C, the underlying
topological space of a Noetherian scheme is Noetherian. Hence by Exercise 3.6.U,
all open subsets of a Noetherian scheme are quasicompact.

5.3.A. E XERCISE . Show that locally Noetherian schemes are quasiseparated.

5.3.B. E XERCISE . Show that a Noetherian scheme has a finite number of irreducible
components. (Hint: Proposition 3.6.15.) Show that a Noetherian scheme has a finite
number of connected components, each a finite union of irreducible components.

5.3.C. E XERCISE . Show that a Noetherian scheme X is integral if and only if X is

nonempty and connected and all stalks OX,p are integral domains. Thus in “good
situations”, integrality is the union of local (stalks are integral domains) and global
(connected) conditions. Hint: Recall that integral = irreducible + reduced (Exer-
cise 5.2.G). If a scheme’s stalks are integral domains, then it is reduced (reducedness
is a stalk-local condition, Definition 5.2.1). If a scheme X has underlying topological
space that is Noetherian, then X has finitely many irreducible components (by the
previous exercise); if two of them meet at a point p, then OX,p is not an integral
domain. (You can readily extend this from Noetherian schemes to locally Noether-
ian schemes, by showing that a connected scheme is irreducible if and only if it
is nonempty and has a cover by open irreducible subsets. But some Noetherian
hypotheses are necessary, see [Stacks, tag 0568].)

5.3.5. Unimportant caution. The ring of global sections of a Noetherian scheme need
not be Noetherian, see Exercise 19.11.H.

5.3.6. Schemes over a given field k, or more generally over a given ring A (A-
schemes). You may be particularly interested in working over a particular field,
such as C or Q, or over a ring such as Z. Motivated by this, we define the notion
of A-scheme, or scheme over A, where A is a ring, as a scheme where all the
rings of sections of the structure sheaf (over all open sets) are A-algebras, and
all restriction maps are maps of A-algebras. (Like some earlier notions such as
quasiseparatedness, this will later in Exercise 7.3.I be properly understood as a
“relative notion”; it is the data of a morphism X → Spec A.) Suppose now X is an
A-scheme. If X can be covered by affine open sets Spec Bi where each Bi is a finitely
generated A-algebra, we say that X is locally of finite type over A, or that it is a
locally finite type A-scheme. (This is admittedly cumbersome terminology; it will
make more sense later, once we know about morphisms in §8.3.9.) If furthermore
X is quasicompact, X is (of) finite type over A, or a finite type A-scheme. Note
that a scheme locally of finite type over k or Z (or indeed any Noetherian ring) is
locally Noetherian, and similarly a scheme of finite type over any Noetherian ring
is Noetherian. As our key “geometric” examples: (i) Spec C[x1 , . . . , xn ]/I is a finite
type C-scheme; and (ii) Pn C is a finite type C-scheme. (The field C may be replaced
by an arbitrary ring A.)

5.3.D. E XERCISE .
(a) (quasiprojective implies finite type) If X is a quasiprojective A-scheme (Defini-
tion 4.5.10), show that X is of finite type over A. If A is furthermore assumed to be
Noetherian, show that X is a Noetherian scheme, and hence has a finite number of
164 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

irreducible components.
(b) Suppose U is an open subscheme of a projective A-scheme. Show that U is
locally of finite type over A. If A is Noetherian, show that U is quasicompact,
and hence quasiprojective over A, and hence by (a) of finite type over A. Show
this need not be true if A is not Noetherian. Better: give an example of an open
subscheme of a projective A-scheme that is not quasicompact, necessarily for some
non-Noetherian A. (Hint: Silly example 4.5.12.)

5.3.7. Varieties. We now make a connection to the classical language of varieties.

An affine scheme that is reduced and of finite type over k is said to be an affine
variety (over k), or an affine k-variety. A reduced (quasi)projective k-scheme is
a (quasi)projective variety (over k), or a (quasi)projective k-variety. (Warning:
in the literature, it is sometimes also assumed in the definition of variety that the
scheme is irreducible, or that k is algebraically closed.)

5.3.E. E XERCISE .
(a) Show that Spec k[x1 , . . . , xn ]/I is an affine k-variety if and only if I ⊂ k[x1 , . . . , xn ]
is a radical ideal.
(b) Suppose I ⊂ k[x0 , . . . , xn ] is a radical graded ideal. Show that Proj k[x0 , . . . , xn ]/I
is a projective k-variety. (Caution: The example of I = (x20 , x0 x1 , . . . , x0 xn ) shows
that Proj k[x0 , . . . , xn ]/I can be a projective k-variety without I being radical.)
We will not define varieties in general until §11.2.9; we will need the notion of
separatedness first, to exclude abominations like the line with the doubled origin
(Example 4.4.5). But many of the statements we will make in this section about
affine k-varieties will automatically apply more generally to k-varieties.

5.3.F. E XERCISE . Show that a point of a locally finite type k-scheme is a closed
point if and only if the residue field of the stalk of the structure sheaf at that point
is a finite extension of k. Show that on a locally finite type k-scheme, the closed
points are dense. Hint: §3.6.9. (Warning: closed points need not be dense even on
quite reasonable schemes, see Exercise 3.6.J(b).)

5.3.8. Definition. The degree of a closed point p of a locally finite type k-scheme
(e.g., a variety over k) is the degree of the field extension κ(p)/k. For example,
in A1k = Spec k[t], the point [(p(t))] (p(t) ∈ k[t] irreducible) is deg p(t). If k is
algebraically closed, the degree of every closed point is 1.

5.3.9. Proof of Proposition 5.3.3. We divide each part into (i) and (ii) following the
statement of the Affine Communication Lemma 5.3.2.
Proof of Proposition 5.3.3(a). (i) Let ϕ : A → Af be the localization map. If
I1 ⊊ I2 ⊊ I3 ⊊ · · · is a strictly increasing chain of ideals of Af , then we can
verify that J1 ⊊ J2 ⊊ J3 ⊊ · · · is a strictly increasing chain of ideals of A, where
Jj = ϕ−1 (Ij ). (We think of Jj as Ij ∩ A, except in general A needn’t inject into Afi .)
Clearly Jj is an ideal of A. If x/fn ∈ Ij+1 \ Ij where x ∈ A, then x ∈ Jj+1 , and x ∈ / Jj
(or else x(1/f)n ∈ Ij as well).
(ii) Suppose I1 ⊊ I2 ⊊ I3 ⊊ · · · is a strictly increasing chain of ideals of A. Then
for each 1 ≤ i ≤ n,

Ii,1 ⊂ Ii,2 ⊂ Ii,3 ⊂ · · ·

 © 2024 Ravi Vakil. Published by Princeton University Press. 165

is an increasing chain of ideals in Afi , where Ii,j = Ij ⊗A Afi . It remains to show

that for each j, Ii,j ⊊ Ii,j+1 for some i; the result will then follow.

Finish the proof of Proposition 5.3.3(a). (Hint: if M := Ij /Ij−1 ̸= 0,

5.3.G. E XERCISE . Q
then from M ,→ Mfi (4.1.6.1), one of the Mfi is nonzero.)
Proof of Proposition 5.3.3(b). Part (i) is clear: if A is generated over B by r1 , . . . ,
rn , then Af is generated over B by r1 , . . . , rn , 1/f. P
(ii) Here is the idea. As the fi generate A, we can write 1 = ci fi for ci ∈ A.
kj
We have generators of Afi : rij /fi , where rij ∈ A. I claim that {fi }i ∪ {ci } ∪ {rij }ij
generate A as a B-algebra. Here is why. Suppose you have any r ∈ A. Then in
Afi , we can write r as some polynomial in the rij ’s and fi , divided by some huge
power of fi . So “in each Afi , we have described r in the desired way”, except for
this annoying denominator. Now use a “partition of unity” type argument as in
the proof of Theorem 4.1.2 to combine all of these into a single expression, killing
the denominator. Show that the resulting expression you build still agrees with r in
each of the Afi . Thus it is indeed r (by the identity axiom for the structure sheaf).

5.3.H. E XERCISE . Make this argument precise.

This concludes the proof of Proposition 5.3.3. □

5.4 Normality and factoriality

5.4.1. Normality.
We can now define a property of schemes that says that they are “not too far
from smooth”, called normality, which will come in very handy. We will see later
that “locally Noetherian normal schemes satisfy Hartogs’s Lemma” (Algebraic
Hartogs’s Lemma 13.5.19 for Noetherian normal schemes): functions defined away
from a set of codimension 2 extend over that set. (We saw a first glimpse of this in
§4.4.2.) As a consequence, rational functions (defined in Exercise 5.2.I) that have no
poles (certain sets of codimension one where the function isn’t defined) are defined
everywhere. We need definitions of dimension and poles to make this precise.
See §13.8.7 and §26.3.5 for the fact that “smoothness” (really, “regularity”) implies
normality.
Recall that an integral domain A is integrally closed if the only zeros in K(A)
to any monic polynomial in A[x] must lie in A itself. The basic example is Z (see
Exercise 5.4.F for a reason). We say a scheme X is normal if all of its stalks OX,p are
normal, i.e., are integral domains, and integrally closed in their fraction fields. (So
by definition, normality is a stalk-local property.) As reducedness is a stalk-local
property (Definition 5.2.1), normal schemes are reduced. (One might say that a ring
A is a normal ring if Spec A is normal, thereby extending the notion of “integral
closure” to rings that are not integral domains.)

5.4.A. E XERCISE . Show that integrally closed domains behave well under local-
ization: if A is an integrally closed domain, and S is a multiplicative subset not
containing 0, show that S−1 A is an integrally closed domain. (Hint: assume that
166 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

xn + an−1 xn−1 + · · · + a0 = 0 where ai ∈ S−1 A has a root in the fraction field. Turn
this into another equation in A[x] that also has a root in the fraction field.)
It is no fun checking normality at every single point of a scheme. Thanks to this
exercise, we know that if A is an integrally closed domain, then Spec A is normal.
Also, for schemes that are quasicompact or locally finite type over a field, normality
can be checked at closed points, thanks to this exercise, and the fact that for such
schemes, any point is a generization of a closed point (Exercises 5.1.E and 5.3.F).
It is not true that normal schemes are integral. ` For example, the disjoint
union of two normal schemes is normal. Thus Spec k Spec k = ∼ Spec(k × k) = ∼
Spec k[x]/(x(x − 1)) is normal, but its ring of global sections is not an integral
domain.

5.4.B. E XERCISE . Show that a Noetherian scheme is normal if and only if it is the
finite disjoint union of integral Noetherian normal schemes. (Hint: Exercise 5.3.C.)
We are close to proving a useful result in commutative algebra, so we may as
well go all the way.

5.4.2. Proposition. — If A is an integral domain, then the following are equivalent.

(i) A is integrally closed.
(ii) Ap is integrally closed for all prime ideals p ⊂ A.
(iii) Am is integrally closed for all maximal ideals m ⊂ A.

Proof. Exercise 5.4.A shows that integral closedness is preserved by localization, so

(i) implies (ii). Clearly (ii) implies (iii).
It remains to show that (iii) implies (i). This argument involves a pretty con-
struction that we will use again. Suppose A is not integrally closed. We show that
there is some m such that Am is also not integrally closed. Suppose

(5.4.2.1) xn + an−1 xn−1 + · · · + a0 = 0

(with ai ∈ A) has a solution s in K(A) \ A. Let I be the ideal of denominators of s:

(5.4.2.2) I := {r ∈ A : rs ∈ A}.

(Note that I is clearly an ideal of A.) Now I ̸= A, as 1 ∈ / I. Thus there is some

maximal ideal m containing I. Then s ∈ / Am , so equation (5.4.2.1) in Am [x] shows
that Am is not integrally closed as well, as desired. □

5.4.C. U NIMPORTANT E XERCISE . If A is an integral domain, show that A = ∩Am ,

where the intersection runs over all maximal ideals of A. (We won’t use this exercise,
but it gives good practice with the ideal of denominators.)

5.4.D. U NIMPORTANT E XERCISE RELATING TO THE IDEAL OF DENOMINATORS .

One might naively hope from experience with unique factorization domains that the
ideal of denominators is principal. This is not true. As a counterexample, consider
our new friend A = k[w, x, y, z]/(wz − xy) (which we first met in Example 4.4.12,
and which we will later recognize as the cone over the quadric surface), and
w/y = x/z ∈ K(A). Show that the ideal of denominators of this element of K(A) is
(y, z).
 © 2024 Ravi Vakil. Published by Princeton University Press. 167

We will see that the I in the above exercise is not principal (Exercise 13.1.D —
you may be able to show it directly, using the fact that I is a homogeneous ideal of a
graded ring). But we will also see that in good situations (Noetherian, normal), the
ideal of denominators is “pure codimension 1” — this is the content of Algebraic
Hartogs’s Lemma 13.5.19. In its proof, §13.5.21, we give a geometric interpretation
of the ideal of denominators.

5.4.3. Factoriality.
If all the stalks of a scheme X are unique factorization domains, we say that X
is factorial. (Unimportant remark: the locus of points on an affine variety over an
algebraically closed field that are factorial is an open subset, [BGS, p. 1].)

5.4.E. E XERCISE . Show that any nonzero localization of a unique factorization

domain is a unique factorization domain.

5.4.4. Thus if A is a unique factorization domain, then Spec A is factorial. The

converse need not hold, see Exercise 5.4.N. In fact, we will see that elliptic curves
are factorial, yet no affine open set is the Spec of a unique factorization domain,
§19.11.1. Hence one can show factoriality by finding an appropriate affine cover,
but there need not be such a cover of a factorial scheme. (One might reasonably call
a ring A such that Spec A is factorial, a factorial ring; this is a strictly weaker notion
than unique factorization domain. We won’t need this terminology.)

5.4.5. Remark: How to check if a ring is a unique factorization domain. There are very
few means of checking that a Noetherian integral domain is a unique factorization
domain. Some useful ones are: (0) elementary means: rings with a Euclidean
algorithm such as Z, k[t], and Z[i]; polynomial rings over a unique factorization
domain, by Gauss’s Lemma (see e.g., [Lan, IV.2.3]). (1) the localization of a unique
factorization domain is also a unique factorization domain (Exercise 5.4.E). (2)
height 1 prime ideals are principal (Proposition 12.3.7). (3) normal and Cl = 0
(Exercise 15.5.P). (4) Nagata’s Lemma (Exercise 15.5.Q). (Caution: even if A is
a unique factorization domain, A[[t]] need not be; see [Mat2, p. 165]. The first
example, due to P. Salmon, was A = k(u)[[x, y, z]]/(z2 + x3 + uz6 ), see [Sa, Dan].)

5.4.6. Factoriality implies normality. One reason we like factoriality is that it implies
normality.

5.4.F. I MPORTANT E XERCISE . Show that unique factorization domains are integrally
closed. Hence factorial schemes are normal, and if A is a unique factorization
domain, then Spec A is normal. (However, rings can be integrally closed without
being unique factorization domains, as we will see in Exercise 5.4.L. Another
example is given without proof in Exercise 5.4.N; in that example, Spec of the ring
is factorial. A variation on Exercise 5.4.L will show that schemes can be normal
without being factorial, see Exercise 13.1.E.)

5.4.7. Examples.

5.4.G. E ASY E XERCISE . Show that the following schemes are normal: An n
k , Pk ,
Spec Z. (As usual, k is a field. Although it is true that if A is integrally closed then
A[x] is as well — see [Bo, Ch. 5, §1, no. 3, Cor. 2] or [E, Ex. 4.18] — this is not an
easy fact, so do not use it here.)
168 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

5.4.H. H ANDY E XERCISE ( YIELDING MANY ENLIGHTENING EXAMPLES LATER ).

Suppose A is a unique factorization domain with 2 invertible, and z2 − f is irre-
ducible in A[z].
(a) Show that if f ∈ A has no repeated prime factors, then Spec A[z]/(z2 − f) is
normal. Hint: B := A[z]/(z2 − f) is an integral domain, as (z2 − f) is prime in
A[z]. Suppose we have monic F(T ) ∈ B[T ] so that F(T ) = 0 has a solution α in
K(B) \ K(A). Then by replacing F(T ) by F(T )F(T ), we can assume F(T ) ∈ A[T ]. Also,
α = g + hz where g, h ∈ K(A). Now α is the solution of Q(T ) = 0 for monic
Q(T ) = T 2 − 2gT + (g2 − h2 f) ∈ K(A)[T ], so we can factor F(T ) = P(T )Q(T ) in
K(A)[T ]. By Gauss’s lemma, 2g, g2 − h2 f ∈ A. Say g = r/2, h = s/t (s and t have no
common factors, r, s, t ∈ A). Then g2 −h2 f = (r2 t2 −4s2 f)/4t2 . Then t is invertible.
(b) Show that if f ∈ A has repeated prime factors, then Spec A[z]/(z2 − f) is not
normal.

5.4.I. E XERCISE . Show that the following schemes are normal:

(a) Spec Z[x]/(x2 − n) where n is a square-free integer congruent to 3 modulo
4. Caution: the hypotheses of Exercise 5.4.H do not apply, so you will have
to do this directly. (Your argument may also show the √ result when 3 is
replaced by 2. A similar argument shows that Z[(1 + n)/2] is integrally
closed if n ≡ 1 (mod 4) is square-free.)
(b) Spec k[x1 , . . . , xn ]/(x21 + x22 + · · · + x2m ) where char k ̸= 2, n ≥ m ≥ 3.
(c) Spec k[w, x, y, z]/(wz − xy) where char k ̸= 2. This is our cone over a
quadric surface example from Example 4.4.12 and Exercise 5.4.D. Hint:
Exercise 5.4.J may help. (The result also holds for char k = 2, but don’t
worry about this.)

5.4.J. E XERCISE ( DIAGONALIZING QUADRATIC FORMS ). Suppose k is an alge-

braically closed field of characteristic not 2. (The hypothesis that k is algebraically
closed is not necessary, so feel free to deal with this more general case.)
(a) Show that any quadratic form in n variables can be “diagonalized” by changing
coordinates to be a sum of at most n squares (e.g., uw − v2 = ((u + w)/2)2 + (i(u −
w)/2)2 + (iv)2 ), where the linear forms appearing in the squares are linearly independent.
(Hint: use induction on the number of variables, by “completing the square” at
each step.)
(b) Show that the number of squares appearing depends only on the quadratic form.
For example, x2 + y2 + z2 cannot be written as a sum of two squares. (Possible
approach: given a basis x1 , . . . , xn of the linear forms, write the quadratic form as
 
x1

 . 
x1 ··· xn M  .. 
xn

where M is a symmetric matrix. Determine how M transforms under a change of

basis, and show that the rank of M is independent of the choice of basis.)
The rank of the quadratic form is the number of (“linearly independent”)
squares needed. If the number of squares equals the number of variables, the
quadratic form is said to be full rank or (of) maximal rank.
 © 2024 Ravi Vakil. Published by Princeton University Press. 169

5.4.K. E ASY E XERCISE ( RINGS CAN BE INTEGRALLY CLOSED BUT √ NOT UNIQUE
FACTORIZATION DOMAINS , ARITHMETIC VERSION ). Show that Z[ −5] is integrally

√ but not a√unique factorization domain. (Hints: Exercise 5.4.I(a) and 2 × 3 =

closed
(1 + −5)(1 − −5).)

5.4.L. E ASY E XERCISE ( RINGS CAN BE INTEGRALLY CLOSED BUT NOT UNIQUE
FACTORIZATION DOMAINS , GEOMETRIC VERSION ). Suppose char k ̸= 2. Let
A = k[w, x, y, z]/(wz − xy), so Spec A is the cone over the smooth quadric surface
(cf. Exercises 4.4.12 and 5.4.D).
(a) Show that A is integrally closed. (Hint: Exercises 5.4.I(c) and 5.4.J.)
(b) Show that A is not a unique factorization domain. (Clearly wz = xy. But why
are w, x, y, and z irreducible? Hint: A is a graded integral domain. Show that if a
homogeneous element factors, the factors must be homogeneous.)
The previous two exercises look similar, but there is a difference. The cone
over the quadric surface is normal (by √ Exercise 5.4.L) but not factorial; see Exer-
cise 13.1.E. On the other hand, Spec Z[ −5] is factorial — all of its stalks are unique
√
factorization domains. (You will later be able to show this by showing that Z[ −5]
is a Dedekind domain, §13.5.11, whose stalks are necessarily unique factorization
domains by Theorem 13.5.8(f).)

5.4.M. E XERCISE . Suppose A is a k-algebra, and l/k is a finite extension of fields.

(Your proof might not use finiteness; this hypothesis is included to avoid distraction
by infinite-dimensional vector spaces.) Show that if A ⊗k l is a normal integral
domain, then A is a normal integral domain as well. (Although we won’t need this,
a version of the converse is true if l/k is separable, [Gr-EGA, IV2 .6.14.2].) Hint: fix
a k-basis for l, b1 = 1, . . . , bd . Explain why 1 ⊗ b1 , . . . , 1 ⊗ bd forms a free A-basis
for A ⊗k l. Explain why we have injections

A / K(A)

 
A ⊗k l / K(A) ⊗k l.

Show that K(A) ⊗k l = K(A ⊗k l). (Idea: A ⊗k l ⊂ K(A) ⊗k l ⊂ K(A ⊗k l). Why is
K(A) ⊗k l a field?) Show that (A ⊗k l) ∩ K(A) = A. Now assume P(T ) ∈ A[T ] is
monic and has a root α ∈ K(A), and proceed from there.

5.4.N. E XERCISE (UFD- NESS IS NOT AFFINE - LOCAL ). Let A = (Q[x, y]x2 +y2 )0
denote the homogeneous degree 0 part of the ring Q[x, y]x2 +y2 . In other words,
it consists of quotients f(x, y)/(x2 + y2 )n , where f has pure degree 2n. Show
2 2
that the distinguished open sets D( x2x+y2 ) and D( x2y+y2 ) cover Spec A. (Hint:
the sum of those two fractions is 1.) Show that A x2 and A y2 are unique
x2 +y2 x2 +y2
factorization domains. (Hint: show that both rings are isomorphic to Q[t]t2 +1 ; this
is a localization of the unique factorization domain Q[t].) Finally, show that A is
not a unique factorization domain. Possible hint:
2
x2 y2
   
xy
= .
x 2 + y2 x 2 + y2 x2 + y2
170 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

(This is generalized in Exercise 15.5.H . It is also related to Exercise 15.5.M.)

√
Number theorists may prefer the example of Exercise 5.4.K: Z[ −5] is not a
unique factorization domain, but it turns out that you can cover it with two affine
open subsets D(2) and D(3), each corresponding
√ to√unique factorization domains.
(For number theorists: to show that Z[ −5]2 and √ Z[ −5]3 are unique factorization
domains, first show that the class group of Z[ −5] is Z/2 using the geometry√ of
numbers,
√ as in [Ar3, Ch. 11, Thm. 7.9]. Then show that the ideals (1 + √−5, 2) and
(1 + −5, 3) are not principal, using the usual norm in C.) The ring Z[ −5] is an
example of a Dedekind domain, as we will discuss in §13.5.11.

5.4.8. Remark. For an example of k-algebra A that is not a unique factorization

domain, but becomes one after a particular field extension, see Exercise 15.5.J.
 CHAPTER 6

Rings are to modules as schemes are to ...

When considering the notion of rings for the first time, one is quickly led to
define the notion of module. For example, any ring morphism R → S expresses S
as an R-module. An ideal I of a ring R is also an R-module, as is the quotient R/I;
and the exact sequence 0 → I → R → R/I → 0 is a first example of the utility of the
abelian category structure of ModR .
With a scheme X, you may think that we already have the right analog of
modules: the category of OX -modules ModOX , which also forms an abelian category
(§2.6). But the right analog for a scheme X turns out to be better-behaved subset
(subcategory!) of ModOX , called quasicoherent OX -modules, or quasicoherent sheaves.

6.1 Quasicoherent sheaves

Given an A-module M, we defined an O-module M f on Spec A in §4.1.5 — the
sections over D(f) were Mf . (Now is a good time to complete the exercises in §4.1.5
if you have not done so already.) These are our “local models” for quasicoherent
sheaves. In the same way that a scheme is defined by “gluing together rings”, a
quasicoherent sheaf over that scheme is obtained by “gluing together modules over
those rings”.

6.1.1. Definition: the category of quasicoherent sheaves. If X is a scheme, then an

OX -module F is a quasicoherent sheaf if for every affine open subset Spec A ⊂ X,
∼M
F |Spec A = f for some A-module M. We make quasicoherent sheaves on a scheme
X into category QCohX by taking the morphisms to be morphisms as OX -modules.

6.1.2. Theorem. — Let X be a scheme, and F an OX -module. Suppose P is the property

∼M
of affine open subschemes Spec A ⊂ X that F |Spec A = f for some A-module M. Then P
satisfies the two hypotheses of the Affine Communication Lemma 5.3.2.
Thus to check quasicoherence of an O-module, it suffices to check a collec-
tion of affine open sets covering X. For example, M
f is a quasicoherent sheaf on
Spec A. Exercise 4.1.G shows that the category ModA and the category QCohSpec A
∼
are essentially the same (we have described an equivalence ModA ←→ QCohSpec A ).

Proof. As usual, the first hypothesis of the Affine Communication Lemma 5.3.2 is
easier: if Spec A has property P, then so does the distinguished open Spec Af : if M

This material is published by Princeton University Press as The Rising Sea: Foundations of Algebraic
Geometry by Ravi Vakil. This version is free to view and download for personal use only. Not for
redistribution, resale, or use in derivative works. © 2024 Ravi Vakil.

171
172 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

is an A-module, then M| f Spec A = ∼ M

gf as sheaves of OSpec A -modules (both sides
f f
agree on the level of distinguished open sets and their restriction maps).
We next show the second hypothesis of the Affine Communication Lemma 5.3.2.
We are given:
• an OSpec A -module F ,
• elements f1 , . . . , fn generating A,
• Afi -modules Mi ,
∼
• and isomorphisms ϕi : F |D(fi ) −→ M fi .
The isomorphisms
F |D(fi )∩D(fj )
∼ ∼
v (
fi |D(f )∩D(f )
M fj |D(f )∩D(f )
M
i j i j

∼
yield (by taking sections over D(fi ) ∩ D(fj )) isomorphisms ϕij : (Mi )fj −→ (Mj )fi
of Afi fj -modules, satisfying the cocycle condition (2.5.4.1). As suggestive notation,
let Mij := (Mi )fj (identified with (Mj )fi via ϕij ).
f ←→ ∼
We seek an A-module M with an isomorphism M F , and (4.1.6.1) sug-
gests that we should define M by

(6.1.2.1) 0 /M / M1 × · · · × Mn γ
/ M12 × M13 × · · · × M(n−1)n .

(The discussion at (4.1.6.1) suggests how the map γ should be defined, with ap-
propriate signs.) Translation: M = ker(γ). If we can describe an isomorphism
fD(f ) −→∼ fi on each D(fi ) (or equivalently, an isomorphism Mf −→ ∼
M i
M i
Mi of
Afi -modules), such that triangle

M|
f D(f )∩D(f )
i j

∼ ∼

w '
ϕij
/M
g
fi |D(f )∩D(f )
M fj |D(f )∩D(f )
i j ∼ i j

of sheaves on Spec Afi fj commutes (for all i, j), then Exercise 2.5.E (on gluing
f ←→ ∼
sheaves) gives our desired isomorphism M F.
∼
By Exercise 4.1.G, we need only describe an isomorphism Mfi −→ Mi such
that the triangles
(6.1.2.2) Mfi fj
∼ ∼

z ϕij $
(Mi )fj / (Mj )f
i

commute.
For notational convenience, we assume i = 1. Because localization is exact,
from (6.1.2.1) we have an exact sequence
γf1
0 / Mf / (M1 )f × · · · × (Mn )f / (M12 )f × (M13 )f × · · · × (M(n−1)n )f
1 1 1 1 1 1
 © 2024 Ravi Vakil. Published by Princeton University Press. 173

We will show that

(6.1.2.3)
γf1
0 / M1 β
/ (M1 )f × · · · × (Mn )f / (M12 )f × (M13 )f × · · · × (M(n−1)n )f
1 1 1 1 1

is an exact sequence (do you see how β should be defined?), yielding our desired
∼
isomorphism Mf1 −→ M1 . We rewrite (6.1.2.3) as
(6.1.2.4) 0 / M1 / (M1 )f × (M1 )f × · · · × (M1 )f
1 2 n

α / (M1 )f f × · · · × (M1 )f fn × (M1 )f f × · · · × (M1 )f fn .

1 2 1 2 3 n−1

But this is precisely the exact sequence (4.1.6.1), except the ring A is replaced by
Af1 , and the module M is replaced by M1 .

6.1.A. E XERCISE . Tie up the last loose end: Why does the triangle (6.1.2.2) com-
mute?
□
⋆⋆ Remark for experts: The proof of Theorem 6.1.2, phrased slightly more care-
fully, shows that quasicoherent sheaves satisfy faithfully flat descent.

6.1.3. Stray concluding topics.

6.1.B. U NIMPORTANT E XERCISE ( NOT EVERY OX - MODULE IS A QUASICOHERENT

SHEAF ).
(a) Suppose X = Spec k[t]. Let F be the skyscraper sheaf supported at the origin
[(t)], with group k(t) and the usual k[t]-module structure. Show that this is an
OX -module that is not a quasicoherent sheaf. (More generally, if X is an integral
scheme, and p ∈ X is not the generic point, we could take the skyscraper sheaf at p
with group the function field of X. Except in silly circumstances, this sheaf won’t
be quasicoherent.) See Exercises 9.1.G and 6.2.H for more (pathological) examples
of OX -modules that are not quasicoherent.
(b) Suppose X = Spec k[t]. Let F be the skyscraper sheaf supported at the generic
point [(0)], with group k(t). Give this the structure of an OX -module. Show that
this is a quasicoherent sheaf. Describe the restriction maps in the distinguished
topology of X. (Remark: Your argument will apply more generally, for example
when X is an integral scheme with generic point η, and F is the skyscraper sheaf
iη,∗ K(X).)

6.1.4. Torsion-free sheaves (a stalk-local condition) and torsion sheaves. An A-

module M is said to be torsion-free if am = 0 implies that either a is a zerodivisor
in A or m = 0.
In the case where A is an integral domain, which is basically the only context in
which we will use this concept, the definition of torsion-freeness can be restated as
am = 0 only if a = 0 or m = 0. In this case, the torsion submodule of M, denoted
Mtors , consists of those elements of M annihilated by some nonzero element of
A. (If A is not an integral domain, this construction needn’t yield an A-module.)
Clearly M is torsion-free if and only if Mtors = 0. We say a module M over an
integral domain A is torsion if M = Mtors ; this is equivalent to M ⊗A K(A) = 0.
If X is a scheme, then an OX -module F is said to be torsion-free if Fp is a
torsion-free OX,p -module for all p.
174 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

6.1.5. Definition: torsion quasicoherent sheaves on reduced schemes. Motivated by the

definition of Mtors above, we say that a quasicoherent sheaf on a reduced scheme
is torsion if its stalk at the generic point of every irreducible component is 0. We
will mainly use this for coherent sheaves on regular curves, where this notion is
very simple indeed (see Exercise 14.3.F(b)), but in the literature it comes up in more
general situations.

6.2 Characterizing quasicoherence using the distinguished affine

base

Because quasicoherent sheaves are locally of a very special form, in order

to “know” a quasicoherent sheaf, we need only know what the sections are over
every affine open set, and how to restrict sections from an affine open set U to a
distinguished affine open subset of U. We make this precise by defining what we
will call the distinguished affine base of the Zariski topology — not a base in the
usual sense. This is a refinement of the notion of a sheaf on base. The point of this
discussion is to give a useful characterization of quasicoherence, but you may wish
to just jump to §6.2.3.
The open sets of the distinguished affine base are the affine open subsets of
X. We have already observed that this forms a base. But forget that fact. We like
distinguished open sets Spec Af ,→ Spec A, and we don’t really understand open
embeddings of one random affine open subset in another. So we just remember the
“nice” inclusions.

6.2.1. Definition. The distinguished affine base of a scheme X is the data of the
affine open sets and the distinguished inclusions.
In other words, we remember only some of the open sets (the affine open sets),
and only some of the morphisms between them (the distinguished morphisms). For
experts: if you think of a topology as a category (the category of open sets), we
have described a subcategory.
We can define a sheaf on the distinguished affine base in the obvious way: we
have a set (or abelian group, or ring) for each affine open set, and we know how to
restrict to distinguished open sets. (You should think through the statement of the
identity and gluability axioms yourself.)
Given a sheaf F on X, we get a sheaf on the distinguished affine base. You can
guess where we are going: we will show that all the information of the sheaf is
contained in the information of the sheaf on the distinguished affine base.
As a warm-up, we can recover stalks as follows. (We will be implicitly using
only the following fact. We have a collection of open subsets, and some inclusions
among these subsets, such that if we have any p ∈ U, V where U and V are in our
collection of open sets, there is some W containing p, and contained in U and V
such that W ,→ U and W ,→ V are both in our collection of inclusions. In the case
we are considering here, this is the key Proposition 5.3.1 that given any two affine
open sets Spec A, Spec B in X, Spec A ∩ Spec B could be covered by affine open sets
that were simultaneously distinguished in Spec A and Spec B. In fancy language:
the category of affine open sets, and distinguished inclusions, forms a filtered set.)
 © 2024 Ravi Vakil. Published by Princeton University Press. 175

The stalk Fp is the colimit colim(f ∈ F (U)) where the colimit is over all open
sets contained in X. We compare this to colim(f ∈ F (U)) where the colimit is
over all affine open sets, and all distinguished inclusions. You can check that the
elements of one correspond to elements of the other. (Think carefully about this!)

6.2.A. E XERCISE . Show that a section of a sheaf on the distinguished affine base is
determined by the section’s germs.

6.2.2. Theorem. —
(a) A sheaf on the distinguished affine base F b determines a unique (up to unique
isomorphism) sheaf F which when restricted to the affine base is F b . (Hence
if you start with a sheaf, and take the sheaf on the distinguished affine base, and
then take the induced sheaf, you get the sheaf you started with.)
(b) A morphism of sheaves on a distinguished affine base uniquely determines a
morphism of sheaves.
(c) An OX -module “on the distinguished affine base” yields an OX -module.
This proof is identical to our argument of §2.5 showing that sheaves are (es-
sentially) the same as sheaves on a base, using the “sheaf of compatible germs”
construction. The main reason for repeating it is to let you see that all that is needed
is for the open sets to form a filtered set (or in the current case, that the category of
open sets and distinguished inclusions is filtered).
For experts: (a) and (b) are describing an equivalence of categories between
sheaves on the Zariski topology of X and sheaves on the distinguished affine base
of X.

Proof. (a) Suppose F b is a sheaf on the distinguished affine base. Then we can
define stalks.
For any open set U of X, define the sheaf of compatible germs

F (U) := {(fp ∈ Fpb )p∈U : for all p ∈ U, there exists Up with p ∈ Up ⊂ U,

s ∈ F b (Up ) such that sq = fq for all q ∈ Up }

where each Up is in our base, and sq means “the germ of s at q”. (As usual, those
who want to worry about the empty set are welcome to.)
This really is a sheaf: convince yourself that we have restriction maps, identity,
and gluability, really quite easily.
I next claim that if U is in our base, that F (U) = F b (U). We clearly have a map
F (U) → F (U). This is an isomorphism on stalks, and hence an isomorphism by
b

Exercise 2.4.D.

6.2.B. E XERCISE . Prove (b) (cf. Exercise 2.5.C).

6.2.C. E XERCISE . Prove (c) (cf. Remark 2.5.3).

The proof of Theorem 6.2.2 is now complete. □

6.2.3. A characterization of quasicoherent sheaves in terms of distinguished

inclusions. We use this perspective to give a useful characterization of quasico-
herent sheaves among OX -modules. Suppose F is an OX -module, and Spec Af ,→
Spec A ⊂ X is a distinguished open subscheme of an affine open subscheme of
176 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

X. Let ϕ : Γ (Spec A, F ) → Γ (Spec Af , F ) be the restriction map. The source of ϕ

is an A-module, and the target is an Af -module, so by the universal property of
localization (Exercise 1.2.D), ϕ naturally factors as:

Γ (Spec A, F )
ϕ
/ Γ (Spec Af , F )
6
⊗A Af α
(
Γ (Spec A, F )f

6.2.D. V ERY IMPORTANT EXERCISE . Show that an OX -module F is quasicoherent

if and only if for each such distinguished Spec Af ,→ Spec A, α is an isomorphism.
Thus a quasicoherent sheaf is (equivalent to) the data of one module for each
affine open subset (a module over the corresponding ring), such that the module
over a distinguished open set Spec Af is given by localizing the module over
Spec A.

6.2.E. E XERCISE ( GOOD PRACTICE : THE SHEAF OF NILPOTENTS ). If A is a ring, and

f ∈ A, show that N(Af ) =∼ N(A)f . Use this to define/construct the quasicoherent
sheaf of nilpotents on any scheme X. This is an example of an ideal sheaf (of OX ).

6.2.4. Tensor products. Tensor product in the category of quasicoherent sheaves on

X can also be cleanly described in terms of affine open sets.

6.2.F. E XERCISE . If F and G are quasicoherent sheaves on a scheme X, show

that F ⊗ G is a quasicoherent sheaf described by the following information: If
Spec A ⊂ X is an affine open subset, and Γ (Spec A, F ) = M and Γ (Spec A, G ) = N,
then Γ (Spec A, F ⊗ G ) = M ⊗A N, and the restriction map Γ (Spec A, F ⊗ G ) →
Γ (Spec Af , F ⊗ G ) is precisely the localization map M ⊗A N → (M ⊗A N)f = ∼
∼
Mf ⊗Af Nf . (We are using the algebraic fact that (M ⊗A N)f = Mf ⊗Af Nf . You can
prove this by universal property if you want, or by using the explicit construction.)

Note that thanks to the machinery behind the distinguished affine base, sheafi-
fication is taken care of. This is a feature we will use often: constructions involving
quasicoherent sheaves that involve sheafification for general sheaves don’t require
sheafification when considered on the distinguished affine base. Along with the fact
that injectivity, surjectivity, kernels, tensor products and so on may be computed
on affine opens, this is the reason that it is particularly convenient to think about
quasicoherent sheaves in terms of affine open sets. (There is a slight caveat in the
case of Hom , see Exercise 14.3.A.)

6.2.5. ⋆⋆ Elegant side remark, but perhaps too fancy for now. In Exercise 6.2.F, the
tensor product is in the category of quasicoherent sheaves. But in fact this is even
the tensor product in the category of OX -modules. To show that “F ⊗QCohX G
satisfies the universal property of F ⊗OX G ”, first show the following. Let M be
a quasicoherent sheaf on X, and N any OX -module. Let Spec A ⊂ X be an affine
open subset, and M = M (Spec A). Then the “global sections” map

HomSpec A (M |Spec A , N |Spec A ) / HomA (M, N (Spec A))

 © 2024 Ravi Vakil. Published by Princeton University Press. 177

is an isomorphism. Our slogan: on an affine scheme, maps from a quasicoherent

sheaf to any O-module are determined by global sections. You will hear an echo of
this slogan in the parenthetical comment in Exercise 7.3.G.

6.2.6. Definition. Given a section s of F and a section t of G , we have a section

s ⊗ t of F ⊗ G .

6.2.7. Xf and the Qcqs lemma. The next result will be useful in the future in
showing certain constructions yield quasicoherent sheaves.

6.2.8. Definition: Xf . If X is a scheme, and f ∈ Γ (X, OX ) is a function, let Xf ⊂ X be

the subset on which f doesn’t vanish. By Exercise 4.3.F(a), Xf is open. (We avoid
the notation D(f) to avoid any suggestion that X is affine.)

6.2.9. The Qcqs Lemma. — Suppose X is a quasicompact and quasiseparated scheme, F

is a quasicoherent sheaf on X, and f ∈ Γ (X, OX ) is a function on X. Then the restriction
map
resXf ⊂X : Γ (X, F ) / Γ (Xf , F )
∼
is precisely localization: there is an isomorphism Γ (X, F )f −→ Γ (Xf , F ) making the
following diagram commute.
resXf ⊂X
Γ (X, F ) / Γ (Xf , F )
8
∼
⊗Γ (X,OX ) (Γ (X,OX )f )
%
Γ (X, F )f

6.2.G. I MPORTANT EXERCISE . Prove the Qcqs Lemma 6.2.9. Hint: The “quasi-
compact quasiseparated” hypothesis translates precisely to “covered by a finite
number of affine open sets, the pairwise intersection of which is also covered by a
finite number of affine open sets”. Apply the exact functor ⊗Γ (X,OX ) Γ (X, OX )f to
the exact sequence
0 / Γ (X, F ) / ⊕i Γ (Ui , F ) / ⊕Γ (Uijk , F ),

where the Ui form a finite affine cover of X and Uijk form a finite affine cover of
Ui ∩ Uj .

6.2.H. L ESS IMPORTANT EXERCISE . Give a counterexample to show that the Qcqs
Lemma 6.2.9 is false without the quasicompactness hypothesis. (Possible hint: take
an infinite disjoint union of affine schemes. The key idea is that infinite products
do not commute with localization.)

6.2.10. ⋆⋆ Grothendieck topologies. The distinguished affine base isn’t a topology

in the usual sense — the union of two affine sets isn’t necessarily affine, for example.
It is however a first new example of a generalization of a topology — the notion of
a site or a Grothendieck topology. We give the definition to satisfy the curious, but
we certainly won’t use this notion. (For a clean statement, see [Stacks, tag 00VH];
our discussion here is intended only as motivation.) The idea is that we should
abstract away only those notions we need to define sheaves. We need the notion
of open set, but it turns out that we won’t even need an underlying set, i.e., we
178 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

won’t even need the notion of points! Let’s think through how little we need. For
our discussion of sheaves to work, we needed to know what the open sets were,
and what the (allowed) inclusions were, and these should “behave well”, and in
particular the data of the open sets and inclusions should form a category. (For
example, the composition of an allowed inclusion with another allowed inclusion
should be an allowed inclusion — in the distinguished affine base, a distinguished
open set of a distinguished open set is a distinguished open set.) So we just require
the data of this category. At this point, we can already define presheaf (as just a
contravariant functor from this category of “open sets”). We saw this idea earlier in
Exercise 2.2.A.
In order to extend this definition to that of a sheaf, we need to know more
information. We want two open subsets of an open set to intersect in an open set,
so we want the category to be closed under fibered products (cf. Exercise 1.2.O). For the
identity and gluability axioms, we need to know when some open sets cover another,
so we also remember this as part of the data of a Grothendieck topology. The data
of the coverings satisfy some obvious properties. Every open set covers itself (i.e.,
the identity map in the category of open sets is a covering). Coverings pull back: if we
have a map Y → X, then any cover of X pulls back to a cover of Y. Finally, a cover of a
cover should be a cover. Such data (satisfying these axioms) is called a Grothendieck
topology or a site. (There are useful variants of this definition in the literature. Again,
we are following [Stacks].) We can define the notion of a sheaf on a Grothendieck
topology in the usual way, with no change. A topos is a scary name for a category
of sheaves of sets on a Grothendieck topology.
Grothendieck topologies are used in a wide variety of contexts in and near
algebraic geometry. Étale cohomology (using the étale topology), a generalization
of Galois cohomology, is a central tool, as are more general flat topologies, such
as the smooth topology. The definition of a Deligne-Mumford or Artin stack uses
the étale and smooth topology, respectively. Tate developed a good theory of non-
archimedean analytic geometry over totally disconnected ground fields such as Qp
using a suitable Grothendieck topology. Work in K-theory (related for example to
Voevodsky’s work) uses exotic topologies.

6.3 Quasicoherent sheaves form an abelian category

Morphisms from one quasicoherent sheaf on a scheme X to another are defined

to be just morphisms as OX -modules. In this way, the quasicoherent sheaves on a
scheme X form a category, denoted QCohX . (By definition it is a full subcategory of
ModOX .) We now show that quasicoherent sheaves on X form an abelian category.
When you show that something is an abelian category, you have to check
many things, because the definition has many parts. However, if the objects you
are considering lie in some ambient abelian category, then it is much easier. You
have seen this idea before: there are several things you have to do to check that
something is a group. But if you have a subset of group elements, it is much easier
to check that it forms a subgroup.
You can look at back at the definition of an abelian category, and you will see
that in order to check that a subcategory is an abelian subcategory, it suffices to
check only the following:
 © 2024 Ravi Vakil. Published by Princeton University Press. 179

(i) 0 is in the subcategory

(ii) the subcategory is closed under finite sums
(iii) the subcategory is closed under kernels and cokernels
In our case of QCohX ⊂ ModOX , the first two are cheap: 0 is certainly quasico-
herent, and the subcategory is closed under finite sums: if F and G are sheaves on
X, and over Spec A, F = ∼M f and G = ∼ N,
e then F ⊕ G = ∼M ⊕ N (Do you see why?),
^
so F ⊕ G is a quasicoherent sheaf.
We now check (iii), using the characterization of Important Exercise 6.2.D.
Suppose α : F → G is a morphism of quasicoherent sheaves. Then on any affine
open set U, where the morphism is given by β : M → N, define (ker α)(U) = ker β
and (coker α)(U) = coker β. Then these behave well under inversion of a single
element: if
0 /K /M /N /P /0
is exact, then so is
0 / Kf / Mf / Nf / Pf / 0,

from which (ker β)f = ∼ ker(βf ) and (coker β)f =∼ coker(βf ). Thus both of these
define quasicoherent sheaves. Moreover, by checking stalks, they are indeed the
kernel and cokernel of α (exactness can be checked stalk-locally). Thus the quasico-
herent sheaves indeed form an abelian category.

6.3.A. E XERCISE . Show that a sequence of quasicoherent sheaves F → G → H on

X is exact if and only if it is exact on every open set in any given affine cover of X.
(In particular, taking sections over an affine open Spec A is an exact functor from
the category of quasicoherent sheaves on X to the category of A-modules. Recall
that taking sections is only left-exact in general, see §2.6.G.) Thus we may check
injectivity or surjectivity of a morphism of quasicoherent sheaves by checking on
an affine cover of our choice.
Caution: If 0 → F → G → H → 0 is an exact sequence of quasicoherent
sheaves, then for any open set
0 / F (U) / G (U) / H (U)

is exact, and exactness on the right is guaranteed to hold only if U is affine. (To set
you up for cohomology: whenever you see left-exactness, you expect to eventually
interpret this as a start of a long exact sequence. So we are expecting H1 ’s on the
right, and now we expect that H1 (Spec A, F ) = 0. This will indeed be the case.)

6.3.B. L ESS IMPORTANT EXERCISE ( CONNECTION TO ANOTHER DEFINITION , AND

QUASICOHERENT SHEAVES ON RINGED SPACES IN GENERAL ). Show that an OX -
module F on a scheme X is quasicoherent if and only if there exists an open cover
by Ui such that on each Ui , F |Ui is isomorphic to the cokernel of a map of two
“free sheaves”:
OU⊕I / O ⊕J / F |U /0
i Ui i

is exact. This is the definition of a quasicoherent sheaf on a ringed space in general.

It is useful in many circumstances, for example in complex analytic geometry.
180 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

6.4 Finite type quasicoherent, finitely presented, and coherent

sheaves

Here are three natural finiteness conditions on an A-module M. If A is a

Noetherian ring, which is the case that almost all of you will ever care about, they
are all the same.
The first is the simplest: a module could be finitely generated. In other words,
there is a surjection A⊕p → M → 0.
The second is reasonable too. It could be finitely presented — it could have a
finite number of generators with a finite number of relations. Translation: there
exists a finite presentation, i.e., an exact sequence

A⊕q / A⊕p /M / 0.

6.4.1. The third notion is frankly a bit surprising. We say that an A-module M is
coherent if (i) it is finitely generated, and (ii) for any map A⊕p → M (not necessarily
surjective!), the kernel is finitely generated.

6.4.2. Proposition. — If A is Noetherian, then these three definitions are the same.

Proof. Clearly coherent implies finitely presented, which in turn implies finitely
generated. So suppose M is finitely generated. Take any α : A⊕p → M. Then ker α
is a submodule of a finitely generated module over A, and is thus finitely generated
by Exercise 3.6.Y. Thus M is coherent. □
Hence most people can think of these three notions as the same.

6.4.3. Proposition. — The coherent A-modules form an abelian subcategory of the

category of A-modules.
The proof in general is a series of short exercises in §6.7. You should try them
only if you are particularly curious.
Proof if the ring A is Noetherian. Recall from our discussion at the start of §6.3 that
we must check three things:
(i) The 0-module is coherent.
(ii) The category of coherent modules is closed under finite sums.
(iii) The category of coherent modules is closed under kernels and cokernels.
The first two are clear. For (iii), suppose that f : M → N is a map of finitely
generated modules. Then coker f is finitely generated (it is the image of N), and
ker f is too (it is a submodule of a finitely generated module over a Noetherian ring,
Exercise 3.6.Y). □

6.4.4. Important remark: Finitely generated modules over a principal ideal domain. We
record for future reference that finitely generated modules over a principal ideal
domain are finite direct sums of cyclic modules (see for example [DF, §12.1, Thm. 5]).
We only mention it here because it needs mentioning somewhere.

6.4.5. We now extend the definitions of these three classes of modules to quasico-
herent sheaves.
 © 2024 Ravi Vakil. Published by Princeton University Press. 181

6.4.A. E XERCISE ( FINITE GENERATION SATISFIES THE HYPOTHESES OF THE A FFINE

C OMMUNICATION L EMMA 5.3.2).
(a) If f ∈ A, and M is a finitely generated A-module, show that Mf is a finitely
generated Af -module.
(b) If (f1 , . . . , fn ) = A, and Mfi is a finitely generated Afi -module for all i, show
that M is a finitely generated A-module.
For the finite presentation case, the following result is handy, and versions
apply in other situations (see Lemma 8.3.R for the “algebra” version). For the
longest time I didn’t even suspect such a thing could be true.

6.4.6. Lemma: Finitely presented implies always finitely presented (module

version). — Suppose M is a finitely presented A-module (for some surjection from a
finite free module, the kernel is finitely generated), and ϕ : A⊕p → M is any surjection.
Then ker ϕ is finitely generated.

Proof. Choose a finite presentation of M:

⊕rk=1 Afk
β
/ ⊕n Axi α /M /0
i=1

so α(x1 ), . . . , α(xn ) generate M. Write the surjection of the hypothesis as ϕ : ⊕p

j=1
Ayj → M. Choose lifts gj ∈ ⊕Axi of ϕ(yj ) ∈ M. Then we can write
M = ⊕Axi / (β(f1 ), . . . , β(fr ))
= (⊕Axi ⊕ Ayj ) / (β(f1 ), . . . , β(fr ), y1 − g1 , . . . , yp − gp )
As the {ϕ(yj )} generate M, for each i = 1, . . . , n we can choose hi ∈ ⊕p
j=1 Ayj (i.e.,
p
a map h : ⊕n
i=1 Ax i → ⊕ j=1 Ay j ) so that α(xi ) = ϕ(h i ). Thus
M = (⊕Axi ⊕ Ayj ) / (β(f1 ), . . . , β(fr ), y1 − g1 , . . . , yp − gp , x1 − h1 , . . . , xn − hn )
= ⊕Ayj / (h(β(f1 )), . . . , h(β(fr )), y1 − h(g1 ), . . . , yp − h(gp ))
□

6.4.B. E XERCISE ( FINITE PRESENTATION SATISFIES THE HYPOTHESES OF THE A FFINE

C OMMUNICATION L EMMA 5.3.2).
(a) If f ∈ A, and if M is a finitely presented A-module, show that Mf is a finitely
presented Af -module.
(b) If (f1 , . . . , fn ) = A, and Mfi is a finitely presented Afi -module for all i, show
that M is a finitely presented A-module. (Hint: use Exercise 6.4.A(b) to show first
that M is finitely generated, so we may write M = coker(α : N ,→ A⊕n ) for some
submodule N ⊂ A⊕n . We wish to show that N is finitely generated. Localize at
the fi , use Lemma 6.4.6 (“finitely presented implies always finitely presented”) to
show Nfi is a finitely generated Afi module, then apply Exercise 6.4.A(b) to N.)

6.4.C. E XERCISE ( COHERENCE SATISFIES THE HYPOTHESES OF THE A FFINE C OM -

MUNICATION L EMMA 5.3.2).
(a) If f ∈ A, and M is a coherent A-module, show that Mf is a coherent Af -module.
(b) If (f1 , . . . , fn ) = A, and Mfi is a coherent Afi -module for all i, show that M is
a coherent A-module. (Hint: if ϕ : A⊕p → M, then (ker ϕ)fi = ker(ϕfi ), which is
finitely generated for all i. Then apply Exercise 6.4.A(b).)
182 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

6.4.7. Definition. A quasicoherent sheaf F is finite type (resp., finitely presented,

coherent) if for every affine open Spec A, Γ (Spec A, F ) is a finitely generated (resp.,
finitely presented, coherent) A-module. Note that coherent sheaves are always
finite type, and that on a locally Noetherian scheme, all three notions are the same
(by Proposition 6.4.2). Proposition 6.4.3 implies that the coherent sheaves on X form
an abelian category, which we denote CohX .
By the Affine Communication Lemma 5.3.2, and Exercises 6.4.A, 6.4.B, and 6.4.C,
it suffices to check “finite typeness” (resp., finite presentation, coherence) on the
open sets in a single affine cover.

6.4.8. Warning. It is not uncommon in the later literature to incorrectly define

coherent as finitely generated. Please only use the correct definition, as the wrong
definition causes confusion. Besides doing this for reasons of honesty, it will also
help you see what hypotheses are actually necessary to prove things. And that
always helps you remember what the proofs are — and hence why things are true.

6.4.9. Why coherence? Proposition 6.4.3 is a good motivation for the definition of
coherence: it gives a small (in a non-technical sense) abelian category in which we
can think about vector bundles.
There are two sorts of people who should care about the details of this definition,
rather than living in a Noetherian world where coherent means finite type. Complex
geometers should care. They consider complex-analytic spaces with the classical
topology. One can define the notion of coherent OX -module in a way analogous to
this (see [Se1, Def. 2]). Then Oka’s Theorem states that the structure sheaf of Cn
(hence of any complex manifold) is coherent, and this is very hard (see [GR, §2.5]
or [Rem, §7.2]).
The second sort of people who should care are the sort of arithmetic people
who may need to work with non-Noetherian rings, see §3.6.21, or work in non-
archimedean analytic geometry.

6.4.10. Remark: Quasicoherent and coherent sheaves on ringed spaces. We will discuss
quasicoherent and coherent sheaves on schemes, but they can be defined more
generally (see Exercise 6.3.B for quasicoherent sheaves, and [Se1, Def. 2] for co-
herent sheaves). Many of the results we state will hold in greater generality, but
because the proofs look slightly different, we restrict ourselves to schemes to avoid
distraction.

6.4.11. ⋆⋆ Coherence is not a good notion in smooth geometry. The following example
from B. Conrad shows that in quite reasonable (but less “rigid”) situations, the
structure sheaf is not coherent over itself. Consider the ring O0 of germs of smooth
(C∞ ) functions at 0 ∈ R, with coordinate x. Now O0 is a local ring. Its maximal ideal
m is generated by x. (Key idea: suppose f ∈ m, and suppose f has a representative
Rt R1
defined on (−ϵ, ϵ). Then for t ∈ (−ϵ, ϵ), f(t) = 0 f ′ (u) du = t 0 f ′ (tv) dv. By
R1
“differentiating under the integral sign” repeatedly, we may check that 0 f ′ (tv) dv
is smooth.)
Let ϕ ∈ O0 be the germ of a smooth function that is 0 for x ≤ 0, and positive
2
for x > 0 (such as ϕ(x) = e−1/x for x > 0). Consider the map ×ϕ : O0 → O0 . The
kernel is the ideal Iϕ of functions vanishing for x ≥ 0. Clearly Iϕ is nonzero (for
 © 2024 Ravi Vakil. Published by Princeton University Press. 183

example, ϕ(−x) ∈ Iϕ ), but as m = (x), Iϕ = xIϕ , so Iϕ cannot be finitely generated

or else Nakayama’s Lemma 8.2.9 would be contradicted. (Essentially the same
argument shows that the sheaf of smooth functions on R is not coherent.) This is
why coherence has no useful meaning for smooth manifolds.

6.5 Algebraic interlude: The Jordan-Hölder package

The Jordan-Hölder Theorem in group theory is part of a more fundamental
and somehow simpler story. The Jordan-Hölder “yoga” is why you can often factor
some sort of algebraic object into primes or irreducibles, uniquely (in an appropriate
sense), where each prime/irreducible appears the same number of times no matter
how you factor. From this point of view, it generalizes unique factorization of
integers; well-definedness of dimension of vector spaces; classification of finitely
generated abelian groups; unique factorization of ideals in a Dedekind domain;
and the traditional Jordan-Hölder Theorem in group theory.
We will be mostly interested in modules over a ring, but there is no harm in
working in a general abelian category C . (This can be readily generalized further,
as in Exercise 6.5.G.)
We say an object M ∈ C is simple (or irreducible) if its only subobjects are 0
and itself. A composition series for M is a (finite) filtration
(6.5.0.1) 0 = X0 ⊊ X1 ⊊ · · · ⊊ Xn−1 ⊊ Xn = M
such that the quotients Xi+1 /Xi are all simple. If M has a composition series, we
say that M has finite length.

6.5.1. The Jordan-Hölder Theorem. — If M has a finite composition series, then all
composition series for M have the same length, and the quotients are all the same, possibly
rearranged.

6.5.2. Definition. We call the length of any of the composition series for M the
length of M, denoted ℓ(M). This notion is well-defined by the Jordan-Hölder
Theorem. But we even have a refined notion: we have the multiplicity with which
each simple object appears in any composition series for M.
If M is not of finite length, we say ℓ(M) = ∞.

6.5.3. Example. In the category of abelian groups, the finite-length objects are the
finite abelian groups. The Jordan-Hölder Theorem in this case, applied to Z/nZ,
can be used to give the unique factorization of n.

6.5.4. Proof of the Jordan-Hölder Theorem 6.5.1.

Suppose we have two finite composition series, (6.5.0.1) and
0 = Y0 ⊊ Y1 ⊊ · · · ⊊ Yp−1 ⊊ Yp = M,
of one object M ∈ C . Make a rectangular array with entries Zi,j := Xi ∩ Yj , as in
Figure 6.1. Figure 6.2 shows this construction applied to two composition series for
the Z-module Z/(12),
(12) ⊊ (6) ⊊ (2) ⊊ (1) and (12) ⊊ (4) ⊊ (2) ⊊ (1).
184 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

F IGURE 6.1. The rectangular array in the proof of the Jordan-

Hölder Theorem

F IGURE 6.2. A sample “Jordan-Hölder table” for two composition

series for Z/(12)

6.5.5. Observe that

• Zi,j ⊂ Zi ′ ,j ′ if i ≤ i ′ and j ≤ j ′ ,
• Zn,j = Yj and Zi,p = Xi ,
• Z0,j = Zi,0 = 0, and
• Zi,j ∩ Zi ′ ,j ′ = Zmin(i,i ′ ),min(j,j ′ ) .

6.5.A. E XERCISE . Show (by descending induction on i) that Zi,j /Zi,j−1 is 0 or

isomorphic to the simple element Yj /Yj−1 .
Similarly, we have the analogous statement for Zi,j /Zi−1,j . In the rectangular
array, draw a thick line between these two (horizontally or vertically adjacent)
 © 2024 Ravi Vakil. Published by Princeton University Press. 185

entries if the quotient is nonzero, and label that line with the (isomorphism class of)
the simple group (again, see Figure 6.1).
Consider any 2 × 2 subsquare of the array:

We will see that the thick lines inside the square form one of the following five
patterns (each of which appears in Figure 6.2).

From A = B ∩ C, we see that (i) if D = B, then A = C, and (ii) if D = C, then

A = B. That takes care of the first three cases.
Suppose next that both D/B and D/C are both nonzero (hence simple). If B = C,
we are in the fourth (“elbow”) case.
Finally, otherwise, we will see that D/B = ∼ C/A (and similarly, D/C = ∼ B/A),
and we are in the fifth case. Consider the map C → D/B. Then A is precisely
the kernel, from A = B ∩ C (§6.5.5). Thus we have an injection C/A ,→ D/B. By
simplicity of D/B, either C/A is zero, or this injection is actually an isomorphism
∼
C/A −→ D/B. □

6.5.B. E XERCISE . Prove the Jordan-Hölder Theorem. Hint: notice that the thickened
lines can be interpreted as paths from the right side of the table to the bottom of the
table, with one left turn. This will give a bijection between the simple quotients of
one filtration, and the simple quotients of the other filtration.

6.5.C. E XERCISE . Show that every subquotient of a finite-length object M is finite

length. Possible approach: suppose the subquotient if M ′′ /M ′ , where M ′ ⊆ M ′′ ⊆
M. Choose a composition series M• for M. Make a new rectangular table in a
similar way, using the composition series M• , and the filtration 0 ⊆ M ′ ⊆ M ′′ ⊆ M.
Think suitably about paths, similar to the proof of the Jordan-Hölder Theorem.

6.5.D. E XERCISE . Show that length is additive in exact sequences: if 0 → M ′ →

M → M ′′ → 0 is an exact sequence, then ℓ(M) = ℓ(M ′ ) + ℓ(M ′′ ). (Do not assume
these objects have finite length.)

6.5.E. E XERCISE . Show that any filtration of a finite length module can be refined
into a composition series.

6.5.F. U NIMPORTANT EXERCISE . Show that the finite length objects in C form a full
subcategory of C .
The category of groups does not form an abelian category, so Theorem 6.5.1
can’t immediately imply the traditional Jordan-Hölder Theorem for groups. How-
ever, the same proof applies without change, with only one additional input.
186 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

6.5.G. E XERCISE ( THAT WE WON ’ T USE ). Prove the Jordan-Hölder Theorem for
groups. You will need the second isomorphism theorem: if N1 and N2 are normal
∼ (N1 N2 )/N1 .
subgroups, then N1 N2 forms a normal subgroup, and N2 /(N1 ∩ N2 ) =

6.5.6. Additional facts particular to modules over a ring.

We now apply these concepts specifically to the category ModA .

6.5.H. E XERCISE . Show that the simple objects of ModA are precisely the objects of
the form A/m, where m is a maximal ideal of A.

6.5.I. E XERCISE . Suppose M is a finite length A-module, and (6.5.0.1) is a compo-

sition series for M, with Mi /Mi−1 =∼ A/mi (where the mi are maximal ideals, not
necessarily distinct). Show that M is annihilated by m1 · · · mn . Equivalently, M is
an A/(m1 · · · mn )-module.
Suppose now that the list (m1 , . . . , mn ) consists of the distinct maximal ideals
n1 , . . . , ns , appearing with multiplicity ℓ1 , . . . , ℓs . (These are the “refined” lengths
mentioned in Definition 6.5.2.)
By the Chinese Remainder Theorem,

∼ A/nℓ1 · · · nℓs =
A/m1 . . . mn = ∼ A/nℓ1 × · · · A/nℓs .
1 s 1 s

For 1 ≤ i ≤ s, let ei ∈ A be an element of A so that ei ≡ 1 (mod nℓii ) and

ℓ
ei ≡ 0 (mod nj j ) for i ̸= j. The ei exist by the Chinese Remainder Theorem.

6.5.J. E XERCISE . Show that M = ∼ e1 M × · · · × es M, and ei M is a finite-length

module where all the simple quotients are A/ni . Thus M is a product of pieces,
each with composition series with only one type of “simple factor”.

6.5.K. E XERCISE . Suppose M is a finite length A-module. Show that M is finitely

generated, and Supp M consists of finitely many points of Spec A, all closed. (A
converse under Noetherian hypotheses will be proved in Exercise 6.6.X(a).) We
thus have a notion of the “length of M at each of these closed points”.

6.5.7. Definition. If A has finite length as a module over itself, we say A is an

Artinian ring. If A is furthermore a local ring, we say A is an Artin or Artinian
local ring.

6.5.8. Applying this language to schemes. We next consider the category QCohX of
quasicoherent sheaves on a scheme X. We have the notion of the length ℓ(F ) of a
finite-length quasicoherent sheaf on X.

6.5.L. E XERCISE . Show that the simple objects of QCohX are those isomorphic to
the structure sheaves of closed points.

6.5.M. E XERCISE . Suppose that F is a finite length element of QCohX . Show that
F is finite type, and Supp F consists of finitely many points of X, all closed. (A
converse under Noetherian hypotheses will be proved in Exercise 6.6.X(b).) Explain
how to define the length of a F at one of the points of Supp F .

6.5.9. Definition. The length of a scheme X is the length of the structure sheaf OX
(in QCohX ). A scheme X is finite length or Artinian if OX is finite length.
 © 2024 Ravi Vakil. Published by Princeton University Press. 187

6.6 Visualizing schemes: Associated points and zerodivisors

The theory of associated points of a module refines the notion of support (§4.1.7).
Associated points will help us understand and visualize nilpotents, and generalize
the notion of “rational functions” to non-integral schemes. They are useful in ways
we won’t use, for example through their connection to primary decomposition.
They might be most useful for us in helping us understand and visualize (non-
)zerodivisors, which will come up repeatedly, through effective Cartier divisors
and line bundles, regular sequences, depth and Cohen-Macaulayness, and more.
There is no particular reason to discuss associated points now, and this section
can be read independently, at your leisure. But it is a good opportunity to practice
visualizing geometry, and to learn some useful algebra.

6.6.1. Motivation. Figure 6.3 is a sketch of a scheme X. We see two connected com-
ponents, and three irreducible components. The irreducible components of X have
dimensions 2, 1, and 1, although we won’t be able to make sense of “dimension”
until Chapter 12. Both connected components are nonreduced.
We see a little more in this picture, which we will make precise in this section,
in terms of “associated points”. The reducible connected component seems to have
different amounts of nonreduced behavior on different loci. The scheme X has six
associated points, which are the generic points of the irreducible subsets “visible”
in the picture. A function on X is a zerodivisor if its zero locus contains any of these
six irreducible subvarieties.

F IGURE 6.3. This scheme has six associated points, of which three
are embedded points. A function is a zerodivisor if it vanishes at
any of these six points.

Suppose M is a finitely generated module over a Noetherian ring A. For

example, M could be A itself. Then there are some special points of Spec A that
are particularly crucial to understanding M. These are the associated points of
M (or equivalently, the associated prime ideals of M — we will use these terms
interchangeably). As motivation, we give a zillion properties of associated points,
and leave it to you to verify them from the theory developed in the rest of this
section.
As you read this section, you may wish to keep in mind
M = A = k[x, y]/(y2 , xy)
(Figure 4.4) as a running example.
188 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

6.6.2. A zillion properties of associated points. Here are some of the properties of
associated points that we will prove.
There are finitely many associated points AssA M ⊂ Spec A.
The support of M is the closure of the associated points of M: Supp M =
AssA M. The support of any submodule of M is the closure of some subset of the
associated points of M. The support of any element of M is the closure of some
subset of the associated points.
The associated points of M are precisely the generic points of irreducible
components of Supp m for all m ∈ M. The associated points of M are precisely
the generic points of those Supp m which are irreducible (as m runs over the
elements of M). The associated primes of M are precisely those prime ideals that
are annihilators of some element of M.
Taking “associated points” commutes with localization. Hence this notion
is “geometric in nature”, which will (in §6.6.2) allow us to extend the notion to
coherent sheaves on locally Noetherian schemes.
Associated points behave fairly well in exact sequences. For example, the
associated points of a submodule are a subset of the associated points of the
module.
If I ⊂ A is an ideal, the associated primes p of A/I are precisely those p such
that a p-primary ideal appears in the primary decomposition of I.
We will repeatedly use the fact that an element of A is a zerodivisor if and only if it
vanishes at an associated point.
An element of A is nilpotent if and only if it vanishes at every associated point.
The locus of points [p] of Spec A where the stalk Ap is nonreduced is the closure of
some subset of the associated points.
An associated point that is in the closure of another associated point is said
to be an embedded point. If A is reduced, then Spec A has no embedded points.
Hypersurfaces in An k have no embedded points. We will later see that complete
intersections have no embedded points (§26.2.7).
Elements of M are determined by their localization at the associated points.
Sections of the corresponding sheaf M f (Exercise/Definition 4.1.D) are determined
by their germs at the associated points.
This discussion immediately implies a notion of associated point for a coherent
sheaf on a locally Noetherian scheme, with all the good properties described here.
The phrase associated point of a locally Noetherian scheme X (without explicit
mention of a coherent sheaf) means “associated point of OX ”, and similarly for
embedded points.
We now establish these zillion facts.

6.6.3. More on the notion of support.

The notion of associated points of an A-module M refines the notion of support (in
the case where M is finitely generated over a Noetherian ring A). (In what follows,
we make no assumptions that A is Noetherian or that M is finitely generated until
we need to.) To set this up, recall (§4.1.7) that the support of m ∈ M,

Supp m = {[p] ∈ Spec A : mp ̸= 0},

is a closed subset, and thus of the form V(I) for some I. Exercise 6.6.A gives the
“best such” I. Define the annihilator ideal AnnA m ⊂ A of an element m of an
 © 2024 Ravi Vakil. Published by Princeton University Press. 189

A-module M by:
×m
AnnA m := {a ∈ A : am = 0} = ker(A −−→ M).
The subscript A is omitted if it is clear from context.

6.6.A. E ASY I MPORTANT E XERCISE . Show that Supp m = V(Ann m).

Recall (Definition 2.7.6) that
f = {p ∈ Spec A : M
Supp M fp ̸= 0},

and the analogous Definition 4.1.7 of the support of the module M,

(6.6.3.1) Supp M := {p ∈ Spec A : Mp ̸= 0},
f If M is a principal module generated by m ∈ M, then
so Supp M = Supp M.
Supp M = Supp Am = Supp m = V(Ann m).
The notions of support and associated points behave well in exact sequences,
and under localization. We begin to explain this now.

6.6.4. The notion of support behaves well in exact sequences.

6.6.B. E XERCISE . Suppose that 0 → M ′ → M → M ′′ → 0 is a short exact sequence

of A-modules.
(a) Show that Supp M = Supp M ′ ∪ Supp M ′′ .
(b) Show that if M is a finitely generated module, then Supp M is a closed
subset of Spec A. (Hint: induction on the number of generators.)
Warning: Supp M need not be closed in general; consider A = Z and M =
⊕p prime Z/(p).

6.6.C. E XERCISE . Suppose M is a finitely generated A-module, and x ∈ A has

value 0 at all the points of Supp M, i.e., x is contained in all of the primes where M
is supported. Show that some power xn of x annihilates every element of M. (Hint:
annihilate a generating set.)

6.6.D. E XERCISE ( FOR USE IN §18.6). Suppose M is a finitely generated A-module,

and x ∈ A. Show that Supp(M/xM) = (Supp M) ∩ V(x). Here M/xM is defined
by the exact sequence

M
×x
/M / M/xM / 0.

6.6.5. Definition: Associated points and associated primes.

Define the associated prime ideals of an A-module M to be those prime ideals
of A of the form AnnA (m) for some m ∈ M. Define the associated points of M to
be the corresponding points of Spec A; we use the terminology “associated points”
and “associated primes” interchangeably. The set of associated points is denoted
AssA M ⊂ Spec A. The subscript A is dropped if it is clear from the context. (To
help remember the definition and the notation, some call these the assassins, as
they are the primes that can ruthlessly annihilate elements of the module, without
remorse. But we will not use this term.)
190 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

6.6.6. The associated primes of a ring A are the associated primes of A considered
as an A-module (i.e., M = A).

6.6.E. E ASY E XERCISE ( ASSOCIATED POINTS OF INTEGRAL DOMAINS ). If A is an

integral domain, show that Ass A = {[(0)]} — the zero ideal is the only associated
prime.

6.6.F. E XERCISE ( ASSOCIATED POINTS OF HYPERSURFACES ). Given f ∈ k[x1 , . . . , xn ],

show that the associated primes of k[x1 , . . . , xn ]/(f) are those principal ideals gen-
erated by the prime factors of f. (Your argument will apply more generally to any
f ∈ A where A is a Unique Factorization Domain.)

6.6.7. The observation that [p] ∈ AssA (M) if and only if there is an injection A/p ,→ M
of A-modules will be essential. The next exercise might drive this home.

6.6.G. E XERCISE .
(a) Suppose M ′ ⊂ M. Show that Ass M ′ ⊂ Ass M. (The corresponding statement
for “support” is implicit in Exercise 6.6.B(a).)
(b) Show that Ass M ⊂ Supp M.
If M is finitely generated, then Supp M is closed (Exercise 6.6.B), so Ass M ⊂
Supp M. Equality will be shown in Proposition 6.6.24, when A is Noetherian.

6.6.8. Nonzero modules over Noetherian rings have associated points.

Suppose m is a nonzero element of an A-module M. Observe that for any
nonzero multiple xm of m, Ann m ⊆ Ann xm ⊊ A.

6.6.H. E XERCISE . Suppose A is Noetherian. Show that there is some multiple

n = xm such that any nonzero multiple yn ̸= 0 of n satisfies Ann yn = Ann n.

6.6.9. Proposition. — Continuing the notation of the previous exercise, we have that
Ann n is a prime ideal.

Proof. Suppose ab ∈ Ann n, so abn = 0. Then either bn = 0 (in which case

b ∈ Ann n), or else a ∈ Ann bn = Ann n. □
We have thus proved the following.

6.6.10. Proposition (nonzero modules over Noetherian rings have associated

primes). — If M is a nonzero module over a Noetherian ring A, then AssA M is
nonempty. More precisely, for any m ̸= 0 in M, there is an associated prime p containing
Ann m, and p = Ann xm for some x ∈ A.

6.6.11. Localizations at the associatedQprimes.

Recall the useful fact that M → p∈Spec A Mp is an injection (Exercise 4.1.F).
Our current situation is much better: we can take the product over only the local-
ization at associated primes.

6.6.I. E XERCISE . Suppose M is a module over a Noetherian ring A. Show that the
natural map

(6.6.11.1) M /Q Mp
p∈Ass M
 © 2024 Ravi Vakil. Published by Princeton University Press. 191

is an injection. Hint: if the kernel K is nonzero, then K has an associated prime p,

which is the annihilator of some m ∈ K ⊂ M, and m is nonzero in Mp .
Clearly we need only the maximal among the associated primes in (6.6.11.1).

6.6.12. Zerodivisors = elements of associated primes.

6.6.13. Proposition. — Suppose f ∈ A, with A Noetherian. Then f is a zerodivisor on M

if and only if f vanishes at an associated point of M. Translation: the set of zerodivisors is
the union of the associated prime ideals.
Again, we need only the maximal among the associated primes. For example,
if (A, m) is a local ring, then m is an associated prime if and only if every element
of m is a zerodivisor.

Proof. Suppose f vanishes at an associated point [p] of M. Choose m with Ann m = p.

Then fm = 0 while m ̸= 0, so f is a zerodivisor.
Conversely, if f vanishes at no associated point, consider the commuting dia-
gram
 /Q
M Mp
p∈Ass M

×f ×f
  
M /Q Mp ,
p∈Ass M

where the rows are the maps of (6.6.11.1). The vertical arrow on the right (multipli-
cation by f) is an injection by hypothesis, so the vertical arrow on the left must be
an injection as well. □

6.6.14. Associated points behave fairly well in exact sequences.

6.6.15. Proposition. — Suppose

(6.6.15.1) 0 / M′ /M / M ′′ /0

is a short exact sequence of A-modules. Then

(6.6.15.2) Ass M ′ ⊂ Ass M ⊂ Ass M ′ ∪ Ass M ′′ .
We come to our first complicated proof of the section.

Proof. The first inclusion of (6.6.15.2) was shown in Exercise 6.6.G(a).

Suppose next that [p] ∈ Ass M, so there is some m ∈ M with Ann m = p. We
wish to find a submodule of M ′ or M ′′ isomorphic to A/p. If this proposition were
true, we would expect to find such a submodule in the “part of (6.6.15.1) spanned
by m”. So we consider instead the exact sequence

0 / Am ∩ M ′ / Am / Am/(Am ∩ M ′ ) / 0,

noting that the three modules appearing here are submodules of the corresponding
modules in (6.6.15.1). So by Exercise 6.6.G(a) it suffices to prove the result in this
“special case”, which can be rewritten as

0 / I/p / A/p / A/I /0

192 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

where I is the annihilator of m considered as an element of the module Am/(Am ∩

M ′ ). For convenience, let B = A/p (an integral domain), so we rewrite the exact
sequence further as the top row of

0 / J / B / B/J /0
_ _ _

  
0 / M′ /M / M ′′ /0

Now localize the top row of B-modules at (0) ⊂ B, so it becomes an exact se-
quence of vector spaces over the fraction field K(B), and the central element is
one-dimensional:
0 / J ⊗ K(B) / K(B) / (B/J) ⊗ K(B) / 0.

Thus one of the outside terms J ⊗ K(B) and (B/J) ⊗ K(B) has a nonzero element,
which (tracing our argument backwards) gives an element of M ′ or M ′′ whose
annihilator is precisely p. □

6.6.16. Cautionary example. The short exact sequence of Z-modules

0 /Z ×2
/Z / Z/2 /0

(and Easy Exercise 6.6.E) shows it is not always true that Ass M = Ass M ′ ∪Ass M ′′ .
However, sometimes we can still ensure some associated primes of M ′′ lift to
associated primes of M, as we shall see in §6.6.20.

6.6.17. Finitely generated modules over Noetherian rings have finitely many
associated points/primes.

6.6.J. I MPORTANT E XERCISE . Suppose that M is a finitely generated module over a

Noetherian ring A. Show that M has a (finite) filtration
(6.6.17.1)
0 = M0 ⊂ M1 ⊂ · · · ⊂ Mn = M where Mi+1 /Mi = ∼ A/pi for some prime pi .

Hint: Build (6.6.17.1) inductively from left to right, using Proposition 6.6.10, and
show the process terminates.

6.6.K. E XERCISE . Suppose an A-module M has a finite filtration (6.6.17.1), with

no assumptions of finite generation or Noetherianity. Show that every associated
prime of M appears as one of the pi . In particular, under this hypothesis (for
example, if M is finitely generated over a Noetherian ring) M has finitely many
associated points/primes. (Hint: Exercise 6.6.E and Proposition 6.6.15.)

6.6.18. Caution: Non-associated prime ideals may unavoidably appear among the quotients
in (6.6.17.1). Example 6.6.16 shows that the non-associated prime ideals may be
among the quotients in (6.6.17.1), although in that case it is because the filtration
was chosen unwisely. But better choices will not always remedy the problem:

6.6.L. E XERCISE . Consider the module M = (x, y) ⊂ A = k[x, y] over the ring A.
Show that any filtration (6.6.17.1) of M contains a quotient A/pi where pi is not an
associated prime.
 © 2024 Ravi Vakil. Published by Princeton University Press. 193

6.6.19. Remark. However, not all is lost: Exercise 6.6.P will show that for any
quotient A/pi in any filtration (6.6.17.1) of M, pi must contain an associated prime
of M.

6.6.20. Associated points behave fairly well in exact sequences, continued.

6.6.21. Proposition. — Suppose A is a Noetherian ring. We continue to consider the

short exact sequence

0 / M′ /M / M ′′ /0

of A-modules. Suppose p ∈ Ass M ′′ , but p ∈

/ Supp M ′ (a stronger hypothesis than
p∈/ Ass M ′ ). Then p ∈ Ass M.
We come to our second complicated proof of the section.

Proof. Choose m ′′ ∈ M ′′ with p = Ann m ′′ in M ′′ . Choose a lift of m ∈ M of

m ′′ ∈ M ′′ . We apply a strategy similar to that of our proof of Proposition 6.6.15.
Consider the “inclusion of short exact sequences”

0 / ker (Am → Am ′′ ) / Am
_ / Am ′′ /0
_ _

  
0 / M′ /M / M ′′ / 0.

As Supp(ker(Am → Am ′′ )) ⊂ Supp M ′ , Ass(Am) ⊂ Ass M, and Ass(Am ′′ ) ⊂

Ass M ′′ (Exercises 6.6.B(a) and 6.6.G(a)), we have reduced to considering the top
row instead of the bottom row. The top row can be conveniently rewritten as

0 / p/I / A/I / A/p /0

(here I = Ann(m)) where our hypothesis translates to [p] ∈

/ Supp(p/I). For conve-
nience, let B = A/I so we may now consider the sequence

0 /q /B / B/q / 0,

where q is prime, with the confusion-inducing hypothesis [q] ∈ / Supp q.

From the confusing hypothesis, there is an element b of B that vanishes on
Supp q but doesn’t vanish at [q]. Translation: (i) b lies in all the primes of Supp q,
/ q. Then from (i) there is some power bn of b that annihilates all elements
but (ii) b ∈
of q (Exercise 6.6.C, which applies because q is finitely generated, which in turn
follows from Noetherianity of A). From (ii), bn ∈ / q.
Then Ann(bn ) contains q from (i), but does not contain any element of B \ q
from (ii), so Ann(bn ) = q. hence q is an associated prime of B, which (unwinding
our argument) shows that p is an associated prime of M. □

6.6.22. Minimal primes are associated.

6.6.M. E XERCISE : MINIMAL PRIMES (“ IRREDUCIBLE COMPONENTS ”) ARE ASSOCI -

ATED . Suppose M is a finitely generated module over Noetherian A, and p ⊂ A
is a prime ideal corresponding to an irreducible component of Supp M ⊂ Spec A.
Show that [p] ∈ Ass M. Hint: Exercise 6.6.J and Proposition 6.6.21.
194 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

6.6.23. Non-Noetherian Remark. By combining Proposition 6.6.13 with Exercise 6.6.M,

we see that if A is a Noetherian ring, then any element of any minimal prime p is
a zerodivisor. This is true without Noetherian hypotheses: suppose s ∈ p. Then
by minimality of p, pAp is the unique prime ideal in Ap , so the element s/1 of
Ap is nilpotent (because it is contained in all prime ideals of Ap , Theorem 3.2.13).
Thus for some t ∈ A \ p, tsn = 0, so s is a zerodivisor in A. We will use this in
Exercise 12.1.H.

6.6.24. Proposition. — Suppose M is a finitely generated module over a Noetherian ring

A. Then Supp M = Ass M.

Proof. Combine Exercises 6.6.G(b) and 6.6.M. □

6.6.N. E XERCISE . Show that the locus of points [p] where Ap is nonreduced is the
support of the nilradical Supp N. Hence show that the “reduced locus” of a locally
Noetherian scheme is open.
The following justifies a simple way of thinking about associated primes of a
ring.

6.6.O. E XERCISE . Show that a prime ideal p ⊂ A is an associated prime of a

Noetherian ring A if and only if there is f ∈ A such that Supp f = V(p) = [p].

6.6.P. E XERCISE , PROMISED IN R EMARK 6.6.19. Show that for each quotient in the
filtration (6.6.17.1) of M, Supp A/pi = [pi ] ⊂ Supp M, and that every pi contains a
minimal prime, and hence an associated prime.

6.6.25. “Support” and “associated points” commute with localization.

Suppose S is a multiplicative subset of A, and p ⊂ A is a prime ideal not
meeting S, so (abusing notation slightly) [p] ∈ Spec S−1 A ⊂ Spec A (§3.2.9).

6.6.Q. E XERCISE (Supp COMMUTES WITH LOCALIZATION ). Show that for any
A-module M, SuppA M ∩ Spec S−1 A = SuppS−1 A S−1 M.

6.6.26. Proposition (Ass commutes with localization). — Let M be a finitely gener-

ated module over a Noetherian ring A. Then AssA M ∩ Spec S−1 A = AssS−1 A S−1 M.

Proof. We first show that AssA M ∩ Spec S−1 A ⊂ AssS−1 A S−1 M. If p ∈ AssA M,
then we have an injection A/p ,→ M. Localizing by S (which preserves injectivity),
we have (S−1 A)/(S−1 p) ,→ M. (Where did we use p ∈ Spec S−1 A?)
We next show that AssS−1 A S−1 M ⊂ AssA M ∩ Spec S−1 A. Suppose q :=
S p ∈ AssS−1 A S−1 M, so q = AnnS−1 A (m/s), for some s ∈ S, and m ∈ M.
−1

Since the elements of S are units in S−1 A, we have that q = AnnS−1 A m. As

support commutes with localization, V(p) must be an irreducible component of
Supp m (as Supp m ∩ Spec S−1 M contains [q], but no generizations of [q]). Then by
Exercise 6.6.M, p is an associated prime of M. □

6.6.27. Embedded points/primes.

6.6.28. Definition. The associated points that are not the generic points of irreducible
components of Supp M are called embedded points, and their corresponding
 © 2024 Ravi Vakil. Published by Princeton University Press. 195

primes are called embedded primes. The motivation for this language is their
geometric incarnation as embedded points (§6.6.32 shortly). For example, the
scheme of Figure 6.3 has three embedded points (the corresponding ring has three
embedded primes).

6.6.29. Remark. Exercise 6.6.F translates to “hypersurfaces in An

k have no embedded
points”. More generally, if A is a unique factorization domain, then Spec A/(f) has
no embedded points for any f ∈ A. Generalizing in a different direction, we will
see that “complete intersections have no embedded points” in §26.2.7.

6.6.R. E XERCISE . Suppose A is a reduced ring (i.e., A has no nonzero nilpotents).

Show that Spec A has no embedded primes. (Hints: if p = Ann a is an embedded
prime, show that you can find an element b of p not contained in any of the minimal
primes of A. From ab = 0, show that a is contained in all the minimal primes. Show
that a vanishes at all points of Spec A, and hence by Theorem 3.2.13 is nilpotent.)
Thus reduced rings have no embedded primes. Even better: the only elements of a
ring that an embedded prime can annihilate are nilpotents.

6.6.30. Remark. The converse to Exercise 6.6.R is false. Rings without embedded
primes can still have nilpotents — witness k[x]/(x2 ).

6.6.S. E XERCISE . Show that if p is an embedded prime of a ring A, then Ap is

nonreduced.

6.6.31. Get your hands dirty: Explicit algebraic exercises.

6.6.T. E XERCISE . (See Figure 4.4.)

(a) Suppose f is a function on Spec k[x, y]/(y2 , xy), i.e., f ∈ k[x, y]/(y2 , xy). Show
that Supp f is either the empty set, or the origin, or the entire space. Hence find the
associated points of Spec k[x, y]/(y2 , xy).
(b) Show explicitly by hand that f ∈ k[x, y]/(y2 , xy) is a zerodivisor if and only if
f(0, 0) = 0.

6.6.U. E XERCISE ( PRACTICE WITH FUZZY PICTURES ). Suppose X = Spec C[x, y]/I,
and that the associated points of X are [(y − x2 )], [(x − 1, y − 1)], and [(x − 2, y − 2)].
(Exercise 6.6.U will verify that such an X actually exists.)
(a) Sketch X as a subset of A2C = Spec C[x, y], including fuzz.
(b) Do you have enough information to know if X is reduced?
(c) Do you have enough information to know if x + y − 2 is a zerodivisor? How
about x + y − 3? How about y − x2 ?
(d) Let I = (y − x2 )3 ∩ (x − 1, y − 1)15 ∩ (x − 2, y − 2). Show that X = Spec C[x, y]/I
satisfies the hypotheses of this exercise. (Rhetorical question: Is there a “smaller”
example? Is there a “smallest”?)

6.6.32. Geometric definitions.

6.6.V. E XERCISE /D EFINITION . Define the associated points and embedded points
of a coherent sheaf on a locally Noetherian scheme. (Idea/hint: do it affine-
locally.)
196 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

6.6.W. E XERCISE . Suppose X is a locally Noetherian scheme, and U ⊂ X is an open

subscheme. Show that the natural map

(6.6.32.1) Γ (U, OX ) /Q OX,p

p∈Ass X∩U

(cf. (6.6.11.1)) is an injection.

6.6.33. Scheme-theoretic density (and the adjective “scheme-theoretic” more generally).

An open subscheme U of a scheme X is said to be scheme-theoretically dense if
any function on any open set V is 0 if it restricts to 0 on U ∩ V. This is stronger than
density in the usual (topological) sense. For example, if X is locally Noetherian, then
you can use the injection (6.6.32.1) (with M = A) to show that an open subscheme
U ⊂ X is scheme-theoretically dense if and only if it contains all the associated
points of X.
More generally, the adjective scheme-theoretic typically indicates an ideal-
theoretic definition, enriching and refining a more naive set-theoretic definition.
Examples will include:
• scheme-theoretically dense (here)
• scheme-theoretic intersection (Figure 4.5, Exercise 9.1.I, Exercise 10.4.B)
• scheme-theoretic support (Definition 9.1.4)
• scheme-theoretic image (§9.4.1)
• scheme-theoretic closure (§9.4.8)
• scheme-theoretic preimage (§10.3.1)
• scheme-theoretic fiber (Definition 10.3.3)

6.6.34. Revisiting the notion of length.

6.6.X. E XERCISE .
(a) (Noetherian converse to Exercise 6.5.K) Suppose A is a Noetherian ring. Show
that any finitely generated module whose support consists of finitely many points
of Spec A, all closed, has finite length.
(b) (Noetherian converse to Exercise 6.5.M) Suppose F is a finite type quasicoherent
sheaf on a locally Noetherian scheme X, supported at finitely many points of X, all
closed. Show that F has finite length.

6.6.Y. E XERCISE .
(a) Show that each finite length scheme (Definition 6.5.9) has a finite number of
points, with the discrete topology.
(b) If X is locally finite type over k, of finite length ℓ, show that dimk Γ (X, OX ) = ℓ.

6.6.35. Generalizing the fraction field: the total fraction ring.

6.6.36. Definitions: Rational functions on locally Noetherian schemes. A rational

function on a locally Noetherian scheme is an element of the image of Γ (U, OU )
in (6.6.32.1) for some U containing all the associated points. Equivalently, the
set of rational functions is the colimit of OX (U) over all open sets containing the
associated points.
x−2
For example, on Spec k[x, y]/(y2 , xy) (Figure 4.4), (x−1)(x−3) is a rational func-
x−2
tion, but x(x−1) is not.
 © 2024 Ravi Vakil. Published by Princeton University Press. 197

A rational function has a maximal domain of definition, because any two

actual functions on an open set (i.e., sections of the structure sheaf over that open
set) that agree as “rational functions” (i.e., on small enough open sets containing
associated points) must be the same function, by the injectivity of (6.6.32.1). We say
that a rational function f is regular at a point p if p is contained in this maximal
domain of definition (or equivalently, if there is some open set containing p where
x−2
f is defined). For example, on Spec k[x, y]/(y2 , xy), the rational function (x−1)(x−3)
has domain of definition consisting of everything but 1 and 3 (i.e., [(x − 1)] and
[(x − 3)]), and is regular away from those two points. A rational function is regular
if it is regular at all points. (Unfortunately, “regular” is a regularly overused word
in mathematics, and in algebraic geometry in particular.)
The complement of the domain of definition of a rational function f is called
the indeterminacy locus of f (a phrase we’ll see again in §11.3.3).
The rational functions form a ring, called the total fraction ring or total quotient
ring of X. If X = Spec A is affine, then this ring is called the total fraction (or
quotient) ring of A. If X is integral, the total fraction ring is the function field K(X)
— the stalk at the generic point — so this extends our earlier Definition 5.2.I of K(·).

6.6.Z. E XERCISE . Show that the ring of rational functions on a locally Noetherian
scheme is the colimit of the functions over all open sets containing the associated
points:
colim{U : Ass X⊂U} O(U).
Slightly better (but slightly different): show that a rational function is the data
of a function f defined on an open set U containing all associated points, where
two such data (U, f) and (U ′ , f ′ ) define the same rational function if and only if
′
fU∩U ′ = fU∩U ′ (cf. §1.3.9). If X is reduced, show that this is the same as requiring

that they are defined on an open set of each of the irreducible components.

6.6.37. Remark: Associated points of integral schemes. In order for some of our
discussion elsewhere to make sense in non-Noetherian settings, we note that the
notion of associated points for integral schemes works perfectly, because it works for
integral domains — only the generic point is associated. In particular, the definition
above of rational functions on an integral scheme X agrees with Definition 5.2.I, as
precisely the elements of the function field K(X).

6.6.38. ⋆⋆ Aside: Primary ideals and associated primes. Primary decomposition

was introduced by the world chess champion E. Lasker, and axiomatized by world
math champion E. Noether. We won’t need it, but here is the beginning of the story,
for the curious reader. An ideal I ⊂ A in a ring is primary if √I ̸= A, and xy ∈ I
implies either x ∈ I or yn ∈ I for some n > 0. In this case, I is prime, and I
is said to be p-primary. Equivalently, if I ⊂ A then I is p-primary if and only if
A/I has only one associated prime p. If I is an ideal of a Noetherian ring A, then
the associated prime ideals A/I turn out to be precisely the radicals of ideals in a
primary decomposition. See [E, §3.3], for example, for more of this important story.

6.7 ⋆⋆ Coherent modules over non-Noetherian rings

198 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

This section is intended for people who might work with non-Noetherian rings,
or who otherwise might want to understand coherent sheaves in a more general
setting. Even these people should read this much later. Read this only if you really
want to!
Suppose A is a ring. Recall the definition of when an A-module M is finitely
generated, finitely presented, and coherent (§6.4). The reason we like coherence is
that coherent modules form an abelian category. Here are some accessible exercises
working out why these notions behave well. Some repeat earlier discussion in
order to keep this section self-contained.
The notion of coherence of a module is only interesting in the case that a ring is
coherent over itself. Similarly, coherent sheaves on a scheme X will be interesting
only when OX is coherent (“over itself”). In this case, coherence is clearly the same
as finite presentation. An example where non-Noetherian coherence comes up
is the ring R⟨x1 , . . . , xn ⟩ of “restricted power series” over a valuation ring R of
a non-discretely valued K (for example, a completion of the algebraic closure of
Qp ). This is relevant to Tate’s theory of non-archimedean analytic geometry over
K (which you can read about in [BCDKT], for example). The importance of the
coherence of the structure sheaf underlines the importance of Oka’s Theorem in
complex geometry (stated in §6.4.8).

6.7.A. E XERCISE . Show that coherent implies finitely presented implies finitely
generated. (This was discussed at the start of §6.4.)

6.7.B. E XERCISE . Show that 0 is coherent.

Suppose for Exercises 6.7.C–6.7.I that

(6.7.0.1) 0 /M /N /P /0

is an exact sequence of A-modules. In this series of exercises, we will show that if

two of {M, N, P} are coherent, the third is as well, which will prove very useful.

6.7.1. Hint †. The following hint applies to several of the exercises: try to write

0 / A⊕p / A⊕(p+q) / A⊕q /0

  
0 /M /N /P /0

and possibly use the Snake Lemma 1.6.5.

6.7.C. E XERCISE . Show that N finitely generated implies P finitely generated. (You
will only need right-exactness of (6.7.0.1).)

6.7.D. E XERCISE . Show that M, P finitely generated implies N finitely generated.

Possible hint: †. (You will only need right-exactness of (6.7.0.1).)

6.7.E. E XERCISE . Show that N, P finitely generated need not imply M finitely
generated. (Hint: if I is an ideal, we have 0 → I → A → A/I → 0.)

6.7.F. E XERCISE . Show that N coherent, M finitely generated implies M coherent.

(You will only need left-exactness of (6.7.0.1).)
 © 2024 Ravi Vakil. Published by Princeton University Press. 199

6.7.G. E XERCISE . Show that N, P coherent implies M coherent. Hint for (i) in the
definition of coherence (§6.4.1):
A⊕q

#
A⊕p

  !
0 /M /N /P /0

  
0 0 0
(You will only need left-exactness of (6.7.0.1).)

6.7.H. E XERCISE . Show that M finitely generated and N coherent implies P

coherent. Hint for (ii) in the definition of coherence (§6.4.1): †.

6.7.I. E XERCISE . Show that M, P coherent implies N coherent. (Hint: †.)

6.7.J. E XERCISE . Show that a finite direct sum of coherent modules is coherent.

6.7.K. E XERCISE . Suppose M is finitely generated and N is coherent. If ϕ : M → N

is any map, show that Im ϕ is coherent.

6.7.L. E XERCISE . Show that the kernel and cokernel of maps of coherent modules
are coherent.
At this point, we have verified that coherent A-modules form an abelian sub-
category of the category of A-modules. (Things you have to check: 0 should be in
this set; it should be closed under finite sums; and it should be closed under taking
kernels and cokernels.)

6.7.M. E XERCISE . Suppose M and N are coherent submodules of the coherent

module P. Show that M + N and M ∩ N are coherent. (Hint: consider the right
map M ⊕ N → P.)

6.7.N. E XERCISE . Show that if A is coherent (as an A-module) then finitely

presented modules are coherent. (Of course, if finitely presented modules are
coherent, then A is coherent, as A is finitely presented!)

6.7.O. E XERCISE . If M is finitely presented and N is coherent, show that Hom(M, N)

is coherent. (Hint: Hom is left-exact in its first argument.)

6.7.P. E XERCISE . If M is finitely presented, and N is coherent, show that M ⊗A N

is coherent.
 Part III

Morphisms of schemes
 CHAPTER 7

Morphisms of schemes

We now describe the morphisms between schemes. We will define some easy-
to-state properties of morphisms, but leave more subtle properties for later.
Recall that a scheme is (i) a set, (ii) with a topology, (iii) and a (structure) sheaf of
rings, and that it is sometimes helpful to think of the definition as having three steps.
In the same way, the notion of morphism of schemes X → Y may be defined (i) as a
map of sets, (ii) that is continuous, and (iii) with some further information involving
the sheaves of functions. In the case of affine schemes, we have already seen the
map of sets (§3.2.10) and later saw that this map is continuous (Exercise 3.4.H).

7.1 Motivations for the “right” definition of morphism of schemes

Here are two motivations for how morphisms should behave. The first is
algebraic, and the second is geometric.

7.1.1. Algebraic motivation. We will want morphisms of affine schemes Spec A →

Spec B to be precisely the ring maps B → A. We have already seen that ring maps
B → A induce maps of topological spaces in the opposite direction (Exercise 3.4.H);
the main new ingredient will be to see how to add the structure sheaf of functions
into the mix. Then a morphism of schemes should be something that “on the level
of affine open sets, looks like this”.

7.1.2. Geometric motivation. Motivated by the theory of manifolds (§3.1.1), which

like schemes are ringed spaces, we want morphisms of schemes at the very least to
be morphisms of ringed spaces; we now motivate what these are. (We will formalize
this in the next section.) Notice that if π : X → Y is a map of differentiable manifolds,
then a smooth function on Y pulls back to a smooth function on X. More precisely,
given an open subset U ⊂ Y, there is a natural map Γ (U, OY ) → Γ (π−1 (U), OX ).
This behaves well with respect to restriction (restricting a function to a smaller open
set and pulling back yields the same result as pulling back and then restricting),
so in fact we have a map of sheaves on Y: OY → π∗ OX . Similarly a morphism of
schemes π : X → Y should induce a map OY → π∗ OX . But in fact in the category
of differentiable manifolds a continuous map π : X → Y is a map of differentiable
manifolds precisely when smooth functions on Y pull back to smooth functions on
X (i.e., the pullback map from smooth functions on Y to functions on X in fact lies
in the subset of smooth functions, i.e., the continuous map π induces a pullback of

This material is published by Princeton University Press as The Rising Sea: Foundations of Algebraic
Geometry by Ravi Vakil. This version is free to view and download for personal use only. Not for
redistribution, resale, or use in derivative works. © 2024 Ravi Vakil.

203
204 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

smooth functions, which can be interpreted as a map OY → π∗ OX ), so this map of

sheaves characterizes morphisms in the differentiable category. So we could use this
as the definition of morphism in the differentiable category (see Exercise 3.1.A).
But how do we apply this to the category of schemes? In the category of
differentiable manifolds, a continuous map π : X → Y induces a pullback of (the
sheaf of) functions, and we can ask when this induces a pullback of smooth functions.
However, functions are odder on schemes, and we can’t recover the pullback map
just from the map of topological spaces. The right patch is to hardwire this into
the definition of morphism, i.e., to have a continuous map π : X → Y, along with a
pullback map π♯ : OY → π∗ OX . This leads to the definition of the category of ringed
spaces.
One might hope to define morphisms of schemes as morphisms of ringed
spaces. This isn’t quite right, as then Motivation 7.1.1 isn’t satisfied: as desired,
to each morphism A → B there is a morphism Spec B → Spec A, but there can be
additional morphisms of ringed spaces Spec B → Spec A not arising in this way
(see Exercise 7.2.G). A revised definition as morphisms of ringed spaces that locally
look of this form will work, but to avoid the awkwardness of sorting out the details,
we take a different approach, using locally ringed spaces, which corresponds to
asking not just that functions pull back, but also that values of functions pull back.
However, we will check that our eventual definition actually is equivalent to this
(Exercise 7.3.D).
We begin by formally defining morphisms of ringed spaces.

7.2 Morphisms of ringed spaces

7.2.1. Definition. A morphism of ringed spaces π : X → Y is a continuous

map of topological spaces (which we unfortunately also call π) along with a map
OY → π∗ OX , which we think of as a “pullback map”. By adjointness (§2.7.1), this is
the same as a map π−1 OY → OX . (It can be convenient to package this information
as in the diagram (2.7.2.1).) There is an obvious notion of composition of morphisms,
so ringed spaces form a category. Hence we have notion of automorphisms and
isomorphisms. You can easily verify that an isomorphism of ringed spaces means
the same thing as it did before (Definition 4.3.1).
If U ⊂ Y is an open subset, then there is a natural morphism of ringed spaces
(U, OY |U ) → (Y, OY ) (which implicitly appeared earlier in §2.2.13). More precisely,
if U → Y is an isomorphism of U with an open subset V of Y, and we are given
∼ ∼
an isomorphism (U, OU ) −→ (V, OY |V ) (via the isomorphism U −→ V), then the
resulting map of ringed spaces is called an open embedding (or open immersion)
of ringed spaces, and the morphism U → Y is often written U ,→ Y.

7.2.A. E XERCISE ( MORPHISMS OF RINGED SPACES GLUE ). Suppose (X, OX ) and

(Y, OY ) are ringed spaces, X = ∪i Ui is an open cover of X, and we have morphisms
of ringed spaces πi : Ui → Y that “agree on the overlaps”, i.e., πi |Ui ∩Uj = πj |Ui ∩Uj .
Show that there is a unique morphism of ringed spaces π : X → Y such that π|Ui =
πi . (Exercise 2.2.F essentially showed this for topological spaces.)
 © 2024 Ravi Vakil. Published by Princeton University Press. 205

7.2.B. E XERCISE . Show that open embeddings of ringed spaces are monomor-
phisms (in the category of ringed spaces).

7.2.C. E ASY IMPORTANT EXERCISE : PUSHING FORWARD O - MODULES . Given a

morphism of ringed spaces π : X → Y, show that sheaf pushforward induces a
functor π∗ : ModOX → ModOY .

7.2.D. I MPORTANT EXECISE FOR LATER : PULLING BACK O - MODULES . (You may
wish to ignore this exercise until when you need it, because you may not appreciate
it until then.) Suppose π : (X, OX ) → (Y, OY ) is a morphism of ringed spaces. A
slight variation on the inverse image allows us to pull back an OY -module G to get
an OX -module π∗ G .
(a) Show that (X, π−1 OY ) is a ringed space, and that the morphism of ringed
spaces π factors into (X, OX ) → (X, π−1 OY ) → (Y, OY ).
(b) Show that π−1 G is a π−1 OY -module, and describe how π−1 G ⊗π−1 OY OX
is an OX -module, which we call π∗ G , the pullback of G .
(c) Show that π∗ is a covariant functor ModOY → ModOX .
(d) If furthermore ρ : (W, OW ) → (X, OX ) is another morphism of ringed
spaces, show that “ρ∗ π∗ = (π ◦ ρ)∗ ” (or more precisely, describe a natural
∼
isomorphism of functors ρ∗ π∗ ←→ (π ◦ ρ)∗ ).
∗
(e) (Cf. Exercise 2.7.B.) Show that π is left-adjoint to π∗ . More precisely, for
any F ∈ ModOX , describe a natural bijection
∼
Hom(X,OX ) (π∗ G , F ) o / Hom(Y,O ) (G , π∗ F )
Y

that is functorial in both F ∈ ModOX and G ∈ ModOY .

(f) (the push-pull map for O-modules) Suppose
β′
W /X

α′ α
 
Y
β
/Z
is a commutative (not necessarily Cartesian) diagram of ringed spaces, and
F is an OX -module. Explain how to define the push-pull map
β∗ α∗ F / α∗′ (β ′ )∗ F

(cf. (2.7.4.2)) of OY -modules, by following the discussion of §2.7.4 for the

push-pull map for sheaves.

7.2.E. E ASY IMPORTANT EXERCISE . Given a morphism of ringed spaces π : X → Y

with π(p) = q, show that there is a map of stalks (OY )q → (OX )p .

7.2.F. K EY E XERCISE . Suppose π♯ : B → A is a morphism of rings. Define a

morphism of ringed spaces π : Spec A → Spec B as follows. The map of topo-
logical spaces was given in Exercise 3.4.H. To describe a morphism of sheaves
OSpec B → π∗ OSpec A on Spec B, it suffices to describe a morphism of sheaves on the
distinguished base of Spec B. On D(g) ⊂ Spec B, we define

OSpec B (D(g)) / OSpec A (π−1 D(g)) = OSpec A (D(π♯ g))

206 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

by Bg → Aπ♯ g . Verify that this makes sense (e.g., is independent of g), and that
this describes a morphism of sheaves on the distinguished base. (This is the third
in a series of exercises. We saw that a morphism of rings induces a map of sets in
§3.2.10, a map of topological spaces in Exercise 3.4.H, and now a map of ringed
spaces here.)
The map of ringed spaces of Key Exercise 7.2.F is really not complicated. Here
is an example. Consider the ring map C[y] → C[x] given by y 7→ x2 (see Figure 3.6).
We are mapping the affine line with coordinate x to the affine line with coordinate
y. The map is (on closed points) a 7→ a2 . For example, where does [(x − 3)] go to?
Answer: [(y − 9)], i.e., 3 7→ 9. What is the preimage of [(y − 4)]? Answer: those
prime ideals in C[x] containing [(x2 − 4)], i.e., [(x − 2)] and [(x + 2)], so the preimage
of 4 is indeed ±2. This is just about the map of sets, which is old news (§3.2.10), so
let’s now think about functions pulling back. What is the pullback of the function
3/(y − 4) on D(y − 4) = A1 − {4}? Of course it is 3/(x2 − 4) on A1 \ {−2, 2}.
The construction of Key Exercise 7.2.F will soon be an example of a morphism
of schemes. In fact we could make that definition right now. Before we do, we
point out (via the next exercise) that not every morphism of ringed spaces between
affine schemes is of the form of Key Exercise 7.2.F. (In the language of §7.3, this
morphism of ringed spaces is not a morphism of locally ringed spaces.) We won’t
use this exercise, but it answers a question that everyone will have.

7.2.G. E XERCISE . Recall (Exercise 3.4.K) that Spec k[y](y) has two points, [(0)] and
[(y)], where the second point is closed, and the first is not. Describe a map of ringed
spaces Spec k(x) → Spec k[y](y) sending the unique point of Spec k(x) to the closed
point [(y)], where the pullback map on global sections sends k to k by the identity,
and sends y to x. Show that this map of ringed spaces is not of the form described
in Key Exercise 7.2.F.

7.2.2. Tentative Definition we won’t use (cf. Motivation 7.1.1). A morphism of

schemes π : (X, OX ) → (Y, OY ) is a morphism of ringed spaces that “locally looks
like” the maps of affine schemes described in Key Exercise 7.2.F. Precisely, for each
choice of affine open sets Spec A ⊂ X, Spec B ⊂ Y, such that π(Spec A) ⊂ Spec B,
the induced map of ringed spaces should be of the form shown in Key Exercise 7.2.F.

We would like this definition to be checkable on any affine cover, and we might
hope to use the Affine Communication Lemma to develop the theory in this way.
This works, but it will be more convenient to use a clever trick: in the next section,
we will use the notion of locally ringed spaces, and then once we have used it, we
will discard it like yesterday’s garbage.

7.3 From locally ringed spaces to morphisms of schemes

In order to prove that morphisms behave in a way we hope, we will use

the notion of a locally ringed space. It will not be used later, although it is useful
elsewhere in geometry. The notion of locally ringed spaces (and maps between
them) is inspired by what we know about manifolds (see Exercise 3.1.B). If π : X → Y
is a morphism of manifolds, with π(p) = q, and f is a function on Y vanishing at q,
 © 2024 Ravi Vakil. Published by Princeton University Press. 207

then the pulled back function π♯ (f) on X should vanish on p. Put differently: germs
of functions (at q ∈ Y) vanishing at q should pull back to germs of functions (at
p ∈ X) vanishing at p.

7.3.1. Definition. Recall (Definition 4.3.7) that a locally ringed space is a ringed space
(X, OX ) such that the stalks OX,p are all local rings. A morphism of locally ringed
spaces π : X → Y is a morphism of ringed spaces such that the induced map of
stalks π♯ : OY,q → OX,p (Exercise 7.2.E) sends the maximal ideal mY,q of the former
into the maximal ideal mX,p of the latter (a “morphism of local rings”). (Note in
particular that locally ringed spaces form a category.) This means something rather
concrete and intuitive: “if p 7→ q, and g is a function vanishing at q, then it will
pull back to a function vanishing at p.” You would also want: “if p 7→ q, and g is a
function not vanishing at q, then it will pull back to a function not vanishing at p.”
This is a consequence of the following exercise.

7.3.A. E XERCISE ( VALUES OF FUNCTIONS ON LOCALLY RINGED SPACES PULL BACK ).

If π : X → Y is a morphism of locally ringed spaces, and π(p) = q, show that π
induces an inclusion κ(q) ,→ κ(p) of residue fields. This captures the fact pull-
back sends the locus where a function is zero (respectively, nonzero) to the locus
where the pulled back function is zero (respectively, nonzero). In particular, by
Exercise 4.3.F(b), the pullback of invertible functions are invertible functions.
To summarize: we use the notion of locally ringed space only to define mor-
phisms of schemes, and to show that morphisms have reasonable properties. The
main things you need to remember about locally ringed spaces are (i) that the
functions have values at points, and (ii) that given a map of locally ringed spaces,
values of functions “pull back”, and in particular, zeros of functions “pull back”.

7.3.B. E XERCISE . Show that morphisms of locally ringed spaces glue (cf. Exer-
cise 7.2.A). (Hint: your solution to Exercise 7.2.A may work without change.)

7.3.C. E ASY IMPORTANT EXERCISE . Recall (§4.3.7) that Spec A is a locally ringed
space. Show that the morphism of ringed spaces π : Spec A → Spec B defined by a
ring morphism π♯ : B → A (Exercise 7.2.F) is a morphism of locally ringed spaces.

7.3.2. Key Proposition. — If π : Spec A → Spec B is a morphism of locally ringed

spaces then it is the morphism of locally ringed spaces induced by the map π♯ : B =
Γ (Spec B, OSpec B ) → Γ (Spec A, OSpec A ) = A as in Exercise 7.3.C.
(Aside: Exercise 4.3.A is a special case of Key Proposition 7.3.2. You should
look back at your solution to Exercise 4.3.A, and see where you implicitly used
ideas about locally ringed spaces.)

Proof. Suppose π : Spec A → Spec B is a morphism of locally ringed spaces. We

wish to show that it is determined by its map on global sections π♯ : B → A. We first
need to check that the map of points is determined by global sections. Now a point
p of Spec A can be identified with the prime ideal of global functions vanishing on
it. By Exercise 7.3.A, π(p) as an ideal is the kernel of the natural map B → κ(p).
(Here we use the fact that π is a map of locally ringed spaces.) This is precisely the
way in which the map of sets Spec A → Spec B induced by a ring map B → A was
defined (§3.2.10).
208 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Note in particular that if g ∈ B, π−1 (D(g)) = D(π♯ g), again using the hypothe-
sis that π is a morphism of locally ringed spaces.
It remains to show that π♯ : OSpec B → π∗ OSpec A is the morphism of sheaves
given by Exercise 7.2.F (cf. Exercise 7.3.C). It suffices to check this on the distin-
guished base (Exercise 2.5.C(a)). We now want to check that for any map of locally
ringed spaces inducing the map of sheaves OSpec B → π∗ OSpec A , the map of sections
on any distinguished open set D(g) ⊂ Spec B (g ∈ B) is determined by the map of
global sections B → A.
Consider the commutative diagram

π♯Spec B
B Γ (Spec B, OSpec B ) / Γ (Spec A, OSpec A ) A
resSpec B,D(g) resSpec A,D(π♯ g)
 π♯D(g) 
Bg Γ (D(g), OSpec B ) / Γ (D(π♯ g), OSpec A ) Aπ♯ g = A ⊗B Bg .

The vertical arrows (restrictions to distinguished open sets) are localizations by

g, so the lower horizontal map π♯D(g) is determined by the upper map (it is just
localization by g). □
We are ready for our definition.

7.3.3. Definition. If X and Y are schemes, then a morphism π : X → Y as locally

ringed spaces is called a morphism of schemes. We have thus defined the category
of schemes, which we denote Sch. (We then have notions of isomorphism — just
the same as before, §4.3.1 — and automorphism. The target Y of π is sometimes
called the base scheme or the base, when we are interpreting π as a “family of
schemes parametrized by Y” — this may become clearer once we have defined the
fibers of morphisms in §10.3.2.)
The definition in terms of locally ringed spaces easily implies Tentative Defini-
tion 7.2.2:

7.3.D. I MPORTANT E XERCISE . Show that a morphism of schemes π : X → Y is a

morphism of ringed spaces that looks locally like morphisms of affine schemes.
Precisely, if Spec A is an affine open subset of X and Spec B is an affine open subset
of Y, and π(Spec A) ⊂ Spec B, then the induced morphism of ringed spaces is a
morphism of affine schemes. (In case it helps, note: if W ⊂ X and Y ⊂ Z are both
open embeddings of ringed spaces, then any morphism of ringed spaces X → Y
induces a morphism of ringed spaces W → Z, by composition W → X → Y → Z.)
Show that it suffices to check on a set (Spec Ai , Spec Bi ) where the Spec Ai form an
open cover of X and the Spec Bi form an open cover of Y.
In practice, we will use the affine cover interpretation, and forget completely
about locally ringed spaces. In particular, put imprecisely, the category of affine
schemes is the category of rings with the arrows reversed. More precisely:

7.3.E. E XERCISE . Show that the category of rings and the opposite category of
affine schemes are equivalent (see §1.1.16 and §1.1.21 to read about opposite and
equivalent categories, respectively).
 © 2024 Ravi Vakil. Published by Princeton University Press. 209

In particular, there can be many different maps from one point to another! For
example, here are two different maps from the point Spec C to the point Spec C: the
identity (corresponding to the identity C → C), and complex conjugation. (There
are even more such maps!)
It is clear (from the corresponding facts about locally ringed spaces) that mor-
phisms of schemes glue (Exercise 7.3.B), and the composition of two morphisms is a
morphism. Isomorphisms in this category are precisely what we defined them to
be earlier (§4.3.7).

7.3.F. E NLIGHTENING E XERCISE . (This exercise can give you some practice with
understanding morphisms of schemes by cutting up into affine open sets.) Make
sense of the following sentence: “An+1
k \ {⃗0} → Pn
k given by

(x0 , x1 , . . . , xn )  / [x0 , x1 , . . . , xn ]

is a morphism of schemes.” Caution: you can’t just say where points go; you have
to say where functions go. So you may have to divide these up into affines, and
describe the maps, and check that they glue. (Can you generalize to the case where
k is replaced by a general ring B? See Exercise 7.3.O for an answer.)

7.3.4. Morphisms to affine schemes.

The following result shows that it is easy to describe morphisms to an affine
scheme without working hard to cover the source with affine open sets.

7.3.G. E SSENTIAL EXERCISE . Show that morphisms X → Spec A are in natural

bijection with ring morphisms A → Γ (X, OX ). Hint: Show that this is true when X
is affine. Use the fact that morphisms glue, Exercise 7.3.B. (This is even true in the
category of locally ringed spaces. You are free to prove it in this generality, but it is
easier in the category of schemes. You might find in this more general statement an
echo of the slogan of double-starred Remark 6.2.5.)
In particular, there is a canonical morphism from a scheme to Spec of its ring
of global sections. (Warning: Even if X is a finite type k-scheme, the ring of
global sections might be nasty! In particular, it might not be finitely generated, see
19.11.13.) The canonical morphism X → Spec Γ (X, OX ) is an isomorphism if and
only if X is affine (i.e., isomorphic to Spec A for some ring A), and in this case it is
the isomorphism hinted at in Remark 4.3.2.

7.3.H. E ASY E XERCISE . If S• is a finitely generated graded A-algebra, describe a

natural “structure morphism” Proj S• → Spec A.

7.3.5. Remark. From Essential Exercise 7.3.G, it is one small step to show that some
products of schemes exist: if A and B are rings, then Spec A×Spec B = Spec(A⊗Z B);
and if A and B are C-algebras, then Spec A ×Spec C Spec B = Spec(A ⊗C B). But we
are in no hurry, so we wait until Exercise 10.1.A to discuss this properly.

7.3.6. ⋆⋆ Side fact for experts: Γ and Spec are adjoints. We have a contravariant functor
Spec from rings to locally ringed spaces, and a contravariant functor Γ from locally
ringed spaces to rings. In fact (Γ, Spec) is an adjoint pair! (Caution: we have only
discussed adjoints for covariant functors; if you care, you will have to figure out
how to define adjoints for contravariant functors.) Thus we could have defined
210 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Spec by requiring it to be right-adjoint to Γ . (Fun but irrelevant side question: if

you used ringed spaces rather than locally ringed spaces, Γ again has a right adjoint.
What is it?)
Our ability to easily describe morphisms to affine schemes will allow us to
revisit our earlier discussions of schemes over a given field or ring (§5.3.6).

7.3.7. The category of complex schemes (or more generally the category of k-
schemes where k is a field, or more generally the category of A-schemes where A
is a ring, or more generally the category of S-schemes where S is a scheme). The
category of S-schemes SchS (where S is a scheme) is defined as follows. The objects
(S-schemes) are morphisms of the form
X


S
(The morphism to S is called the structure morphism. A motivation for this termi-
nology is the fact that if S = Spec A, the structure morphism gives the functions on
each open set of X the structure of an A-algebra, cf. §5.3.6.) The morphisms in the
category of S-schemes are defined to be commutative diagrams
X /Y

 
S
= /S
which is more conveniently written as a commutative diagram
X /Y

 
S
When there is no confusion (if the base scheme is clear), simply the top row of the
diagram is given. In the case where S = Spec A, where A is a ring, we get the notion
of an A-scheme, which is the same as the same definition as in §5.3.6 (Exercise 7.3.I),
but in a more satisfactory form. For example, complex geometers may consider the
category of C-schemes.

7.3.I. E ASY E XERCISE . Show that this definition of A-scheme given in §7.3.7 agrees
with the earlier definition of §5.3.6.

7.3.J. E ASY E XERCISE . Show that Spec Z is the final object in the category of
schemes. In other words, if X is any scheme, there exists a unique morphism to
Spec Z. (Hence the category of schemes is isomorphic to the category of Z-schemes.)
If A is any ring (for example, a field k), show that Spec A is the final object in the
category of A-schemes.

7.3.8. Coproducts of schemes.

7.3.K. E ASY
` E XERCISE . Suppose X and Y are schemes. Show that their disjoint
union X Y (defined in Exercise 4.3.E) is the coproduct of X and Y in the category of
 © 2024 Ravi Vakil. Published by Princeton University Press. 211

`
schemes, justifying the use of the coproduct symbol . (See §3.6.3 for a surprising
connection to idempotents.)

7.3.9. Morphisms from (some) affine schemes.

Morphisms from affine schemes are not quite as simple as morphisms to affine
schemes, but some cases are worth pointing out.

7.3.L. E XERCISE .
(a) Suppose p is a point of a scheme X. Describe a canonical (choice-free) morphism
Spec OX,p → X. (Hint: do this for affine X first. But then for general X be sure to
show that your morphism is independent of choice.)
(b) Define a canonical morphism Spec κ(p) → X. (This is often written p → X; one
gives p the obvious interpretation as a scheme, Spec κ(p).)

7.3.M. E XERCISE ( MORPHISMS FROM Spec OF A LOCAL RING TO X). Suppose

X is a scheme, and (A, m) is a local ring. Suppose we have a scheme morphism
π : Spec A → X sending [m] to p. Show that any open set containing p contains the
image of π. Show that there is a bijection between Mor(Spec A, X) and

{(p ∈ X, some local homomorphism OX,p → A)} .

(Possible hint: Exercise 7.3.L(a).)

These exercises lead us to the notion of field-valued points, and more generally
ring-valued points, and more generally still, scheme-valued points.

7.3.10. Definition: The functor of points, and scheme-valued points (and ring-
valued points, and field-valued points) of a scheme. If Z is a scheme, then
Z-valued points of a scheme X, denoted X(Z), are defined to be maps Z → X. If A
is a ring, then A-valued points of a scheme X, denoted X(A), are defined to be the
(Spec A)-valued points of the scheme. (The most common case of this is when A is
a field.)
If you are working over a base scheme B — for example, complex algebraic
geometers will consider only schemes and morphisms over B = Spec C — then in
the above definition, there is an implicit structure map Z → B (or Spec A → B in
the case of X(A)). For example, for a complex geometer, if X is a scheme over C, the
C(t)-valued points of X correspond to commutative diagrams of the form

Spec C(t) /X

π
ξ
% |
Spec C

where π : X → Spec C is the structure map for X, and ξ corresponds to the obvious
inclusion of rings C → C(t). (Warning: a k-valued point of a k-scheme X is
sometimes called a “rational point” of X, which is dangerous, as for most of the
world, “rational” refers to Q. We will use the safer phrase “k-valued point” of X.
Another safe choice is “k-rational point” of X or k-point.)
The terminology “Z-valued point” (and A-valued point) is unfortunate, because
we earlier defined the notion of points of a scheme, and Z-valued points (and A-
valued points) are not (necessarily) points! But these usages are well-established in
212 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

the literature. (Look in the index under “point” to see even more inconsistent use
of adjectives that modify this word.)

7.3.N. E XERCISE .
(a) (easy) Show that a morphism of schemes X → Y induces a map of Z-valued
points X(Z) → Y(Z).
(b) Note that morphisms of schemes X → Y are not determined by their “underlying”
map of points. (What is an example?) Show that they are determined by their
induced maps of Z-valued points, as Z varies over all schemes. (Hint: pick Z = X.
In the course of doing this exercise, you will largely prove Yoneda’s Lemma in the
guise of Exercise 10.1.B.)

7.3.11. Furthermore, we will see that “products of Z-valued points” behave as you
might hope (§10.1.3). A related reason this language is suggestive: the notation X(Z)
suggests the interpretation of X as a (contravariant) functor hX from schemes to
sets — the functor of (scheme-valued) points of the scheme X (cf. Example 1.1.20).
Here is a low-brow reason A-valued points are a useful notion: the A-valued
points of an affine scheme Spec Z[x1 , . . . , xn ]/(f1 , . . . , fr ) (where fi ∈ Z[x1 , . . . , xn ] are
relations) are precisely the solutions to the equations
f1 (x1 , . . . , xn ) = · · · = fr (x1 , . . . , xn ) = 0
in the ring A. For example, the rational solutions to x2 + y2 = 16 are precisely the
Q-valued points of Spec Z[x, y]/(x2 + y2 − 16). The integral solutions are precisely
the Z-valued points. So A-valued points of an affine scheme (finite type over Z)
can be interpreted simply. In the special case where A is local, A-valued points of a
general scheme have a good interpretation too (Exercises 7.3.L and 7.3.M).
On the other hand, Z-valued points of projective space can be subtle. There are
some maps we can write down easily, as shown by applying the next exercise in
the case X = Spec A, where A is a B-algebra.

7.3.O. E ASY ( BUT SURPRISINGLY ENLIGHTENING ) E XERCISE ( CF. E XERCISE 7.3.F).

(a) Suppose B is a ring. If X is a B-scheme, and f0 , . . . , fn are n + 1 functions on X
with no common zeros, then show that [f0 , . . . , fn ] gives a morphism of B-schemes
X → Pn B.
(b) Suppose g is a nowhere vanishing function on X, and fi are as in part (a). Show
that the morphisms [f0 , . . . , fn ] and [gf0 , . . . , gfn ] to Pn
B are the same.

7.3.12. Example (Exercise 7.3.F revisited). Consider the n + 1 functions x0 , . . . , xn

on An+1 (otherwise known as n + 1 sections of the trivial bundle). They have no
common zeros on An+1 \ {⃗0}. Hence they determine a morphism An+1 \ {⃗0} → Pn .

7.3.13. You might hope that Exercise 7.3.O(a) gives all morphisms to projective space
(over B). But this isn’t the case. Indeed, even the identity morphism X = P1k → P1k
isn’t of this form, as the source P1 has no nonconstant global functions with which
to build this map. (There are similar examples with an affine source.) However,
there is a correct generalization (characterizing all maps from schemes to projective
schemes) in Theorem 15.2.2. This result roughly states that this works, so long as the
fi are not quite functions, but sections of a line bundle. Our desire to understand
maps to projective schemes in a clean way will be one important motivation for
understanding line bundles.
 © 2024 Ravi Vakil. Published by Princeton University Press. 213

We will see more ways to describe maps to projective space in the next sec-
tion. A different description directly generalizing Exercise 7.3.O(a) will be given
in Exercise 15.2.A, which will turn out (in Theorem 15.2.2) to be a “universal”
description.
Incidentally, before Grothendieck, it was considered a real problem to figure
out the right way to interpret points of projective space with “coordinates” in a ring.
These difficulties were due to a lack of functorial reasoning. And the clues to the
right answer already existed (the same problems arise for maps from a manifold to
RPn ) — if you ask such a geometric question (for projective space is geometric),
the answer is necessarily geometric, not purely algebraic!

7.3.14. Visualizing morphisms: Picturing maps of schemes when nilpotents are present.
You now know how to visualize the points of schemes (§3.3), and nilpotents (§4.2
and §6.6). The following imprecise exercise will give you some sense of how to
visualize maps of schemes when nilpotents are involved. Suppose a ∈ C. Consider
the map of rings C[x] → C[ϵ]/(ϵ2 ) given by x 7→ aϵ. Recall that Spec C[ϵ]/(ϵ2 ) may
be pictured as a point with a tangent vector (§4.2). How would you picture this
map if a ̸= 0? How does your picture change if a = 0? (The tangent vector should
be “crushed” in this case.)
Exercise 13.1.G will extend this considerably. You may enjoy reading its state-
ment now.

7.3.15. ⋆⋆ For readers with appropriate background: Analytification of complex

algebraic varieties. Warning: Any discussion of analytification is only for readers
who are familiar with the notion of complex analytic varieties, or willing to develop
it on their own in parallel with our development of schemes.

7.3.P. E XERCISE ( ANALYTIFYING COMPLEX ALGEBRAIC VARIETIES ). Suppose X

is a reduced, finite type C-scheme. Define the corresponding complex analytic
prevariety Xan . (The definition of an analytic prevariety is the same as the definition
of a variety without the Hausdorff condition.) Caution: your definition should not
depend on a choice of an affine cover of X. (Hint: First explain how to analytify
reduced finite type affine C-schemes. Then glue.) Give a bijection between the
closed points of X and the points of Xan , using the weak Nullstellensatz 3.2.5. (In fact
one may construct a continuous map of sets Xan → X generalizing Exercise 3.2.I.)

7.3.Q. E XERCISE ( THE ANALYTIFICATION FUNCTOR ). Recall the analytification

construction of Exercise 7.3.16. For each morphism of reduced finite type C-schemes
π : X → Y (over C), define a morphism of complex analytic prevarieties πan : Xan →
Yan (the analytification of π). Show that analytification gives a functor from the
category of reduced finite type C-schemes to the category of complex analytic
prevarieties.

7.3.16. Remark. Two nonisomorphic varieties can have isomorphic analytifications.

For example, Serre described two different algebraic structures on the complex
manifold C∗ × C∗ , see [Ha2, p. 232] and [MO68421]; one is “the obvious one”, and
the other is a P1 -bundle over an elliptic curve, with a section removed. For an
example of a smooth complex surface with infinitely many algebraic structures,
see §19.11.3. On the other hand, a compact complex variety can have only one
algebraic structure (see [Se3, §19]).
214 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

7.3.17. Further facts about analytification. For more on analytification, see Exer-
cises 11.2.K (Hausdorffness), 14.1.J, 18.4.H, 18.4.U (line bundles), §18.6.5, §20.1.9
(degree), and Exercise 21.2.W (manifolds). For background on complex analytic
spaces, see [GR].

7.4 Maps of graded rings and maps of projective schemes

As maps of rings correspond to maps of affine schemes in the opposite direction,
maps of graded rings (over a base ring A) sometimes give maps of projective
schemes in the opposite direction. This is an imperfect generalization: not every
map of graded rings gives a map of projective schemes (§7.4.2); not every map
of projective schemes comes from a map of graded rings (§19.11.9); and different
maps of graded rings can yield the same map of schemes (Exercise 7.4.C).
You may find it helpful to think through Examples 7.4.1 and 7.4.2 while working
through the following exercise.

7.4.A. E SSENTIAL EXERCISE . Suppose that ϕ : S• / R• is a morphism of (Z≥0 -

)graded rings. (By map of graded rings, we mean a map of rings that preserves
the grading as a map of “graded abelian semigroups” (or “graded commutative
monoids”). In other words, there is a d > 0 such that Sn maps to Rdn for all n.)
Show that this induces a morphism of schemes
(Proj R• ) \ V(ϕ(S+ )) → Proj S• .
(Hint: Suppose f is a homogeneous element of S+ . Define a map D(ϕ(f)) → D(f).
Show that they glue together (as f runs over all homogeneous elements of S+ ).
Show that this defines a map from all of Proj R• \ V(ϕ(S+ )).) In particular, if
(7.4.0.1) V(ϕ(S+ )) = ∅,
then we have a morphism Proj R• → Proj S• . From your solution, it will be clear
that if ϕ is furthermore a morphism of A-algebras, then the induced morphism
Proj R• \ V(ϕ(S+ )) → Proj S• is a morphism of A-schemes.

7.4.1. Example. Let’s see Exercise 7.4.A in action. We will scheme-theoretically

interpret the map of complex projective manifolds CP1 to CP2 given by
CP1 / CP2

[s, t]  / [s20 , s9 t11 , t20 ]

Notice first that this is well-defined: [λs, λt] is sent to the same point of CP2 as [s, t].
The reason for it to be well-defined is that the three polynomials s20 , s9 t11 , and t20
are all homogeneous of degree 20.
Algebraically, this corresponds to a map of graded rings in the opposite direc-
tion
C[x, y, z] / C[s, t]
given by x 7→ s20 , y 7→ s9 t11 , z 7→ t20 . You should interpret this in light of your
solution to Exercise 7.4.A, and compare this to the affine example of §3.2.11.
 © 2024 Ravi Vakil. Published by Princeton University Press. 215

7.4.2. Example. Notice that there is no map of complex manifolds CP2 → CP1
given by [x, y, z] 7→ [x, y], because the map is not defined when x = y = 0. This
corresponds to the fact that the map of graded rings C[s, t] → C[x, y, z] given by
s 7→ x and t 7→ y, doesn’t satisfy hypothesis (7.4.0.1).

7.4.B. E XERCISE . Show that if ϕ : S• → R• satisfies (ϕ(S+ )) = R+ , then hypoth-

p

esis (7.4.0.1) is satisfied. (Hint: Exercise 4.5.L.) This algebraic formulation of the
more geometric hypothesis can sometimes be easier to verify.

7.4.C. U NIMPORTANT EXERCISE . This exercise shows that different maps of

graded rings can give the same map of schemes. Let R• = k[x, y, z]/(xz, yz, z2 )
and S• = k[a, b, c]/(ac, bc, c2 ), where every variable has degree 1. Show that
∼ Proj S• =
Proj R• = ∼ P1 . Show that the maps S• → R• given by (a, b, c) 7→ (x, y, z)
k
and (a, b, c) 7→ (x, y, 0) give the same (iso)morphism Proj R• → Proj S• . (The real
reason is that all of these constructions are insensitive to what happens in a finite
number of degrees. This will be made precise in a number of ways later, most
immediately in Exercise 7.4.F.)

7.4.3. Unimportant remark. Exercise 15.2.H shows that not every morphism of
schemes Proj R• → Proj S• comes from a map of graded rings S• → R• , even in
quite reasonable circumstances.

7.4.4. Veronese subrings.

Here is a useful construction. Suppose S• is a finitely generated graded ring.
Define the nth Veronese subring of S• by Sn• = ⊕∞ j=0 Snj . (The “old degree” n is
“new degree” 1.) The geometric interpretation of this construction is the important
Veronese embedding, discussed in §9.3.6.

7.4.D. E XERCISE . Show that the map of graded rings Sn• ,→ S• induces an isomor-
∼
phism Proj S• −→ Proj Sn• . (Hint: if f ∈ S+ is homogeneous of degree divisible by
n, identify D(f) on Proj S• with D(f) on Proj Sn• . Why do such distinguished open
sets cover Proj S• ?)

7.4.E. E XERCISE . If S• is generated in degree 1, show that Sn• is also generated in

degree 1. (You may want to consider the case of the polynomial ring first.)

7.4.F. E XERCISE . Show that if R• and S• are the same finitely generated graded rings
except in a finite number of nonzero degrees (make this precise!), then Proj R• = ∼
Proj S• .

7.4.G. E XERCISE . Suppose S• is generated over S0 by f1 , . . . , fn . Find a d such that

Sd• is finitely generated in “new” degree 1 (= “old” degree d). (This is surprisingly
tricky, so here is a hint. Suppose there are generators x1 , . . . , xn of degrees d1 , . . . ,
dn respectively.
Q Show that any monomial xa 1 an
1 · · · xn of degree at least nd1 . . . dn
has ai ≥ ( j dj )/di for some i. Show that the (nd1 . . . dn )th Veronese subring is
generated by elements in “new” degree 1.)
Exercise 7.4.G, in combination with Exercise 7.4.D, shows that there is little
harm in assuming that finitely generated graded rings are generated in degree 1, as
after a regrading (or more precisely, keeping only terms of degree a multiple of d,
then dividing the degree by d), this is indeed the case. This is handy, as it means
216 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

that, using Exercise 7.4.D, we can assume that any finitely generated graded ring is
generated in degree 1. Exercise 9.3.F will later imply as a consequence that we can
embed every Proj in some projective space.

7.4.H. L ESS IMPORTANT EXERCISE . Suppose S• is a finitely generated graded ring.

Show that Sn• is a finitely generated graded ring. (Possible approach: use the
previous exercise, or something similar, to show there is some N such that SnN• is
generated in degree 1, so the graded ring SnN• is finitely generated. Then show
that for each 0 < j < N, SnN•+nj is a finitely generated module over SnN• .)

7.5 Rational maps from reduced schemes

Informally speaking, a “rational map” is “a morphism defined almost every-

where”, much as a rational function (Definition 6.6.36, Exercise 5.2.I) is a name for a
function defined almost everywhere. We will later see that in good situations, just
as with rational functions, where a rational map is defined, it is uniquely defined
(the Reduced-to-Separated Theorem 11.3.2), and has a largest “domain of definition”
(§11.3.3). For this section only, we assume X to be reduced. A key example will be
irreducible varieties (§7.5.6), and the language of rational maps is most often used
in this case.

7.5.1. Definition. A rational map π of reduced schemes, from X to Y, denoted

π : X 99K Y, is the data of a morphism α : U → Y from a dense open set U ⊂ X, with
the equivalence relation (α : U → Y) ∼ (β : V → Y) if there is a dense open set Z ⊂
U ∩ V such that α|Z = β|Z . (In §11.3.3, we will improve this to: if α|U∩V = β|U∩V
in good circumstances — when Y is separated.) People often use the word “map”
for “morphism”, which is quite reasonable, except that a rational map need not be
a map. So to avoid confusion, when one means “rational map”, one should never
just say “map”.
We will also discuss rational maps of S-schemes for a scheme S. The definition
is the same, except now X and Y are S-schemes, and α : U → Y is a morphism of
S-schemes.

7.5.2. ⋆ Rational maps more generally. Just as with rational functions, Definition 7.5.1
can be extended to where X is not reduced, as is (using the same name, “rational
map”), or in a version that imposes some control over what happens over the
nonreduced locus (pseudo-morphisms, [Stacks, tag 01RX]). We will see in §11.3 that
rational maps from reduced schemes to separated schemes behave particularly
well, which is why they are usually considered in this context. The reason for the
definition of pseudo-morphisms is to extend these results to when X is nonreduced.
We will not use the notion of pseudo-morphism.

7.5.3. An obvious example of a rational map is a morphism. Another important

example is the projection Pn n−1
A 99K PA given by [x0 , · · · , xn ] 7→ [x0 , · · · , xn−1 ].
(How precisely is this a rational map in the sense of Definition 7.5.1?)

7.5.A. E XERCISE . Suppose X is a reduced scheme. Describe how rational maps

X 99K A1 should be identified with rational functions on X.
 © 2024 Ravi Vakil. Published by Princeton University Press. 217

A rational map π : X 99K Y is dominant (or in some sources, dominating) if for

some (and hence every) representative U → Y, the image is dense in Y. A morphism
is a dominant morphism (or dominating morphism) if it is dominant as a rational
map.

7.5.B. E XERCISE . Show that a rational map π : X 99K Y of irreducible schemes is

dominant if and only if π sends the generic point of X to the generic point of Y.
A little thought will convince you that you can compose (in a well-defined
way) a dominant map π : X 99K Y from an irreducible scheme X to an irreducible
scheme Y with a rational map ρ : Y 99K Z. Furthermore, the composition ρ ◦ π
will be dominant if ρ is dominant. Integral schemes and dominant rational maps
between them form a category which is geometrically interesting.

7.5.C. E ASY E XERCISE . Show that dominant rational maps of integral schemes
give morphisms of function fields in the opposite direction.
In “suitably classical situations” (integral finite type k-schemes — and in par-
ticular, irreducible varieties, to be defined in §11.2.9) — this is reversible: dominant
rational maps correspond to inclusions of function fields in the opposite direc-
tion. We make this precise in §7.5.6 below. But it is not true that morphisms of
function fields always give dominant rational maps, or even rational maps. For
example, Spec k[x] and Spec k(x) have the same function field k(x), but there is no
corresponding rational map Spec k[x] 99K Spec k(x) of k-schemes. Reason: such a
rational map would correspond to a morphism from an open subset U of Spec k[x],
say Spec k[x, 1/f(x)], to Spec k(x). But there is no map of rings k(x) → k[x, 1/f(x)]
(sending k identically to k and x to x) for any one f(x).
(If you want more evidence that the topologically-defined notion of dominance
is simultaneously algebraic, you can show that if ϕ : A → B is a ring morphism,
then the corresponding morphism Spec B → Spec A is dominant if and only if ϕ
has kernel contained in the nilradical of A.)

7.5.4. Definition. A rational map π : X 99K Y is said to be birational if it is dominant,

and there is another rational map (a “rational inverse”) ψ that is also dominant, such
that π ◦ ψ is (in the same equivalence class as) the identity on Y, and ψ ◦ π is (in the
same equivalence class as) the identity on X. This is the notion of isomorphism in
the category of integral schemes and dominant rational maps. (In the differentiable
category, this is not particularly interesting.) A morphism is a birational morphism
if it is birational as a rational map. (Note that the “inverse” to a birational morphism
may be only a rational map, not a morphism.)
We say X and Y are birational (to each other) if there exists a birational map
X 99K Y. If X and Y are irreducible, then birational maps induce isomorphisms
of function fields. The fact that maps of function fields correspond to rational
maps in the opposite direction for integral finite type k-schemes, to be proved in
Proposition 7.5.7, shows that a map between integral finite type k-schemes that
induces an isomorphism of function fields is birational. An integral finite type
k-scheme is said to be rational if it is birational to An k for some n.

7.5.5. Proposition. — Suppose X and Y are reduced schemes. Then X and Y are birational
if and only if there is a dense open subscheme U of X and a dense open subscheme V of Y
such that U = ∼ V.
218 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

(We have “integral” hypotheses for simplicity. The hypotheses can be dropped
with sufficient care and modified definitions.)
Proposition 7.5.5 tells you how to think of birational maps. Just as a rational
map is a “mostly-defined function”, two birational reduced schemes are “mostly
isomorphic”. For example, a reduced finite type k-scheme (such as an affine k-
variety) is rational if and only if it has a dense open subscheme isomorphic to an
open subscheme of An k.

Proof. The “if” direction is immediate, so we prove the “only if” direction. We
basically follow our nose, and use what we are given (see Figure 7.1).

g◦f f◦g
f g◦f =id as: g f◦g =id as:

∼
rat’l
map
rat’l
map
rat’l
map XO ^ o /Y
@ O
rat’l
map
rat’l
map
rat’l
map

? ?
morphism morphism
” ” XO 1 @ YO 1 ” ”

?
morphism
” ” XO 2

? ?
” ” morphism
XO 3 o YO 2 ” morphism
”

? 
X4 o
∼ / Y? 3 ” ” morphism

F IGURE 7.1. Outline of the proof of Proposition 7.5.5

We have inverse rational maps F : X 99K Y and G : Y 99K X. Choose repre-

sentative morphisms f : X1 → Y (where X1 ⊂ X) and g : Y1 → X (where Y1 ⊂ Y)
for the rational maps F and G respectively. If X2 = (f|X1 )−1 (Y1 ) ⊂ X1 , then g ◦ f
is a morphism from X2 to X that is the identity as a rational map. Thus there is a
dense open subset X3 ⊂ X2 such that the morphism g ◦ f : X3 → X is the identity
morphism onto X3 (or more precisely, it is the open embedding X3 ,→ X).
Similarly, let Y2 = (g|Y1 )−1 (X3 ) ⊂ Y1 , so f ◦ g is a morphism from Y2 to Y that
is the identity as a rational map. Then let Y3 ⊂ Y2 be a dense open subset such that
f ◦ g : Y3 → Y is the inclusion (the “identity”).
Finally, if X4 = (f|X3 )−1 (Y3 ) ⊂ X3 , then (g ◦ f)|X4 is the identity morphism on
X4 (by way of Y3 ), and (f ◦ g)|Y3 is the identity morphism on Y3 (by way of X4 ),
so we have found our isomorphism of open sets that is a “representative” for our
birational map. □

7.5.6. Rational maps of irreducible varieties.

 © 2024 Ravi Vakil. Published by Princeton University Press. 219

7.5.7. Proposition. — Suppose X is an integral k-scheme and Y is an integral finite type

k-scheme, and we are given an extension of function fields ϕ♯ : K(Y) ,→ K(X) preserving k.
Then there exists a dominant rational map of k-schemes ϕ : X 99K Y inducing ϕ♯ .

Proof. By replacing Y with an open subset, we may assume that Y is affine, say
Spec B, where B is generated over k by finitely many elements y1 , . . . , yn . Since
we only need to define ϕ on an open subset of X, we may similarly assume that
X = Spec A is affine. Then ϕ♯Q gives an inclusion ϕ♯ : B ,→ K(A). Write ϕ♯ (yi ) as
fi /gi (fi , gi ∈ A), and let g := gi . Then ϕ♯ further induces an inclusion B ,→ Ag .
Therefore ϕ : Spec Ag → Spec B induces ϕ♯ . The morphism ϕ is dominant because
the inverse image of the zero ideal under the inclusion B ,→ Ag is the zero ideal, so
ϕ takes the generic point of X to the generic point of Y. □

7.5.D. E XERCISE . Let K be a finitely generated field extension of k. (Recall that a

field extension K over k is finitely generated if there is a finite “generating set” x1 ,
. . . , xn in K such that every element of K can be written as a rational function in x1 ,
. . . , xn with coefficients in k.) Show that there exists an irreducible affine k-variety
with function field K. (Hint: Consider the map k[t1 , . . . , tn ] → K given by ti 7→ xi ,
and show that the kernel is a prime ideal p, and that k[t1 , . . . , tn ]/p has fraction
field K. Interpreted geometrically: consider the map Spec K → Spec k[t1 , . . . , tn ]
given by the ring map ti 7→ xi , and take the closure of the one-point image.)

7.5.E. E XERCISE . Describe equivalences of categories among the following.

(a) the category with objects “integral affine k-varieties”, and morphisms
“dominant rational maps defined over k”; and
(b) the opposite (“arrows-reversed”) category with objects “finitely generated
field extensions of k”, and morphisms “inclusions extending the identity
on k”.
(Once we define varieties in general in §11.2.9, you can add in: (c) the category with
objects “integral k-varieties”, and morphisms “dominant rational maps defined
over k”.)
In particular, an integral affine k-variety X is rational if its function field K(X) is
a purely transcendental extension of k, i.e., K(X) = ∼ k(x1 , . . . , xn ) for some n. (This
needs to be said more precisely: the map k ,→ K(X) induced by X → Spec k should
agree with the “obvious” map k ,→ k(x1 , . . . , xn ) under this isomorphism.)

7.5.8. More examples of rational maps.

A recurring theme in these examples is that domains of definition of rational
maps to projective schemes extend over regular codimension one points. We will
make this precise in the Curve-to-Projective Extension Theorem 15.3.1, when we
discuss curves.
The first example is the classical formula for Pythagorean triples, and its deriva-
tion by “stereographic projection”. Suppose you are looking for rational points on
the circle C given by x2 + y2 = 1 (Figure 7.2). One rational point is p = (1, 0). If q
is another rational point, then pq is a line of rational (non-infinite) slope. This gives
a rational map from the conic C (now interpreted as Spec Q[x, y]/(x2 + y2 − 1))
to A1Q , given by (x, y) 7→ y/(x − 1). (Something subtle just happened: we were
talking about Q-points on a circle, and ended up with a rational map of schemes.)
220 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

F IGURE 7.2. Finding primitive Pythagorean triples using geometry

Conversely, given a line of slope m through p, where m is rational, we can recover q

by solving the equations y = m(x − 1), x2 + y2 = 1. We substitute the first equation
into the second, to get a quadratic equation in x. We know that we will have a
solution x = 1 (because the line meets the circle at (x, y) = (1, 0)), so we expect to
be able to factor this out, and find the other factor. This indeed works:

x2 + (m(x − 1))2 = 1
=⇒ (m2 + 1)x2 + (−2m2 )x + (m2 − 1) = 0
=⇒ (x − 1)((m2 + 1)x − (m2 − 1)) = 0

The other solution is x = (m2 − 1)/(m2 + 1), which gives y = −2m/(m2 + 1). Thus
we get a birational map between the conic C and A1 with coordinate m, given by
f : (x, y) 7→ y/(x − 1) (which is defined for x ̸= 1), and with inverse rational map
given by m 7→ ((m2 − 1)/(m2 + 1), −2m/(m2 + 1)) (which is defined away from
m2 + 1 = 0).
We can extend this to a rational map C 99K P1Q via the “inclusion” A1Q → P1Q
(which we later call an open embedding). Then f is given by (x, y) 7→ [y, x − 1].
We then have an interesting question: what is the domain of definition of f? It
appears to be defined everywhere except for where y = x − 1 = 0, i.e., everywhere
but p. But in fact it can be extended over p! Note that (x, y) 7→ [x + 1, −y] (where
(x, y) ̸= (−1, 0)) agrees with f on their common domains of definition, as [x +
1, −y] = [y, x − 1]. Hence this rational map can be extended farther than we
at first thought. This will be a special case of the Curve-to-Projective Extension
Theorem 15.3.1.

7.5.F. E XERCISE . Use the above to find a “formula” yielding all Pythagorean triples.

7.5.G. E XERCISE . Show that the conic x2 + y2 = z2 in P2k is isomorphic to P1k for any
field k of characteristic not 2. (See Exercise 4.5.S for the definition of “x2 + y2 = z2
in P2k ” if you need it.)

7.5.9. In fact, any conic in P2k with a k-valued point (i.e., a point with residue field
k) of rank 3 (after base change to k, so “rank” makes sense, see Exercise 5.4.J) is
 © 2024 Ravi Vakil. Published by Princeton University Press. 221

isomorphic to P1k . (The hypothesis of having a k-valued point is certainly necessary:

x2 + y2 + z2 = 0 over k = R is a conic that is not isomorphic to P1k .)

7.5.H. E XERCISE . Find all rational solutions to y2 = x3 + x2 , by finding a birational

map to A1Q , mimicking what worked with the conic. Hint: what point should you
project from? (In Exercise 19.10.I, we will see that these points basically form a
group, and that this is a degenerate elliptic curve.)
You will obtain a rational map to P1Q that is not defined over the “node” x =
y = 0, and cannot be extended over this codimension 1 set. This is an example of
the limits of our future result, the Curve-to-Projective Extension Theorem 15.3.1,
showing how to extend rational maps to projective space over codimension 1 sets:
the codimension 1 sets have to be regular.

7.5.I. E XERCISE . Use a similar idea to find a birational map from the quadric
surface Q = {x2 + y2 = w2 + z2 } ⊂ P3Q to P2Q . Use this to find all rational points on
Q. (This illustrates a good way of solving Diophantine equations. You will find
a dense open subset of Q that is isomorphic to a dense open subset of P2 , where
you can easily find all the rational points. There will be a closed subset of Q where
the rational map is not defined, or not an isomorphism, but you can deal with this
subset in an ad hoc fashion.)

7.5.J. E XERCISE ( THE C REMONA TRANSFORMATION , A USEFUL CLASSICAL CON -

STRUCTION ). Consider the rational map P2 k 99K Pk , given by [x, y, z] → [1/x, 1/y, 1/z].
2

What is the domain of definition? (It is bigger than the locus where xyz ̸= 0!) You
will observe that you can extend it over “codimension 1 sets” (ignoring the fact
that we don’t yet know what codimension means). This again foreshadows the
Curve-to-Projective Extension Theorem 15.3.1.

7.5.10. ⋆ Complex curves that are not rational (fun but inessential).
We now describe two examples of curves C that do not admit a nonconstant
rational map from P1C . (Admittedly, we do not yet know what “curve” means, but
no matter.) Both proofs are by Fermat’s method of infinite descent. These results
can be interpreted (as you will later be able to check using Theorem 16.3.3) as the
fact that these curves have no “nontrivial” C(t)-valued points, where by this, we
mean that any C(t)-valued point is secretly a C-valued point. You may notice that
if you consider the same examples with C(t) replaced by Q (and where C is a curve
over Q rather than C), you get two fundamental questions in number theory and
geometry. The analog of Exercise 7.5.M is the question of rational points on elliptic
curves, and you may realize that the analog of Exercise 7.5.K is even more famous.
Also, the arithmetic analog of Exercise 7.5.M(a) is the “Four Squares Theorem”
(there are not four integer squares in arithmetic progression), first stated by Fermat.
These examples will give you a glimpse of how and why facts over number fields
are often paralleled by facts over function fields of curves. This parallelism is a
recurring deep theme in the subject.

7.5.K. E XERCISE . If n > 2, show that P1C has no dominant rational maps to the
“Fermat curve” xn + yn = zn in P2C . Hint: reduce this to showing that there is
no “nonconstant” solution (f(t), g(t), h(t)) to f(t)n + g(t)n = h(t)n , where f(t),
g(t), and h(t) are rational functions in t (that is, elements of C(t)). By clearing
222 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

denominators, reduce this to showing that there is no nonconstant solution where

f(t), g(t), and h(t) are relatively prime polynomials. For this, assume there is a
solution, and consider one of the lowest positive degree.
QThen use the fact that C[t]
n
is a unique factorization domain, and h(t)n − g(t)n = i=1 (h(t) − ζi g(t)), where
ζ is a primitive nth root of unity. Argue that each h(t) − ζi g(t) is an nth power.
Then use

(h(t) − g(t)) + α (h(t) − ζg(t)) = β h(t) − ζ2 g(t)

for suitably chosen α and β to get a solution of smaller degree. (How does this
argument fail for n = 2?)

7.5.L. E XERCISE . Give two smooth complex curves X and Y so that no nonempty
open subset of X is isomorphic to a nonempty open subset of Y. (Try not to be both-
ered by the fact that we have not yet defined “smoothness”.) Hint: Exercise 7.5.K.

7.5.M. E XERCISE . Suppose a, b, and c are distinct complex numbers. By the

following steps, show that if x(t) and y(t) are two rational functions of t such that

(7.5.10.1) y(t)2 = (x(t) − a)(x(t) − b)(x(t) − c),

then x(t) and y(t) are constants (x(t), y(t) ∈ C). (Here C may be replaced by any
field K of characteristic not 2; slight extra care is needed if K is not algebraically
closed.)
(a) Suppose P, Q ∈ C[t] are relatively prime polynomials such that four linear combi-
nations of them are perfect squares, no two of which are constant multiples of each
other. Show that P and Q are constant (i.e., P, Q ∈ C). Hint: By renaming P and Q,
show that you may assume that the perfect squares are P, Q, P −Q, P −λQ (for some
λ ∈ C). Define
√ u and
√ v to be square roots of P and Q respectively. Show that u − v,
u+v, u− λv, u+ λv are perfect squares, and that u and v are relatively prime. If P
and Q are not both constant, note that 0 < max(deg u, deg v) < max(deg P, deg Q).
Assume from the start that P and Q were chosen as a counterexample with minimal
max(deg P, deg Q) to obtain a contradiction. (Aside: It is possible to have three
distinct linear combinations that are perfect squares. Such examples essentially
correspond to primitive Pythagorean triples in C(t) — can you see how?)
(b) Suppose (x, y) = (p/q, r/s) is a solution to (7.5.10.1), where p, q, r, s ∈ C[t],
and p/q and r/s are in lowest terms. Clear denominators to show that r2 q3 =
s2 (p − aq)(p − bq)(p − cq). Show that s2 |q3 and q3 |s2 , and hence that s2 = δq3 for
some δ ∈ C. From r2 = δ(p − aq)(p − bq)(p − cq), show that (p − aq), (p − bq),
(p − cq) are perfect squares. Show that q is also a perfect square, and then apply
part (a).
A much better geometric approach to Exercises 7.5.K and 7.5.M is given in
Exercise 21.4.H.

7.6 ⋆ Representable functors and group schemes

 © 2024 Ravi Vakil. Published by Princeton University Press. 223

7.6.1. Maps to A1 correspond to functions. If X is a scheme, there is a bijection

between the maps X → A1 and global sections of the structure sheaf: by Exer-
cise 7.3.G, maps π : X → A1Z correspond to maps of rings π♯ : Z[t] → Γ (X, OX ), and
π♯ (t) is a function on X; this is reversible.
This map is very natural in an informal sense: you can even picture this map
to A1 as being given by the function. (By analogy, a function on a manifold is a
map to R.) But it is natural in a more precise sense: this bijection is functorial in X.
We will ponder this example at length, and see that it leads us to two important
sophisticated notions: representable functors and group schemes.

7.6.A. E ASY E XERCISE . Suppose X is a C-scheme. Verify that there is a natural

bijection between maps X → A1C in the category of C-schemes and functions on X.
(Here the base ring C can be replaced by any ring A.)
This interpretation can be extended to rational maps, as follows.

7.6.B. U NIMPORTANT E XERCISE . Interpret rational functions on an integral scheme

(Exercise 5.2.I)) as rational maps to A1Z .
(If you wish, you can extend your argument to rational maps on locally Noe-
therian schemes, see Definition 6.6.36.)

7.6.2. Representable functors. We restate the bijection of §7.6.1 as follows. We

have two different contravariant functors from Sch to Sets: maps to A1 (i.e., H : X 7→
Mor(X, A1Z )), and functions on X (F : X 7→ Γ (X, OX )). The “naturality” of the bijection
— the functoriality in X — is precisely the statement that the bijection gives a natural
isomorphism of functors (§1.1.21): given any π : X → X ′ , the diagram

H(π)
H(X ′ ) / H(X)

 F(π) 
F(X ′ ) / F(X)

(where the vertical maps are the bijections given in §7.6.1) commutes.
More generally, if Y is an element of a category C (we care about the special
case C = Sch), recall the contravariant functor hY : C → Sets defined by hY (X) =
Mor(X, Y) (Example 1.1.20). We say a contravariant functor from C to Sets is
represented by Y if it is naturally isomorphic to the functor hY . We say it is
representable if it is represented by some Y.
The bijection of §7.6.1 may now be restated as: the global section functor is
represented by A1 .

7.6.C. I MPORTANT E ASY E XERCISE ( REPRESENTING OBJECTS ARE UNIQUE UP TO

UNIQUE ISOMORPHISM ). Show that if a contravariant functor F is represented
∼
by Y and by Z, then we have a unique isomorphism Y −→ Z induced by the
∼
natural isomorphism of functors hY −→ hZ . Hint: this is a version of the universal
property arguments of §1.2: once again, we are recognizing an object (up to unique
isomorphism) by maps to that object. This exercise is essentially Exercise 1.2.Z(b).
(This extends readily to Yoneda’s Lemma in this setting, Exercise 10.1.B. You are
welcome to try that now.)
224 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

You have implicitly seen this notion before: you can interpret the existence of
products and fibered products in a category as examples of representable functors.
∼
(You may wish to work out how a natural isomorphism hY×Z −→ hY × hZ induces
the projection maps Y × Z → Y and Y × Z → Z.)

7.6.D. E XERCISE ( WARM - UP ). Suppose F is the contravariant functor Sch → Sets

defined by F(X) = {Grothendieck} for all schemes X. Show that F is representable.
(What is it representable by?)

7.6.E. E XERCISE . In this exercise, Z may be replaced by any ring.

(a) (Affine n-space represents the functor of n functions.) Show that the contravariant
functor from (Z-)schemes to Sets
 / {(f1 , . . . , fn ) : fi ∈ Γ (X, OX )}
X

1 ∼
is represented by An 1 2 2
Z . Show that AZ ×Z AZ = AZ , i.e., that A satisfies the functorial
1 1
description ofP A × A . (You will undoubtedly be able to immediately show that
Q mi ∼ m
AZ ←→ AZ i .)
(b) (The functor of invertible functions is representable.) Show that the contravariant
functor from (Z-)schemes to Sets taking X to invertible functions on X is repre-
sentable by Spec Z[t, t−1 ].

7.6.3. Definition. The scheme defined in Exercise 7.6.E(b) is called the multi-
plicative group (or multiplicative group scheme) Gm . “Gm over a field k” (“the
multiplicative group over k”) means Spec k[t, t−1 ], with the same group operations.
Better: it represents the group of invertible functions in the category of k-schemes.
We can similarly define Gm over an arbitrary ring or even arbitrary scheme.

7.6.F. L ESS IMPORTANT EXERCISE . Fix a ring A. Consider the functor H from the
category of locally ringed spaces to Sets given by H(X) = {A → Γ (X, OX )}. Show
that this functor is representable (by Spec A). This gives another (admittedly odd)
motivation for the definition of Spec A, closely related to that of §7.3.6.

7.6.4. ⋆⋆ Group schemes (or more generally, group objects in a category).

(The rest of §7.6 should be read only for entertainment.) We return again to
Example 7.6.1. Functions on X are better than a set: they form a group. (Indeed
they even form a ring, but we will worry about this later.) Given a morphism
X → Y, pullback of functions Γ (Y, OY ) → Γ (X, OX ) is a group homomorphism. So
we should expect A1 to have some group-like structure. This leads us to the notion
of group scheme, or more generally a group object in a category, which we now define.
Suppose C is a category with a final object Z and with products. (We know
that Sch has a final object Z = Spec Z, by Exercise 7.3.J. We will later see that it has
products, §10.1. But in Exercise 7.6.K we will give an alternative characterization of
group objects that applies in any category, so we won’t worry about this.)
A group object in C is an object X along with three morphisms:
• multiplication: m : X × X → X
• inverse: i : X → X
• identity element: e : Z → X (not the identity map)
These morphisms are required to satisfy several conditions.
 © 2024 Ravi Vakil. Published by Princeton University Press. 225

(i) associativity axiom:

X×X×X
m×id
/ X×X

id×m m
 
X×X
m /X
commutes. (Here id means the equality X → X.)
(ii) identity axiom:
∼ / Z×X e×id
/ X×X m /X
X
and
∼ / id×e
/ X×X m / X
X X×Z
are both the identity map X → X. (This corresponds to the group axiom: “multipli-
cation by the identity element is the identity map”.)
(iii) inverse axiom: X
i×id
/ X×X / X and X id×i / X × X m / X are
m

both the map that is the composition X /Z e /X.

As motivation, you can check that a group object in the category of sets is in
fact the same thing as a group. (This is symptomatic of how you take some notion
and make it categorical. You write down its axioms in a categorical way, and if all
goes well, if you specialize to the category of sets, you get your original notion. You
can apply this to the notion of “rings” in an exercise below.)
A group scheme is defined to be a group object in the category of schemes. A
group scheme over a ring A (or a scheme S) is defined to be a group object in the
category of A-schemes (or S-schemes).

7.6.G. E XERCISE . Give A1Z the structure of a group scheme, by describing the
three structural morphisms, and showing that they satisfy the axioms. (Hint: the
morphisms should not be surprising. For example, inverse is given by t 7→ −t.
Note that we know that the product A1Z × A1Z exists, by Exercise 7.6.E(a).)

7.6.H. E XERCISE . Show that if G is a group object in a category C , then for any
X ∈ C , Mor(X, G) has the structure of a group, and the group structure is preserved
by pullback (i.e., Mor(·, G) is a contravariant functor to the category of groups,
Groups).

7.6.I. E XERCISE . Show that the group structure described by the previous exercise
translates the group scheme structure on A1Z to the group structure on Γ (X, OX ), via
the bijection of §7.6.1.

7.6.J. E XERCISE . Define the notion of abelian group scheme, and ring scheme.
(You will undoubtedly at the same time figure out how to define the notion of
abelian group object and ring object in any category C . You may discover a more
efficient approach to such questions after reading §7.6.5.)

7.6.5. Group schemes, more functorially. There was something unsatisfactory about
our discussion of the “group-respecting” nature of the bijection in §7.6.1: we
observed that the right side (functions on X) formed a group, then we developed
the axioms of a group scheme, then we cleverly figured out the maps that made
226 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

A1Z into a group scheme, then we showed that this induced a group structure on
the left side of the bijection (Mor(X, A1 )) that precisely corresponded to the group
structure on the right side (functions on X).
The picture is more cleanly explained as follows. The language of scheme-
valued points (Definition 7.3.10) has the following advantage: notice that the
points of a group scheme need not themselves form a group (consider A1Z ). But
Exercise 7.6.H shows that the Z-valued points of a group scheme (where Z is any
given scheme) indeed form a group.

7.6.K. E XERCISE . Suppose we have a contravariant functor F from Sch (or indeed
any category) to Groups. Suppose further that F composed with the forgetful functor
Groups → Sets is represented by an object Y. Show that the group operations on F(X)
(as X varies through Sch) uniquely determine m : Y × Y → Y, i : Y → Y, e : Z → Y
satisfying the axioms defining a group scheme, such that the group operation on
Mor(X, Y) is the same as that on F(X).
In particular, the definition of a group object in a category was forced upon
us by the definition of group. More generally, you should expect that any class of
objects that can be interpreted as sets with additional structure should fit into this
picture.
You should apply this exercise to A1Z , and see how the explicit formulas you
found in Exercise 7.6.G are forced on you.

7.6.L. E XERCISE . Work out the maps m, i, and e in the group schemes of Exer-
cise 7.6.E.

7.6.M. E XERCISE . Explain why the product of group objects in a category can be
naturally interpreted as a group object in that category.

7.6.N. E XERCISE .
(a) Define morphism of group schemes.
(b) Recall that if A is a ring, then GLn (A) (the general linear group over A) is the
group of invertible n × n matrices with entries in the ring A. Figure out the right
definition of the group scheme GLn (over a ring A), and describe the determinant
map det : GLn → Gm .
(c) Make sense of the statement: “(·n ) : Gm → Gm given by t 7→ tn is a morphism
of group schemes.”
The language of Exercise 7.6.N(a) suggests that group schemes form a category;
feel free to prove this if you want. In fact, the category of group schemes has a zero
object. What is it?

7.6.O. E XERCISE ( KERNELS OF MAPS OF GROUP SCHEMES ). Suppose F : G1 → G2 is

a morphism of group schemes. Consider the contravariant functor Sch → Groups
given by X 7→ ker(Mor(X, G1 ) → Mor(X, G2 )). If this is representable, by a group
scheme G0 , say, show that G0 → G1 is the kernel of F in the category of group
schemes.

7.6.P. E XERCISE . Show that the kernel of (·n ) (Exercise 7.6.N) is representable. If
n > 0, show that over a field k of characteristic p dividing n, this group scheme is
 © 2024 Ravi Vakil. Published by Princeton University Press. 227

nonreduced. (This group scheme, denoted µn , is important, although we will not

use it.)

7.6.Q. E XERCISE . Show that the kernel of det : GLn → Gm is representable. This is
the group scheme SLn . (You can do this over Z, or over a field k, or even over an
arbitrary ring A; the algebra is the same.)

7.6.R. E XERCISE . Show (as easily as possible) that A1k is a ring k-scheme. (Here k
can be replaced by any ring.)

7.6.S. E XERCISE .
(a) Define the notion of a (left) group scheme action (of a group scheme on a
scheme).
(b) Suppose A is a ring. Show that specifying an integer-valued grading on A
is equivalent to specifying an action of Gm on Spec A. (This interpretation of a
grading is surprisingly enlightening. Caution: there are two possible choices of
the integer-valued grading, and there are reasons for both. Both are used in the
literature.)

7.6.6. Aside: Hopf algebras. Here is a notion that we won’t use, but it is easy
enough to define now. Suppose G = Spec A is an affine group scheme, i.e., a group
scheme that is an affine scheme. The categorical definition of group scheme can be
restated in terms of the ring A. (This requires thinking through Remark 7.3.5; see
Exercise 10.1.A.) Then these axioms define a Hopf algebra. For example, we have a
“comultiplication map” A → A ⊗ A.

7.6.T. E XERCISE . As A1Z is a group scheme, Z[t] has a Hopf algebra structure.
Describe the comultiplication map Z[t] → Z[t] ⊗Z Z[t].

7.7 ⋆⋆ The Grassmannian: First construction

The Grassmannian is a useful geometric construction that is “the geometric
object underlying linear algebra”. In (classical) geometry over a field K = R or
C, just as projective space parametrizes one-dimensional subspaces of a given n-
dimensional vector space, the Grassmannian parametrizes k-dimensional subspaces
of n-dimensional space. The Grassmannian G(k, n) is a manifold of dimension
k(n − k) (over the field). The manifold structure is given as follows. Given a
basis (v1 , . . . , vn ) of n-space, “most” k-planes can be described as the span of the k
vectors
X X X
* n n n
+
(7.7.0.1) v1 + a1i vi , v2 + a2i vi , . . . , vk + aki vi .
i=k+1 i=k+1 i=k+1

(Can you describe which k-planes are not of this form? Hint: row reduced echelon
form. Aside: the stratification of G(k, n) by normal form is the decomposition of
the Grassmannian into Schubert cells. You may be able to show using the normal
form that each Schubert cell is isomorphic to an affine space.) Any k-plane of
this form can be described in such a way uniquely. We use this to identify those
k-planes of this form with the manifold Kk(n−k) (with coordinates aji ). This is
228 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

a large affine patch on the Grassmannian (called the “open Schubert cell” with
respect to this basis). As the vi vary, these patches cover the Grassmannian (why?),
and the manifold structures agree (a harder fact).
We now define the Grassmannian in algebraic geometry, over a ring A. Suppose
v = (v1 , . . . , vn ) is a basis for A⊕n . More precisely: vi ∈ A⊕n , and the map
A⊕n → A⊕n given by (a1 , . . . , an ) 7→ a1 v1 + · · · + an vn is an isomorphism.

7.7.A. E XERCISE . Show that any two bases are related by an invertible n × n matrix
over A — a matrix with entries in A whose determinant is an invertible element of
A.
∼ Ak(n−k) , with coordinates
For each such basis v, we consider the scheme Uv = A
aji (k + 1 ≤ i ≤ n, 1 ≤ j ≤ k), which we imagine as corresponding to the k-plane
spanned by the vectors (7.7.0.1).

7.7.B. E XERCISE . Given two bases v and w, explain how to glue Uv to Uw along
appropriate open sets. You may find it convenient to work with coordinates aji
where i runs from 1 to n, not just k + 1 to n, but imposing aji = δji (i.e., 1 when
i = j and 0 otherwise) when i ≤ k. This convention is analogous to coordinates xi/j
on the patches of projective space (§4.4.9). Hint: the relevant open subset of Uv will
be where a certain determinant doesn’t vanish.

7.7.C. E XERCISE /D EFINITION . By checking triple intersections, verify that these

patches (over all possible bases) glue together to a single scheme (Exercise 4.4.A).
This is the Grassmannian G(k, n) over the ring A. Because it can be interpreted as
a space of linear “Pk−1 n−1
A ’s” in PA , it is often also written G(k − 1, n − 1). (You will
see that this is wise notation in Exercise 12.2.N, for example.)
Although this definition is pleasantly explicit (it is immediate that the Grass-
mannian is covered by Ak(n−k) ’s), and perhaps more “natural” than our original
definition of projective space in §4.4.9 (we aren’t making a choice of basis; we use all
bases), there are several things unsatisfactory about this definition of the Grassman-
nian. In fact the Grassmannian is projective; this isn’t obvious with this definition.
n
Furthermore, the Grassmannian comes with a natural morphism to P( k )−1 (which
is an example of a closed embedding, a notion we will meet in Definition 9.1.1), called
the Plücker embedding. Finally, there is an action of GLn on the space of k-planes
in n-space, so we should be able to see this in our algebraic incarnation. We will
address these issues in §16.4, by giving a better description, as a moduli space.

7.7.1. (Partial) flag varieties. Just as the Grassmannian “parametrizes” k-planes in

n-space, the flag variety parametrizes “flags”: nested sequences of subspaces of
n-space
F0 ⊂ F1 ⊂ · · · ⊂ Fn
where dim Fi = i. Generalizing both of these is the notion of a partial flag variety
associated to data 0 ≤ a1 < · · · < aℓ ≤ n, parametrizing nested subspaces of
n-space
Fa1 ⊂ · · · ⊂ Faℓ
where dim Fai = ai . You should be able to generalize all of the discussion in §7.7
to this setting.
 CHAPTER 8

Useful classes of morphisms of schemes

We now define an excessive number of types of morphisms. Some (often

finiteness properties) are useful because most “reasonable” morphism has such
properties, and they will be used in proofs in obvious ways. Others correspond to
geometrically meaningful behavior, and you should have a picture of what each
means.

8.0.1. Change of perspective: morphisms are more fundamental than objects. One of
Grothendieck’s lessons is that things that we often think of as properties of objects
are better understood as properties of morphisms. One way of turning properties of
objects into properties of morphisms is as follows. If P is a property of schemes, we
often (but not always) say that a morphism π : X → Y has P if for every affine open
subset U ⊂ Y, π−1 (U) has P. We will see this for P = quasicompact, quasiseparated,
affine, and more. (As you might hope, in good circumstances, P will satisfy the
hypotheses of the Affine Communication Lemma 5.3.2, so we don’t have to check
every affine open subset.) Informally, you can think of such a morphism as one
where all the fibers have P. (You can quickly define the fiber of a morphism as a
topological space, but once we define fibered product, we will define the scheme-
theoretic fiber, and then this discussion will make sense.) But it means more than
that: it means that “being P” is really not just fiber-by-fiber, but behaves well as the
fiber varies. (For comparison, “submersion of manifolds” means more than that
the fibers are smooth.)

8.1 “Reasonable” classes of morphisms (such as open embeddings)

8.1.1. You will notice that almost all classes of morphisms that are useful have many
properties in common, which follow from the following three basic properties. We
call any class of morphisms satisfying these properties a “reasonable” class of
morphisms of schemes. (To avoid any stupidities, we assume that the class includes
all isomorphisms; but this requirement doesn’t deserve a number.)
(i) The class is preserved by composition: if π : X → Y and ρ : Y → Z are both
in this class, then so is ρ ◦ π.
(ii) The class is preserved by “base change” (or “pullback” or “fibered prod-
uct”). Precisely: if π : X → Y is in this class, then for any Y ′ → Y, the

This material is published by Princeton University Press as The Rising Sea: Foundations of Algebraic
Geometry by Ravi Vakil. This version is free to view and download for personal use only. Not for
redistribution, resale, or use in derivative works. © 2024 Ravi Vakil.

229
230 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

induced map X ×Y Y ′ → Y ′ is also in this class. (Implicit in this statement

is that the fibered product X ×Y Y ′ exists; but we will soon see that all
fibered products of schemes exist, §10.1.)
(iii) The class is local on the target. In other words, (a) if π : X → Y is
in the class, then for any open subset V of Y, the restricted morphism
π−1 (V) → V is in the class; and (b) for a morphism π : X → Y, if there is
an open cover {Vi } of Y for which each restricted morphism π−1 (Vi ) → Vi
is in the class, then π is in the class. In particular, as schemes are built
out of affine schemes, properties are often easy to verify on any affine
cover (the properties are “affine-local” on the target), as described in §8.0.1.
(Stalk-local properties are automatically local on the target.)
Properties (i) and (ii) imply a useful additional property — (iv) “reasonable”
classes of morphisms are preserved by product:

8.1.A. E XERCISE . Suppose P is a property of morphisms preserved by composition

and base change ((i) and (ii) above), and X → Y and X ′ → Y ′ are two morphisms
of S-schemes with property P. Assume that X ×S X ′ and Y ×S Y ′ exist (which is
indeed true, §10.1). Show that X ×S X ′ → Y ×S Y ′ has property P as well. Hint:

X ×S X ′ / X ×S Y ′ / Y ×S Y ′

{
X′ / Y′ y %
X
#/
Y

(v) Another extremely important consequence of (i) and (ii) is the Cancellation
Theorem for such morphisms, Theorem 11.1.1.
When you learn of a new class of morphisms, you should immediately ask
whether these properties (i)–(iii) (and hence (iv) and (v)) hold. The answer will
almost always be “yes”; and if it is “no”, then you should exercise caution, and
possibly figure out how to make the definition better (see for example §10.5).
To prepare you to think about these properties for schemes, you may want to
verify that the analogous properties to (i)–(iii) hold in the category of topological
spaces with the classes “injections”, “surjections”, “open embeddings” (open em-
bedding of topological spaces = isomorphism with an open subset), and “closed
embeddings” (closed embedding of topological spaces = isomorphism with a closed
subset).
As a first example in the category of schemes:

8.1.B. E ASY E XERCISE . Show that “isomorphisms” of schemes are “reasonable” in

this sense, i.e., that they satisfy the three properties (i)–(iii) above.
We are now ready to introduce a fundamental class of morphisms of schemes.

8.1.2. Definition. A morphism π : X → Y of schemes is an open embedding (or

open immersion) if it is an open embedding as ringed spaces (§7.2.1). In other
words, a morphism π : (X, OX ) → (Y, OY ) of schemes is an open embedding if π
factors as

(X, OX )
ρ
/ (U, OY |U )   τ / (Y, OY )
∼
 © 2024 Ravi Vakil. Published by Princeton University Press. 231

where ρ is an isomorphism, and τ : U ,→ Y is an inclusion of an open set. It is

immediate that isomorphisms are open embeddings. The symbol ,→ is often used
to indicate that a morphism is an open embedding (or more generally, a locally
closed embedding, see §9.2).
If X is actually a subset of Y (and π is the inclusion, i.e., ρ is the identity), then
we say (X, OX ) is an open subscheme of (Y, OY ). The difference between open
embeddings and open subschemes is a bit confusing, and not too important: at the
level of sets, open subschemes are subsets, while open embeddings are bijections
onto subsets.
“Open subschemes” are scheme-theoretic analogs of open subsets. (“Closed
subschemes” are scheme-theoretic analogs of closed subsets, but they have a sur-
prisingly different flavor, as we will see in §9.1.)
The next two Exercises verify that the class of open embeddings is “reasonable”
in the sense described above.

8.1.C. E XERCISE . Verify that the class of open embeddings satisfies properties (i)
and (iii) of §8.1.
Property (ii) is the content of the next exercise.

8.1.D. I MPORTANT BUT EASY EXERCISE : FIBERED PRODUCTS WITH OPEN EMBED -
DINGS EXIST. Verify that the class of open embeddings satisfies property (ii) of §8.1.
More specifically: suppose i : U → Z is an open embedding, and ρ : Y → Z is any
morphism. Show that U ×Z Y exists and U ×Z Y → Y is an open embedding. (Hint:
I’ll even tell you what U ×Z Y is: (ρ−1 (U), OY |ρ−1 (U) ).) In particular, if U ,→ Z and
V ,→ Z are open embeddings, U ×Z V = ∼ U ∩ V: “intersection of open embeddings
is fiber product”.

8.1.E. E ASY EXERCISE . Show that open embeddings of schemes are monomor-
phisms (in the category of schemes). Hint: Exercise 7.2.B.

8.1.F. E ASY E XERCISE . Suppose π : X → Y is an open embedding. Show that if

Y is locally Noetherian, then X is too. Show that if Y is Noetherian, then X is too.
However, show that if Y is quasicompact, X need not be. (Hint: let Y be affine but
not Noetherian, see Exercise 3.6.G(b).)

8.1.3. Definition. In analogy with “local on the target” (§8.1.1), we define what it
means for a property P of morphisms to be local on the source: if π : X → Y has
P, and i : U ,→ X is an open subset, then π ◦ i : U → Y has P; and to check if a
morphism π : X → Y has P, it suffices to check on any open cover {Ui } of X. We then
define affine-local on the source (and affine-local on the target) in the obvious
way. For example, to check if π : X → Y has P it suffices to check on any affine open
cover of the source (resp., target). The use of “affine-local” rather than “local” is to
emphasize that the criterion on affine schemes is simple to describe. (Clearly the
Affine Communication Lemma 5.3.2 will be very handy.)

8.1.G. E XERCISE ( PRACTICE WITH THE CONCEPT ). Show that the notion of “open
embedding” is not local on the source.
232 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

8.2 Another algebraic interlude: Lying Over and Nakayama

Algebra is the offer made by the devil to the mathematician. The devil says: I will give
you this powerful machine, it will answer any question you like. All you need to do is give
me your soul: give up geometry and you will have this marvelous machine.
— M. Atiyah, [At2, p. 659]; but see the Atiyah quote at the start of §0.3
To set up our discussion in the next section on integral morphisms, we develop
some algebraic preliminaries. A clever trick we use can also be used to show
Nakayama’s Lemma, so we discuss this as well.
Suppose ϕ : B → A is a ring morphism. We say a ∈ A is integral over B if a
satisfies some monic polynomial

an + ?an−1 + · · · + ? = 0

where the coefficients lie in ϕ(B). A ring morphism ϕ : B → A is integral if every

element of A is integral over ϕ(B). An integral ring morphism ϕ is an integral
extension if ϕ is an inclusion of rings. You should think of integral morphisms
and integral extensions as ring-theoretic generalizations of the notion of algebraic
extensions of fields.

8.2.A. E XERCISE . Show that if ϕ : B → A is a ring morphism, (b1 , . . . , bn ) = 1

in B, and Bbi → Aϕ(bi ) is integral for all i, then ϕ is integral. Hint: replace B by
ϕ(B) to reduce to the case where B is a subring of A. Suppose a ∈ A. Show that
there is some t and m such that bti am ∈ B + Ba + Ba2 + · · · + Bam−1 for some t
and m independent of i. Use a “partition of unity” argument as in the proof of
Theorem 4.1.2 to show that am ∈ B + Ba + Ba2 + · · · + Bam−1 .

8.2.B. E XERCISE .
(a) Show that the property of a morphism ϕ : B → A being integral is always
preserved by localization and quotient of B, and quotient of A, but not localiza-
tion of A. More precisely: suppose ϕ is integral. Show that the induced maps
T −1 B → ϕ(T )−1 A, B/J → A/ϕ(J)A, and B → A/I are integral (where T is a multi-
plicative subset of B, J is an ideal of B, and I is an ideal of A), but B → S−1 A need
not be integral (where S is a multiplicative subset of A). (Hint for the latter: show
that k[t] → k[t] is an integral ring morphism, but k[t] → k[t](t) is not.)
(b) Show that the property of ϕ being an integral extension is preserved by localiza-
tion of B, but not localization or quotient of A. (Hint for the latter: k[t] → k[t] is an
integral extension, but k[t] → k[t]/(t) is not.)
(c) In fact the property of ϕ being an integral extension (as opposed to integral
morphism) is not preserved by taking quotients of B either. (Let B = k[x, y]/(y2 ) and
A = k[x, y, z]/(z2 , xz−y). Then B injects into A, but B/(x) doesn’t inject into A/(x).)
But it is in some cases. Suppose ϕ : B → A is an integral extension, and J ⊂ B is
the restriction of an ideal I ⊂ A. (Side Remark: you can show that this holds if J is
prime.) Show that the induced map B/J → A/JA is an integral extension. (Hint:
show that the composition B/J → A/JA → A/I is an injection.)
The following lemma uses a useful but sneaky trick.

8.2.1. Lemma. — Suppose ϕ : B → A is a ring morphism. Then a ∈ A is integral over B

if and only if it is contained in a subalgebra of A that is a finitely generated B-module.
 © 2024 Ravi Vakil. Published by Princeton University Press. 233

Proof. If a satisfies a monic polynomial equation of degree n, then the B-submodule

of A generated by 1, a, . . . , an−1 is closed under multiplication, and hence a
subalgebra of A.
Assume conversely that a is contained in a subalgebra A ′ of A that is a finitely
generated B-module. P Choose a finite generating set m1 , . . . , mn of A ′ (as a B-
module). Then ami = bij mj , for some bij ∈ B. Thus
  
m1 0
(8.2.1.1) (aIdn×n − [bij ]ij )  ...  =  ..  ,
  
. 
mn 0

where Idn is the n × n identity matrix in A. We can’t invert the matrix (aIdn×n −
[bij ]ij ), but we almost can. Recall that an n × n matrix M has an adjugate matrix
adj(M) such that adj(M)M = det(M)Idn . (The (i, j)th entry of adj(M) is the
determinant of the matrix obtained from M by deleting the ith column and jth row,
times (−1)i+j . You have likely seen this in the form of a formula for M−1 when
there is an inverse; see for example [DF, p. 440].) The coefficients of adj(M) are
polynomials in the coefficients of M. Multiplying (8.2.1.1) by adj(aIdn×n − [bij ]ij ),
we get
   
m1 0
 ..   .. 
det(aIdn×n − [bij ]ij )  .  =  .  .
mn 0

So det(aI − [bij ]) annihilates the generating elements mi , and hence every element
of A ′ , i.e., det(aI − [bij ]) = 0. But expanding the determinant yields an integral
equation for a with coefficients in B. □

8.2.2. Corollary (finite implies integral). — If A is a finite B-algebra (a finitely

generated B-module), then ϕ is an integral ring morphism.
The converse is false: integral does not imply finite, as Q ,→ Q is an integral
ring morphism, but Q is not a finite Q-module. (A field extension is integral if it is
algebraic.)

8.2.C. E XERCISE . Show that if C → B and B → A are both integral ring morphisms,
then so is their composition.

8.2.D. E XERCISE . Suppose ϕ : B → A is a ring morphism. Show that the elements

of A integral over B form a subalgebra of A.

8.2.3. Remark: Transcendence theory. These ideas lead to the main facts about
transcendence theory we will need for a discussion of dimension of varieties, see
Exercise/Definition 12.2.A.

8.2.4. The Lying Over and Going-Up Theorems. The Lying Over Theorem is a
useful property of integral extensions.

8.2.5. The Lying Over Theorem. — Suppose ϕ : B → A is an integral extension. Then

for any prime ideal q ⊂ B, there is a prime ideal p ⊂ A such that p ∩ B = q.
234 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

To be clear on how weak the hypotheses are: B need not be Noetherian, and A
need not be finitely generated over B.

8.2.6. Geometric translation: if B → A is an integral extension, then Spec A → Spec B

is surjective. Although the Lying Over Theorem 8.2.5 is a theorem in algebra, the
name can be interpreted geometrically: the theorem asserts that the corresponding
morphism of schemes is surjective. (A map of schemes is surjective if the under-
lying map of sets is surjective.) Translation: “above” every prime q “downstairs”,
there is a prime p “upstairs”, see Figure 8.1. (For this reason, it is often said that p
“lies over” q if p ∩ B = q.)

F IGURE 8.1. A picture of the Lying Over Theorem 8.2.5: if ϕ : B →

A is an integral extension, then Spec A → Spec B is surjective

The following exercise sets up the proof.

8.2.E. ⋆ E XERCISE . Show that the special case where A is a field translates to: if
B ⊂ A is a subring with A integral over B, then B is a field. Prove this. (Hint: you
must show that all nonzero elements in B have inverses in B. Here is the start: If
b ∈ B, then 1/b ∈ A, and this satisfies some integral equation over B.)
⋆ Proof of the Lying Over Theorem 8.2.5. We first make a reduction: by localizing at
q (preserving integrality by Exercise 8.2.B(b)), we can assume that (B, q) is a local
ring. Then let p be any maximal ideal of A. Consider the following diagram.
AO / / A/p field
O

? ?
B / / B/(p ∩ B)

The right vertical arrow is an integral extension by Exercise 8.2.B(c). By Exer-

cise 8.2.E, B/(p ∩ B) is a field too, so p ∩ B is a maximal ideal, hence it is q. □

8.2.F. I MPORTANT E XERCISE ( THE Going-Up Theorem).

(a) Suppose ϕ : B → A is an integral ring morphism (not necessarily an integral
extension). Show that if q1 ⊂ q2 ⊂ · · · ⊂ qn is a chain of prime ideals of B, and
p1 ⊂ · · · ⊂ pm is a chain of prime ideals of A such that pi “lies over” qi (and
1 ≤ m < n), then the second chain can be extended to p1 ⊂ · · · ⊂ pn so that this
 © 2024 Ravi Vakil. Published by Princeton University Press. 235

remains true. (Hint: reduce to the case m = 1, n = 2; reduce to the case where
q1 = (0) and p1 = (0); use the Lying Over Theorem 8.2.5.)
(b) Draw a picture of this theorem (akin to Figure 8.1).
There are analogous “Going-Down” results (requiring quite different hypothe-
ses); see for example Theorem 12.2.12 and Exercise 24.5.E.

8.2.7. Nakayama’s Lemma.

The trick in the proof of Lemma 8.2.1 can be used to quickly prove Nakayama’s
Lemma, which we will use repeatedly in the future. This name is used for several
different but related results, which we discuss here. (A geometric interpretation
will be given in Exercise 14.3.D.) We may as well prove it while the trick is fresh in
our minds.

8.2.8. Nakayama’s Lemma version 1. — Suppose A is a ring, I is an ideal of A, and

M is a finitely generated A-module, such that M = IM. Then there exists an a ∈ A with
a ≡ 1 (mod I) with aM = 0. (Equivalently, there is some i ∈ I for which multiplication
by i induces the identity on M: im = m for all m ∈ M.)

Proof.
P Say M is generated by m1 , . . . , mn . Then as M = IM, we have mi =
j aij mj for some aij ∈ I. Thus
 
m1
 ..
(8.2.8.1) (Idn − Z)  . =0


mn
where Z = (aij ). Multiplying both sides of (8.2.8.1) on the left by adj(Idn − Z), we
obtain  
m1
det(Idn − Z)  ...  = 0.
 

mn
But when you expand out det(Idn − Z), as Z has entries in I, you get something
that is 1 (mod I). □
Here is why you care. Suppose I is contained in all maximal ideals of A. (The
intersection of all the maximal ideals is called the Jacobson radical, but we won’t
use this phrase. For comparison, recall that the nilradical was the intersection of
the prime ideals of A.) Then any a ≡ 1 (mod I) is invertible. (We are not using
Nakayama yet!) Reason: otherwise (a) ̸= A, so the ideal (a) is contained in some
maximal ideal m — but a ≡ 1 (mod m), contradiction. As a is invertible, we have
the following.

8.2.9. Nakayama’s Lemma version 2. — Suppose A is a ring, I is an ideal of A

contained in all maximal ideals, and M is a finitely generated A-module. (The most
interesting case is when A is a local ring, and I is the maximal ideal.) Suppose M = IM.
Then M = 0.

8.2.G. E XERCISE (N AKAYAMA’ S L EMMA VERSION 3). Suppose A is a ring, and I

is an ideal of A contained in all maximal ideals. Suppose M is a finitely generated
A-module, and N ⊂ M is a submodule. If N/IN → M/IM is surjective, then
M = N.
236 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

8.2.H. I MPORTANT E XERCISE (N AKAYAMA’ S L EMMA VERSION 4: GENERATORS OF

M/mM LIFT TO GENERATORS OF M). Suppose (A, m) is a local ring. Suppose M is
a finitely generated A-module, and f1 , . . . , fn ∈ M, with (the images of) f1 , . . . , fn
generating M/mM. Then f1 , . . . , fn generate M. (In particular, taking M = m, if
we have generators of m/m2 , they also generate m.)

8.2.I. I MPORTANT E XERCISE GENERALIZING L EMMA 8.2.1. Recall that a B-module

N is said to be faithful if the only element of B acting on N by the identity is 1 (or
equivalently, if the only element of B acting as the 0-map on N is 0). Suppose S is a
subring of a ring A, and r ∈ A. Suppose there is a faithful S[r]-module M that is
finitely generated as an S-module. Show that r is integral over S. (Hint: change a
few words in the proof of version 1 of Nakayama, Lemma 8.2.8.)

8.2.J. E XERCISE . Suppose A is an integral domain, and A e is the integral closure of

A in K(A), i.e., those elements of K(A) integral over A, which form a subalgebra by
Exercise 8.2.D. Show that A e is integrally closed in K(A)
e = K(A).

8.3 A gazillion finiteness conditions on morphisms

By the end of this section, you will have seen the following types of morphisms:
quasicompact, quasiseparated, affine, finite, integral, closed, (locally) of finite type,
quasifinite — and possibly, (locally) of finite presentation.

8.3.1. Quasicompact and quasiseparated morphisms.

A morphism π : X → Y of schemes is quasicompact if for every open affine
subset U of Y, π−1 (U) is quasicompact. (Equivalently, the preimage of any quasi-
compact open subset is quasicompact. This is the right definition in other parts of
geometry.)
We will like this notion because (i) finite sets have advantages over infinite sets
(e.g., a finite set of integers has a maximum; also, things can be proved inductively),
and (ii) most reasonable schemes will be quasicompact.
Along with quasicompactness comes the weird notion of quasiseparatedness.
A morphism π : X → Y is quasiseparated if for every affine open subset U of Y,
π−1 (U) is a quasiseparated scheme (§5.1.1). (Equivalently, the preimage of any
quasiseparated open subset is quasiseparated, although we won’t worry about
proving this. This is the definition that extends to other parts of geometry.) This will
be a useful hypothesis in theorems, usually in conjunction with quasicompactness.
(For this reason, “quasicompact and quasiseparated” is often abbreviated as qcqs, cf.
§5.1.2.) Various interesting kinds of morphisms (locally Noetherian source, affine,
separated, see Exercises 8.3.B(b), 8.3.D, and 11.2.D resp.) are quasiseparated, and
having the word “quasiseparated” will allow us to state theorems more succinctly.
Important remark: we will give an equivalent definition of quasiseparatedness in
§11.1.4 (“quasiseparated = quasicompact diagonal”), which will be much simpler
to use.

8.3.A. E ASY E XERCISE . Show that the composition of two quasicompact morphisms
is quasicompact. (It is also true — but not easy — that the composition of two
quasiseparated morphisms is quasiseparated. You are free to show this directly, but
 © 2024 Ravi Vakil. Published by Princeton University Press. 237

will in any case follow easily once we understand it in a more sophisticated way,
see §11.1.5.)
Following Grothendieck’s philosophy of thinking that the important notions
are properties of morphisms, not of objects (§8.0.1), we can restate the definition of
a quasicompact (resp., quasiseparated) scheme as a scheme that is quasicompact
(resp., quasiseparated) over the final object Spec Z in the category of schemes
(Exercise 7.3.J).

8.3.B. E ASY E XERCISE .

(a) Show that any morphism from a Noetherian scheme is quasicompact.
(b) Show that any morphism from a quasiseparated scheme is quasiseparated. Thus
by Exercise 5.3.A, any morphism from a locally Noetherian scheme is quasisep-
arated. Thus readers working only with locally Noetherian schemes may take
quasiseparatedness as a standing hypothesis.

8.3.2. Caution. The two parts of the Exercise 8.3.B may lead you to suspect
that any morphism π : X → Y with quasicompact source and target is necessarily
quasicompact. This is false, and you may verify that the following is a counterexam-
ple. Let Z be the nonquasiseparated scheme constructed in Exercise 5.1.K, and let
X = Spec k[x1 , x2 , . . . ] as in Exercise 5.1.K. The obvious open embedding π : X → Z
(identifying X with one of the two pieces glued together to get Z) is not quasicom-
pact. (But once you see the Cancellation Theorem 11.1.1, you will quickly see that
any morphism from a quasicompact source to a quasiseparated target is necessarily
quasicompact.)

8.3.C. E XERCISE . (Obvious hint for both parts: the Affine Communication
Lemma 5.3.2.)
(a) (quasicompactness is affine-local on the target) Show that a morphism π : X → Y is
quasicompact if there is a cover of Y by affine open sets Ui such that π−1 (Ui ) is
quasicompact.
(b) (quasiseparatedness is affine-local on the target) Show that a morphism π : X → Y is
quasiseparated if there is a cover of Y by affine open sets Ui such that π−1 (Ui ) is
quasiseparated.

8.3.3. Affine morphisms.

A morphism π : X → Y is affine if for every affine open set U of Y, π−1 (U) (in-
terpreted as an open subscheme of X) is an affine scheme. Trivially, the composition
of two affine morphisms is affine.

8.3.D. FAST E XERCISE . Show that affine morphisms are quasicompact and qua-
siseparated. (Hint for the second: Exercise 5.1.G.)

8.3.4. Proposition (the property of “affineness” of a morphism is affine-local on

the target). — A morphism π : X → Y is affine if there is a cover of Y by affine open sets
U such that π−1 (U) is affine.

Proof. We apply the Affine Communication Lemma 5.3.2 to the condition “π is affine
over”. We check our two criteria. First, suppose π : X → Y is affine over Spec B, i.e.,
π−1 (Spec B) = Spec A. Then for any s ∈ B, π−1 (Spec Bs ) = Spec Aπ♯ s . (Do you see
why?)
238 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Second, suppose we are given π : X → Spec B and (s1 , . . . , sn ) = B with Xπ♯ si

affine (Spec Ai , say). (Recall from Definition 6.2.8 that Xπ♯ si is the open subset of X
where π♯ si doesn’t vanish.) We wish to show that X is affine too. Let A = Γ (X, OX ).
Then X → Spec B factors through the tautological map α : X → Spec A (arising from
the (iso)morphism A → Γ (X, OX ), Exercise 7.3.G).

∪i Xπ♯ si = X
α / Spec A

π
( w β

∪i D(si ) = Spec B

We want to show that α is an isomorphism. Now β−1 (D(si )) = D(β♯ si ) = ∼

Spec Aβ♯ si (the preimage of a distinguished open set is a distinguished open set),
and π−1 (D(si )) = Spec Ai . Furthermore, X is quasicompact and quasiseparated
by the affine-locality of these notions (Exercise 8.3.C), so the hypotheses of the
Qcqs Lemma 6.2.9 are satisfied. Hence by the Qcqs Lemma 6.2.9, we have an
induced isomorphism of Aβ♯ si = Γ (X, OX )β♯ si =∼ Γ (Xβ♯ s , OX ) = Ai . Thus α in-
i
∼
duces an isomorphism Spec Ai −→ Spec Aβ♯ si (an isomorphism of rings induces
an isomorphism of affine schemes, Exercise 4.3.A). Thus α is an isomorphism over
each Spec Aβ♯ si , which cover Spec A, and hence α is an isomorphism. Therefore
X → Spec A is an isomorphism, so X is affine as desired. □
The affine-locality of affine morphisms (Proposition 8.3.4) has some nonobvious
consequences, as shown in the next exercise.

8.3.E. U SEFUL E XERCISE . Suppose Z is a closed subset of an affine scheme Spec A

locally cut out by one equation. (In other words, Spec A can be covered by smaller
open sets, and on each such set Z is cut out by one equation.) Show that the
complement Y of Z is affine. (This is clear if Z is globally cut out by one equation f,
even set-theoretically; then Y = Spec Af . However, Z is not always of this form, see
§19.11.10.)

8.3.5. Finite and integral morphisms.

Before defining finite and integral morphisms, we give an example to keep in
mind. If L/K is a field extension, then Spec L → Spec K (i) is always affine; (ii) is
integral if L/K is algebraic; and (iii) is finite if L/K is finite.
Recall that if we have a ring morphism B → A such that A is a finitely generated
B-module then we say that A is a finite B-algebra. This is stronger than being a
finitely generated B-algebra. (The similarity of the terminology “finite” and “finitely
generated” B-algebra is unfortunate.) We say a morphism π : X → Y is finite if for
every affine open set Spec B of Y, π−1 (Spec B) is the spectrum of a finite B-algebra.
By definition, finite morphisms are affine.

8.3.F. E XERCISE ( THE PROPERTY OF FINITENESS IS AFFINE - LOCAL ON THE TARGET ).

Show that a morphism π : X → Y is finite if there is a cover of Y by affine open sets
Spec A such that π−1 (Spec A) is the spectrum of a finite A-algebra.
The following four examples will give you some feeling for finite morphisms.
In each example, you will notice two things. In each case, the maps are always
finite-to-one (as maps of sets). We will verify this in general in Exercise 8.3.J. You
 © 2024 Ravi Vakil. Published by Princeton University Press. 239

will also notice that the morphisms are closed as maps of topological spaces, i.e.,
the images of closed sets are closed. We will show that finite morphisms are always
closed in Exercise 8.3.L (and give a second proof in §9.3.5). Intuitively, you should
think of finite as being closed plus finite fibers, although this isn’t quite true. We
will make this precise in Theorem 28.5.2.
Example 1: Branched covers. Consider the morphism Spec k[t] → Spec k[u] given
by u 7→ p(t), where p(t) ∈ k[t] is a degree n polynomial (see Figure 8.2). This is
finite: k[t] is generated as a k[u]-module by 1, t, t2 , . . . , tn−1 .

F IGURE 8.2. The “branched cover” A1k → A1k of the “u-line” by

the “t-line” given by u 7→ p(t) is finite

Example 2: Closed embeddings (to be defined soon, in §9.1.1). If I is an ideal of a ring

A, consider the morphism Spec A/I → Spec A given by the obvious map A → A/I
(see Figure 8.3 for an example, with A = k[t], I = (t)). This is a finite morphism
(A/I is generated as a A-module by the element 1 ∈ A/I).

F IGURE 8.3. The “closed embedding” Spec k → Spec k[t] given by

t 7→ 0 is finite

Example 3: Normalization (to be defined in §10.7). Consider the morphism

Spec k[t] → Spec k[x, y]/(y2 −x2 −x3 ) corresponding to k[x, y]/(y2 −x2 −x3 ) → k[t]
given by x 7→ t2 − 1, y 7→ t3 − t (check that this is a well-defined ring map!), see
Figure 8.4. This is a finite morphism, as k[t] is generated as a (k[x, y]/(y2 −x2 −x3 ))-
module by 1 and t. (The figure suggests that this is an isomorphism away from the
“node” of the target. You can verify this, by checking that it induces an isomorphism
between D(t2 − 1) in the source and D(x) in the target. We will meet this example
again repeatedly!)

8.3.G. I MPORTANT E XERCISE (E XAMPLE 4, FINITE MORPHISMS TO Spec k). Show

that if X → Spec k is a finite morphism, then X is a finite union of points with the
discrete topology, each point with residue field a finite extension of k, see Figure 8.5.
(An example is Spec(F8 × F4 [x, y]/(x2 , y4 ) × F4 [t]/(t9 ) × F2 ) → Spec F2 .) Do not
just quote some fancy theorem! Possible approach: By Exercise 3.2.G, any integral
domain which is a finite k-algebra must be a field. If X = Spec A, show that every
prime p of A is maximal. Show that the irreducible components of Spec A are
240 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

F IGURE 8.4. The “normalization” of the nodal cubic Spec k[t] →

Spec k[x, y]/(y2 − x2 − x3 ) given by (x, y) 7→ (t2 − 1, t3 − t) is finite

closed points. Show Spec A is finite, then discrete. Show that the residue fields A/m
of A are finite extensions of k. (See Exercise 8.4.D for an extension to quasifinite
morphisms.)

F IGURE 8.5. A picture of a finite morphism to Spec k. Bigger fields

are depicted as bigger points.

8.3.H. E ASY E XERCISE ( CF. E XERCISE 8.2.C). Show that the composition of two
finite morphisms is also finite.

8.3.I. E XERCISE (“ FINITE MORPHISMS TO Spec A ARE PROJECTIVE ”). If R is an

A-algebra, define a graded ring S• by S0 = A, and Sn = R for n > 0. (What is the
multiplicative structure? Hint: you know how to multiply elements of R together,
and how to multiply elements of A with elements of R.) Describe an isomorphism
∼
Proj S• −→ Spec R. Show that if R is a finite A-algebra (finitely generated as an
A-module) then S• is a finitely generated graded ring over A, and hence that Spec R
is a projective A-scheme (§4.5.10).
Hence finite morphisms are affine (by definition) and projective. The converse
is also true, see Corollary 18.1.5.

8.3.J. I MPORTANT E XERCISE . Show that finite morphisms have finite fibers. (This is
a useful exercise, because you will have to figure out how to get at points in a fiber
of a morphism: given π : X → Y, and q ∈ Y, what are the points of π−1 (q)? This
will be easier to do once we discuss fibers in greater detail, see Remark 10.3.5, but
it will be enlightening to do it now.) Hint: if X = Spec A and Y = Spec B are both
affine, and q = [q], then we can throw out everything in B outside q by modding
out by q; show that the preimage is Spec(A/π♯ qA). Then you have reduced to the
 © 2024 Ravi Vakil. Published by Princeton University Press. 241

case where Y is the Spec of an integral domain B, and [q] = [(0)] is the generic point.
We can throw out the rest of the points of B by localizing at (0). Show that the
preimage is Spec of A localized at π♯ (B \ {0}). Show that the condition of finiteness
is preserved by the constructions you have done, and thus reduce the problem to
Exercise 8.3.G.
The previous two exercises show that finite morphisms are projective and have
finite fibers. The converse is also true, see Theorem 18.1.6.
There is more to finiteness than finite fibers, and three examples to keep in mind
are Example 8.3.6, Exercise 8.3.K, and Figure 19.1 (a variant of Figure 8.4 showing a
morphism that is affine, closed, one-to-one on points and even a monomorphism,
but not finite).

8.3.6. Example. The open embedding A2k \ {(0, 0)} → A2k has finite fibers, but is not
affine (as A2k − {(0, 0)} isn’t affine, §4.4.1) and hence not finite.

8.3.K. E ASY E XERCISE . Show that the open embedding A1C \ {0} → A1C has finite
fibers and is affine, but is not finite.

8.3.7. Definition. A morphism π : X → Y of schemes is integral if π is affine, and

for every affine open subset Spec B ⊂ Y, with π−1 (Spec B) = Spec A, the induced
map B → A is an integral ring morphism. This is an affine-local condition on the
target by Exercises 8.2.A and 8.2.B, and the Affine Communication Lemma 5.3.2.
It is preserved by composition by Exercise 8.2.C. Integral morphisms are mostly
useful because finite morphisms are integral by Corollary 8.2.2. Note that the
converse implication doesn’t hold (witness Spec Q → Spec Q, as discussed after the
statement of Corollary 8.2.2).

8.3.L. E XERCISE . Prove that integral morphisms are closed, i.e., that the image
of closed subsets are closed. (Hence finite morphisms are closed. A second proof
will be given in §9.3.5.) Hint: Reduce to the affine case. If π♯ : B → A is a ring map,
inducing the integral morphism π : Spec A → Spec B, then suppose I ⊂ A cuts out
a closed set of Spec A, and J = (π♯ )−1 (I), then note that B/J ⊂ A/I, and apply the
Lying Over Theorem 8.2.5 here.

8.3.M. U NIMPORTANT EXERCISE . Suppose B → A is integral. Show that for any

ring morphismP B → C, the induced map C → A ⊗B C is integral. (Hint: We wish to
n
show that any i=1 ai ⊗ ci ∈ A ⊗B C is integral over C. Use the fact that each of the
finitely many ai are integral over B, and then Exercise 8.2.D.) Once we know what
“base change” is, this will imply that the property of integrality of a morphism is
preserved by base change, Exercise 10.4.D(e).

8.3.8. Fibers of integral morphisms. Unlike finite morphisms (Exercise 8.3.J), inte-
gral morphisms don’t always have finite fibers. (Can you think of an example?)
However, once we make sense of fibers as topological spaces (or even schemes) in
§10.3.2, you can check (Exercise 12.1.E) that the fibers have the property that no
point is in the closure of any other point.

8.3.9. Morphisms (locally) of finite type.

A morphism π : X → Y is locally of finite type if for every affine open set Spec B
of Y, and every affine open subset Spec A of π−1 (Spec B), the induced morphism
242 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

B → A expresses A as a finitely generated B-algebra. By the affine-locality of finite

typeness of B-schemes (Proposition 5.3.3(b)), this is equivalent to: π−1 (Spec B) can
be covered by affine open subsets Spec Ai so that each Ai is a finitely generated
B-algebra.
A morphism π is of finite type if it is locally of finite type and quasicompact.
Translation: for every affine open set Spec B of Y, π−1 (Spec B) can be covered with
a finite number of open sets Spec Ai so that the induced morphism B → Ai expresses
Ai as a finitely generated B-algebra.

8.3.10. Linguistic curiosity. It is a common practice to name properties as follows: P

equals locally P plus quasicompact. Two exceptions are “ringed space” (§7.3) and
“finite presentation” (§8.3.13).
Example: the “structure morphism” Pn A → Spec A is of finite type, as PA is
n

covered by n + 1 open sets of the form Spec A[x1 , . . . , xn ].

Our earlier definition of schemes of “finite type over k” (or “finite type k-
schemes”) from §5.3.6 is now a special case of this more general notion: the phrase
“a scheme X is of finite type over k” means that we are given a morphism X → Spec k
(the “structure morphism”) that is of finite type.
Here are some properties enjoyed by morphisms of finite type.

8.3.N. E XERCISE ( THE NOTIONS “ LOCALLY FINITE TYPE ” AND “ FINITE TYPE ” ARE
AFFINE - LOCAL ON THE TARGET ). Show that a morphism π : X → Y is locally of
finite type if there is a cover of Y by affine open sets Spec Bi such that π−1 (Spec Bi )
is locally finite type over Bi . Hence the notions of “locally finite type” and “finite
type” are both affine-local on the target.

8.3.O. E XERCISE ( FINITE = INTEGRAL + FINITE TYPE ).

(a) (easier) Show that finite morphisms are of finite type.
(b) Show that a morphism is finite if and only if it is integral and of finite type.

8.3.P. E XERCISES ( NOT HARD , BUT IMPORTANT ).

(a) Show that every open embedding is locally of finite type, and hence that every
quasicompact open embedding is of finite type. Show that every open embedding
into a locally Noetherian scheme is of finite type.
(b) Show that the composition of two morphisms locally of finite type is locally
of finite type. (Hence as the composition of two quasicompact morphisms is
quasicompact, Easy Exercise 8.3.A, the composition of two morphisms of finite type
is of finite type.)
(c) Suppose π : X → Y is locally of finite type, and Y is locally Noetherian. Show
that X is also locally Noetherian. If π : X → Y is a morphism of finite type, and Y is
Noetherian, show that X is Noetherian.

8.3.Q. E XERCISE . Suppose X is a scheme over Fp . Explain how to define (without

choice) an endomorphism F : X → X such that for each affine open subset Spec A ⊂
X, F corresponds to the map A → A given by f 7→ fp for all f ∈ A. (The morphism F
is called the absolute Frobenius morphism.) Prove that if X is locally of finite type
over Fp , then F is finite.
 © 2024 Ravi Vakil. Published by Princeton University Press. 243

8.3.11. Less important definition. A morphism π : X → Y is quasifinite if it is of finite

type, and for all q ∈ Y, π−1 (q) is a finite set. The main point of this definition is
the “finite fiber” part; the “finite type” hypothesis will ensure that this notion is
“preserved by fibered product,” Exercise 10.4.F.
Combining Exercise 8.3.J with Exercise 8.3.O(a), we see that finite morphisms
are quasifinite. There are quasifinite morphisms which are not finite, such as
A2k \ {(0, 0)} → A2k (Example 8.3.6). However, we will soon see that quasifinite
morphisms to Spec k are finite (Exercise 8.4.D). A key example of a morphism with
finite fibers that is not quasifinite is Spec C(t) → Spec C. Another is Spec Q →
Spec Q. (For interesting behavior caused by the fact that Spec Q → Spec Q is not of
finite type, see Warning 10.1.5.)

8.3.12. How to picture quasifinite morphisms. If π : X → Y is a finite morphism of

locally Noetherian schemes, then for any quasicompact open subset U ⊂ X, the
induced morphism π|U : U → Y is quasifinite. In fact every reasonable quasifinite
morphism arises in this way. (This simple-sounding statement is in fact a deep and
important result — a form of Zariski’s Main Theorem, see Exercise 28.5.F.) Thus
the right way to visualize quasifiniteness is as a finite map with some (closed locus
of) points removed.

8.3.13. ⋆⋆ (Locally) finitely presented morphisms. The following variant of

“locally of finite type” is useful in non-Noetherian situations.

8.3.14. Definition. A ring A is a finitely presented B-algebra (or B → A is finitely

presented) if “A has a finite number of generators and a finite number of relations
over B”:
A=∼ B[x1 , . . . , xn ]/(f1 (x1 , . . . , xn ), . . . , fr (x1 , . . . , xn )).

If B is Noetherian, then finitely presented is the same as finite type (“finite number
of relations” comes for free, Exercise 3.6.S), so normal people will not care.

8.3.R. E XERCISE : F INITELY PRESENTED IMPLIES ALWAYS FINITELY PRESENTED ( AL -

GEBRA VERSION ). Suppose A = B[x1 , . . . , xn ]/(f1 , . . . , fr ), where fi ∈ B[x1 , . . . , xn ].
Suppose also that Y1 , . . . , Ym generate A as an B-algebra, so we have a surjec-
tion ϕ : B[y1 , . . . , ym ] / / A where yj 7→ Yj . Show that the ideal ker(ϕ) ⊂
B[y1 , . . . , ym ] is finitely generated. Hint: follow the proof of Lemma 6.4.6.

8.3.15. Lemma (finite presentation of A over B is an affine-local property in A). —

The condition of a B-algebra being finitely presented satisfies the hypotheses of the Affine
Communication Lemma 5.3.2.

Proof. (i) The first hypothesis of the Affine Communication Lemma is the content of
the following Exercise.

8.3.S. E XERCISE . Suppose A = B[x1 , . . . , xn ]/(f1 , . . . , fr ), and g ∈ A. Show that Ag

is a finitely presented B-algebra. (Hint: it has one more generator and one more
relation.)
(ii) (This argument is the hardest part of our discussion on finite presentation.)
We suppose that A is a B-algebra, and g1 , . . . , gm ∈ A with (g1 , . . . , gm ) = A, and
Agi finitely presented over B. We wish to show that A is finitely presented over B.
244 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Now the Agi are finitely generated over B, so by Proposition 5.3.3(b), A is finitely
generated over B, say ϕ : B[x1 , . . . , xn ] / / A . We will show that I := ker(ϕ) is
a finitely generated ideal of B[x1 , . . . , xn ].
As (g1 , . . . , gm ) = A, we may choose hj ∈ A so that 1 = h1 g1 + · · · + hm gm .
Choose lifts Hj and Gj of the hj and gj respectively to B[x1 , . . . , xn ]. Define G0 =
1 − H1 G1 − · · · − Hm Gm , so G0 ∈ I, and (G0 , G1 , . . . , Gm ) = 1 in B[x1 , . . . , xn ].

8.3.T. E XERCISE . Show that IGj is a finitely generated module over B[x1 , . . . , xn ]Gj =
B[x1 , . . . , xn , yj ]/(yj Gj − 1). (The case j = 0 is different but easy — IG0 = (1)!) Hint:
Lemma 8.3.R.
Then by Exercise 6.4.A (“finite generation of modules is an affine-local condi-
tion”), I is a finitely generated B[x1 , . . . , xn ]-module, i.e., a finitely generated ideal
of B[x1 , . . . , xn ], as desired. □

8.3.16. Definition. We say a B-scheme X → Spec B is locally finitely presented over

B if there exists an affine cover X = ∪ Spec Ai such that Ai is finitely presented over
B. This is equivalent (by Lemma 8.3.15) to A being finitely presented over B for all
affine open subsets Spec A of X.

8.3.U. E XERCISE ( THE NOTION OF “ LOCALLY FINITELY PRESENTED MORPHISM ” IS

AFFINE - LOCAL IN THE TARGET ). Suppose π : X → Y is a morphism of schemes.
Show as Spec B ⊂ Y runs through the affine open subsets of Y, the condition
that π : π−1 (Spec B) → Spec B is locally finitely presented over Spec B satisfies the
hypotheses of the Affine Communication Lemma. (This is notably easier than
Lemma 8.3.15.)

8.3.17. Definition. A morphism π : X → Y is locally finitely presented (or locally of

finite presentation) if for every affine open set Spec B of Y, π−1 (Spec B) → Spec B
is locally finitely presented over Spec B. By Exercise 8.3.U, it suffices to check a
single affine cover of Y. (Hence this subsumes Definition 8.3.16.)

8.3.V. E XERCISE . Suppose π : X → Y is a morphism of schemes.

(a) Show that if π is locally of finite presentation, then for any open subscheme
Spec B of Y and any affine open subscheme Spec A of X with π(Spec A) ⊂ Spec B,
A is a finitely presented B-algebra.
(b) Suppose ∪ Spec Ai = X is an affine open cover of X, and Spec Bi are affine open
subsets of Y with π(Spec Ai ) ⊂ Spec Bi . Suppose Ai is finitely presented over Bi .
Show that π is locally finitely presented.

8.3.W. E XERCISE . Show that open embeddings are locally finitely presented.

8.3.X. E XERCISE . Show that the composition of two locally finitely presented
morphisms is locally finitely presented.

8.3.18. ⋆⋆ Remark. A morphism π : X → Y is locally finitely presented if and only

if for every projective system of Y-schemes {Sλ }λ∈I with each Sλ an affine scheme,
the natural map

colimλ HomY (Sλ , X) / HomY (lim Sλ , X)

λ
 © 2024 Ravi Vakil. Published by Princeton University Press. 245

is a bijection (see [Gr-EGA, IV3 .8.14.2]). This characterization of locally finitely

presented morphisms as “limit-preserving” can be useful.

8.3.19. Definition. A morphism is finitely presented (or of finite presentation) if it

is locally finitely presented, quasicompact, and quasiseparated. In particular, the
notion of finite presented morphisms is affine-local on the target.
This definition violates the principle that erasing “locally” is the same as adding
“quasicompact and” (Remark 8.3.10). But it is well motivated: finite presentation
means “finite in all possible ways” (the algebra corresponding to each affine open
set has a finite number of generators, and a finite number of relations, and a finite
number of such affine open sets cover, and their intersections are also covered by a
finite number affine open sets) — it is all you would hope for in a scheme without
it actually being Noetherian. Exercise 10.3.H makes this precise, and explains how
this notion often arises in practice.
If Y is locally Noetherian, then locally of finite presentation is the same as
locally of finite type, and finite presentation is the same as finite type. So if you are
a Noetherian person, you don’t need to worry about this notion.

8.4 Images of morphisms: Chevalley’s Theorem and elimination

theory

In this section, we will answer a question that you may have wondered about
long before hearing the phrase “algebraic geometry”. If you have a number of
polynomial equations in a number of variables with indeterminate coefficients, you
would reasonably ask what conditions there are on the coefficients for a (common)
solution to exist. Given the algebraic nature of the problem, you might hope that the
answer should be purely algebraic in nature — it shouldn’t be “random”, or involve
bizarre functions like exponentials or cosines. You should expect the answer to be
given by “algebraic conditions”. This is indeed the case, and it can be profitably
interpreted as a question about images of maps of varieties or schemes, in which
guise it is answered by Chevalley’s Theorem 8.4.2 (see §8.4.8 for a more precise
proof). As a consequence we get an immediate proof of the Nullstellensatz 3.2.6
(§8.4.3).
In certain cases, the image is nicer still. For example, the image of a finite
morphism is always closed (Exercise 8.3.L). We will prove a classical result, the
Fundamental Theorem of Elimination Theory 8.4.10, which essentially generalizes
this (as explained in §9.3.5) to maps from projective space.
In a different direction, in the distant future we will see that in certain good
circumstances (“flat” plus a bit more, see Exercise 24.5.G), morphisms are open (the
image of open subsets is open); one example (which you can try to show directly)
B → Spec B.
is An

8.4.1. Chevalley’s Theorem.

If π : X → Y is a morphism of schemes, the notion of the image of π as sets is
clear: we just take the points in Y that are the image of points in X. We know that
the image can be open (open embeddings), and we have seen examples where it
is closed, and more generally, locally closed. But it can be weirder still: consider
246 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

the morphism A2k → A2k given by (x, y) 7→ (x, xy). The image is the plane, with
the y-axis removed, but the origin put back in (Figure 8.6). We make a definition
to capture this phenomenon. A constructible subset of a Noetherian topological
space is a subset which belongs to the smallest family of subsets such that (i) every
open set is in the family, (ii) a finite intersection of family members is in the family,
and (iii) the complement of a family member is also in the family. For example, the
image of (x, y) 7→ (x, xy) is constructible.

F IGURE 8.6. The image of (x, y) 7→ (x, xy) = (u, v).

8.4.A. E XERCISE : CONSTRUCTIBLE SUBSETS ARE FINITE DISJOINT UNIONS OF LO -

CALLY CLOSED SUBSETS . Recall that a subset of a topological space X is locally
closed if it is the intersection of an open subset and a closed subset. (Equivalently,
it is an open subset of a closed subset, or a closed subset of an open subset. We
will later have trouble extending this to open and closed and locally closed sub-
schemes, see Exercise 9.2.C.) Show that a subset of a Noetherian topological space
X is constructible if and only if it is the finite disjoint union of locally closed subsets.

As a consequence, if X → Y is a continuous map of Noetherian topological

spaces, then the preimage of a constructible set is a constructible set. (Important
remark: the only reason for the hypothesis of the topological space in question
being Noetherian is because this is the only setting in which we have defined
constructible sets. An extension of the notion of constructibility to more general
topological spaces is mentioned in Exercise 10.3.I.)

8.4.B. E ASY E XERCISE ( USED IN §8.4.3). Show that the generic point of A1k does
not form a constructible subset of A1k (where k is a field).

8.4.C. E XERCISE ( USED IN E XERCISE 24.5.G).

(a) Show that a constructible subset of a Noetherian scheme is closed if and only if
it is “stable under specialization”. More precisely, if Z is a constructible subset of a
Noetherian scheme X, then Z is closed if and only if for every pair of points y1 and
y2 with y1 ∈ y2 , if`y2 ∈ Z, then y1 ∈ Z. Hint for the “if” implication: show that Z
n
can be written as i=1 ` Ui ∩ Zi where Ui ⊂ X is open and Zi ⊂ X is closed. Show
n
that Z can be written as i=1 Ui ∩ Zi (with possibly different n, Ui , Zi ) where each
 © 2024 Ravi Vakil. Published by Princeton University Press. 247

Zi is irreducible and meets Ui . Now use “stability under specialization” and the
generic point of Zi to show that Zi ⊂ Z for all i, so Z = ∪Zi .
(b) Show that a constructible subset of a Noetherian scheme is open if and only if it
is “stable under generization”. (Hint: this follows in one line from (a).)
The image of a morphism of schemes can be stranger than a constructible set.
Indeed if S is any subset of a scheme Y, it can be the image of a morphism: let X
be the disjoint union of spectra of the residue fields of all the points of S, and let
π : X → Y be the natural map. This is quite pathological, but in any reasonable
situation, the image is essentially no worse than what arose in the previous example
of (x, y) 7→ (x, xy). This is made precise by Chevalley’s Theorem.

8.4.2. Chevalley’s Theorem. — If π : X → Y is a finite type morphism of Noetherian

schemes, the image of any constructible set is constructible. In particular, the image of π is
constructible.
(For the minority who might care: see §10.3.8 for an extension to locally finitely
presented morphisms.) We discuss the proof after giving some important con-
sequences that may seem surprising, in that they are algebraic corollaries of a
seemingly quite geometric and topological theorem. The first is a proof of the
Nullstellensatz.

8.4.3. Proof of the Nullstellensatz 3.2.6. We wish to show that if K is a field extension
of k that is finitely generated as a k-algebra, say by x1 , . . . , xn , then it is a finite
extension of fields. It suffices to show that each xi is algebraic over k. But if xi is not
algebraic over k, then we have an inclusion of rings k[xi ] → K, corresponding to a
dominant morphism π : Spec K → A1k of finite type k-schemes. Of course Spec K is
a single point, so the image of π is one point. By Chevalley’s Theorem 8.4.2 and
Exercise 8.4.B, the image of π is not the generic point of A1k , so im(π) is a closed
point of A1k , and thus π is not dominant. □
A similar idea can be used in the following exercise.

8.4.D. E XERCISE ( QUASIFINITE MORPHISMS TO A FIELD ARE FINITE ). Suppose

π : X → Spec k is a quasifinite morphism. Show that π is finite. (Hint: deal first
with the affine case, X = Spec A, where A is finitely generated over k. Suppose
A contains an element x that is not algebraic over k, i.e., we have an inclusion
k[x] ,→ A. Exercise 8.3.G may help.)

8.4.E. E XERCISE . Suppose X is a scheme of finite type over Spec Z, and p is a closed
point of X. Show that the residue field κ(p) is a finite field.

8.4.F. E XERCISE ( FOR MAPS OF VARIETIES , SURJECTIVITY CAN BE CHECKED ON

CLOSED POINTS ). Assume Chevalley’s Theorem 8.4.2. Show that a morphism of
affine k-varieties π : X → Y is surjective if and only if it is surjective on closed points
(i.e., if every closed point of Y is the image of a closed point of X). (Once we define
varieties in general, in Definition 11.2.9, you will see that your argument works
without change with the adjective “affine” removed.)

8.4.4. Proof of Chevalley’s Theorem 8.4.2.

248 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

8.4.G. E XERCISE .
(a) Reduce the proof of Chevalley’s Theorem to the case where Y is affine, say
Y = Spec B.
(b) Reduce further to the case where X is affine, say X = Spec B[t1 , . . . , tn ]/I.
(c) Reduce further to the case where X = An B = Spec B[t1 , . . . , tn ].
(d) Reduce further to the case where X = A1B = Spec B[t].
We have simplified the task to showing that given π : X = Spec B[t] → Spec B =
Y, the image of any constructible subset of X is constructible in Y. Because con-
structible sets are finite unions of locally closed subsets (Exercise 8.4.A — we are in
the Noetherian situation), we need only show that the image of any locally closed
subset Z of X is constructible in Y.

8.4.5. An important special case. We begin with the case when Z is a closed subset of
X = A1B , say
Z = V(f1 (t), . . . , fr (t)), where fi (t) = aidi tdi + · · · + ai1 t + ai0 ,
with aij ∈ B, so fi (t) ∈ B[t] and di = deg fi (t).
Now Y = Spec B is the finite disjoint union of locally closed subsets Y⃗e indexed
by ⃗e = (e1 , . . . , er ), where ei ∈ Z, −1 ≤ ei ≤ di , and Y⃗e is the locus where
f1 (t), . . . , fr (t) have “true degree” e1 , . . . , er respectively: the coefficient of tei in
fi (t) is nonzero, and the “higher” coefficients are zero (at the points of Y⃗e ). More
precisely: Y⃗e ⊂ Y is the locus where aiei ̸= 0, and aij = 0 for j > ei . (The zero
polynomial has “true degree” −1 under this convention.) More precisely still, we
define Y⃗e := Spec B⃗e , where
 
Q
B i aiei \
B⃗e := , so Y⃗e = (∩i D(aiei )) ∩  V(aij ) .
 
(aij )j>ei i
j>ei

If ⃗e = (−1, . . . , −1), then

(8.4.5.1) π−1 (Y⃗e )

π / Y⃗e

is surjective (the map (8.4.5.1) is Spec B(−1,...,−1) [t] → Spec B(−1,...,−1) ), and if
⃗e ̸= (−1, . . . , −1), (8.4.5.1) is a finite morphism corresponding to the map of rings

/ B⃗e [t]/ te1 + a1(e1 −1) te1 −1 + · · · + a10 , . . . , ter + ar(er −1) ter −1 + · · · +
 
ar0
B⃗e a1e a1e are arer .
1 1 r

(The ring on the right is generated as a B⃗e -module by 1, t, . . . , tmax(ei ) .) Because

the images of finite morphisms are closed (Exercise 8.3.L), π(Z) ∩ Y⃗e (the image of
(8.4.5.1)) is a locally closed subset of Y, so π(Z) is indeed constructible when Z is
closed. We have completed our important special case.

8.4.6. Note that the locus of points q ∈ Y where the entire fiber π−1 (q) is contained
in Z, {q ∈ Y : π−1 (q) ⊂ Z}, is a closed subset of Y: it is Spec B(−1,...,−1) =
Spec B/(aij ). We call this FL(Z) ⊂ Y, the “fibral locus for Z”, only for the rest of
this proof.

8.4.7. Finally, if Z is locally closed, then Z ⊂ X is closed, so π(Z) is a constructible

subset of Y = Spec B. Define δZ := Z \ Z, a closed subset of X; FL(Z) and FL(δZ)
 © 2024 Ravi Vakil. Published by Princeton University Press. 249

are both closed subsets of Y by §8.4.6. Since FL(Z) \ FL(δZ) is contained in π(Z),
and FL(δZ) is contained in the complement of π(Z), it suffices to prove the result
for the open subset Y ′ := Y \ (FL(Z) ∪ FL(δZ)) ⊂ Y. Let π ′ : X ′ = A1Y ′ → Y ′ be the
restriction of π above Y ′ , and let Z ′ = Z ∩ A1Y ′ (which is quasifinite above Y ′ ), and
δZ ′ = Z ′ \ Z ′ ⊂ A1Y ′ (where Z ′ is the closure of Z ′ in A1Y ′ ).

8.4.H. E XERCISE . Show that the closed subset δZ ′ ⊂ A1Y ′ does not meet any generic
fiber of π ′ , i.e., the preimage of any generic point of Y ′ . Hence show that there is a
dense open subset of Y ′ not intersecting the constructible set π ′ (δZ ′ ).
We have now proved the result above a dense open subset of Spec B.

8.4.I. E XERCISE . Finish the proof of Chevalley’s Theorem 8.4.2 by invoking

Noetherian induction (in a suitably rigorous manner).
□

8.4.8. ⋆ Elimination of quantifiers. A basic sort of question that arises in any

number of contexts is when a system of equations has a solution. Suppose for
example you have some polynomials in variables x1 , . . . , xn over an algebraically
closed field k, some of which you set to be zero, and some of which you set to
be nonzero. (This question is of fundamental interest even before you know any
scheme theory!) Then there is an algebraic condition on the coefficients which will
n
tell you if there is a solution. Define the Zariski topology on k in the obvious way:
closed subsets are cut out by equations. (A mild generalization of this appears in
Exercise 12.2.L.)

8.4.J. E XERCISE ( ELIMINATION OF QUANTIFIERS , OVER AN ALGEBRAICALLY CLOSED

FIELD ). Fix an algebraically closed field k. Suppose

f1 , . . . , fp , g1 , . . . , gq ∈ k[W1 , . . . , Wm , X1 , . . . Xn ]
m
are given. Show that there is a (Zariski-)constructible subset Y of k such that
(8.4.8.1) f1 (w1 , . . . , wm , X1 , . . . , Xn ) = · · · = fp (w1 , . . . , wm , X1 , . . . , Xn ) = 0
and
(8.4.8.2) g1 (w1 , . . . , wm , X1 , . . . , Xn ) ̸= 0 ··· gq (w1 , . . . , wm , X1 , . . . , Xn ) ̸= 0
n
has a solution (X1 , . . . , Xn ) = (x1 , . . . , xn ) ∈ k if and only if (w1 , . . . , wm ) ∈ Y.
Hints: if Z is a finite type scheme over k, and the closed points are denoted Zcl (“cl”
is for either “closed” or “classical”), then under the inclusion of topological spaces
Zcl ,→ Z, the Zariski topology on Z induces the Zariski topology on Zcl . Note that
p
we can identify (Akp )cl with k by the Nullstellensatz (Exercise 5.3.F). If X is the
locally closed subset of Am+n cut out by the equalities and inequalities (8.4.8.1)
and (8.4.8.2), we have the diagram
 
Xcl  / X / Am+n
loc. cl.

πcl π
  {
m 
k  / Am
250 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

where Y = im πcl . By Chevalley’s Theorem 8.4.2, im π is constructible, and hence so

m m
is (im π) ∩ k . It remains to show that (im π) ∩ k = Y (= im πcl ). You might use
the Nullstellensatz.
This is called “elimination of quantifiers” because it gets rid of the quantifier
“there exists a solution”. The analogous statement for real numbers, where inequali-
ties are also allowed, is a special case of Tarski’s celebrated theorem of elimination
of quantifiers for real closed fields (see, for example, [Ta]).

8.4.9. The Fundamental Theorem of Elimination Theory.

In the case of projective space (and later, projective morphisms), one can do
better than Chevalley (at least in describing images of closed subsets).

8.4.10. Theorem (Fundamental Theorem of Elimination Theory). — The morphism

A → Spec A is closed (sends closed sets to closed sets).
π : Pn
Note that no Noetherian hypotheses are needed.
A great deal of classical algebra and geometry is contained in this theorem as
special cases. Here are some examples.
First, let A = k[a, b, c, . . . , i], and consider the closed subset of P2A (taken
with coordinates x, y, z) corresponding to ax + by + cz = 0, dx + ey + fz = 0,
gx + hy + iz = 0. Then we are looking for the locus in Spec A where these equations
have a nontrivial solution. This indeed corresponds to a Zariski-closed set — where
 
a b c
det d e f  = 0.
g h i

Thus the idea of the determinant is embedded in elimination theory.

As a second example, let A = k[a0 , a1 , . . . , am , b0 , b1 , . . . , bn ]. Now consider
the closed subset of P1A (taken with coordinates x and y) corresponding to a0 xm +
a1 xm−1 y + · · · + am ym = 0 and b0 xn + b1 xn−1 y + · · · + bn yn = 0. Then there is
a polynomial in the coefficients a0 , . . . , bn (an element of A) which vanishes if and
only if these two polynomials have a common nonzero root — this polynomial is
called the resultant.
More generally, given a number of homogeneous equations in n + 1 variables
with indeterminate coefficients, Theorem 8.4.10 implies that one can write down
equations in the coefficients that precisely determine when the equations have a
nontrivial solution.

8.4.11. Proof of the Fundamental Theorem of Elimination Theory 8.4.10. Suppose

Z ,→ Pn A is a closed subset. We wish to show that π(Z) is closed.
By the definition of the Zariski topology on Proj A[x0 , . . . , xn ] (§4.5.7), Z is cut
out (set-theoretically) by some homogeneous elements f1 , f2 , . . . ∈ A[x0 , . . . , xn ].
We wish to show that the points p ∈ Spec A that are in π(Z) form a closed subset.
Equivalently, we want to show that those p for which f1 , f2 , . . . have a common
zero in Proj κ(p)[x0 , . . . , xn ] form a closed subset of Spec A.
To motivate our argument, we consider a related question. Suppose that S• :=
k[x0 , . . . , xn ], and that g1 , g2 , . . . ∈ k[x0 , . . . , xn ] are homogeneous polynomials.
How can we tell if g1 , g2 , . . . have a common zero in Proj S• = Pn k?
 © 2024 Ravi Vakil. Published by Princeton University Press. 251

They would have a common zero if and only if in

An+1
k = Spec S• ,
they cut out (set-theoretically) more than the origin, i.e., if set-theoretically
V(g1 , g2 , . . . ) ̸⊂ V(x0 , . . . , xn ).
By the inclusion-reversing bijection between closed subsets and radical ideals
(Theorem 3.7.1), this is true if and only if
p p
(g1 , g2 , . . . ) ̸⊃ (x0 , . . . , xn ).
But (x0 , . . . , xn ) is radical, so this is true if and only if
p
(g1 , g2 , . . . ) ̸⊃ (x0 , . . . , xn ).
This isptrue if and only if not all of the generators x0 , . . . , xn of the ideal (x0 , . . . , xn )
are in (g1 , g2 , . . . ), which is true if only if there is some i such that no power of
xi is in (g1 , g2 , . . . ). This is true if and only if
(x0 , . . . , xn )N ̸⊂ (g1 , g2 , . . . ) for all N.
(Do you see why this implication goes both ways?) This is equivalent to
SN ̸⊂ (g1 , g2 , . . . ) for all N,
which may be rewritten as
(8.4.11.1) SN ̸⊂ g1 SN−deg g1 ⊕ g2 SN−deg g2 ⊕ · · ·
for all N. In other words, this is equivalent to the statement that the k-linear map
SN−deg g1 ⊕ SN−deg g2 ⊕ · · · / SN

is not surjective. This map is given by a matrix with dim SN rows. (It may have an
infinite number of columns, but this will not bother us.) To check that this linear
map is not surjective, we need only check that all the “maximal” (dim SN × dim SN )
determinants are zero. (Of course, we need to check this for all N.) Thus the
condition that g1 , g2 , . . . have no common zeros in Pn
k is the same as checking some
(admittedly infinite) number of equations. In other words, it is a Zariski-closed
condition on the coefficients of the polynomials g1 , g2 , . . . .

8.4.K. E XERCISE . Complete the proof of the Fundamental Theorem of Elimination

Theory 8.4.10. (Hint: follow precisely the same argument, with k replaced by A,
and the gi replaced by the fi . How and why does this prove the theorem? If the
previous sentence is mysterious, you may want to wait until you reach §10.3.2, or
perhaps glance there now.) □
Notice that projectivity was crucial to the proof: we used graded rings in an
essential way. Notice also that the proof is essentially just linear algebra.
 CHAPTER 9

Closed embeddings and related notions

The scheme-theoretic analog of closed subsets has a surprisingly different fla-

vor from the analog of open sets (open subschemes/embeddings). However, just
as open subschemes (the scheme-theoretic version of open set) are locally modeled
on open sets U ⊂ Y, the analog of closed subsets also has a local model. This was
foreshadowed by our understanding of closed subsets of Spec B as roughly corre-
sponding to ideals. If I ⊂ B is an ideal, then Spec B/I ,→ Spec B is a morphism of
schemes, and we have checked that on the level of topological spaces, this describes
Spec B/I as a closed subset of Spec B, with the subspace topology (Exercise 3.4.I).
This morphism is our “local model” of a closed subscheme/embedding.

9.1 Closed embeddings and closed subschemes

9.1.1. Definition. A morphism π : X → Y is a closed embedding (or closed

immersion) if it is an affine morphism, and for every affine open subset Spec B ⊂ Y,
with π−1 (Spec B) = ∼ Spec A, the map B → A is surjective (i.e., of the form B → B/I,
our desired local model). (“Embedding” is preferable to “immersion”, because the
differential geometric notion of immersion is closer to what algebraic geometers
call unramified, which we will define in §21.7.) The symbol ,→ often is used to
indicate that a morphism is a closed embedding (or more generally, a locally closed
embedding, §9.2).
If X is a subset of Y (and π on the level of sets is the inclusion), we say that X is a
closed subscheme of Y. A closed embedding is the same thing as an isomorphism
with a closed subscheme.

9.1.A. E XERCISE . Show that a closed embedding identifies the topological space
of X with a closed subset of the topological space of Y. (Caution: The closed
embeddings Spec k[x]/(x) ,→ Spec k[x] and Spec k[x]/(x2 ) ,→ Spec k[x] show that
the closed subset does not determine the closed subscheme. The “infinitesimal”
information, or “fuzz”, is lost.)

9.1.B. E ASY E XERCISE . Show that closed embeddings are finite morphisms, hence
of finite type.

This material is published by Princeton University Press as The Rising Sea: Foundations of Algebraic
Geometry by Ravi Vakil. This version is free to view and download for personal use only. Not for
redistribution, resale, or use in derivative works. © 2024 Ravi Vakil.

253
254 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

9.1.C. E ASY E XERCISE . Show that the composition of two closed embeddings is a
closed embedding.

9.1.D. E XERCISE . Show that the property of being a closed embedding is affine-
local on the target.
In particular, if B → A is a surjection of rings, then the induced morphism
Spec A → Spec B is a closed embedding.

9.1.2. Important: Closed subschemes correspond to quasicoherent sheaves of

ideals. We call a sub-OY -module of OY an ideal sheaf (or sheaf of ideals) on Y. A
closed subscheme π : X ,→ Y gives a surjection of OY -modules OY → π∗ OX . The
kernel IX/Y of this map is then a sheaf of ideals on Y. On each open subset U, it
gives an ideal IX/Y (U) of the ring OY (U); on the affine open subset Spec B, it gives
the ideal I of B determining the surjection B → A = B/I in the definition of “closed
subscheme”.

9.1.E. E XERCISE . Show that π∗ OX is a quasicoherent sheaf on Y. (One way of

proceeding is to use Exercise 6.2.D.)
Quasicoherent sheaves form an abelian category, so IX/Y = ker(OY → π∗ OX )
is also a quasicoherent sheaf on Y. Thus every closed subscheme of Y yields a
quasicoherent sheaf of ideals on Y. We call the exact sequence (of quasicoherent
sheaves on Y)

(9.1.2.1) 0 / IX/Y / OY / π∗ OX /0

the closed subscheme exact sequence for π : X ,→ Y.

We recover X (and its map π to Y) from the ideal sheaf IX/Y via (9.1.2.1): for
each affine open subset Spec B ⊂ Y, the sections over Spec B give us an exact
sequence
0 /I /B / B/I / 0,

and Spec B/I is π−1 (Spec B).

The converse holds:

9.1.F. I MPORTANT E XERCISE . Prove that every quasicoherent sheaf of ideals on Y

produces a closed subscheme on Y in this way.
This is a hard exercise, so as a hint, here are two different ways of proceed-
ing; some combination of them may work for you. Approach 1. For each affine
open Spec B, we have a closed subscheme Spec B/I(B) ,→ Spec B. (i) For any
two affine open subschemes Spec A and Spec B, show that the two closed sub-
schemes Spec A/I(A) ,→ Spec A and Spec B/I(B) ,→ Spec B restrict to the same
closed subscheme of their intersection. (Hint: cover their intersection with open
sets simultaneously distinguished in both affine open sets, Proposition 5.3.1.) Thus
for example we can glue these two closed subschemes together to get a closed
subscheme of Spec A ∪ Spec B. (ii) Use Exercise 4.4.A on gluing schemes (or the
ideas therein) to glue together the closed embeddings in all affine open subschemes
simultaneously. You will only need to worry about triple intersections. Approach
2. (i) Describe X first as a subset of Y. (ii) Check that X is closed. (iii) Define the
sheaf of functions OX on this subset, perhaps using compatible germs. (iv) Check
 © 2024 Ravi Vakil. Published by Princeton University Press. 255

that this resulting ringed space is indeed locally the closed subscheme given by
Spec B/I ,→ Spec B.

9.1.G. U NIMPORTANT EXERCISE : A SHEAF OF IDEALS ( AS O - MODULES ) THAT IS

NOT QUASICOHERENT. Let X = Spec k[x](x) , the “germ of the affine line at the
origin”, which has two points, the closed point and the generic point η. Define
I (X) = {0} ⊂ OX (X) = k[x](x) , and I (η) = k(x) = OX (η). Show that this sheaf of
ideals does not correspond to a closed subscheme.

9.1.H. ⋆⋆ H ARD ( AND FRANKLY UNIMPORTANT ) E XERCISE ( NOT USED LATER ). In

the literature, the usual definition of a closed embedding is a morphism π : X → Y
such that π induces a homeomorphism of the underlying topological space of X onto
a closed subset of the topological space of Y, and the induced map π♯ : OY → π∗ OX
of sheaves on Y is surjective. (By “surjective” we mean that the ring morphism
on stalks is surjective.) Show that this definition agrees with the one given above.
(To show that our definition involving surjectivity on the level of affine open sets
implies this definition, you can use the fact that surjectivity of a morphism of
sheaves can be checked on a suitably chosen base, Exercise 2.5.D.)
We have now defined the analog of open subsets and closed subsets in the
land of schemes. Their definition is slightly less “symmetric” than in the classical
topological setting: the “complement” of a closed subscheme is a unique open
subscheme, but there are many “complementary” closed subschemes to a given
open subscheme in general. (We will soon define one that is “best”, that has a
reduced structure, §9.4.9.)

9.1.3. Examples and properties.

9.1.I. I MPORTANT E XERCISE .

(a) In analogy with closed subsets, define the notion of a finite (scheme-theoretic)
union of closed subschemes of X, and an arbitrary (not necessarily finite) (scheme-
theoretic) intersection of closed subschemes of X. (Exercise 9.1.F may help.)
Hint: If X is affine, then you might expect that the union of closed subschemes
corresponding to I1 and I2 would be the closed subscheme corresponding to either
I1 ∩ I2 or I1 I2 — but which one? We would want the union of a closed subscheme
with itself to be itself, so the right choice is I1 ∩ I2 .
(b) Describe the scheme-theoretic intersection of V(y − x2 ) and V(y) in A2 . See
Figure 4.5 for a picture. (For example, explain informally how this corresponds
to two curves meeting at a single point with multiplicity 2 — notice how the 2 is
visible in your answer. Alternatively, what is the nonreducedness telling you —
both its “size” and its “direction”?) Describe their scheme-theoretic union.
(c) Show that the underlying set of a finite union of closed subschemes is the finite
union of the underlying sets, and similarly for arbitrary intersections.
(d) Describe the scheme-theoretic intersection of V(y2 − x2 ) and V(y) in A2 . Draw
a picture. (Did you expect the intersection to have multiplicity one or multiplicity
two?) Hence show that if X, Y, and Z are closed subschemes of W, then (X ∩ Z) ∪
(Y ∩ Z) ̸= (X ∪ Y) ∩ Z in general. In particular, not all properties of intersection and
union carry over from sets to schemes.
256 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

9.1.J. E XERCISE . Show that closed embeddings (like open embeddings, Easy
Exercise 8.1.E) are monomorphisms.
In many cases, loci which are a priori closed sets often have enriched versions
as closed subschemes. The remaining exercises are examples of this. (As described
in §6.6.33, we often indicate that we are considering scheme-theoretic enrichments
by adding the adjective scheme-theoretic — for example, we say “scheme-theoretic
intersection”, “scheme-theoretic union”, and so forth.)

9.1.K. I MPORTANT E XERCISE /D EFINITION : THE VANISHING S CHEME ( CF. E XER -

CISE 4.5.S).
(a) Suppose Y is a scheme, and s ∈ Γ (Y, OY ). Define the closed subscheme cut out
by s. We call this the vanishing scheme V(s) of s, as it is the scheme-theoretic
version of our earlier (set-theoretical) version of V(s) (§3.4.1). (Hint: on affine open
Spec B, we just take Spec B/(sB ), where sB is the restriction of s to Spec B. Use
Exercise 9.1.F to show that this yields a well-defined closed subscheme.)
(b) If u is an invertible function, show that V(s) = V(su).
(c) If S is a set of functions, define V(S).

9.1.4. Definition: scheme-theoretic support. If M is an A-module, and m ∈ M, we

call the closed subscheme Spec(A/ Ann m) of Spec A the scheme-theoretic support
of m. (This is often described as the “smallest closed subscheme on which m is
defined”. Can you make this precise?)

9.1.L. E XERCISE . Show that the underlying set of the scheme-theoretic support of
m is the usual “set-theoretic” support, Supp m.

9.1.M. E XERCISE . Define the scheme-theoretic support of a finitely generated

A-module M to be the scheme-theoretic intersection of all closed subschemes
Spec A/I ,→ Spec A for which M is an (A/I)-module. Show that it is the scheme-
theoretic union of the scheme-theoretic supports of any finite generating set of
M.

9.1.N. E XERCISE ( GLOBALIZING THE PREVIOUS EXERCISE ). Suppose X is a scheme,

and F is a finite type quasicoherent sheaf on X. How will you define the scheme-
theoretic support of F ?

9.2 Locally closed embeddings and locally closed subschemes

Now that we have defined analogs of open and closed subsets, it is natural to
define the analog of locally closed subsets. Recall that locally closed subsets are
intersections of open subsets and closed subsets. Hence they are closed subsets of
open subsets, or equivalently open subsets of closed subsets. The analog of these
equivalences will be a little problematic in the land of schemes. (Simiarly, the notion
of subgroup and quotient group are not quite complementary. Rhetorical question:
should “subquotient” of a group G be defined as a subgroup of a quotient of G , or
a quotient of a subgroup of G, or are these the same? We will confront the same
issue when defining locally closed embeddings.)
 © 2024 Ravi Vakil. Published by Princeton University Press. 257

We say a morphism π : X → Y is a locally closed embedding (or locally closed

immersion) if π can be factored into
 ρ /  τ /
X Z open Y
closed

where (as indicated) ρ is a closed embedding and τ is an open embedding. For

example, the morphism Spec k[t, t−1 ] → Spec k[x, y] given by (x, y) 7→ (t, 0) is a
locally closed embedding (Figure 9.1). The symbol ,→ is often used to indicate
that a morphism is a locally closed embedding. Because closed embeddings and
open embeddings are monomorphisms (Exercises 9.1.J and 8.1.E respectively), and
monomorphisms are preserved by composition, locally closed embeddings are also
monomorphisms.

F IGURE 9.1. The locally closed embedding Spec k[t, t−1 ] →

Spec k[x, y] (t 7→ (t, 0) = (x, y), i.e., (x, y) → (t, 0))

(Warning: The term immersion is often used instead of locally closed embedding
or locally closed immersion, but this is unwise terminology, for reasons that already
arose for closed embeddings in Definition 9.1.1: The differential geometric notion
of immersion is closer to what algebraic geometers call unramified, which we will
define in §21.7. Also, the naked term embedding should be avoided, because it is
needlessly imprecise.)

9.2.1. If X is a subset of Y (and π on the level of sets is the inclusion), we say X is a

locally closed subscheme of Y. (Warning: the naked term subscheme is often used
to mean locally closed subscheme, but we will avoid this unhappy usage.)

9.2.A. E ASY EXERCISE . Show that locally closed embeddings are locally of finite
type.

9.2.B. E XERCISE . Suppose π : X → Y is a locally closed embedding whose image is

a closed subset of Y. Show that π is a closed embedding. Possible hint: the property
of being a closed embedding is affine-local on the target, Exercise 9.1.D. (This is
philosophically useful, see for example §11.2.2.)
At this point, you could define the intersection of two locally closed embed-
dings in a scheme X (which will also be a locally closed embedding in X). But it
would be awkward, as you would have to show that your construction is indepen-
dent of the factorizations of each locally closed embedding into a closed embedding
and an open embedding. Instead, we wait until Exercise 10.2.C, when recognizing
the intersection as a fibered product will make this easier.
258 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Clearly an open subscheme U of a closed subscheme V of X can be interpreted

as a closed subscheme of an open subscheme: as the topology on V is induced from
the topology on X, the underlying set of U is the intersection of some open subset
U ′ on X with V. We can take V ′ = V ∩ U ′ , and then V ′ → U ′ is a closed embedding,
and U ′ → X is an open embedding.
It is not clear that a closed subscheme V ′ of an open subscheme U ′ can be
expressed as an open subscheme of a closed subscheme V. In the category of
topological spaces, we would take V as the closure of V ′ , so we are now motivated
to define the analogous construction, which will give us an excuse to introduce
several related ideas, in §9.4. We will then resolve this issue in good cases (e.g., if X
is Noetherian) in Exercise 9.4.C.
We formalize our discussion in an exercise.

9.2.C. E XERCISE . Suppose V → X is a morphism. Consider three conditions:

(i) V is the intersection of an open subscheme of X and a closed subscheme
of X (you will have to define the meaning of “intersection” here, see
Exercise 8.1.D, or else see the hint below).
(ii) V is an open subscheme of a closed subscheme of X, i.e., it factors into an
open embedding followed by a closed embedding.
(iii) V is a closed subscheme of an open subscheme of X, i.e., V is a locally
closed embedding.
Show that (i) and (ii) are equivalent, and both imply (iii). (Remark: (iii) does not
always imply (i) and (ii), see the pathological example [Stacks, tag 01QW].) Hint: It
may be helpful to think of the problem as follows. You might hope to think of a
locally closed embedding as a Cartesian diagram.


V  _
open emb.
/ K
_
closed emb. closed emb.
  
U /X
open emb.

(We already know that fibered products with open embeddings exist, by Exer-
cise 8.1.D.) Interpret (i) as the existence of the diagram. Interpret (ii) as this diagram
minus the lower left corner. Interpret (iii) as the diagram minus the upper right
corner.

9.2.D. E XERCISE . Show that the composition of two locally closed embeddings is
a locally closed embedding. (Hint: you might use (ii) implies (iii) in the previous
exercise.)

9.2.2. Unimportant but fun remark. It may feel odd that in the definition of a
locally closed embedding, we had to make a choice (as a composition of a closed
embedding followed by an open embedding, rather than vice versa), but this type
of issue comes up earlier: a subquotient of a group can be defined as the quotient of
a subgroup, or a subgroup of a quotient. Which is the right definition? Or are they
the same? (Hint: the subquotient of a subquotient should certainly be a subquotient,
cf. Exercise 9.2.D.)
 © 2024 Ravi Vakil. Published by Princeton University Press. 259

9.3 Important examples from projective geometry

We now interpret closed embeddings in terms of graded rings. Don’t worry;

most of the annoying foundational discussion of graded rings is complete, and we
now just take advantage of our earlier work.

9.3.1. Example: Closed embeddings in projective space Pn A . Recall the definition

of projective space Pn A given in §4.4.10 (and the terminology defined there). Any
homogeneous polynomial f in x0 , . . . , xn defines a closed subscheme. (Thus even if f
doesn’t make sense as a function, its vanishing scheme still makes sense.) On the
open set Ui , the closed subscheme is V(f(x0/i , . . . , xn/i )), which we privately think
deg f
of as V(f(x0 , . . . , xn )/xi ). On the overlap

Ui ∩ Uj = Spec A[x0/i , . . . , xn/i , x−1

j/i ]/(xi/i − 1),

these functions on Ui and Uj don’t exactly agree, but they agree up to a nonva-
nishing scalar, and hence cut out the same closed subscheme of Ui ∩ Uj (Exer-
cise 9.1.K(b)):
deg f
f(x0/i , . . . , xn/i ) = xj/i f(x0/j , . . . , xn/j ).

Similarly, a collection of homogeneous polynomials in A[x0 , . . . , xn ] cuts out a

closed subscheme of Pn A . (Exercise 15.7.I will show that all closed subschemes of
PnA are of this form.)

9.3.2. Definition. As usual, let k be a field. A closed subscheme of Pn k cut out by

a single (nonzero, homogeneous) equation is called a hypersurface (in Pn k ). (Be
careful: the word “hypersurface” can be used to mean slightly different things in
algebraic geometry, for entirely reasonable reasons, see for example §12.1.5.) Of
course, a hypersurface is not cut out by a single global function on Pn k : there are
no nonconstant global functions (Exercise 4.4.E). The degree of a hypersurface is
the degree of the polynomial. (Implicit in this is that this notion can be determined
from the subscheme itself. You may have the tools to prove this now, but we won’t
formally prove it until Exercise 18.6.G.) A hypersurface of degree 1 (resp., degree
2, 3, . . . ) is called a hyperplane (resp., quadric, cubic, quartic, quintic, sextic,
septic, octic, . . . hypersurface). If n = 2, hypersurfaces are called curves; a degree
1 hypersurface is called a line, and a degree 2 hypersurface is called a conic curve,
or a conic for short. If n = 3, a hypersurface is called a surface. (In Chapter 12, we
will justify the terms curve and surface.)

9.3.A. E XERCISE .
(a) Show that wz = xy, x2 = wy, y2 = xz describes an irreducible closed subscheme
in P3k . In fact it is a curve, a notion we will define once we know what dimension
is. This curve is called the twisted cubic. (The twisted cubic is a good nontrivial
example of many things, so you should make friends with it as soon as possible. It
implicitly appeared earlier in Exercise 3.6.F.)
(b) Show that the twisted cubic is isomorphic to P1k .
We now extend this discussion to projective schemes in general.
260 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

9.3.B. E XERCISE . Suppose that S• / / R• is a surjection of graded rings. Show

that the domain of the induced morphism (Exercise 7.4.A) is Proj R• , and that the
induced morphism Proj R• → Proj S• is a closed embedding.
Exercise 15.7.I will show the converse to Exercise 9.3.B: every closed embedding
X ,→ Proj S• into a projective A-scheme (S• is a finitely generated graded A-algebra)
is of the form Proj(S• /I), where I is a homogeneous ideal, of “projective functions”
vanishing on X. (You may be able to prove this now.)

9.3.C. E XERCISE . Show that an injective linear map of k-vector spaces V ,→ W

induces a closed embedding PV ,→ PW. (This is another justification for the
definition of PV in Example 4.5.13 in terms of the dual of V.)

9.3.3. Definition. The closed subscheme defined in Exercise 9.3.C is called a linear
space. Once we know about dimension, we will call this closed subscheme a
linear space of dimension dim V − 1 = dim PV. More explicitly, a linear space of
dimension n in PN is any closed subscheme cut out by N−n k-linearly independent
homogeneous linear polynomials in x0 , . . . , xN . A linear space of dimension 1
(resp., 2, n, dim PV − 1) is called a line (resp., plane, n-plane, hyperplane). (If the
linear map in the previous exercise is not injective, then the hypothesis (7.4.0.1) of
Exercise 7.4.A fails.)

9.3.D. E XERCISE ( A SPECIAL CASE OF B ÉZOUT ’ S T HEOREM ). Suppose X ⊂ Pn k

is a degree d hypersurface cut out by f = 0, and ℓ is a line not contained in X. A
very special case of Bézout’s Theorem (Exercise 18.6.J) implies that X and ℓ meet
with multiplicity d, “counted correctly”. Make sense of this, by restricting the
homogeneous degree d polynomial f to the line ℓ, and using the fact that a degree
d polynomial in k[x] has d roots, counted properly. (If it makes you feel better,
assume k = k.)

9.3.E. E XERCISE . Show that the map of graded rings k[w, x, y, z] → k[s, t] given
by (w, x, y, z) 7→ (s3 , s2 t, st2 , t3 ) induces a closed embedding P1k ,→ P3k , which
yields an isomorphism of P1k with the twisted cubic (defined in Exercise 9.3.A —
in fact, this will solve Exercise 9.3.A(b)). Doing this in a hands-on way will set
you up well for the general Veronese construction of §9.3.6; see Exercise 9.3.I for a
generalization.

9.3.4. A particularly nice case: when S• is generated in degree 1.

Suppose S• is a finitely generated graded ring generated in degree 1. Then S1
is a finitely generated S0 -module, and the irrelevant ideal S+ is generated in degree
1 (cf. Exercise 4.5.G(a)).

9.3.F. E XERCISE . Show that if S• is generated (as an A-algebra) in degree 1 by n + 1

elements x0 , . . . , xn , then Proj S• may be described as a closed subscheme of Pn A as
⊕(n+1)
follows. Consider A as a free module with generators t0 , . . . , tn associated
 © 2024 Ravi Vakil. Published by Princeton University Press. 261

to x0 , . . . , xn . The surjection of

Sym• A⊕(n+1) = A[t0 , t1 , . . . , tn ] / / S•

ti  / xi

implies S• = A[t0 , t1 , . . . tn ]/I, where I is a homogeneous ideal. (In particular,

Proj S• can always be interpreted as a closed subscheme of some Pn A if S• is finitely
generated in degree 1. Then using Exercises 7.4.D and 7.4.G, you can remove the
hypothesis of generation in degree 1.)
This is analogous to the fact that if R is a finitely generated A-algebra, then
choosing n generators of R as an algebra is the same as describing Spec R as a closed
subscheme of An A . In the affine case this is “choosing coordinates”; in the projective
case this is “choosing projective coordinates”.
Recall (Exercise 4.4.F) that if k is algebraically closed, then we can interpret the
closed points of Pn as the lines through the origin in (n + 1)-space. The following
exercise states this more generally.

9.3.G. E XERCISE . Suppose S• is a finitely generated graded ring over an al-

gebraically closed field k, generated in degree 1 by x0 , . . . , xn , inducing closed
embeddings Proj S• ,→ Pn and Spec S• ,→ An+1 . Give a bijection between the
closed points of Proj S• and the “lines through the origin” in Spec S• ⊂ An+1 .

9.3.5. A second proof that finite morphisms are closed. This interpretation of Proj S• as
a closed subscheme of projective space (when it is generated in degree 1) yields the
following second proof of the fact (shown in Exercise 8.3.L) that finite morphisms
are closed. Suppose π : X → Y is a finite morphism. The question is local on the
target, so it suffices to consider the affine case Y = Spec B. It suffices to show that
π(X) is closed. Then by Exercise 8.3.I, X is a projective B-scheme, and hence by the
Fundamental Theorem of Elimination Theory 8.4.10, its image is closed.

9.3.6. Important classical construction: The Veronese embedding.

Suppose S• = k[x, y], so Proj S• = P1k . Then S2• = k[x2 , xy, y2 ] ⊂ k[x, y] (see
§7.4.4 on the Veronese subring). We identify this subring as follows.
∼ k[u, v, w]/(uw−v2 ),
9.3.H. E XERCISE . Let u = x2 , v = xy, w = y2 . Show that S2• =
2 2
by mapping u, v, w to x , xy, y , respectively.
We have a graded ring generated by three elements in degree 1. Thus we think
of it as sitting “in” P2 , via the construction of §9.3.F. This can be interpreted as “P1
as a conic in P2 ”.

9.3.7. Thus if k is algebraically closed of characteristic not 2, using the fact that
we can diagonalize quadratics (Exercise 5.4.J), the conics in P2 , up to change of
coordinates, come in only a few flavors: sums of three (and no fewer) squares
(e.g., our conic of the previous exercise), sums of two (and no fewer) squares (e.g.,
y2 − x2 = 0, the union of two lines), a single (nonzero) square (e.g., x2 = 0, which
looks set-theoretically like a line, and is nonreduced), and 0 (perhaps not a conic at
all). Thus we have proved: any plane conic (over an algebraically closed field of
262 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

characteristic not 2) that can be written as the sum of three nonzero squares (and
no fewer) is isomorphic to P1 . (See Exercise 7.5.G for a closely related fact.)
We now soup up this example.

9.3.I. E XERCISE . We continue to take S• = k[x, y]. Show that Proj Sd• is given by
the equations that
 
y0 y1 · · · yd−1
y1 y2 · · · yd
is rank 1 (i.e., that all the 2 × 2 minors vanish). This is called the degree d rational
normal curve “in” Pd . You did the twisted cubic case d = 3 in Exercises 9.3.A and
9.3.E.

9.3.8. Definition. More generally, if S• = k[x0 , . . . , xn ], then νd : Proj Sd• →

PN−1 (where N is the dimension of the vector space of homogeneous degree d
polynomials in x0 , . . . , xn ) is called the d-uple embedding or the (d-uple) Veronese
embedding. (Combining Exercise 7.4.E with Exercise 9.3.F shows that νd is a closed
embedding.)

9.3.J. C OMBINATORIAL EXERCISE ( CF. R EMARK 4.5.3). Show that N = n+d

d .

9.3.K. U NIMPORTANT EXERCISE . Find six linearly independent quadratic equations

vanishing on the Veronese surface Proj S2• where S• = k[x0 , x1 , x2 ], which sits
naturally in P5 . (You needn’t show that these equations generate all the equations
cutting out the Veronese surface, although this is in fact true.) Possible hint: use the
identity
 
x0 x0 x0 x1 x0 x2
det  x1 x0 x1 x1 x1 x2  = 0.
x2 x0 x2 x1 x2 x2

9.3.9. Rulings on the quadric surface. We return to rulings on the quadric surface,
which first appeared in the optional (starred) section §4.4.12.

9.3.L. U SEFUL GEOMETRIC EXERCISE : THE RULINGS ON THE QUADRIC SURFACE

wz = xy. This exercise is about the lines on the quadric surface X given by
wz − xy = 0 in P3k (where the projective coordinates on P3k are ordered w, x, y, z).
This construction arises all over the place in nature.
(a) Suppose a0 and b0 are given elements of k, not both zero. Make sense of the
statement: as [c, d] varies in P1 , [a0 c, b0 c, a0 d, b0 d] is a line in the quadric surface.
(This describes “a family of lines parametrized by P1 ”, although we can’t yet make
this precise.) Find another family of lines. These are the two rulings of the smooth
quadric surface.
(b) Show that through every k-valued point of the quadric surface X, there passes
one line from each ruling.
(c) Show there are no other lines. (There are many ways of proceeding. At risk
of predisposing you to one approach, here is a germ of an idea. Suppose L is a
line on the quadric surface, and [1, x, y, z] and [1, x ′ , y ′ , z ′ ] are distinct points on
it. Because they are both on the quadric, z = xy and z ′ = x ′ y ′ . Because all of L
is on the quadric, (1 + t)(z + tz ′ ) − (x + tx ′ )(y + ty ′ ) = 0 for all t. After some
algebraic manipulation, this translates into (x − x ′ )(y − y ′ ) = 0. How can this be
 © 2024 Ravi Vakil. Published by Princeton University Press. 263

made watertight? Another possible approach uses Bézout’s Theorem, in the form
of Exercise 9.3.D.)

F IGURE 9.2. One of the two rulings on the quadric surface S :=

V(wz − xy) ⊂ P3 . One ruling contains the line V(w, x), and the
other contains the line V(w, y). (The quadric surface sketched is
actually a2 + b2 = c2 + 1. Can you see why this an affine chart for
S? What are a, b, and c in terms of w, x, y, and z? Hint: Exer-
cise 5.4.J.)

Hence by Exercise 5.4.J, if we are working over an algebraically closed field

of characteristic not 2, we have shown that all rank 4 quadric surfaces have two
rulings of lines. (In Example 10.6.2, we will recognize this quadric as P1 × P1 .)

9.3.10. Side Remark: Rulings on quadric hypersurfaces. The existence of these two
rulings is the first chapter of a number of important and beautiful stories. The
second chapter is often the following. If k is an algebraically closed field, then
a maximal rank (Exercise 5.4.J) quadric hypersurface X of dimension m contains
no linear spaces of dimension greater than m/2. (We will see in Exercise 13.3.D
that the maximal rank quadric hypersurfaces are the “smooth” quadrics.) If m =
2a + 1, then X contains an irreducible a+2


2 -dimensional family of a-planes. If
a+1


m = 2a, then X contains two irreducible 2 -dimensional families of a-planes,
and furthermore two a-planes Λ and Λ ′ are in the same family if and only if
dim(Λ ∩ Λ ′ ) ≡ a (mod 2). These families of linear spaces are also called rulings.
(For more information, see [GH1, §6.1, p. 735, Prop.]. You already know enough
to think through the examples of m = 0, 1, and 2. The case m = 2, rulings on
a quadric surface, turn up in §4.4.12, Figure 9.2, Exercise 9.3.L, §10.6.2, §12.2.19,
§13.2.12, Exercise 15.5.Q, and Exercise 19.10.M. The case m = 3 is discussed in
Exercise 16.4.K.)

9.3.11. Affine and projective cones (Figure 9.3).

If S• is a finitely generated graded ring, then the affine cone of Proj S• is Spec S• .
Caution: this terminology is not ideal, as this construction depends on S• , not just
on Proj S• . As motivation, consider the graded ring S• = C[x, y, z]/(z2 − x2 − y2 ).
Figure 9.4 is a sketch of Spec S• . (Here we draw the “real picture” of z2 = x2 + y2
in R3 .) It is a cone in the traditional sense; the origin (0, 0, 0) is the “cone point”.
This gives a useful way of picturing Proj (even over arbitrary rings, not just
C). Intuitively, you could imagine that if you discarded the origin, you would
264 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

F IGURE 9.3. The affine and projective cones of Proj S•

get something that would project onto Proj S• . The following exercise makes that
precise.

9.3.M. E XERCISE . If Proj S• is a projective scheme over a field k, describe a natural

morphism Spec S• \ V(S+ ) → Proj S• . (Can you see why V(S+ ) is a single point,
and should reasonably be called the origin?)
This readily generalizes to the following exercise, which again motivates the
terminology “irrelevant”.

9.3.N. E ASY E XERCISE . If S• is a finitely generated graded ring over a base ring A,
describe a natural morphism Spec S• \ V(S+ ) → Proj S• .
In fact, it can be made precise that Proj S• is the quotient (by the multiplicative
group of scalars) of the affine cone minus the origin.

9.3.12. Definition. The projective cone of Proj S• is Proj S• [T ], where T is a

new variable of degree 1. For example, the cone corresponding to the conic
Proj k[x, y, z]/(z2 − x2 − y2 ) is Proj k[x, y, z, T ]/(z2 − x2 − y2 ). The projective cone
is sometimes called the projective completion of Spec S• . (Note: this depends on
S• , not just on Proj S• .)

9.3.O. L ESS IMPORTANT EXERCISE ( CF. §4.5.1). Show that the “projective cone”
Proj S• [T ] of Proj S• has a closed subscheme isomorphic to Proj S• (informally, cor-
responding to T = 0), whose complement (the distinguished open set D(T )) is
isomorphic to the affine cone Spec S• .
This construction can be usefully pictured as the affine cone union some points
“at infinity”, and the points at infinity form the Proj. (The reader may wish to start
with Figure 9.4, and try to visualize the conic curve “at infinity”, and then compare
this visualization to Figure 4.9.)
We have thus completely described the algebraic analog of the classical picture
of §4.5.1.

9.3.13. Definition. If we put a nonstandard weighting on the variables of

k[x0 , . . . , xn ] — say we give xi degree di — then Proj k[x0 , . . . , xn ] is called weighted
projective space P(d0 , d1 , . . . , dn ).

9.3.P. E XERCISE . (a) Show that P(m, n) is isomorphic to P1 . (b) Show that
∼
P(1, 1, 2) = Proj k[u, v, w, z]/(uw − v2 ). Hint: do this by looking at the even-
graded parts of k[x0 , x1 , x2 ], cf. Exercise 7.4.D. (This is a projective cone over a
 © 2024 Ravi Vakil. Published by Princeton University Press. 265

conic curve. Over a field of characteristic not 2, it is isomorphic to the traditional

cone x2 + y2 = z2 in P3 , see Figure 9.4.)

F IGURE 9.4. The cone x2 + y2 = z2 in A3 .

9.3.Q. E XERCISE ( COMBINING A NUMBER OF THE ABOVE CONCEPTS ). Show that

P(1, 1, n) is isomorphic to the projective cone over the degree n rational normal
curve.

9.4 The (closed sub)scheme-theoretic image

We now define a series of notions that are all of the form “the smallest closed
subscheme such that something or other is true”. One example will be the notion
of scheme-theoretic closure of a locally closed embedding, which will allow us to
interpret locally closed embeddings in three equivalent ways (open subscheme
intersect closed subscheme; open subscheme of closed subscheme; and closed
subscheme of open subscheme — cf. Exercise 9.2.C).

9.4.1. Scheme-theoretic image.

We start with the notion of scheme-theoretic image. Set-theoretic images are
badly behaved in general (§8.4.1), and even with reasonable hypotheses such as
those in Chevalley’s Theorem 8.4.2, things can be confusing. For example, there is
no reasonable way to impose a scheme structure on the image of A2k → A2k given by
(x, y) 7→ (x, xy). It will be useful (e.g., Exercise 9.4.C) to define a notion of a closed
subscheme of the target that “best approximates” the image. This will incorporate
the notion that the image of something with nonreduced structure (“fuzz”) can
also have nonreduced structure. As usual, we will need to impose reasonable
hypotheses to make this notion behave well (see Theorem 9.4.4 and Corollary 9.4.5).

9.4.2. Definition. Suppose i : Z ,→ Y is a closed subscheme, giving an exact

sequence 0 → IZ/Y → OY → i∗ OZ → 0. We say that the image of π : X → Y lies
in Z if the composition IZ/Y → OY → π∗ OX is zero. Informally, locally, functions
vanishing on Z pull back to the zero function on X. If the image of π lies in some
subschemes Zj (as j runs over some index set), it clearly lies in their intersection
(cf. Exercise 9.1.I(a) on intersections of closed subschemes). We then define the
scheme-theoretic image of π, a closed subscheme of Y, as the “smallest closed
subscheme containing the image”, i.e., the intersection of all closed subschemes
containing the image. In particular (and in our first examples), if Y is affine, the
266 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

scheme-theoretic image is cut out by functions on Y that are 0 when pulled back to
X.
Other reasonable names for this concept are the closed subscheme-theoretic image,
or image closed subscheme. They have the advantage that they remind the reader that
the scheme-theoretic image is a closed subscheme, but we won’t use these phrases.
Example 1. Consider π : Spec k[ϵ]/(ϵ2 ) → Spec k[x] = A1k given by x 7→ ϵ. Then
the scheme-theoretic image of π is given by Spec k[x]/(x2 ) (the polynomials pulling
back to 0 are precisely multiples of x2 ). Thus the image of the fuzzy point still has
some fuzz.
Example 2. Consider π : Spec k[ϵ]/(ϵ2 ) → Spec k[x] = A1k given by x 7→ 0. Then
the scheme-theoretic image is given by Spec k[x]/(x): the image is reduced. In this
picture, the fuzz is “collapsed” by π.
Example 3. Consider π : Spec k[t, t−1 ] = A1 − {0} → A1 = Spec k[u] given by
u 7→ t. Any function g(u) which pulls back to 0 as a function of t must be the
zero-function. Thus the scheme-theoretic image is everything. The set-theoretic
image, on the other hand, is the distinguished open set A1 \ {0}. Thus in not-too-
pathological cases, the underlying set of the scheme-theoretic image is not the
set-theoretic image. But the situation isn’t terrible: the underlying set of the scheme-
theoretic image must be closed, and indeed it is the closure of the set-theoretic
image. We might imagine that in reasonable cases this will be true, and in even nicer
cases, the underlying set of the scheme-theoretic image will be the set-theoretic
image. We will later see that this is indeed the case (§9.4.6).
But sadly pathologies can sometimes happen in, well, pathological situations.
`
Example 4 (Figure 9.5). Let X = Spec k[ϵn ]/((ϵn )n ) (a scheme which appeared
in the hint to Exercise 5.2.F) and Y = Spec k[x], and define X → Y by x → ϵn on
the nth component of X. If a function g(x) on Y pulls back to 0 on X, then its
Taylor expansion is 0 to order n (by examining the pullback to the nth component
of X) for all n, so g(x) must be 0. (This argument will be vastly generalized in
Exercise 13.9.A(b).) Thus the scheme-theoretic image is V(0) on Y, i.e., Y itself,
while the set-theoretic image is easily seen to be just the origin (the closed point 0).
(This morphism implicitly arises in Caution/Example 9.4.11.)

F IGURE 9.5. Yuck (Example 4 of Definition 9.4.2)

9.4.3. Criteria for computing scheme-theoretic images affine-locally. Example 4 clearly is

weird though, and we can show that in “reasonable circumstances” such pathology
doesn’t occur.
 © 2024 Ravi Vakil. Published by Princeton University Press. 267

In the special case where the target Y is affine (pay attention to how this
argument works in Examples 1 through 3), say Y = Spec B, almost by definition the
scheme-theoretic image of π : X → Y is cut out by the ideal I ⊂ B of functions on Y
which pull back to zero on X.
It would be great to use this to compute the scheme-theoretic image affine by
affine (affine-locally). On the affine open set Spec B ⊂ Y, define the ideal I(B) ⊂ B of
functions which pull back to 0 on X. Formally, I(B) := ker(B → Γ (Spec B, π∗ (OX )).
If for each such B, and each g ∈ B, the map ϕ : I(B)g → I(Bg ) is an isomorphism,
then we will have defined the scheme-theoretic image as a closed subscheme (see
Exercise 9.1.F). Injectivity of ϕ is straightforward: if r ∈ B and r/gn pulls back to
0 on π−1 (Spec Bg ), then as g is nonvanishing at all points of Spec Bg = D(g), we
have that π♯ (r) is 0 on π−1 (Spec Bg ) as well.
So the question is the surjectivity of ϕ, which translates to: given a function
r/gn on D(g) that pulls back to zero on π−1 (D(g)), is it true that for some m,
rgm = 0 when pulled back to π−1 (Spec B)?
(i) We first show that the answer is yes if π−1 (Spec B) is reduced. In this case
we can take advantage of Exercise 5.2.A, that a function on a reduced scheme is
zero if it has value zero at every point. We may then take m = 1 (as π♯ (r) van-
ishes on π−1 (D(g)) and g vanishes on V(g), so π♯ (rg) vanishes on π−1 (Spec B) =
π−1 (D(g)) ∪ π−1 (V(g)).)
(ii) The answer is also yes if π−1 (Spec B) is affine, say Spec A: if r ′ = π♯ (r) and
g ′ = π♯ (g) in A, then if r ′ = 0 on D(g ′ ), then there is an m such that r ′ (g ′ )m = 0
(as the statement r ′ = 0 in D(g ′ ) means precisely this fact — the functions on D(g ′ )
are Ag ′ ).
(ii)’ More generally, the answer is yes if π−1 (Spec B) is quasicompact: cover
−1
π (Spec B) with finitely many affine open sets. For each one there will be some
mi so that rgmi = 0 when pulled back to this open set. Then let m = max(mi ).
(We see again that quasicompactness is our friend!)
In conclusion, we have proved the following (subtle) theorem.

9.4.4. Theorem. — Suppose π : X → Y is a morphism of schemes. If X is reduced or π

is quasicompact, then the scheme-theoretic image of π may be computed affine-locally: on
Spec A ⊂ Y, it is cut out by the functions (elements of A) that pull back to the function 0
(on π−1 (Spec A)).

9.4.5. Corollary. — Under the hypotheses of Theorem 9.4.4, the closure of the set-theoretic
image of π is the underlying set of the scheme-theoretic image.
(Example 4 above shows how we need these hypotheses.)

9.4.6. In particular, if the set-theoretic image is closed (e.g., if π is finite or projective),

the set-theoretic image is the underlying set of the scheme-theoretic image, as
promised in Example 3 above.

9.4.7. Proof of Corollary 9.4.5. The set-theoretic image is in the underlying set
of the scheme-theoretic image. (Check this!) The underlying set of the scheme-
theoretic image is closed, so the closure of the set-theoretic image is contained
in the underlying set of the scheme-theoretic image. On the other hand, if U is
the complement of the closure of the set-theoretic image, π−1 (U) = ∅. Under
268 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

these hypotheses, the scheme theoretic image can be computed locally, so the
scheme-theoretic image is the empty set on U. □
We conclude with a few stray remarks.

9.4.A. E ASY E XERCISE . If X is reduced, show that the scheme-theoretic image of

π : X → Y is also reduced.
More generally, you might expect there to be no unnecessary nonreduced
structure on the image not forced by nonreduced structure on the source. The
following makes this precise in the locally Noetherian case (when we can talk about
associated points).

9.4.B. ⋆ U NIMPORTANT EXERCISE . If π : X → Y is a quasicompact morphism of

locally Noetherian schemes, show that the associated points of the image sub-
scheme
` are a subset of the image of the associated points of X. (The example of
a∈C Spec C[t]/(t − a) → Spec C[t] shows what can go wrong if you give up qua-
sicompactness — note that reducedness of the source doesn’t help.) Hint: reduce to
the case where X and Y are affine. (Can you develop your geometric intuition so
that this becomes plausible to you?)

9.4.8. Scheme-theoretic closure of a locally closed subscheme.

We define the scheme-theoretic closure of a locally closed embedding π : X ,→
Y as the scheme-theoretic image of π. (A shorter phrase for this is schematic closure,
although this more elegant nomenclature has not caught on.)

9.4.C. E XERCISE . If a locally closed embedding V ,→ X is quasicompact (e.g., if V

is Noetherian, Exercise 8.3.B(a)), or if V is reduced, show that (iii) implies (i) and (ii)
in Exercise 9.2.C. Thus in this fortunate situation, a locally closed embedding can be
thought of in three different ways, whichever is convenient. (Hint: Corollary 9.4.5.)

9.4.D. U NIMPORTANT EXERCISE , USEFUL FOR INTUITION . If π : X ,→ Y is a locally

closed embedding into a locally Noetherian scheme (so X is also locally Noetherian),
then the associated points of the scheme-theoretic closure are (naturally in bijection
with) the associated points of X. (Hint: Exercise 9.4.B.) Informally, the only nonre-
duced structure on the scheme-theoretic closure is that “forced by” nonreduced
structure on X.

9.4.9. The (reduced) subscheme structure on a closed subset.

Suppose Xset is a closed subset of a scheme Y. Then we can define a canonical
scheme structure X on Xset that is reduced. We could describe it as being cut out by
those functions whose values are zero at all the points of Xset . On the affine open
set Spec B of Y, if the set Xset corresponds to the radical ideal I = I(Xset ∩ Spec B)
(recall the I(·) function from §3.7), the scheme X corresponds to Spec B/I. You can
quickly check that this behaves well with respect to any distinguished inclusion
Spec Bf ,→ Spec B. We could also consider this construction as an example of a
scheme-theoretic image in the following crazy way: let W be the scheme that is a
disjoint union of all the points of Xset , where the point corresponding to p in Xset
is Spec of the residue field of OY,p . Let ρ : W → Y be the “canonical” map sending
“p to p”, and giving an isomorphism on residue fields. Then the scheme structure
 © 2024 Ravi Vakil. Published by Princeton University Press. 269

on X is the scheme-theoretic image of ρ. A third definition: it is the smallest closed

subscheme whose underlying set contains Xset .

9.4.E. E XERCISE . Show that all three definitions are the same.
This construction is called the (induced) reduced subscheme structure on the
closed subset Xset . (Vague exercise: Make a definition of the reduced subscheme
structure precise and rigorous to your satisfaction.)

9.4.F. E XERCISE . Show that the underlying set of the induced reduced subscheme
X ,→ Y is indeed the closed subset Xset . Show that X is reduced.

9.4.10. Reduced version of a scheme.

In the main interesting case where Xset is all of Y, we obtain a reduced closed
subscheme Y red → Y, called the reduction of Y. On the affine open subset Spec B ,→ Y,
Y red ,→ Y corresponds to (the quotient by) the nilradical N(B) of B. The reduction
of a scheme is the “reduced version” of the scheme, and informally corresponds to
“shearing off the fuzz”.
An alternative equivalent definition of the reduction: on the affine open subset
Spec B ,→ Y, the reduction of Y corresponds to the nilradical N(B) ⊂ B. We easily
check that for any f ∈ B, N(B)f = N(Bf ). By Exercise 9.1.F this defines a closed
subscheme, which we call the reduction.

` U ⊂ Y, Γ (U,nOY red )
9.4.11. ⋆ Caution/Example. It is not true that for every open subset
is Γ (U, OY ) modulo its nilpotents. For example, on Y = Spec k[x]/(x ), the
function x is not nilpotent, but is 0 on Y red , as it is “locally nilpotent”. This may
remind you of Example 4 after Definition 9.4.2.

9.5 Slicing by effective Cartier divisors, regular sequences and

regular embeddings
We now introduce regular embeddings, an important class of locally closed
embeddings. Locally closed embeddings of regular schemes in regular schemes
are one important example of regular embeddings (Exercise 13.2.M(b)). Effective
Cartier divisors, basically the codimension 1 case, will turn out to be repeatedly
useful as well, with repeated “slicing by effective Cartier divisors” providing the
inductive step in many arguments. (When we say slicing in many future contexts,
we will mean slicing by effective Cartier divisors.)

9.5.1. Locally principal closed subschemes, and effective Cartier divisors. A

closed subscheme is locally principal if on each open set in a small enough open
cover it is cut out by a single equation (i.e., by a principal ideal, hence the ter-
minology). More specifically, a locally principal closed subscheme is a closed
subscheme π : X ,→ Y for which there is an open cover {Ui } of Y for which for each
i, π−1 (Ui ) ,→ Ui the closed subscheme V(si ) of Ui for some si ∈ Γ (Ui , OX ). (This
open cover can clearly chosen to be by affine open subsets.)
For example, hypersurfaces in Pn k (Definition 9.3.2) are locally principal: each
homogeneous polynomial in n + 1 variables defines a locally principal closed
subscheme of Pn A . (Warnings: unlike “local principality”, “principality” is not an
270 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

affine-local condition, see §19.11.10! Also, the example of a projective hypersurface,

§9.3.1, shows that a locally principal closed subscheme need not be cut out by
a globally-defined function. Finally, we unfortunately use the phrase “locally
principal” in a different way in §15.4.7.)
If the ideal sheaf is locally generated near every point by a function that is
not a zerodivisor, we call the closed subscheme an effective Cartier divisor. More
precisely: if π : X → Y is a closed embedding, and there is a cover Y by affine open
subsets Spec Ai ⊂ Y, and there exist non-zerodivisors ti ∈ Ai with V(ti ) = X|Spec Ai
(scheme-theoretically — i.e., the ideal sheaf of X over Spec Ai is generated by ti ),
then we say that X is an effective Cartier divisor on Y. (We will not explain the
origin of the phrase, as it is not relevant for this point of view.) We will often use the
evocative phrase slicing by an effective Cartier divisor when we mean “restrict to
an effective Cartier divisor”.
The notion of an effective Cartier divisor is central. The notion of a locally
principal closed subscheme is much less so.

9.5.A. E XERCISE . Suppose t ∈ A is a non-zerodivisor. Show that t is a non-

zerodivisor in Ap for each prime p.

9.5.2. Caution. If D is an effective Cartier divisor on an affine scheme Spec A, it is

not necessarily true that D = V(t) for some t ∈ A (see Exercise 15.5.I — §19.11.10
gives a different flavor of example). In other words, the condition of a closed
subscheme being an effective Cartier divisor can be verified on an affine cover, but
cannot be checked on an arbitrary affine cover — it is not an affine-local condition
in this obvious a way.

9.5.B. E XERCISE . Suppose X is a locally Noetherian scheme, If t ∈ Γ (X, OX ) is

a function on X, show that t (or more precisely the closed subscheme V(t)) is an
effective Cartier divisor if and only if it doesn’t vanish on any associated point of X.
Show that a locally principal closed subscheme of X is an effective Cartier divisor if
and only if it doesn’t contain any associated points of X.

9.5.C. E XERCISE . Suppose V(t) = V(t ′ ) ,→ Spec A is an effective Cartier divisor,

with t and t ′ non-zerodivisors in A. Show that t is an invertible function times t ′ .
The idea of effective Cartier divisors leads us to the notion of regular sequences.
(We will close the loop in Exercise 9.5.H, where we will interpret effective Cartier
divisors on any reasonable scheme as regular embeddings of codimension 1.)

9.5.3. Regular sequences.

The notion of “regular sequence”, like so much else, is due to Serre, [Se2]. The
definition of regular sequence is the algebraic version of the following geometric
idea: locally, we take an effective Cartier divisor (a non-zerodivisor); then an
effective Cartier divisor on that; then an effective Cartier divisor on that; and so
on, a finite number of times. A little care is necessary; for example, we might want
this to be independent of the order of the equations imposed, and this is true only
when we say this in the right way.
We make the definition of regular sequence for a ring A, and more generally
for an A-module M.
 © 2024 Ravi Vakil. Published by Princeton University Press. 271

9.5.4. Definition. If M is an A-module, a sequence x1 , . . . , xr ∈ A is called an

M-regular sequence (or a regular sequence for M) if the following two conditions
are satisfied.

(i) For each i, xi is not a zerodivisor for M/(x1 , . . . , xi−1 )M. (The case i = 1
should be interpreted as: “x1 is not a zerodivisor of M.” Recall that
zerodivisors were defined in Exercise 1.2.C.)
(ii) We have a proper inclusion (x1 , . . . , xr )M ⊊ M.

In the case most relevant to us, when M = A, this should be seen as a reasonable
approximation of a “complete intersection”, and indeed we will use this as the
definition (§9.5.7).
We say r is the length of the regular sequence x1 , . . . , xr ∈ A. An A-regular
sequence is just called a regular sequence.

9.5.D. E XERCISE . If M is an A-module, show that an M-regular sequence con-

tinues to satisfy condition (i) of the Definition 9.5.4 of regular sequence upon any
localization. (Once you know what flatness is, you will see that your argument
shows that condition (i) is preserved by any flat base change.)

9.5.E. E XERCISE . If x, y is an M-regular sequence, show that xN , y is an M-

regular sequence. Hint: the difficult part is showing that y is not a zerodivisor of
M/(xN M). Show this by induction on N. If y is a zerodivisor of M/(xN M), then
ym ≡ 0 (mod xN ) for some m ∈ M \ xN M. Hence ym = xN k for some k ∈ M.
Use the fact that x, y is a regular sequence to show that m is a multiple of x. (Your
argument will easily extend to show more generally that if x1 , . . . , xn is a regular
sequence, and a1 , . . . , an ∈ Z+ , then xa an
1 , . . . , xn is a regular sequence .)
1

9.5.5. The regularity of a sequence depends on its order. We now give an example
([E, Example 17.3]) showing that the order of a regular sequence matters. Suppose
A = k[x, y, z]/(x − 1)z, so X = Spec A is the union of the z = 0 plane and the x = 1
plane — X is reduced and has two components (see Figure 9.6). You can readily
verify that x is a non-zerodivisor of A (x = 0 misses one component of X, and
doesn’t vanish entirely on the other), and that the corresponding effective Cartier
divisor X ′ = Spec k[x, y, z]/(x, z) (on X) is integral. Then (x − 1)y gives an effective
Cartier divisor on X ′ (it doesn’t vanish entirely on X ′ ), so x, (x − 1)y is a regular
sequence for A. However, (x − 1)y is not a non-zerodivisor of A, as it does vanish
entirely on one of the two components. Thus (x − 1)y, x is not a regular sequence.
The reason that reordering the regular sequence x, (x − 1)y ruins regularity is clear:
there is a locus on which (x − 1)y isn’t effective Cartier, but it disappears if we
enforce x = 0 first. The problem is one of “nonlocality” — “near” x = y = z = 0
there is no problem. This may motivate the fact that in the (Noetherian) local
situation, this problem disappears. We now make this precise.

9.5.6. Theorem. — Suppose (A, m) is a Noetherian local ring, and M is a finitely

generated A-module. Then any M-regular sequence (x1 , . . . , xr ) in m remains a regular
sequence upon any reordering.
(In [Di], Dieudonné gives a half-page example showing that Noetherian hy-
potheses are necessary in Theorem 9.5.6. Also, in Exercise 24.6.C, we give a different,
somehow simpler proof in the case where A contains a field.)
272 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

F IGURE 9.6. Order matters in a regular sequence (in the “non-

local” situation), for unsurprising reasons

Before proving Theorem 9.5.6 (in Exercise 9.5.F), we prove the first nontrivial
case, when r = 2. This discussion is secretly a baby case of the Koszul complex
(which is briefly mentioned in §24.4.1).
Suppose x, y ∈ m, and x, y is an M-regular sequence. Translation: x ∈ m is a
non-zerodivisor on M, and y ∈ m is a non-zerodivisor on M/xM.
Consider the double complex
×(−x)
(9.5.6.1) M /M
O O
×y ×y

M
×x
/M

where the bottom left is considered to be in position (0, 0). (The only reason for the
minus sign in the top row is solely our arbitrary preference for anti-commuting
rather than commuting squares in §1.6.1, but it really doesn’t matter.)
We compute the cohomology of the total complex using a (simple) spectral
sequence, beginning with the rightward orientation. (The gratuitous use of spectral
sequences here, as in many of our other applications, is overkill; we do this partially
in order to get practice with the machine.) On the first page, we have
×x
ker(M −−→ M) M/xM
O O
×y ×y

×x
ker(M −−→ M) M/xM
×x
The entries ker(M −−→ M) in the first column are 0, as x is a non-zerodivisor on M.
Taking homology in the vertical direction to obtain the second page, we find

(9.5.6.2) 0 M/(x, y)M

0 0
using the fact that y is a non-zerodivisor on M/xM. The sequence clearly converges
here. Thus the original double complex (9.5.6.1) only has nonzero cohomology in
degree 2, where it is M/(x, y)M.
 © 2024 Ravi Vakil. Published by Princeton University Press. 273

Now we run the spectral sequence on (9.5.6.1) using the upward orientation.
The first page of the sequence is:

×(−x)
M/yM / M/yM

×y
ker(M −−→ M)
×x
/ ker(M −
×y
−→ M)

The sequence must converge to (9.5.6.2) after the next step. From the top row, we
see that multiplication by x must be injective on M/yM, so x is a non-zerodivisor
on M/yM. From the bottom row, multiplication by x gives an isomorphism
×y
of ker(M −−→ M) with itself. As x ∈ m, by version 2 of Nakayama’s Lemma
×y
(Lemma 8.2.9), this implies that ker(M −−→ M) = 0, so y is a non-zerodivisor on
M. Thus we have shown that y, x is a regular sequence on M — the r = 2 case of
Theorem 9.5.6.

9.5.F. E ASY E XERCISE . Prove Theorem 9.5.6. (Hint: show it first in the case of
a reordering where only two adjacent xi are swapped, using the r = 2 case just
discussed.) Where are the Noetherian hypotheses used?

9.5.7. Regular embeddings.

Suppose π : X ,→ Y is a locally closed embedding. We say that π is a regular
embedding (of codimension r) at a point p ∈ X if in the local ring OY,p , the ideal
of X is generated by a regular sequence (of length r). We say that π is a regular
embedding (of codimension r) if it is a regular embedding (of codimension r) at
all p ∈ X. (Another reasonable name for a regular embedding might be “local
complete intersection”. Unfortunately, “local complete intersection morphism”, or
“lci morphism”, is already used for a related notion, see [Stacks, tag 068E].)
Our terminology uses the word “codimension”, which we have not defined as
a word on its own. The reason for using this word will become clearer once you
meet Krull’s Principal Ideal Theorem 12.3.3 and Krull’s Height Theorem 12.3.10.

9.5.G. E XERCISE ( THE CONDITION OF A LOCALLY CLOSED EMBEDDING BEING

A REGULAR EMBEDDING IS OPEN ). Show that if a locally closed embedding
π : X ,→ Y of locally Noetherian schemes is a regular embedding at p, then it is
a regular embedding in some open neighborhood of p in X. Hint: reduce to the
case where π is a closed embedding, and then where Y (hence X) is affine — say
Y = Spec B, X = Spec B/I, and p = [p] — and there are f1 , . . . , fr such that in OY,p ,
the images of the fi are a regular sequence generating Ip . We wish to show that
(f1 , . . . , fr ) = I “in an open neighborhood of p”. Prove the following fact in algebra:
if I and J are ideals of a Noetherian ring A, and p ⊂ A is a prime ideal such that
Ip = Jp , show that there exists a ∈ A \ p such that Ia = Ja in Aa . To do this, show
that it suffices to consider the special case I ⊂ J, by considering I ∩ J and J instead
of I and J. To show this special case, let K = J/I, a finitely generated module, and
show that if Kp = 0 then Ka = 0 for some a ∈ A \ p.
Hence if X is locally of finite type over a field (e.g., a variety), then to check that
a closed embedding π is a regular embedding it suffices to check at closed points of
X (Exercise 5.3.F).
274 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Exercise 13.1.F(b) will show that not all closed embeddings are regular embed-
dings.

9.5.H. E XERCISE . Show that a closed embedding X ,→ Y of locally Noetherian

schemes is a regular embedding of codimension 1 if and only if X is an effective
Cartier divisor on Y. Unimportant remark: the Noetherian hypotheses can be
replaced by requiring OY to be coherent, and essentially the same argument applies.
It is interesting to note that “effective Cartier divisor” implies “regular embedding
of codimension 1” always, but that the converse argument requires Noetherian(-
like) assumptions. (See [MO129242] for a counterexample to the converse.)

9.5.8. Definition. A codimension r complete intersection in a scheme Y is a closed

subscheme X that can be written as the scheme-theoretic intersection of r effective
Cartier divisors D1 , . . . , Dr , such that at every point p ∈ X, the equations corre-
sponding to D1 , . . . , Dr form a regular sequence. The phrase complete intersection
means “codimension r complete intersection for some r”.
 CHAPTER 10

Fibered products of schemes, and base change

Fibered products have an unexpectedly central place in algebraic geometry.

Experience will gradually teach you why they play such an outsized role. Until
you have acquired that experience, you should pay closer attention to them than
you think they might deserve.

10.1 They exist

Before we get to products, we note that coproducts exist in the category of
schemes: just as with the category of sets (Exercise 1.2.T), coproduct is disjoint
union.
We will now construct the fibered product in the category of schemes.

10.1.1. Theorem: Fibered products exist. — Suppose α : X → Z and β : Y → Z are

morphisms of schemes. Then the fibered product
α′ /Y
X ×Z Y
β′ β
 
X
α /Z
exists in the category of schemes.
Note: if A is a ring, people often sloppily write ×A for ×Spec A . If B is an A-
algebra, and X is an A-scheme, people often write XB or X ×A B for X ×Spec A Spec B.

10.1.2. Warning: products of schemes aren’t products of sets. Before showing existence,
here is a warning: the product of schemes isn’t a product of sets (and more generally
for fibered products). We have made a big deal about schemes being sets, endowed
with a topology, upon which we have a structure sheaf. So you might think that we
will construct the product in this order: take the product set, endow it with the
product topology, and then figure out what the structure sheaf must be. But we
won’t, because scheme products behave oddly on the level of sets. You may have
checked (Exercise 7.6.E(a)) that the product of two affine lines over your favorite
algebraically closed field k is the affine plane: A1k ×k Ak1 =∼ A2 . But the underlying
k
set of the latter is not the underlying set of the former —- we get additional points,
corresponding to curves in A2 that are not lines parallel to the axes!

This material is published by Princeton University Press as The Rising Sea: Foundations of Algebraic
Geometry by Ravi Vakil. This version is free to view and download for personal use only. Not for
redistribution, resale, or use in derivative works. © 2024 Ravi Vakil.

275
276 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

10.1.3. On the other hand, W-valued points (where W is a scheme, Definition 7.3.10)
do behave well under (fibered) products (as mentioned in §7.3.11). This is just the
universal property definition of fibered product: a W-valued point of a scheme X
is defined as an element of Hom(W, X), and the fibered product is defined (via
Yoneda, see §1.2.1) by

(10.1.3.1) Hom(W, X ×Z Y) = Hom(W, X) ×Hom(W,Z) Hom(W, Y).

This is one justification for making the definition of scheme-valued point. For
this reason, those classical people preferring to think only about varieties over an
algebraically closed field k (or more generally, finite type schemes over k), and
preferring to understand them through their closed points — or equivalently, the
k-valued points, by the Nullstellensatz (Exercise 5.3.F) — needn’t worry: the closed
points of the product of two finite type k-schemes over k are (naturally identified
with) the product of the closed points of the factors. This will follow from the fact
that the product is also finite type over k, which we verify in Exercise 10.2.E. This
is one of the reasons that varieties over algebraically closed fields can be easier
to work with. But over a nonalgebraically closed field, things become even more
interesting; Example 10.2.3 is a first glimpse.

10.1.4. Fancy aside. You may feel that (i) “products of topological spaces are
products on the underlying sets” is natural, while (ii) “products of schemes are
not necessarily products on the underlying sets” is weird. But really (i) is the
lucky consequence of the fact that the underlying set of a topological space can
be interpreted as set of p-valued points, where p is a point, so it is best seen as a
consequence of paragraph 10.1.3, which is the “more correct” — i.e., more general
— fact.

10.1.5. Warning on Noetherianness. The fibered product of Noetherian schemes

need not be Noetherian. You will later be able to verify that Exercise 10.2.F gives an
example, i.e., that A := Q ⊗Q Q is not Noetherian, as follows. By Exercise 12.1.H(a),
dim A = 0. A Noetherian dimension 0 scheme has a finite number of points
(Exercise 12.1.C). But by Exercise 10.2.F, Spec A has an infinite number of points.
On the other hand, the fibered product of finite type k-schemes over finite type
k-schemes is a finite type k-scheme (Exercise 10.2.E), so this pathology does not
arise for varieties.

10.1.6. Philosophy behind the proof of Theorem 10.1.1. The proof of Theo-
rem 10.1.1 can be confusing. The following comments may help a little.
We already basically know existence of fibered products in two cases: the case
where X, Y, and Z are affine (stated explicitly below), and the case where β : Y → Z
is an open embedding (Exercise 8.1.D).

10.1.A. E XERCISE ( PROMISED IN R EMARK 7.3.5). Use Exercise 7.3.G (HomSch (W, Spec A) =
HomRings (A, Γ (W, OW ))) to show that given ring maps C → A and C → B,
∼ Spec A ×Spec C Spec B.
Spec(A ⊗C B) =

(Interpret tensor product as the “fibered coproduct” in the category of rings.) Hence
the fibered product of affine schemes exists (in the category of schemes). (This
generalizes the fact that the product of affine lines exist, Exercise 7.6.E(a).)
 © 2024 Ravi Vakil. Published by Princeton University Press. 277

The main theme of the proof of Theorem 10.1.1 is that because schemes are
built by gluing affine schemes along open subsets, these two special cases will be
all that we need. The argument will repeatedly use the same ideas — roughly, that
schemes glue (Exercise 4.4.A), and that morphisms of schemes glue (Exercise 7.3.B).
This is a sign that something more structural is going on; §10.1.7 describes this for
experts.
Proof of Theorem 10.1.1. The key idea is this: we cut everything up into affine
open sets, do fibered products there, and show that everything glues nicely. The
conceptually difficult part of the proof comes from the gluing, and the realization
that we have to check almost nothing. We divide the proof up into a number of
bite-sized pieces.
Step 1: fibered products of affine with “almost-affine” over affine. We begin by
combining the affine case with the open embedding case as follows. Suppose X
 ι
and Z are affine, and β : Y → Z factors as Y  / Y′ / Z where ι is an open
′
embedding and Y is affine. Then X ×Z Y exists. This is because if the two small
squares of

W _ / Y
_
open open
 
W′ / Y′

 
X /Z

are Cartesian diagrams, then the “outside rectangle” is also a Cartesian diagram.
(This was Exercise 1.2.Q, although you should be able to see this on the spot.) It will
be important to remember (from Important Exercise 8.1.D) that “open embeddings”
are “preserved by fibered product”: the fact that Y → Y ′ is an open embedding
implies that W → W ′ is an open embedding.
Key Step 2: fibered product of affine with arbitrary over affine exists. We now come
to the key part of the argument: if X and Z are affine, and Y is arbitrary. This is
confusing when you first see it, so we first deal with a special case, when Y is the
union of two affine open sets Y1 ∪ Y2 . Let Y12 = Y1 ∩ Y2 .
Now for i = 1 and 2, X ×Z Yi exists by the affine case, Exercise 10.1.A. Call
this Wi . Also, X ×Z Y12 exists by Step 1 (call it W12 ), and comes with canonical
open embeddings into W1 and W2 (by construction of fibered products with open
embeddings, see the last sentence of Step 1). Thus we can glue W1 to W2 along
W12 ; call this resulting scheme W.
We check that the result is the fibered product by verifying that it satisfies the
universal property. Suppose we have maps α ′′ : V → X, β ′′ : V → Y that compose
(with α and β respectively) to the same map V → Z. We need to construct a unique
278 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

map γ : V → W, so that α ′ ◦ γ = β ′′ and β ′ ◦ γ = α ′′ .

(10.1.6.1) V
β ′′
∃!γ?

W /' Y
α ′′ α′
β′ β
 
X /Z
α

For i = 1 and 2, define Vi := (β ′′ )−1 (Yi ). Define V12 := (β ′′ )−1 (Y12 ) = V1 ∩ V2 .

Then there is a unique map Vi → Wi such that the composed maps Vi → X
and Vi → Yi are as desired (by the universal property of the fibered product
X ×Z Yi = Wi ), hence a unique map γi : Vi → W. Similarly, there is a unique map
γ12 : V12 → W such that the composed maps V12 → X and V12 → Y are as desired.
But the restriction of γi to V12 is one such map, so it must be γ12 . Thus the maps
γ1 and γ2 agree on V12 , and glue together to a unique map γ : V → W. We have
shown existence and uniqueness of the desired γ.
We have thus shown that if Y is the union of two affine open sets, and X and Z
are affine, then X ×Z Y exists.
We now tackle the general case. (You may prefer to first think through the
case where “two” is replaced by “three”.) We now cover Y with open sets Yi , as
i runs over some index set (not necessarily finite!). As before, we define Wi and
Wij . We can glue these together to produce a scheme W along with open sets we
identify with Wi (Exercise 4.4.A — you should check the triple intersection cocycle
condition).
As in the two-affine case, we show that W is the fibered product by showing that
it satisfies the universal property. Suppose we have maps α ′′ : V → X, β ′′ : V → Y
that compose to the same map V → Z. We construct a unique map γ : V → W, so
that α ′ ◦ γ = β ′′ and β ′ ◦ γ = α ′′ . Define Vi = (β ′′ )−1 (Yi ) and Vij := (β ′′ )−1 (Yij ) =
Vi ∩ Vj . Then there is a unique map Vi → Wi such that the composed maps Vi → X
and Vi → Yi are as desired, hence a unique map γi : Vi → W. Similarly, there is
a unique map γij : Vij → W such that the composed maps Vij → X and Vij → Y
are as desired. But the restriction of γi to Vij is one such map, so it must be γij .
Thus the maps γi and γj agree on Vij . Thus the γi glue together to a unique map
γ : V → W. We have shown existence and uniqueness of the desired γ, completing
this step.
Step 3: Z affine, X and Y arbitrary. We next show that if Z is affine, and X and Y
are arbitrary schemes, then X ×Z Y exists. We just follow Step 2, with the roles of X
and Y reversed, using the fact that by the previous step, we can assume that the
fibered product of an affine scheme with an arbitrary scheme over an affine scheme
exists.
Step 4: Z “almost-affine”, X and Y arbitrary. This is akin to Step 1. Let Z ,→ Z ′
be an open embedding into an affine scheme. Then X ×Z Y satisfies the universal
property of X ×Z ′ Y. (If you don’t see this directly, one argument is to use Easy
Exercise 1.2.X and that open embeddings are monomorphisms, Exercise 8.1.E.)
Step 5: the general case. We employ the same trick yet again. Suppose α : X → Z,
β : Y → Z are two morphisms of schemes. Cover Z with affine open subschemes Zi ,
and let Xi = α−1 (Zi ) and Yi = β−1 (Zi ). Define Zij := Zi ∩ Zj , Xij := α−1 (Zij ), and
 © 2024 Ravi Vakil. Published by Princeton University Press. 279

Yij := β−1 (Zij ). Then Wi := Xi ×Zi Yi exists for all i (Step 3), and Wij := Xij ×Zij Yij
exists for all i, j (Step 4), and for each i and j, Wij comes with a canonical open
embedding into both Wi and Wj (see the last sentence in Step 1). As Wi satisfies
the universal property of X ×Z Yi (Do you see why?), we may canonically identify
Wi (which we know to exist by Step 3) with X ×Z Yi . Similarly, we identify Wij
with X ×Z Yij .
We then proceed exactly as in Step 2: the Wi ’s can be glued together along the
Wij (the cocycle condition can be readily checked to be satisfied), and W can be
checked to satisfy the universal property of X ×Z Y (again, exactly as in Step 2). □

10.1.7. ⋆⋆ Describing the existence of fibered products using the fancy language
of representable functors. The proof above can be described more cleanly in the
language of representable functors (§7.6). This will be enlightening only after you
have absorbed the above argument and meditated on it for a long time. It may be
most useful to shed light on representable functors, rather than on the existence of
the fibered product.
Until the end of §10.1 only, by functor, we mean contravariant functor from the
category Sch of schemes to the category of Sets. For each scheme X, we have a functor
hX , taking a scheme Y to the set Mor(Y, X) (§1.1.20). Recall (§1.2.11, §7.6) that
a functor is representable if it is naturally isomorphic to some hX . If a functor is
representable, then the representing scheme is unique up to unique isomorphism
(Exercise 7.6.C). This can be usefully extended as follows:

10.1.B. E XERCISE (Yoneda’s Lemma).

(a) If X and Y are schemes, describe a bijection between morphisms of schemes
X → Y and natural transformations of functors hX → hY . Hence show that
the category of schemes is a full subcategory (defined in §1.1.15) of the “functor
category” (all contravariant functors Sch → Sets).
(b) If X is a scheme and F is a contravariant functor Sch → Sets, give a bijection
between F(X) and the natural transformation of functors hX → F. (You are free to
check that your bijection is “functorial” in both F and X.)
Hint: this question has nothing to do with schemes; your argument will work
in any category. This is the contravariant version of Exercise 1.2.Z(c).
One of Grothendieck’s insights is that we should try to treat such functors as
“geometric spaces”, without worrying about representability. Many notions carry
over to this more general setting without change, and some notions are easier. For
example, fibered products of functors always exist: h ×h ′′ h ′ may be defined by
(h ×h ′′ h ′ )(W) = h(W) ×h ′′ (W) h ′ (W),
where the fibered product on the right is a fibered product of sets, which always
exists. (This isn’t quite enough to define a functor; we have only described where
objects go. You should work out where morphisms go too.) We didn’t use anything
about schemes; this works with Sch replaced by any category.
Then “X ×Z Y exists” translates to “hX ×hZ hY is representable”.

10.1.8. Representable functors are Zariski sheaves. Because “morphisms to schemes

glue” (Exercise 7.3.B), we have a necessary condition for a functor to be repre-
sentable. We know that if {Ui } is an open cover of Y, a morphism Y → X is
determined by its restrictions Ui → X, and given morphisms Ui → X that agree on
280 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

the overlap Ui ∩ Uj → X, we can glue them together to get a morphism Y → X. In

the language of equalizer exact sequences (§2.2.7),

Q Q
· / hX (Y) / hX (Ui ) // hX (Ui ∩ Uj )

is exact. Thus morphisms to X (i.e., the functor hX ) form a sheaf on every scheme
Y. If this holds, we say that the functor is a Zariski sheaf. (You can impress your
friends by telling them that this is a sheaf on the big Zariski site.) We can repeat this
discussion with Sch replaced by the category SchS of schemes over a given base
scheme S. We have proved (or observed) that in order for a functor to be representable,
it is necessary for it to be a Zariski sheaf.
The fibered product passes this test:

10.1.C. E XERCISE . If X, Y → Z are schemes, show that hX ×hZ hY is a Zariski sheaf.

(Do not use the fact that hX ×hZ hY is representable! The point of this section is to
recover representability from a more sophisticated perspective.)
We can make some other definitions that extend notions from schemes to
functors. We say that a map (i.e., natural transformation) of functors h ′ → h
expresses h ′ as a subfunctor of h if for all schemes X, the induced map of sets
h ′ (X) → h(X) is an inclusion. We say that a subfunctor h ′ ,→ h is an open
subfunctor if for all schemes X and maps hX → h (i.e., all elements of h(X)), the
fibered product hX ×h h ′ is representable, by the scheme U say, and hU → hX
corresponds to an open embedding of schemes U → X. The following fibered
diagram may help.

open
hU / hX

 
h′ /h

10.1.D. E XERCISE . Show that a map of representable functors hW → hZ is an open

subfunctor if and only if W → Z is an open embedding, so this indeed extends the
notion of open embedding to (contravariant) functors (Sch → Sets).

10.1.E. E XERCISE ( THE GEOMETRIC NATURE OF THE NOTION OF “ OPEN SUBFUNC -

TOR ”).
(a) Show that an open subfunctor of an open subfunctor is also an open subfunctor.
(b) Suppose h ′ → h and h ′′ → h are two open subfunctors of h. Define the intersec-
tion of these two open subfunctors, which should also be an open subfunctor of h.
(c) Suppose U and V are two open subschemes of a scheme X, so hU → hX and
hV → hX are open subfunctors. Show that the intersection of these two open
subfunctors is, as you would expect, hU∩V .

10.1.F. E XERCISE . Suppose α : X → Z and β : Y → Z are morphisms of schemes, and

U ⊂ X, V ⊂ Y, W ⊂ Z are open embeddings, where U and V map to W. Interpret
hU ×hW hV as an open subfunctor of hX ×hZ hY . (Hint: given a map hT → hX×Z Y ,
what open subset of T should correspond to U ×W V?)
 © 2024 Ravi Vakil. Published by Princeton University Press. 281

A collection hi of open subfunctors of h is said to cover h if for every map

hX → h from a representable subfunctor, the corresponding open subsets Ui ,→ X
cover X.
Given that functors do not have an obvious underlying set (let alone a topology),
it is rather amazing that we are talking about when one is an “open subset” of
another, or when some functors “cover” another!

10.1.G. E XERCISE . Suppose {Zi }i is an affine cover of Z, {Xij }j is an affine cover

of the preimage of Zi in X, and {Yik }k is an affine cover of the preimage of Zi in
Y. Show that {hXij ×hZi hYik }ijk is an open cover of the functor hX ×hZ hY . (Hint:
consider a map hT → hX ×hZ hY , and extend your solution to Exercise 10.1.F.)
We now come to a key point: a Zariski sheaf that is “locally representable”
must be representable:

10.1.H. K EY E XERCISE . If a functor h is a Zariski sheaf that has an open cover by

representable functors (“is covered by schemes”), then h is representable. (Hint:
use Exercise 4.4.A to glue together the schemes representing the open subfunctors.)
This immediately leads to the existence of fibered products as follows. Ex-
ercise 10.1.C shows that hX ×hZ hY is a Zariski sheaf. But hXij ×hZi hYik is rep-
resentable for each i, j, k (fibered products of affines over an affine exist, Exer-
cise 10.1.A), and these functors are an open cover of hX ×hZ hY by Exercise 10.1.G,
so by Key Exercise 10.1.H we are done.

10.2 Computing fibered products in practice

Before giving some examples, we first see how to compute fibered products
in practice. There are four types of morphisms (1)–(4) that it is particularly easy
to take fibered products with, and all morphisms can be built from these atomic
components. More precisely, (1) will imply that we can compute fibered products
locally on the source and target. Thus to understand fibered products in general, it
suffices to understand them on the level of affine sets, i.e., to be able to compute
A ⊗B C given ring maps B → A and B → C. Any map B → A (and similarly
B → C) may be expressed as B → B[t1 , . . . ]/I, so if we know how to base change by
“adding variables” (2) and “taking quotients” (3), we can “compute” any fibered
product (at least in theory). The fourth type of morphism (4), corresponding to
localization, is useful to understand explicitly as well.
(1) Base change by open embeddings.
We have already done this (Exercise 8.1.D), and we used it in the proof that
fibered products of schemes exist.
(2) Adding an extra variable.

10.2.A. E ASY ALGEBRA EXERCISE . Show that A ⊗B B[t] = ∼ A[t], so the following is
a Cartesian diagram. (Your argument might naturally extend to allow the addition
of infinitely many variables, but we won’t need this generality.) Hint: show that
282 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

A[t] satisfies an appropriate universal property.

Spec A[t] / Spec B[t]

 
Spec A / Spec B

10.2.1. Definition (affine space over an arbitrary scheme). If X is any scheme, we define
An n n
X as X ×Z Spec Z[x1 , . . . , xn ]. Clearly ASpec A is canonically the same as AA as
defined in Example 8 of §3.2.4.
(3) Base change by closed embeddings.

10.2.B. E XERCISE . Suppose ϕ : B → A is a ring morphism, and I ⊂ B is an

ideal. Let Ie := ⟨ϕ(i)⟩i∈I ⊂ A be the extension of I to A. Describe a natural
∼
isomorphism A/Ie ←→ A ⊗B (B/I). (Hint: consider I → B → B/I → 0, and use the
right-exactness of ⊗B A, Exercise 1.2.H.)

10.2.2. As an immediate consequence: the fibered product with a closed subscheme

is a closed subscheme of the fibered product in the obvious way. We say that
“closed embeddings are preserved by base change”.

10.2.C. E XERCISE .
(a) Interpret the intersection of two closed embeddings into X (cf. Exercise 9.1.I) as
their fibered product over X.
(b) Show that “locally closed embeddings” are preserved by base change.
(c) Define the intersection of n locally closed embeddings Xi ,→ Z (1 ≤ i ≤ n) by
the fibered product of the Xi over Z (mapping to Z). Show that the intersection of
(a finite number of) locally closed embeddings is also a locally closed embedding.

10.2.D. E XERCISE . Suppose I ⊂ A[x1 , . . . , xm ] and J ⊂ A[y1 , . . . , yn ] are ideals.

Prove the isomorphism

∼
A[x1 , . . . , xm ]/I ⊗A A[y1 , . . . , yn ]/J o / A[x1 , . . . , xm , y1 , . . . , yn ]/(I, J).

Hint: Exercises 10.2.A and 10.2.B.

As an application, we can compute tensor products of finitely generated k-
algebras over k. For example, we have

∼ k[x1 , x2 , y1 , y2 ]/(x2 − x2 , y3 + y3 ).
k[x1 , x2 ]/(x21 − x2 ) ⊗k k[y1 , y2 ]/(y31 + y32 ) = 1 1 2

10.2.E. E XERCISE . Suppose X and Y are locally of finite type A-schemes. Show
that X ×A Y is also locally of finite type over A. Prove the same thing with “locally”
removed from both the hypothesis and conclusion.
 © 2024 Ravi Vakil. Published by Princeton University Press. 283

10.2.3. Example. We can use these ideas to compute C ⊗R C:

C ⊗R C = ∼ C ⊗R (R[x]/(x2 + 1))
∼ (C ⊗R R[x])/(x2 + 1)
= by §10.2(3)
∼ C[x]/(x2 + 1)
= by §10.2(2)
∼ C[x]/ ((x − i)(x + i))
=
∼ C[x]/(x − i) × C[x]/(x + i)
= by the Chinese Remainder Theorem
∼ C×C
=
Thus Spec C ×R Spec C = ∼ Spec C ` Spec C. This example is the first example of
many different behaviors. Notice for example that two points somehow correspond
to the Galois group of C over R; for one of them, x (the “i” in one of the copies of
C) equals i (the “i” in the other copy of C), and in the other, x = −i.

10.2.4. ⋆ Remark. Here is a clue that there is something deep going on behind
Example 10.2.3. If L/K is a (finite) Galois extension with Galois group G, then
L ⊗K L is isomorphic to LG (the product of |G| copies of L). This turns out to be
a restatement of the classical form of linear independence of characters! In the
language of schemes, Spec L ×K Spec L is a union of a number of copies of Spec L
that naturally form a torsor over the Galois group G; but we will not define torsor
here.

10.2.F. ⋆ H ARD BUT FASCINATING EXERCISE FOR THOSE FAMILIAR WITH Gal(Q/Q).
Show that the points of Spec Q ⊗Q Q are in natural bijection with Gal(Q/Q), and
the Zariski topology on the former agrees with the profinite topology on the latter.
(Some hints: first do the case of finite Galois extensions. Relate the topology on
Spec of a direct limit of rings to the inverse limit of Specs. Can you see which point
corresponds to the identity of the Galois group?)
At this point, we can compute any A ⊗B C (where A and C are B-algebras): any
map of rings ϕ : B → A can be interpreted by adding variables (perhaps infinitely
many) to B, and then imposing relations. But in practice (4) is useful, as we will see
in examples.
(4) Base change of affine schemes by localization.

10.2.G. E XERCISE . Suppose ϕ : B → A is a ring morphism, and S ⊂ B is a

multiplicative subset of B, which implies that ϕ(S) is a multiplicative subset of A.
∼
Describe a natural isomorphism ϕ(S)−1 A ←→ A ⊗B (S−1 B).
Informal translation: “the fibered product with a localization is the localization
of the fibered product in the obvious way”. We say that “localizations are preserved
by base change”. This is handy if the localization is of the form B ,→ Bf (correspond-
ing to taking distinguished open sets) or (if B is an integral domain) B ,→ K(B)
(corresponding to taking the generic point), and various things in between.

10.2.5. Examples. These four facts let you calculate lots of things in practice, and
we will use them freely.

10.2.H. E XERCISE : THE THREE IMPORTANT TYPES OF MONOMORPHISMS OF SCHEMES .

Show that the following are monomorphisms (Definition 1.2.10): open embeddings,
284 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

closed embeddings, and localization of affine schemes. As monomorphisms are

preserved by composition (Exercise 1.2.V) compositions of the above are also
monomorphisms — for example, locally closed embeddings, or maps from “Spec
of stalks at points of X” to X. (Caution: if p is a point of a scheme X, the natural
morphism Spec OX,p → X, cf. Exercise 7.3.M, is a monomorphism but is not in
general an open embedding.)

10.2.I. E XERCISE . Recall that An ∼ n

A = AZ ×Spec Z Spec A (§10.2.1). Prove similarly
n ∼ n
that PA = PZ ×Spec Z Spec A. Thus affine space and projective space are pulled back
from their “universal manifestation” over the final object Spec Z.

10.2.6. Extending the base field. One special case of base change is called extending
the base field: if X is a k-scheme, and ℓ is a field extension (often ℓ is the algebraic
closure of k), then X ×Spec k Spec ℓ (sometimes informally written X ×k ℓ or Xℓ )
is an ℓ-scheme. Often properties of X can be checked by verifying them instead
on Xℓ . This is the subject of descent — certain properties “descend” from Xℓ to
X. We have already seen that the property of being the Spec of a normal integral
domain descends in this way (Exercise 5.4.M). Exercises 10.2.J and 10.2.K give other
examples of properties which descend: the property of two morphisms being equal,
and the property of a(n affine) morphism being a closed embedding, both descend
in this way. Those interested in schemes over non-algebraically closed fields will
use this repeatedly, to reduce results to the algebraically closed case.

10.2.J. E XERCISE . Suppose π : X → Y and ρ : X → Y are morphisms of k-

schemes, ℓ/k is a field extension, and πℓ : X ×Spec k Spec ℓ → Y ×Spec k Spec ℓ and
ρℓ : X ×Spec k Spec ℓ → Y ×Spec k Spec ℓ are the induced maps of ℓ-schemes. (Be
sure you understand what this means!) Show that if πℓ = ρℓ then π = ρ. (Hint:
show that π and ρ are the same on the level of sets. To do this, you may use that
X ×Spec k Spec ℓ → X is surjective, which we will soon prove in Exercise 10.4.G. Then
reduce to the case where X and Y are affine.)

10.2.K. E ASY E XERCISE . Suppose π : X → Y is an affine morphism over k, and ℓ/k is

a field extension. Show that π is a closed embedding if and only if π ×k ℓ : X ×k ℓ →
Y ×k ℓ is. (The affine hypothesis is not necessary for this result, but it makes the
proof easier, and this is the situation in which we will most need it.)

10.2.L. E XERCISE ( SEEMINGLY PATHOLOGICAL BEHAVIOR IN NONPATHOLOGICAL

CIRCUMSTANCES ). Suppose k is a field, and A = k(x) ⊗k k(y). Show that A is a
(nonzero) localization of k[x, y], hence an integral domain. Thus (0) is the unique
minimal prime ideal of A. Show that the remaining prime ideals of A correspond
to ideals (f(x, y)) ⊂ k[x, y], where f(x, y) is an irreducible polynomial in k[x, y]
containing both the variables x and y. (Hint: (2) and (4) above.)
This example will come up again in Remarks 12.2.17 and 24.5.9. The idea in the
solution to Exercise 10.2.L also yields the following.

10.2.M. U NIMPORTANT BUT FUN EXERCISE . Show that Spec Q(t) ⊗Q C has closed
points in natural correspondence with the transcendental complex numbers. This
scheme doesn’t come up in nature, but it is certainly neat!
 © 2024 Ravi Vakil. Published by Princeton University Press. 285

10.3 Interpretations: Pulling back families, and fibers of

morphisms

10.3.1. Pulling back families.

Before making any definitions, we give a motivating informal example. Con-
sider the “family of curves”
y2 = x3 + tx
in the xy-plane parametrized by t. Translation: consider

Spec k[x, y, t]/(y2 − x3 − tx) / Spec k[t].

If we “pull back this family” to the uv-plane via uv = t, we get the family
y2 = x3 + uvx.
If instead we set t to 3, we get the curve
y2 = x3 + 3x,
which we interpret as the fiber of the original family above t = 3. You may have
noticed the fiber products underlying in these constructions:

Spec k[x, y, u, v]/(y2 − x3 − uvx) / Spec k[x, y, t]/(y2 − x3 − tx)

 
Spec k[u, v]
uv←[t / Spec k[t]

and
Spec k[x, y]/(y2 − x3 − 3x) / Spec k[x, y, t]/(y2 − x3 − tx)

 
Spec k
3←[t / Spec k[t]
We now formalize this.
Suppose Y → Z is a morphism. We interpret this as a “family of schemes
parametrized by a base scheme (or just plain base) Z.” Then if we have another
morphism ψ : X → Z, we interpret the induced map X ×Z Y → X as the “pulled
back family” (see Figure 10.1).
X ×Z Y /Y

pulled back family family

 
X
ψ
/Z
We sometimes say that X ×Z Y is the scheme-theoretic preimage (or scheme-theoretic
pullback, or scheme-theoretic inverse image, or inverse image scheme) of Y. (Our forth-
coming discussion of fibers may give some motivation for this.) For this reason,
fibered product is often called base change or change of base or pullback. In
addition to the various names for a Cartesian diagram given in §1.2.7, in algebraic
geometry it is often called a base change diagram or a pullback diagram, and
X ×Z Y → X is called the pullback of Y → Z by ψ, and X ×Z Y is called the pullback
of Y by ψ. One often uses the phrase “over X” or “above X” when discussing X×Z Y,
286 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

F IGURE 10.1. A picture of a pulled back family

especially if X is a locally closed subscheme of Z. (Random side remark: scheme-

theoretic preimage always makes sense, while the notion of scheme-theoretic image
is somehow problematic, as discussed in §9.4.1.)

10.3.2. Fibers of morphisms.

(If you did Exercise 8.3.J, that finite morphisms have finite fibers, you will not
find the following discussion surprising.) A special case of pullback is the notion of
a fiber of a morphism. We motivate this with the notion of fiber in the category of
topological spaces.

10.3.A. E XERCISE . Show that if Y → Z is a continuous map of topological spaces,

and X is a point p of Z, then the fiber of Y over p (the set-theoretic fiber, with the
induced topology) is naturally identified with X ×Z Y.
More generally, for any π : X → Z, the fiber of X ×Z Y → X over a point p of X
is naturally identified with the fiber of Y → Z over π(p).

10.3.3. Definition. Motivated by topology, we return to the category of schemes.

Suppose p → Z is the inclusion of a point (not necessarily closed). More precisely, if
p is a point with residue field K, consider the map Spec K → Z sending Spec K to p,
with the natural isomorphism of residue fields. Then if g : Y → Z is any morphism,
the scheme-theoretic preimage of p is called the scheme-theoretic fiber (or fibre)
of g above p, and is denoted g−1 (p). If Z is irreducible, the scheme-theoretic fiber
above the generic point of Z is called the generic fiber (of g). In an affine open
subscheme Spec A containing p, p corresponds to some prime ideal p, and the
morphism Spec K → Z corresponds to the ring map A → Ap /pAp . This is the
composition of localization and closed embedding, and thus can be computed by
the tricks above. (Note that p → Z is a monomorphism, by Exercise 10.2.H.)
 © 2024 Ravi Vakil. Published by Princeton University Press. 287

10.3.B. E XERCISE . Show that the underlying topological space of the scheme-
theoretic fiber of X → Y above a point p is naturally identified with the topological
fiber of X → Y above p.

10.3.C. E XERCISE ( ANALOG OF E XERCISE 10.3.A). Suppose that π : Y → Z and

τ : X → Z are morphisms, and p ∈ X is a point. Show that the fiber of X ×Z Y → X
over p is (isomorphic to) the base change to p of the fiber of π : Y → Z over τ(p).

F IGURE 10.2. The map C → C given by y 7→ y2 (cf. Figure 3.6)

10.3.4. Example (enlightening in several ways). Consider the projection of the parabola
y2 = x to the x-axis over Q, corresponding to the map of rings Q[x] → Q[y], with
x 7→ y2 . If Q alarms you, replace it with your favorite field and see what happens.
(You should look at Figure 10.2, which is a flipped version of the parabola of
Figure 3.6, and figure out how to edit it to reflect what we glean here.) Writing Q[y]
as Q[x, y]/(y2 − x) helps us interpret the morphism conveniently.
(i) Then the preimage of 1 is two points:
Spec Q[x, y]/(y2 − x) ⊗Q[x] Q[x]/(x − 1) = ∼ Spec Q[x, y]/(y2 − x, x − 1)
∼ Spec Q[y]/(y2 − 1)
=
a
∼ Spec Q[y]/(y − 1)
= Spec Q[y]/(y + 1).
(ii) The preimage of 0 is one nonreduced point:
∼ Spec Q[y]/(y2 ).
Spec Q[x, y]/(y2 − x, x) =
(iii) The preimage of −1 is one reduced point, but of “size 2 over the base field”.
∼ Spec Q[y]/(y2 + 1)
Spec Q[x, y]/(y2 − x, x + 1) =
∼ Spec Q[i]
=
= Spec Q(i).
(iv) The preimage of the generic point is again one reduced point, but of “size 2
over the residue field”, as we verify now.
Spec Q[x, y]/(y2 − x) ⊗Q[x] Q(x) = ∼ Spec Q[y] ⊗Q[y2 ] Q(y2 )
288 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

i.e., (informally) the Spec of the ring of polynomials in y divided by polynomials in

y2 . A little thought shows you that in this ring you may invert any polynomial in y,
as if f(y) is any polynomial in y, then
1 f(−y)
= ,
f(y) f(y)f(−y)
and the latter denominator is a polynomial in y2 . Thus
∼ Q(y)
Q[x, y]/(y2 − x) ⊗Q[x] Q(x) =
which is a degree 2 field extension of Q(x) (note that Q(x) = Q(y2 )).
(You might want to work through the preimages of other points of A1Q , such as
[(x − 4)] and [(x2 − 2)].)
3

Notice the following interesting fact: in each of the four cases, the number of
preimages can be interpreted as 2, where you count to two in several ways: you can
count points (as in the case of the preimage of 1); you can get nonreduced behavior
(as in the case of the preimage of 0); or you can have a field extension of degree 2
(as in the case of the preimage of −1 or the generic point). In each case, the fiber
is an affine scheme whose dimension as a vector space over the residue field of
the point is 2. Number theoretic readers may have seen this behavior before. We
will discuss this example again in §16.3.10. This is going to be symptomatic of a
very important kind of morphism (a finite flat morphism, see Remark 24.3.8 and
§24.3.12).
Try to draw a picture of this morphism if you can, so you can develop a
pictorial shorthand for what is going on. A good first approximation is the parabola
of Figure 10.2, but you will want to somehow depict the peculiarities of (iii) and
(iv).

10.3.5. Remark: Finite morphisms have finite fibers. If you haven’t done Exercise 8.3.J,
that finite morphisms have finite fibers, now would be a good time to do it, as you
will find it more straightforward given what you know now.

10.3.D. E XERCISE ( IMPORTANT FOR THOSE WITH MORE ARITHMETIC BACKGROUND ).

What is the scheme-theoretic fiber of Spec Z[i] → Spec Z over the prime (p)? Your
answer will depend on p, and there are four cases, corresponding to the four cases
of Example 10.3.4. (Can you draw a picture?)

10.3.E. E XERCISE . (This exercise will give you practice in computing a fibered
product over something that is not a field.) Consider the morphism of schemes
X = Spec k[t] → Y = Spec k[u] corresponding to k[u] → k[t], u 7→ t2 , where
char k ̸= 2. Show that X ×Y X has two irreducible components. (What happens if
char k = 2? See Exercise 10.5.A for a clue.)

10.3.6. A first view of a blow-up.

10.3.F. I MPORTANT C ONCRETE E XERCISE . (The discussion here immediately

generalizes to An 2 2 1
A .) Define a closed subscheme Bl(0,0) Ak of Ak ×k Pk as follows (see
Figure 10.3). If the coordinates on Ak are x, y, and the projective coordinates on P1k
2

are u, v, this subscheme is cut out in A2k ×k P1k by the single equation xv = yu. (You
may wish to interpret Bl(0,0) A2k as follows. The P1k parametrizes lines through the
origin. The blow-up corresponds to ordered pairs of (point p, line ℓ) such that (0, 0)
 © 2024 Ravi Vakil. Published by Princeton University Press. 289

and p both lie on ℓ.) Describe the fiber of the morphism Bl(0,0) A2k → P1k over each
closed point of P1k . Show that the morphism Bl(0,0) A2k → A2k is an isomorphism
away from (0, 0) ∈ A2k . Show that the fiber over (0, 0) is an effective Cartier divisor
(§9.5.1, a closed subscheme that is locally cut out by a single equation, which is
not a zerodivisor). It is called the exceptional divisor. We will discuss blow-ups
in Chapter 22. This particular example will come up in the motivating example of
§22.1, and in Exercise 20.2.D.

F IGURE 10.3. A first example of a blow-up

We haven’t yet discussed regularity, but here is a hand-waving argument

suggesting that the Bl(0,0) A2k is “smooth”: the preimage above either standard
open set Ui ⊂ P1 is isomorphic to A2 . Thus “the blow-up is a surgery that takes
the smooth surface A2k , cuts out a point, and glues back in a P1 , in such a way that
the outcome is another smooth surface.”

10.3.7. General fibers, generic fibers, generically finite morphisms.

The phrases “generic fiber” and “general fiber” parallel the phrases “generic
point” and “general point” (Definition 3.6.11). Suppose π : X → Y is a morphism of
schemes. When one says the general fiber (or a general fiber) of π a has a certain
property, this means that there exists a dense open subset U ⊂ Y such that the fibers
above any point in U have that property.
When one says the generic fiber of π : X → Y (Definition 10.3.3), this implicitly
means that Y is irreducible, and the phrase refers to the fiber over the generic
point. General fiber and generic fiber are not the same thing! Clearly if something
holds for the general fiber, then it holds for the generic fiber, but the converse
is not always true. However, in good circumstances, it can be — properties of
the generic fiber extend to an honest open neighborhood. For example, if Y is
irreducible and Noetherian, and π is finite type, then if the generic fiber of π is
empty (resp., nonempty), then the general fiber is empty (resp., nonempty), by
Chevalley’s Theorem 8.4.2.
If π : X → Y is finite type, we say π is generically finite if π is finite after base
change to the generic point of each irreducible component (or equivalently, by
Exercise 8.4.D, if the preimage of the generic point of each irreducible component
of Y is a finite set). (The notion of generic finiteness can be defined in more general
circumstances, see [Stacks, tag 073A].)
290 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

10.3.G. E XERCISE (“ GENERICALLY FINITE ” USUALLY MEANS “ GENERALLY FINITE ”).

Suppose π : X → Y is an affine, finite type, generically finite morphism of locally
Noetherian schemes, and Y is reduced. Show that there is an open neighborhood
of each generic point of Y over which π is actually finite. (The hypotheses can be
weakened considerably, see [Stacks, tag 02NW].) Hint: reduce to the case where Y
is Spec B, where B is an integral domain. Then X is affine, say X = Spec A. Write
A = B[x1 , . . . , xn ]/I. Now A ⊗B K(B) is a finite K(B)-module (finite-dimensional
vector space) by hypothesis, so there are monic polynomials fi (t) ∈ K(B)[t] such
that fi (xi ) = 0 in A ⊗B K(B). Let b be the product of the (finite number of)
denominators appearing in the coefficients in the fi (x). By replacing B by Bb , argue
that you can assume that fi (t) ∈ B[t]. Then fi (xi ) = 0 in A ⊗B K(B), meaning that
fi (xi ) is annihilated by some nonzero element of B. By replacing B by its localization
at the product of these n nonzero elements (“shrinking Spec B further”), argue that
fi (xi ) = 0 in A. Then conclude.

10.3.8. ⋆⋆ Finitely presented families (morphisms) are locally pullbacks of partic-

ularly nice families. If you are macho and are embarrassed by Noetherian rings,
the following exercise can be used to extend results from the Noetherian case to
locally finitely presented situations. Exercise 10.3.I, an extension of Chevalley’s
Theorem 8.4.2, is a good example.

10.3.H. E XERCISE . Suppose π : X → Spec B is a finitely presented morphism. Show

that there exists a base change diagram of the form

X / X′

π π′
 
Spec B
ρ
/ Spec Z[x1 , . . . , xN ]

where N is some integer, and π ′ is finitely presented (= finite type as the target
is Noetherian, see §8.3.13). Thus each finitely presented morphism is locally (on
the base) a pullback of a finite type morphism to a Noetherian scheme. Hence
any result proved for Noetherian schemes and stable under base change is auto-
matically proved for locally finitely presented morphisms to arbitrary schemes.
Hint: think about the case where X is affine first. If X = Spec A, then A =
B[y1 , . . . , yn ]/(f1 , . . . , fr ). Choose one variable xi for each coefficient of fi ∈
B[y1 , . . . , yn ]. What is X ′ in this case? Then consider the case where X is the
union of two affine open sets, that intersect in an affine open set. Then consider
more general cases until you solve the full problem. You will need to use every
part of the definition of finite presentation. (Exercise 25.2.B extends this result.)

10.3.I. E XERCISE (C HEVALLEY ’ S T HEOREM FOR LOCALLY FINITELY PRESENTED

MORPHISMS ).
(a) Suppose that A is a finitely presented B-algebra (B not necessarily Noetherian),
so A = B[x1 , . . . , xn ]/(f1 , . . . , fr ). Show that the image of Spec A → Spec B is a
finite union of locally closed subsets of Spec B. Hint: Exercise 10.3.H (the simpler
affine case).
(b) Show that if π : X → Y is a quasicompact locally finitely presented morphism,
and Y is quasicompact, then π(X) is a finite union of locally closed subsets. (For
hardened experts only: [Gr-EGA, 0III .9.1] gives a definition of local constructibility,
 © 2024 Ravi Vakil. Published by Princeton University Press. 291

and of constructibility in more generality. The general form of Chevalley’s Con-

structibility Theorem [Gr-EGA, IV1 .1.8.4] is that the image of a locally constructible
subset, under a finitely presented map, is also locally constructible.)

10.4 Properties preserved by base change

All “reasonable” properties of morphisms are preserved under base change (cf.
§8.1(ii)). We discuss this, and in §10.5.1 we will explain how to fix those that don’t
fit this pattern.
We have already shown that the notion of “open embedding” is preserved by
base change (Exercise 8.1.D). We did this by explicitly describing what the fibered
product of an open embedding is: if Y ,→ Z is an open embedding, and ψ : X → Z
is any morphism, then we checked that the open subscheme ψ−1 (Y) of X satisfies
the universal property of fibered products.
We have also shown that the notion of “closed embedding” is preserved by
base change (§10.2 (3)). In other words, given a Cartesian diagram

W / Y
_
cl. emb.
 
X /Z

where Y ,→ Z is a closed embedding, W → X is as well.

10.4.A. E ASY E XERCISE . Show that the closed embeddings and locally closed
embeddings are both “reasonable” classes of morphisms in the sense of §8.1.

10.4.B. E XERCISE . Suppose X → Z and Y → Z are both locally closed embeddings.

(a) Show that X ×Z Y → Z is a locally closed embedding.
(b) In this way we define the (scheme-theoretic) intersection of any two locally
closed subschemes of Z. Verify that this is in agreement with the previous definition
of intersections of two open subschemes of Z, and the previous definition of the
intersection of two closed subschemes of Z.

10.4.C. E ASY E XERCISE . Show that locally principal closed subschemes (Defini-
tion 9.5.1) pull back to locally principal closed subschemes.
Similarly, other important properties are preserved by base change.

10.4.D. E XERCISE . Show that the following properties of morphisms are preserved
by base change.
(a) quasicompact
(b) affine morphism
(c) finite
(d) integral
(e) locally of finite type
(f) finite type
⋆ (g) locally of finite presentation
292 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

10.4.E. E XERCISE . Show that the properties of morphisms of the previous exercise
are all “reasonable” in the sense of §8.1.

10.4.F. E XERCISE .
⋆ (a) Show that the notion of “quasifinite morphism” (finite type + finite fibers,
Definition 8.3.11) is preserved by base change. (Warning: the notion of “finite fibers”
is not preserved by base change. Spec Q → Spec Q has finite fibers, but Spec Q ⊗Q
Q → Spec Q has one point for each element of Gal(Q/Q), see Exercise 10.2.F.) Hint:
reduce to the case Spec A → Spec B. Reduce to the case ϕ : Spec A → Spec k. By
Exercise 8.4.D, such ϕ are actually finite, and finiteness is preserved by base change.

(b) Show that the notion of “quasifinite morphism” is “reasonable” in the sense of
§8.1.

10.4.G. E XERCISE . (a) Show that surjectivity is preserved by base change. (Surjec-
tivity has its usual meaning: surjective as a map of sets.) You may end up showing
that for any fields k1 and k2 containing k3 , k1 ⊗k3 k2 is nonzero, and using Zorn’s
Lemma to find a maximal ideal in k1 ⊗k3 k2 . (b) Show that surjective morphisms
form a “reasonable” class (§8.1).

10.4.1. On the other hand, injectivity is not preserved by base change — witness the
bijection Spec C → Spec R, which loses injectivity upon base change by Spec C →
Spec R (see Example 10.2.3). This can be rectified (see §10.5.I).

10.4.H. E XERCISE ( CF. E XERCISE 10.2.E). Suppose X and Y are integral finite type
k-schemes. Show that X ×k Y is an integral finite type k-scheme. (Once we define
“variety”, this will become the important fact that the product of irreducible varieties
over an algebraically closed field is an irreducible variety, cf. Proposition 11.2.11.
The fact that the base field k is algebraically closed is important, see §10.5.1. See
Exercises 10.5.O and 10.5.R for improvements.) Hint: reduce to the case where
X and Y are both affine, say X = Spec A and Y = Spec B with A and B integral
domains. P You might P flip ahead
to Easy Exercise 10.5.N to see how to do this.
Suppose ( ai ⊗ bi ) aj′ ⊗ bj′ = 0 in A ⊗k B with ai , aj′ ∈ A, bi , bj′ ∈ B, where
both {bi } and {bj′ } are linearly independent over k, and a1 and a1′ are nonzero.
Show that D(a1 a1′ ) ⊂ Spec A is nonempty. By the Weak Nullstellensatz 3.2.5, there
is a maximal m ⊂ A in
D(a1 a1′ ) with A/m = k. By reducing modulo m, deduce
P P ′
( a i ⊗ bi ) a j ⊗ bj′ = 0 in B, where the overline indicates residue modulo m.
Show that this contradicts the fact that B is an integral domain.

10.5 ⋆ Properties not preserved by base change, and how to fix them

We saw in the previous section that many useful properties of morphisms

are preserved by base change. Indeed, “preserved by base change” is one of the
properties any “reasonable” class of morphisms should have (§8.1). But some
seemingly reasonable notions are not preserved by base change, and in this section
we discuss how to modify them appropriately so that they are. We do this not for
 © 2024 Ravi Vakil. Published by Princeton University Press. 293

our own amusement, but because the resulting notions come up in nature, and
perhaps in retrospect are the notions we should have started with.
The “universal” patch to this problem is as follows. If P is some property of
schemes, then a morphism of schemes is said to be universally P if it remains P
under any base change. Then the class of universally P morphisms is preserved by
base change. (Do you see why?)
An important example (§11.4) is the notion of universally closed morphisms.
Another example is that of universally injective morphisms, which turns out to
generalize (and “geometrize”) purely inseparable field extensions (§10.5.4).
One problem with “universally P” morphisms is that it is a priori hard to
determine whether a given morphism is universally P — you seem to have to check
every single base change of the morphism. Finding other equivalent criteria is thus
essential.

10.5.1. Geometric fibers.

There are some notions that you should reasonably expect to be preserved
by pullback based on your geometric intuition. Given a family in the topological
category, fibers pull back in reasonable ways. So for example, any pullback of a
family in which all the fibers are irreducible will also have this property; ditto for
connected. Unfortunately, both of these fail in algebraic geometry, as Example 10.2.3
shows:
` / Spec C
Spec C Spec C

 
Spec C / Spec R

The family on the right (the vertical map) has irreducible and connected fibers, and
the one on the left doesn’t. The same example shows that the notion of “integral
fibers” also doesn’t behave well under pullback. And we used it in §10.4.1 to show
that injectivity isn’t preserved by base change.

10.5.A. E XERCISE . Suppose k is a field of characteristic p, so k(u)/k(up ) is a purely

inseparable extension. By considering k(u) ⊗k(up ) k(u), show that the notion of
“reduced fibers” does not necessarily behave well under pullback. (We will soon
see that this happens only in characteristic p, in the presence of inseparability.)
We rectify this problem as follows.

10.5.2. A geometric point of a scheme X is defined to be a morphism Spec k → X

where k is an algebraically closed field. Awkwardly, this is now the third kind
of “point” of a scheme! There are just plain points, which are elements of the
underlying set; there are Z-valued points (Z a scheme), which are maps Z → X,
§7.3.10; and there are geometric points. Geometric points are clearly a flavor of
a scheme-valued point, but they are also an enriched version of a (plain) point:
they are the data of a point with an inclusion of the residue field of the point in an
algebraically closed field.
A geometric fiber of a morphism X → Y is defined to be the fiber over a
geometric point of Y, i.e., the fibered product with the geometric point Spec k → Y.
A morphism has connected (resp., irreducible, integral, reduced) geometric fibers
if all its geometric fibers are connected (resp., irreducible, integral, reduced). One
294 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

usually says that the morphism has geometrically connected (resp., geometrically
irreducible, geometrically integral, geometrically reduced) fibers. A k-scheme X is
geometrically connected (resp., geometrically irreducible, geometrically integral,
geometrically reduced) if the structure morphism X → Spec k has geometrically
connected (resp., irreducible, integral, reduced) fibers. We will soon see that to
check any of these conditions, we need only base change to k.
(Warning: in some sources, in the definition of “geometric point”, “algebraically
closed” is replaced by “separably closed”.)

10.5.B. E XERCISE . Show that the notion of “connected (resp., irreducible, integral,
reduced) geometric fibers” behaves well under base change.

10.5.C. E XERCISE FOR THE ARITHMETICALLY- MINDED . Show that for the mor-
phism Spec C → Spec R, all geometric fibers consist of two reduced points. (Cf.
Example 10.2.3.) Thus Spec C is a geometrically reduced but not geometrically
irreducible R-scheme.

10.5.D. E ASY EXERCISE . Give examples of k-schemes that:

(a) are reduced but not geometrically reduced;
(b) are connected but not geometrically connected;
(c) are integral but not geometrically integral.

10.5.E. E XERCISE ( TO CONVINCE GEOMETERS WHY GEOMETRIC FIBERS ARE MEAN -

INGFUL ). Recall Example 10.3.4, the projection of the parabola y2 = x to the
x-axis, corresponding to the map of rings Q[x] → Q[y], with x 7→ y2 . Show that
the geometric fibers of this map are always two points, except for those geometric
fibers “over 0 = [(x)]”. (Note that Spec C → Spec Q[x] and Spec Q → Spec Q[x],
both corresponding to ring maps with x 7→ 0, are both geometric points “above 0”.)
Checking whether a k-scheme is geometrically connected etc. seems annoying:
you need to check every single algebraically closed field containing k. However, in
each of these four cases, the failure of nice behavior of geometric fibers can already
be detected after a finite extension of fields. For example, Spec Q(i) → Spec Q is
not geometrically connected, and in fact you only need to base change by Spec Q(i)
to see this. We make this precise as follows.
Suppose X is a k-scheme. If K/k is a field extension, define XK = X ×k Spec K.
Consider the following twelve statements of the form “XK is [property] for all fields
[condition]”:
• (CK ), (CK=K ), (Ck ), (Cks ),
• (IK ), (IK=K ), (Ik ), (Iks ),
• (RK ), (RK=K ), (Rk ), (Rkp ).
Here C means “connected”, I means “irreducible”, and R means “reduced”. Also,
ks is the separable closure of k, and kp is the perfect closure of k. If char k = 0, then
k = ks and k = kp , so life is simpler.
Thus (CK ) means “XK is connected for all fields K”, (Ik ) means “Xk is irre-
ducible”, and (Rkp ) means “Xkp is reduced”. Geometrically connected (resp., geo-
metrically irreducible, geometrically reduced) translates to (CK=K ) (resp., (IK=K ),
(RK=K )).
 © 2024 Ravi Vakil. Published by Princeton University Press. 295

Trivially (CK ) implies (CK=K ) implies (Ck ), and (CK ) implies (Cks ), and simi-
larly (with appropriate changes) with “connected” replaced by “irreducible” and
“reduced”. (See Figure 10.4.)

F IGURE 10.4. Connectedness conditions that will later turn out to

be equivalent (solid arrows are trivial, dotted arrows are Exer-
cise 10.5.F)

10.5.F. E XERCISE .
(a) Suppose that E/F is a field extension, and A is an F-algebra. Show that A is a
subalgebra of A ⊗F E. (Hint: think of these as vector spaces over F.)
(b) Show that: (CK=K ) implies (CK ) and (Ck ) implies (Cks ).
(c) Show that: (IK=K ) implies (IK ) and (Ik ) implies (Iks ).
(d) Show that: (RK=K ) implies (RK ) and (Rk ) implies (Rkp ).
Thus for example a k-scheme is geometrically integral if and only if it remains
integral under any field extension. Hence “geometrically integral” means “univer-
sally, integral fibers”.

10.5.3. Harder fact. In fact, (CK ) through (Cks ) are all equivalent, and similarly for
the other two properties (irreducibility and reducedness instead of connectedness,
with kp replacing ks in the case of reducedness). We defer the explanation to the
end of this section, in a double-starred discussion (§10.5.5). On a first reading, you
may want to read only Corollary 10.5.12 on connectedness, Proposition 10.5.15
on irreducibility, Proposition 10.5.22 on reducedness, and Proposition 11.2.11 on
varieties, and then to use them to solve Exercise 10.5.Q. You can later come back
and read the proofs, which include some useful tricks turning questions about
general schemes over a field to questions about finite type schemes.
The following exercise may help even the geometrically-minded reader appre-
ciate the utility of these notions. (There is nothing important about the dimension 2
and the degree 4 in this exercise!)

10.5.G. ⋆ E XERCISE . Recall from Remark 4.5.3 that the quartic curves in P2k are
parametrized by a P14k . (This will be made much more precise in §25.3.5.) Show
that the points of P14
k corresponding to geometrically integral curves form an
open subset. Explain the necessity of the modifier “geometrically” (even if k is
algebraically closed).

10.5.4. Universally injective (= radicial) morphisms.

As remarked in §10.4.1, injectivity is not preserved by base change. A better
notion is that of universally injective morphisms: morphisms that are injections of
sets after any base change. In keeping with the traditional agricultural terminology
296 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

(sheaves, germs, . . . , cf. Remark 2.4.3), these morphisms were named radicial after
radishes. As a first example: all locally closed embeddings are universally injective
(as they are injective, and remain locally closed embeddings upon any base change).
If you wish, you can show more generally that all monomorphisms are universally
injective. (Hint: show that monomorphisms are injective, and that the class of
monomorphisms is preserved by base change.)
Suppose you want to determine whether a given morphism is universally in-
jective. A map of sets is injective if and only if each fiber contains at most one point.
Now “the fiber of a base change is the base change of the fiber” (Exercise 10.3.C).
Also, the underlying set of the scheme-theoretic fiber is the fiber of the map on
underlying sets. Hence the question of universal injectivity turns into one about
field theory. We make this precise in Exercise 10.5.I, after first making the field
theory question precise in Exercise 10.5.H. En route, we will see why universal in-
jectivity is the algebro-geometric generalization of the notion of purely inseparable
extensions of fields.

10.5.H. E XERCISE . Suppose E/F is an extension of fields, inducing ϕ : Spec E →

Spec F.
(a) If E contains an element x transcendental over F, show that ϕ is not universally
injective. Hint: Exercise 10.2.L (on k(x) ⊗k k(y)) and the Cartesian diagrams

? / Spec F(x) ⊗F F(y) / Spec F(y)

  
Spec E / Spec F(x) / Spec F.

Now Spec E → Spec F(x) (a map from a point to a point) is surjective, and surjectiv-
ity is preserved by base change (Exercise 10.4.G(a)).
(b) If E \ F contains an element x algebraic over F (with minimal polynomial
mx (t) ∈ F[t], say), and separable over F, show that ϕ is not universally injective.
Hint: the Cartesian diagrams

? / Spec F[t]/(mx (t)) ⊗F F / Spec F

  
Spec E / Spec F[t]/(mx (t)) / Spec F.

(c) Suppose E contains no elements of the above forms, i.e., E/F is a purely insepa-
rable extension (all elements of E are algebraic over F, and have some pN th power
in F). If E ′ /F is any other field extension, show that Spec E ⊗F E ′ contains precisely
one point.

10.5.I. E XERCISE . Suppose π : X → Y is a morphism of schemes. Show that the

following are equivalent.
(i) The morphism π is universally injective.
(ii) The morphism π is injective, and for each p ∈ X, the field extension
κ(p)/κ(π(p)) is purely inseparable.
 © 2024 Ravi Vakil. Published by Princeton University Press. 297

You may already see that the class of universally injective morphisms is also
preserved by composition, and local on the target, and hence forms a “reasonable”
class (§8.1). In any case, we will see this in a different way in §11.2.16.

10.5.5. ⋆⋆ Proof of Harder fact 10.5.3.

We will use one fact that we will not prove until much later. Recall that a map
of topological spaces is an open map if the image of every open set is an open set.

10.5.6. Fact (to be proved in Theorem 24.5.11). — Suppose X is a k-scheme. Then

X → Spec k is universally open, i.e., remains open after any base change.
We could prove Fact 10.5.6 now without too much trouble, but it will follow
much more easily with the magic of flatness.

10.5.7. Lemma. — Suppose the field extension E/F is purely inseparable. Suppose X is
any F-scheme. Then ϕ : XE → X is a homeomorphism.

Proof. The morphism ϕ is a bijection, so we may identify the points of X and XE .

(Reason: for any point p ∈ X, the scheme-theoretic fiber ϕ−1 (p) is a single point, by
the definition of pure inseparability.) The morphism ϕ is continuous (so open sets
in X are also open in XE ), and by Fact 10.5.6, ϕ is open (so open sets in XE are also
open in X). □

10.5.8. Connectedness.
Recall that a connected component of a topological space X is a maximal
connected subset of X (§3.6.12).

10.5.J. E XERCISE ( PROMISED IN R EMARK 3.6.13). Show that every point is con-
tained in a connected component, and that connected components are closed. (Hint:
see the hint for Exercise 3.6.O.)

10.5.K. E ASY TOPOLOGICAL EXERCISE . Suppose ϕ : X → Y is open, and has

nonempty connected fibers. Show that ϕ induces a bijection of connected compo-
nents.

10.5.9. Lemma. — Suppose X is geometrically connected over k. Then for any scheme
Y/k, X ×k Y → Y induces a bijection of connected components.

Proof. Combine Fact 10.5.6 and Exercise 10.5.K. □

We come next to a repeatedly useful algebra-geometry link, promised in Re-
mark 3.6.3: idempotent functions on X on the algebraic side correspond to a decom-
position of X into two open and closed sets on the geometric side.

10.5.L. E XERCISE ( THE IDEMPOTENT- CONNECTEDNESS PACKAGE ). Show that a

scheme X is disconnected if and only if there exists a function e ∈ Γ (X, OX ) that is
an idempotent (e2 = e) distinct from 0 and 1. (Hint: if X is the disjoint union of two
open sets X0 and X1 , let e be the function that is 0 on X0 and 1 on X1 . Conversely,
given such an idempotent, define X0 = V(e) and X1 = V(1 − e).)

10.5.10. Proposition. — Suppose k is separably closed, and A is a k-algebra with Spec A

connected. Then Spec A is geometrically connected over k.
298 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Proof. We wish to show that Spec A ⊗k K is connected for any field extension K/k.
It suffices to assume that K is algebraically closed (as Spec A ⊗k K → Spec A ⊗k K
is surjective). By choosing an embedding k ,→ K and considering the diagram

Spec A ⊗k K / Spec A ⊗k k homeo. / Spec A

by Lem. 10.5.7

  
Spec K / Spec k / Spec k

it suffices to assume k is algebraically closed.

If Spec A ⊗k K is disconnected, then A ⊗k K contains an idempotent e ̸= 0, 1
PN
(by Exercise 10.5.L). Write e = i=1 ai ⊗ ℓi , where ai ∈ A and ℓi ∈ K. Define B as
the subalgebra of K generated over k by the ℓi (as an algebra), so B is an integral
domain of finite type over k. Then K(B) is a finitely generated field extension of k.

Spec A ⊗k K / Spec A ⊗k B / Spec A

universally open
  
Spec K / Spec B / Spec k

Also, e ∈ A ⊗k B, and we recall that e is an idempotent distinct from 0 and 1 (even

in this ring). By the idempotent-connectedness package (Exercise 10.5.L) applied
to A ⊗k` B, Spec A ⊗k B is disconnected, say with nonempty open subsets U and V
with U V = Spec A ⊗k B.
Now ϕ : Spec A ⊗k B → Spec B is an open map (because Spec A → Spec k is
universally open, Fact 10.5.6), so ϕ(U) and ϕ(V) are nonempty open sets of Spec B.
As Spec B is connected, the intersection ϕ(U) ∩ ϕ(V) is a nonempty open set, which
∼ Spec A, and
has a closed point p (with residue field k, as k = k). But then ϕ−1 (p) =
we have covered Spec A with two disjoint open sets, yielding a contradiction. □

10.5.11. Sneaky “tensor-finiteness” trick that was just used. This argument used a
particularly clever trick that we will use again: we conjured a finitely generated
algebra B out of nowhere, and then used classical “geometry over algebraically
closed fields” to prove something far-removed from classical algebraic geometry.
What let us do that was that any single element of a tensor product (such as e in the
argument above) is a finite sum of “elementary tensors”, and so we worked instead
in a subring generated by the “parts” of those “elementary tensors”. We will call
this the tensor-finiteness trick in order to help direct your attention to where it can
be used again.

10.5.12. Corollary. — If k is separably closed, and Y is a connected k-scheme, then Y is

geometrically connected.

Proof. We wish to show that for any field extension K/k, YK is connected. By Propo-
sition 10.5.10, Spec K is geometrically connected over k. Then apply Lemma 10.5.9
with X = Spec K. □

10.5.13. Irreducibility.
 © 2024 Ravi Vakil. Published by Princeton University Press. 299

10.5.14. Proposition. — Suppose k is separably closed, A is a k-algebra with Spec A

irreducible, and K/k is a field extension. Then Spec A ⊗k K is irreducible.

Proof. We follow the philosophy of the proof of Proposition 10.5.10. As in the first
paragraph of that proof, it suffices to assume that k is algebraically closed.
If A ⊗k K is not irreducible, then we can find x and y with V(x), V(y) ̸=
Spec A ⊗k K and V(x) ∪ V(y) = Spec A ⊗k K. As in the second paragraph of the
proof of Proposition 10.5.10,
P we use the “tensor-finiteness
P ′ trick” (§10.5.11). We
choose expressions x = ai ⊗ ℓi and y = aj ⊗ ℓj′ (ai , aj′ ∈ A, ℓi , ℓj′ ∈ K), and
then define B to be the subring of K generated by the ℓi and ℓj′ . Thus B is an integral
domain finitely generated over k = k, and x, y ∈ A ⊗k B. Then D(x) and D(y)
are nonempty open subsets of Spec A ⊗k B, whose image in Spec B are nonempty
open sets (by universal openness of Spec A → Spec k, Fact 10.5.6), and thus their
intersection is nonempty and contains a closed point p (with residue field k = k).
But then ϕ−1 (p) = ∼ Spec A, and we have covered Spec A with two proper closed
sets (the restrictions of V(x) and V(y)), yielding a contradiction. □

10.5.M. E XERCISE . Suppose k is separably closed, and A and B are k-algebras,

both with irreducible Spec (i.e., with one minimal prime). Show that A ⊗k B has
irreducible Spec too. (Hint: reduce to the case where A and B are finite type over k.
Extend the proof of the previous proposition.)

10.5.N. E ASY EXERCISE . Show that a scheme X is irreducible if and only if there
exists an open cover X = ∪Ui with Ui irreducible for all i, and Ui ∩ Uj ̸= ∅ for all
i, j.

10.5.15. Proposition. — Suppose K/k is a field extension of a separably closed field and
Xk is irreducible. Then XK is irreducible.

Proof. Take an open cover Xk = ∪Ui by pairwise intersecting irreducible affine

open subsets. The base change of each Ui to K is irreducible by Proposition 10.5.14,
and they pairwise intersect. The result then follows from Exercise 10.5.N. □

10.5.O. E XERCISE ( GEOMETRICALLY INTEGRAL × INTEGRAL = INTEGRAL ). Sup-

pose B is a k-algebra such that B ⊗k k is an integral domain (Spec B is geometrically
integral), and A is a k-algebra that is an integral domain (Spec A is integral). Show
that A ⊗k B is an integral domain (Spec A ⊗k B is integral). (Once we define “va-
riety”, this will imply that the product of a geometrically integral variety with an
integral variety is an integral variety.) Hint: revisit the proof of Exercise 10.4.H.

10.5.16. Reducedness.
We recall a statement from transcendence theory (the basics of which are
developed in Exercise 12.2.A). Because this is a starred section, we content ourselves
with a reference rather than a proof.

10.5.17. Theorem: finitely generated extensions of perfect fields are separably

generated. — Any finitely generated extension of a perfect field E/F can be factored into
300 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

a finite separable part and a purely transcendental part:

E
finite separable

F(t1 , . . . , tn )
purely transcendental

F.

See [E, Cor. 16.17(b)] or [van, §19.7] for a proof. If you wish to prove it yourself,
see Exercise 10.5.P.

10.5.18. Finitely generated extensions that can be factored in this way are said to be
separably generated. The transcendence basis {t1 , . . . , tn } is said to be a separating
transcendence basis.

10.5.P. ⋆⋆ E XERCISE (P ROOF OF T HEOREM 10.5.17). Suppose E/F is a finitely

generated extension of a perfect field of characteristic p. Fix a transcendence basis
{x1 , . . . , xd } of E/F.
(a) Suppose the inseparable degree n := [E : F(x1 , . . . , xd )]i > 1. Show there
is a ∈ E not separable over F(x1 , . . . , xd ) such that its minimal polynomial f ∈
F(x1 , . . . , xd )[T ] lies in F[x1 , . . . , xd , T ], and prove f is irreducible in F[x1 , . . . , xd , T ]
as well as in F(x1 , . . . , xd−1 , T )[xd ].
∂f
(b) For a and f as in (a), prove ∂x j
̸= 0 for some j. Relabel the xi ’s so j = d. Prove
that {x1 , . . . , xd−1 , a} is a transcendence basis of E over F, and use multiplicativity
of inseparable degree to show [E : F(x1 , . . . , xd−1 , a)]i < n.
(c) Deduce the existence of a separating transcendence basis of E/F.

10.5.19. Proposition (geometrically reduced × reduced = reduced). — Suppose B is

a geometrically reduced k-algebra, and A is a reduced k-algebra. Then A ⊗k B is reduced.
(Compare this to Exercise 10.5.O.)

Proof. Reduce to the case where A is finitely generated over k using the tensor-
finiteness
P trick (§10.5.11). (Suppose we have x ∈ A ⊗k B with xn = 0. Suppose
x= ai ⊗ bi . Let A ′ be the finitely generated subring of A generated by the ai .
Then A ′ ⊗k B ′
Qis a subring of A ⊗k B. Replace A by A .) Then A is a subring of
the product Ki of the function fieldsQ of its irreducible components. (One can
see this as follows: the kernel of A → A/pi is the intersection of the primes pi ,
which is the ideal of nilpotents N by Theorem 3.2.13; but n = 0 as A is reduced.
And A/pi ,→ K(A/pi ) = Ki clearly. Alternatively, you may find this this clear from
our discussion of associated points (§6.6).) So it suffices to prove the result for A a
product of fields. Then it suffices to prove the result when A is a field. But then we
are done, by the definition of geometric reducedness. □

10.5.20. Proposition. — Suppose A is a reduced k-algebra. Then:

(a) A ⊗k k(t) is reduced.
(b) If E/k is a finite separable extension, then A ⊗k E is reduced.
 © 2024 Ravi Vakil. Published by Princeton University Press. 301

Proof. (a) Clearly A ⊗ k[t] is reduced, and localization preserves reducedness (as
reducedness is stalk-local, Exercise 5.2.A).
(b) Working inductively, we can assume E is generated by a single element,
with minimal polynomial p(t). By the tensor-finiteness trick, we can assume A
is finitely generated over k. (Why can we do this? What are the finitely many
elementary tensors in this argument?) Then by the same trick as in the proof of
Proposition 10.5.19, we can replace A by the product of its function fields of its
components, and then we can assume A is a field. But then A[t]/p(t) is reduced by
the definition of separability of p. □

10.5.21. Lemma. — Suppose E/k is a field extension of a perfect field, and A is a reduced
k-algebra. Then A ⊗k E is reduced.

Proof. By the tensor-finiteness trick, we may assume E is finitely generated over k.

(Be sure you see how!) By Theorem 10.5.17, we can factor E/k into extensions of
the forms of Proposition 10.5.20 (a) and (b). We then apply Proposition 10.5.20. □

10.5.22. Proposition. — Suppose E/k is an extension of a perfect field, and X is a reduced

k-scheme. Then XE is reduced.

Proof. Reduce to the case where X is affine. Use Lemma 10.5.21. □

10.5.23. Corollary. — Suppose k is perfect, and A and B are reduced k-algebras. Then
A ⊗k B is reduced.

Proof. By Lemma 10.5.21, A is a geometrically reduced k-algebra. Then apply

Lemma 10.5.19. □

10.5.Q. E XERCISE ( COMPLETING H ARD FACT 10.5.3). Show that (Rkp ) implies
(RK ), (Iks ) implies (IK ), and (Cks ) implies (CK ).

10.5.R. E XERCISE . Suppose that A and B are two integral domains that are k-
algebras. Show that A ⊗k B is an integral domain. (Compare this to Exercise 10.4.H,
which had finite type hypotheses.)

10.6 Products of projective schemes: The Segre embedding

We next describe products of projective A-schemes over A. (The case of greatest

initial interest is if A = k.) To do this, we need only describe Pm n
A ×A PA , because any
m
projective A-scheme has a closed embedding in some PA , and closed embeddings
behave well under base change, so if X ,→ Pm n
A and Y ,→ PA are closed embeddings,
m n
then X ×A Y ,→ PA ×A PA is also a closed embedding, cut out by the equations of
X and Y (§10.2(3)). We will describe Pm n
A ×A PA , and see that it too is a projective
A-scheme. (Hence if X and Y are projective A-schemes, then their product X ×A Y
over A is also a projective A-scheme.)
Before we do this, we will get some motivation from classical projective spaces
(nonzero vectors modulo nonzero scalars, Exercise 4.4.F) in a special case. Our map
302 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

will send [x0 , x1 , x2 ] × [y0 , y1 ] to a point in P5 , whose coordinates we think of as

being entries in the “multiplication table”
[ x0 y0 , x1 y0 , x 2 y0 ,
x 0 y1 , x 1 y1 , x2 y1 ].
This is indeed a well-defined map of sets. Notice that the resulting matrix is rank
one, and from the matrix, we can read off [x0 , x1 , x2 ] and [y0 , y1 ] up to multiplica-
tion by nonzero scalars. For example, to read off the point [x0 , x1 , x2 ] ∈ P2 , we take
the first row, unless it is all zero, in which case we take the second row. (They can’t
both be all zero.) In conclusion: in classical projective geometry, given a point of Pm
and Pn , we have produced a point in Pmn+m+n , and from this point in Pmn+m+n ,
we can recover the points of Pm and Pn .
Suitably motivated, we return to algebraic geometry. We define a map

Pm n / Pmn+m+n
A ×A PA A

by
([x0 , . . . , xm ], [y0 , . . . , yn ])  / [z00 , z01 , . . . , zij , . . . , zmn ]

= [x0 y0 , x0 y1 , . . . , xi yj , . . . , xm yn ].
More explicitly, we consider the map from the affine open set Ui × Vj (where
Ui = D(xi ) and Vj = D(yj ) to the affine open set Wij = D(zij ) by
(x0/i , . . . , xm/i , y0/j , . . . , yn/j ) 7→ (x0/i y0/j , . . . , xi/i yj/j , . . . , xm/i yn/j )
or, in terms of algebras, zab/ij 7→ xa/i yb/j .

10.6.A. E XERCISE . Check that these maps glue to give a well-defined morphism
A ×A PA → PA
mn+m+n
Pm n
.

10.6.1. We next show that this morphism is a closed embedding. We can check
this on an open cover of the target (the notion of being a closed embedding is
affine-local, Exercise 9.1.D). Let’s check this on the open set where zij ̸= 0. The
preimage of this open set in Pm n
A × PA is the locus where xi ̸= 0 and yj ̸= 0, i.e.,
Ui × Vj . As described above, the map of rings is given by zab/ij 7→ xa/i yb/j ; this
is clearly a surjection, as zaj/ij 7→ xa/i and zib/ij 7→ yb/j . (A generalization of this
ad hoc description will be given in Exercise 15.2.E.)
This map is called the Segre morphism or Segre embedding. If A is a field, the
image is called the Segre variety.

10.6.B. E XERCISE . Show that the Segre scheme (the image of the Segre embedding)
is cut out (scheme-theoretically) by the equations corresponding to
 
z00 · · · z0n
rank  ... .. ..  = 1,

. . 
zm0 ··· zmn
i.e., that all 2 × 2 minors vanish. Hint: suppose you have a polynomial in the zij
that becomes zero upon the substitution zij = xi yj . Give a recipe for subtracting
polynomials of the form “monomial times 2 × 2 minor” so that the end result is 0.
(The analogous question for the Veronese embedding in special cases is the content
of Exercises 9.3.I and 9.3.K.)
 © 2024 Ravi Vakil. Published by Princeton University Press. 303

10.6.2. Important Example. Let’s consider the first nontrivial example, when
m = n = 1. We get P1k ×k P1k ,→ P3k (where k is a field). We get a single equation
 
z00 z01
rank = 1,
z10 z11
i.e., z00 z11 − z01 z10 = 0. We again meet our old friend, the quadric surface (§9.3.9)!
Hence: the smooth quadric surface wz − xy = 0 (Figure 9.2) is isomorphic to
P1k ×k P1k . Recall from Exercise 9.3.L that the quadric surface has two families
(rulings) of lines. You may wish to check that one family of lines corresponds to the
image of {x} ×k P1k as x varies, and the other corresponds to the image P1k ×k {y} as
y varies.
If k is an algebraically closed field of characteristic not 2, then by diagonaliz-
ability of quadratics (Exercise 5.4.J), all rank 4 (“full rank”) quadratics in 4 variables
are isomorphic, so all rank 4 quadric surfaces over an algebraically closed field of
characteristic not 2 are isomorphic to P1k ×k P1k .
Note that this is not true over a field that is not algebraically closed. For
example, over R, w2 + x2 + y2 + z2 = 0 (in P3R ) is not isomorphic to P1R ×R P1R .
Reason: the former has no real points, while the latter has lots of real points.
You may wish to do the next two exercises in either order. The second can be
used to show the first, but the first may give you insight into the second.

10.6.C. E XERCISE : A COORDINATE - FREE DESCRIPTION OF THE S EGRE EMBEDDING .

Show that the Segre embedding can be interpreted as PV × PW → P(V ⊗ W) via
the surjective map of graded rings

/ / ⊕∞ Symi V ∨ ⊗ Symi W ∨
   
Sym• (V ∨ ⊗ W ∨ ) i=0

“in the opposite direction”.

10.6.D. E XERCISE : A COORDINATE - FREE DESCRIPTION OF PRODUCTS OF PROJEC -

TIVE A- SCHEMES IN GENERAL . Suppose that S• and T• are finitely generated
graded rings over A. Describe an isomorphism
∼
(Proj S• ) ×A (Proj T• ) o / Proj ⊕∞ (Sn ⊗A Tn )
n=0

(where hopefully the definition of multiplication in the graded ring ⊕∞

n=0 (Sn ⊗A Tn )
is clear).

10.7 Normalization
We discuss normalization now only because the central construction gives
practice with the central idea behind the construction in §10.1 of the fibered product
(see Exercises 10.7.B and 10.7.I).
Normalization is a means of turning a reduced scheme into a normal scheme.
(Unimportant remark: reducedness is not a necessary hypothesis, and is included
only to avoid distraction. If you care, you can follow through the construction,
and realize that the normalization of a scheme factors through the reduction, and
is the normalization of the reduction.) A normalization of a reduced scheme X
is a morphism ν : X e → X from a normal scheme, where ν induces a bijection of
304 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

irreducible components of X e and X, and in reasonable cases (cf. Exercise 10.7.N)

ν gives a birational morphism on each of the irreducible components. It will
satisfy a universal property, and hence it will be unique up to unique isomorphism.
Figure 8.4 is an example of a normalization.
We begin with the case where X is irreducible, and hence (by Exercise 5.2.G)
integral. (We will then deal with a more general case, and also discuss normalization
in a function field extension.) In this case of irreducible X, the normalization
ν: Xe → X is a dominant morphism (not just dominant rational map!) from an
irreducible normal scheme to X, such that any other such morphism factors through
ν:

(10.7.0.1) normal Y
∃! /X
e normal

dominant ν dominant

X

Thus if the normalization exists, then it is unique up to unique isomorphism.

We now have to show that it exists, and we do this in a way that will look familiar.
We deal first with the case where X is affine, say X = Spec A, where A is an integral
domain. Then let A e be the integral closure of A in its fraction field K(A).

10.7.A. E XERCISE . (Recall that A is an integral domain.) Show that ν : Spec A e→

Spec A satisfies the universal property of normalization. (En route, you might show
that the global sections of an irreducible normal scheme are also “normal”, i.e.,
integrally closed.)

10.7.B. I MPORTANT ( BUT SURPRISINGLY EASY ) E XERCISE . Show that normaliza-

tions of integral schemes exist in general. (Hint: Ideas from the existence of fibered
products, §10.1, may help.)

10.7.C. E ASY E XERCISE . Show that normalizations are integral and surjective.
(Hint for surjectivity: the Lying Over Theorem, see §8.2.6.)
We will soon see that normalization of integral finite type k-schemes is always
a birational morphism, in Exercise 10.7.N.

10.7.D. E XERCISE . Explain (by defining a universal property) how to extend the
notion of normalization to the case where X is a reduced scheme, with possibly
more than one component, but under the hypothesis that every affine open subset
of X has finitely many irreducible components. Note that this includes all locally
Noetherian schemes. (If you wish, you can show that the normalization exists in
this case. See [Stacks, tag 035Q] for more.)
Here are some examples.

10.7.E. E XERCISE ( NORMALIZATION OF THE NODAL CUBIC ). Suppose that char k ̸=

2 (only so we can use the word “nodal”). Show that Spec k[t] → Spec k[x, y]/(y2 −
x2 (x+1)) given by (x, y) 7→ (t2 −1, t(t2 −1)) (see Figure 8.4) is a normalization. The
target curve is called the nodal cubic curve. (Hint: show that k[t] and k[x, y]/(y2 −
x2 (x + 1)) have the same fraction field. Show that k[t] is integrally closed. Show
that k[t] is contained in the integral closure of k[x, y]/(y2 − x2 (x + 1)).)
 © 2024 Ravi Vakil. Published by Princeton University Press. 305

You will see from the previous exercise that once we guess what the normal-
ization is, it isn’t hard to verify that it is indeed the normalization. Perhaps a few
words are in order as to where the polynomials t2 − 1 and t(t2 − 1) arose in the
previous exercise. The key idea is to guess t = y/x. (Then t2 = x + 1 and y = xt
quickly.) This idea comes from three possible places. We begin by sketching the
curve, and noticing the “node” at the origin. (“Node” will be formally defined in
§28.2.1.) (a) The function y/x is well-defined away from the node, and at the node,
the two branches have “values” y/x = 1 and y/x = −1. (b) We can also note that if
t = y/x, then t2 is a polynomial, so we will need to adjoin t in order to obtain the
normalization. (c) The curve is cubic, so we expect a general line to meet the cubic
in three points, counted with multiplicity. (We will make this precise when we
discuss Bézout’s Theorem, Exercise 18.6.J, but in this case we have already gotten a
hint of this in Exercise 7.5.H.) There is a P1 parametrizing lines through the origin
(with coordinate equal to the slope of the line, y/x), and most such lines meet the
curve with multiplicity two at the origin, and hence meet the curve at precisely
one other point of the curve. So this “coordinatizes” most of the curve, and we try
adding in this coordinate.

10.7.F. E XERCISE . Find the normalization of the cusp y2 = x3 (see Figure 10.5).
(“Cusp” will be formally defined in Definition 28.2.2.)

F IGURE 10.5. Normalization of a cusp

10.7.G. E XERCISE . Suppose char k ̸= 2. Find the normalization of the tacnode

y2 = x4 , and draw a picture analogous to Figure 10.5. (“Tacnode” will be formally
defined in Definition 28.2.2.)
(Although we haven’t defined “singularity”, “smooth”, “curve”, or “dimen-
sion”, you should still read this.) Notice that in the previous examples, normaliza-
tion “resolves” the singularities (“non-smooth” points) of the curve. In general, it
will do so in dimension one (in reasonable Noetherian circumstances, as normal
local Noetherian integral domains of dimension one are all discrete valuation rings,
§13.5), but won’t do so in higher dimension (the cone z2 = x2 + y2 over a field k of
characteristic not 2 is normal, Exercise 5.4.I(b)).

10.7.H. E XERCISE . Suppose X = Spec Z[15i]. Describe the normalization X e → X.

(Hint: Z[i] is a unique factorization domain, §5.4.5(0), and hence is integrally closed
by Exercise 5.4.F.) Over what points of X is the normalization not an isomorphism?
306 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Another exercise in a similar vein is the normalization of the “pinched plane”,

Exercise 13.5.H.

10.7.I. E XERCISE ( NORMALIZATION IN A FUNCTION FIELD EXTENSION , AN IMPOR -

TANT GENERALIZATION ). Suppose X is an integral scheme. The normalization
e → X of X in a given algebraic field extension L of the function field K(X) of
ν: X
X is a dominant morphism from a normal integral scheme X e with function field L,
such that ν induces the inclusion K(X) ,→ L, and that is universal with respect to
this property.
Spec L Spec K(Y) /Y normal
∼ ∃!
 
Spec K(X)
e /X
e normal

  
Spec K(X) /X

Show that the normalization in a (given) extension of fields exists.

The following two examples, one arithmetic and one geometric, show that this
is an interesting construction.

10.7.J. E XERCISE . Suppose X = Spec Z (with function field Q). Find its integral clo-
sure in the field extension Q(i). (There is no “geometric” way to do this; it is purely
an algebraic problem, although the answer should be understood geometrically.)

10.7.1. Remark: Rings of integers in number fields. A finite extension K of Q is called a

number field, and the integral closure of Z in K the ring of integers in K, denoted
OK . (This notation is awkward given our other use of the symbol O.)
Spec K / Spec OK

 
Spec Q / Spec Z

By the previous exercises, OK is a normal integral domain, and we will see soon
(Theorem 10.7.3(a)) that it is Noetherian, and later (Exercise 12.1.G) that it has
“dimension 1”. This is an example of a Dedekind domain, see §13.5.11. We will think
of it as a “smooth” curve as soon as we define what “smooth” (really, regular) and
“curve” mean.
√
10.7.K. E XERCISE . Find the ring of integers in Q( n), where n is square-free and
n ≡ 3 (mod 4). (Hint: Exercise 5.4.I(a), where you will also be able to figure out
the answer for square-free n in general.)

10.7.L. E XERCISE . Suppose char k ̸= 2 for convenience (although it isn’t necessary).

(a) Suppose X = Spec k[x] (with function field k(x)). Find its normalization in
the field extension k(y), where y2 = x2 + x. (Again we get a Dedekind domain.)
Hint: this can be done without too much pain. Show that Spec k[x, y]/(x2 + x − y2 )
is normal, possibly by identifying it as an open subset of P1k , or possibly using
 © 2024 Ravi Vakil. Published by Princeton University Press. 307

Exercise 5.4.H.
(b) Suppose X = P1 , with distinguished open subscheme Spec k[x]. Find its normal-
ization in the field extension k(x, y), where y2 = x2 +x. (Part (a) involves computing
the normalization over one affine open set; now figure out what happens over the
“other” affine open set, and how to glue.)

10.7.2. Fancy fact: Finiteness of integral closure.

The following fact is useful.

10.7.3. Theorem (finiteness of integral closure). — Suppose A is a Noetherian integral

domain, K = K(A), L/K is a finite extension of fields, and B is the integral closure of A in
L (“the integral closure of A in the field extension L/K”, i.e., those elements of L integral
over A).
(a) If A is integrally closed and L/K is separable, then B is a finitely generated A-module.
(b) (E. Noether) If A is a finitely generated k-algebra, then B is a finitely generated A-
module.
Eisenbud gives a proof in a page and a half: (a) is [E, Prop. 13.14] and (b) is [E,
Cor. 13.13]. A sketch is given in §10.7.5.

10.7.4. Warning. Part (b) does not hold for Noetherian A in general — the integral
closure of a Noetherian ring need not be Noetherian. (See [Rei2, §9.4] or [E, §13.3]
for some discussion.) This is alarming. The existence of such an example is a sign
that Theorem 10.7.3 is not easy.
In the following four exercises, you can replace “integral finite type k-scheme”
by “integral variety” once you know what a variety is.

10.7.M. E XERCISE ( FINITENESS OF NORMALIZATION ).

e→
(a) Show that if X is an integral finite type k-scheme, then its normalization ν : X
X is a finite morphism.
(b) Suppose X is a locally Noetherian integral scheme. Show that if X is normal,
then the normalization in a finite separable field extension is a finite morphism.
Show that if X is an integral finite type k-scheme, then the normalization in a finite
extension of fields is a finite morphism. In particular, once we define “variety”
(Definition 11.2.9), you will see that this implies that the normalization of a variety
(including in a finite extension of the function field) is a variety.

10.7.N. E XERCISE . Show that if X is an integral finite type k-scheme, then the nor-
malization morphism is birational. (Hint: Proposition 7.5.7; or solve Exercise 10.7.O
first.)

10.7.O. E XERCISE . Suppose that X is an integral finite type k-scheme. Show that the
normalization map of X is an isomorphism on an open dense subset of X. (Hint:
Proposition 7.5.5.) Show that the normal points of X form a dense open subset. (By
Exercise 10.7.D, the “integral” hypothesis can be relaxed to “reduced”.)

10.7.P. E XERCISE . Suppose ρ : Z → X is a finite birational morphism from an

integral finite type k-scheme to a normal integral finite type k-scheme. Show that ρ
is an isomorphism.
308 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

10.7.5. ⋆⋆ Sketch of proof of finiteness of integral closure, Theorem 10.7.3. Here is a

sketch to show the structure of the argument. It uses commutative algebra ideas
from Chapter 12, so you should only glance at this to see that nothing fancy is
going on. Part (a): reduce to the case where L/K is Galois, with group {σ1 , . . . , σn }.
Choose b1 , . . . , bn ∈ B forming a K-vector space basis of L. Let M be the matrix
(familiar from Galois theory) with ijth entry σi bj , and let d = det M. Show that
the entries of M lie in B, and that d2 ∈ K (as d2 is Galois-fixed). Show that d ̸= 0
using linear independence of characters. Then complete the proof by showing
that B ⊂ d−2 (Ab1 + · · · + Abn ) (submodules of finitely generated modules over
Noetherian rings Pare also Noetherian, Exercise 3.6.Y) as follows. Suppose b ∈ B,
and write b = ci bi (ci ∈ K). If c is the column vector with entries ci , show
that the ith entry of the column vector Mc is σi b ∈ B. Multiplying Mc on the
left by adj M (see the trick of the proof of Lemma 8.2.1), show that dci ∈ B. Thus
d2 ci ∈ B ∩ K = A (as A is integrally closed), as desired.
For (b), use the Noether Normalization Lemma 12.2.4 to reduce to the case
A = k[x1 , . . . , xn ]. Reduce to the case where L is normally closed over K. Let L ′
be the subextension of L/K so that L/L ′ is Galois and L ′ /K is purely inseparable.
Use part (a) to reduce to the case L = L ′ . If L ′ ̸= K, then for some q, L ′ is generated
over K by the qth root of a finite set of rational functions. Reduce to the case
1/q 1/q
L ′ = k ′ (x1 , . . . , xn ) where k ′ /k is a finite purely inseparable extension. In this
1/q 1/q
case, show that B = k ′ [x1 , . . . , xn ], which is indeed finite over k[x1 , . . . , xn ]. □

10.7.6. Reduction of schemes, revisited. It may be enlightening to revisit the

“reduction” of a scheme using this point of view. The argument is clean and simple,
and by working it out for yourself, you will better appreciate the various moving
parts. We now define “reduction” from scratch, and then show that it agrees with
the definition of §9.4.10.

10.7.7. Definition. Suppose X is a scheme. In analogy with (10.7.0.1), we say

ρ : Xred → X is “the” reduction of X if any morphism W → X from a reduced
scheme W factors uniquely through ρ:

reduced W
∃! / Xred
ρ
! 
X

The “the” is in quotes because this is a slight abuse of notation. In analogy with
fibered products and normalization, we now check that such an Xred exists:

10.7.Q. E XERCISE .
(a) In analogy with Exercise 10.7.A, show that ρ : Spec A/N(A) → Spec A
satisfies the universal property of reduction.
(b) In analogy with Exercise 10.7.B, show that the reduction of schemes exists
in general.
(c) To make sure we haven’t contradicted our earlier selves: Show that the
construction of §9.4.10 satisfies this universal property.
 © 2024 Ravi Vakil. Published by Princeton University Press. 309

10.7.8. Definition. By the universal property of reduction, any morphism of schemes

π : X → Y induces a morphism Xred → Y red , the reduction of π which we quite
reasonably name πred . (You likely see that “reduction” gives a functor from the
category of schemes to the category of reduced schemes.)
 CHAPTER 11

Separated and proper morphisms, and (finally!)

varieties

Separatedness is a fundamental notion. It is the analog of the Hausdorff

condition for manifolds (see Exercise 11.2.B), and as with Hausdorffness, this
geometrically intuitive notion ends up being just the right hypothesis to make
theorems work. Although the definition initially looks odd, in retrospect it is just
perfect.
Let’s review why we like Hausdorffness. Recall that a topological space is
Hausdorff if for every two points x and y, there are disjoint open neighborhoods
of x and y. The real line is Hausdorff, but the “real line with doubled origin” (of
which Figure 4.6 may be taken as a sketch) is not. Many proofs and results about
manifolds use Hausdorffness in an essential way. For example, the classification of
compact one-dimensional manifolds is very simple, but if the Hausdorff condition
were removed, we would have a very wild set.
So once armed with this definition, we can cheerfully exclude the line with
doubled origin from civilized discussion, and we can (finally) define the notion of
a variety, in a way that corresponds to the classical definition.
With our motivation from manifolds, we shouldn’t be surprised that all of
our affine and projective schemes are separated. Subsets of Hausdorff topological
spaces are automatically Hausdorff. Hence if Y is a manifold, and X is a subset
that satisfies all the hypotheses of a manifold except possibly Hausdorffness, X is
a manifold. Similarly, we will see that locally closed embeddings in something
separated are also separated (combine Exercise 11.2.C and Proposition 11.2.3).
Grothendieck taught us that one should try to define properties of morphisms,
not of objects; then we can say that an object has that property if its morphism to
the final object has that property. We discussed this briefly at the start of Chapter 8.
In this spirit, separatedness will be a property of morphisms, not schemes.
As an unexpected bonus, a separated morphism to an affine scheme has the
property that the intersection of two affine open sets in the source is affine (Proposi-
tion 11.2.13). This will make Čech cohomology work very easily on (quasicompact)
schemes (Chapter 18). You might consider this an analog of the fact that in Rn , the
intersection of two convex sets is also convex. As affine schemes are trivial from
the point of view of quasicoherent cohomology, just as convex sets in Rn have no
cohomology, this metaphor is apt.
A lesson arising from the construction is the importance of the diagonal mor-
phism (Definition 1.2.7). More precisely, given a morphism π : X → Y, good

This material is published by Princeton University Press as The Rising Sea: Foundations of Algebraic
Geometry by Ravi Vakil. This version is free to view and download for personal use only. Not for
redistribution, resale, or use in derivative works. © 2024 Ravi Vakil.

311
312 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

consequences can be leveraged from good behavior of the diagonal morphism

δπ : X → X ×Y X (the product of the identity morphism X → X with itself), usually
through fun diagram chases. This lesson applies across many fields of geome-
try. (Another nice gift of the diagonal morphism: it will give us a good algebraic
definition of differentials, in Chapter 21.)
So before we define separatedness, we make some observations about diagonal
morphisms.

11.1 Fun with diagonal morphisms, and quasiseparatedness made

easy

Recall that most classes of morphisms are “reasonable” in the sense of §8.1:
they are (i) preserved by composition, (ii) presesrved by base change, (iii) local on
the target, and as a consequence they are (iv) preserved by product (Exercise 8.1.A).
Another useful consequence is that (v) “reasonable” classes of morphisms have
a “Cancellation Property”. This involves morphisms whose diagonal lies in P, i.e.,
those π : X → Y such that δπ : X → X ×Y X lies in P. We call this auxiliary class of
morphisms Pδ (“P-diagonal”).
The Cancellation Theorem is a strange-looking, but very useful, result. Like
the Diagonal-Base-Change diagram 1.2.S, this result is unexpectedly ubiquitous.

11.1.1. The Cancellation Theorem for a Property P of Morphisms. — Let P be a

class of morphisms that is preserved by base change and composition. (Every “reasonable”
class of morphisms satisfies this, §8.1.) Suppose

X
π /Y

τ ρ
 
Z

is a commuting diagram of schemes. Suppose ρ is in Pδ and τ is in P. Then π is in P.

The following diagram summarizes this important theorem:

X
∴∈P
/Y

∈Pδ
∈P
 
Z

When you plug in different P, you get very different-looking (and nonobvious) con-
sequences, the first of which are given in Exercise 11.2.E. (Here are some facts you
can prove easily, but which can be interpreted as applications of the Cancellation
Theorem in Sets, and which may thus shed light on how the Cancellation Theorem
works. If π : X → Y and ρ : Y → Z are maps of sets, and ρ ◦ π is injective, then so is
π; and if ρ ◦ π is surjective and ρ is injective, then π is surjective.)
 © 2024 Ravi Vakil. Published by Princeton University Press. 313

11.1.A. E XERCISE . Prove the Cancellation Theorem 11.1.1. Possible hint: Interpret
and make use of the following diagram.
∴∈P

/ X ×Z Y / Z ×Z Y (
X X ×Y Y Y

y
Y
| ∈P
/ Y ×Z Y %
X
∈P
/" Z

(vi) If P is a “reasonable” class, so is Pδ:

11.1.2. Theorem. — If P is a “reasonable” class of morphisms, then the class Pδ of

morphisms whose diagonal is in P is also a “reasonable” class.

Proof. We verify that Pδ satisfies the “three axioms of reasonableness”.

11.1.B. E XERCISE . Prove that “Pδ is preserved by composition”. Hint: Interpret

and make use of the following diagram.
∴∈P

#
X
∈P
/ X ×Y X / X ×Z X

 
Y
∈P
/ Y ×Z Y

11.1.C. E XERCISE . Prove that “Pδ is preserved by base change”. Hint: Suppose (a)
below is Cartesian. Show that (b) is also Cartesian.
(a) V /W (b) V / V ×W V

   
X /Y X / X ×Y X

11.1.D. E XERCISE . Prove that “Pδ is local on the target”. Hint: Suppose we have
a cover {Ui → Y} of Y so that we have a Cartesian diagram (a). Then we have
Cartesian diagrams (b).

(a) Vi _
∈Pδ
/ Ui (b) Vi _
∴∈P
/ Vi ×U Vi
_ i

   
X /Y X / X ×Y X
□
We continue our list of useful properties of “reasonable” classes of morphisms.
(vii) Finally, if you have two morphisms from X that are of certain form, the
next exercise shows what more you need to ensure that the map to the product is
also of that form. This time you can find the “proof by diagram” by yourself.
314 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

11.1.E. E XERCISE . Suppose P is a “reasonable” class of morphisms. Suppose

ρ : X → Y and σ : X → Z are morphisms of S-schemes in class P, and X → S is in
Pδ. Show that (ρ, σ) : X → Y ×S Z is in class P as well. (See Exercise 11.2.A for an
application.)

11.1.F. ⋆ E XERCISE . Show that finitely presented morphisms form a “reasonable”

class of morphisms.

11.1.3. Example: Monomorphisms. Recall that a morphism π : X → Y is a

monomorphism if and only if the diagonal δπ : X → X ×Y X is an isomorphism
(Exercise 1.2.W). Because isomorphisms form a “reasonable” class of morphisms,
we have immediately shown that monomorphisms form a “reasonable” class as
well (and accidentally solving Exercise 1.2.V, and proving that monomorphisms
are preserved by base change).

11.1.4. Redefinition: Quasiseparated morphisms.

We (now) say a morphism π : X → Y is quasiseparated if the diagonal mor-
phism δπ : X → X ×Y X is quasicompact.

11.1.G. E XERCISE . Show that this agrees with our earlier definition of quasisep-
arated (§8.3.1): show that π : X → Y is quasiseparated if and only if for any affine
open Spec A of Y, and two affine open subsets U and V of X mapping to Spec A,
U ∩ V is a finite union of affine open sets. (Possible hint: compare this to Propo-
sition 11.2.13. Another possible hint: the Diagonal-Base-Change diagram, Exer-
cise 1.2.S.)

11.1.5. Because quasicompact morphisms are a “reasonable” class (§10.4.E), The-

orem 11.1.2 implies that quasiseparated morphisms are a reasonable class! In
particular, quasiseparated morphisms are preserved by composition and base
change, and the notion is local on the target.

11.1.6. All of this discussion was remarkably removed from geometry. To bring in
the geometry, we now look more closely at the diagonal, which isn’t just any old
kind of morphism.

11.2 Separatedness, and varieties

11.2.1. Proposition. — Let π : X → Y be a morphism of schemes.

(a) If X and Y are affine, then the diagonal morphism δπ : X → X×Y X is a closed embedding.
(b) In general, the diagonal morphism δπ : X → X ×Y X is a locally closed embedding.
The corresponding locally closed subscheme of X ×Y X — the diagonal (locally
closed) subscheme — will be denoted ∆.

Proof. (a) Suppose X = Spec A and Y = Spec B. Then the diagonal morphism
corresponds to the natural ring map A ⊗B A → A given by a1 ⊗ a2 7→ a1 a2 , which
is obviously surjective.
(b) We will describe a union of open subsets of X ×Y X covering the image of X,
such that the image of X is a closed embedding in this union.
 © 2024 Ravi Vakil. Published by Princeton University Press. 315

Say Y is covered with affine open sets Vi and X is covered with affine open sets
Uij , with π : Uij → Vi . Note that Uij ×Vi Uij is an affine open subscheme of the
product X×Y X (basically this is how we constructed the product, by gluing together
affine building blocks). Then the diagonal is covered by these affine open subsets
Uij ×Vi Uij . (Any point p ∈ X lies in some Uij ; then δ(p) ∈ Uij ×Vi Uij . Figure 11.1
may be helpful.) Note that δ−1 (Uij ×Vi Uij ) = Uij : clearly Uij ⊂ δ−1 (Uij ×Vi Uij ),
and because pr1 ◦ δ = idX (where pr1 is the first projection), δ−1 (Uij ×Vi Uij ) ⊂ Uij .
Finally, by part (a), Uij → Uij ×Vi Uij is a closed embedding. □

F IGURE 11.1. An open neighborhood of the diagonal is covered

by Uij ×Vi Uij

The open subsets we described may not cover X ×Y X, so we have not shown
that δ is a closed embedding.

11.2.A. E ASY E XERCISE . Suppose X → Y and X → Z are both locally closed embed-
dings. Prove that X → Y × Z is a locally closed embedding. Hint: Exercise 11.1.E.

11.2.2. Definition. A morphism π : X → Y is separated if the diagonal morphism

δπ : X → X ×Y X is a closed embedding. An A-scheme X is said to be separated
over A if the structure morphism X → Spec A is separated. When people say
that a scheme (rather than a morphism) X is separated, they mean implicitly that
some “structure morphism” is separated. For example, if they are talking about
A-schemes, they mean that X is separated over A.
Thanks to Proposition 11.2.1(b) (and the fact that a locally closed embedding
whose image is closed is actually a closed embedding, Exercise 9.2.B), a morphism
is separated if and only if the diagonal ∆ is a closed subset — a purely topological
condition on the diagonal. This is reminiscent of a definition of Hausdorff, as the
next exercise shows.

11.2.B. E XERCISE ( ONLY FOR THOSE SEEKING TOPOLOGICAL MOTIVATION ). Show

that a topological space X is Hausdorff if and only if the diagonal is a closed subset
of X × X. (The reason separatedness of schemes doesn’t give Hausdorffness —
i.e., that for any two open points x and y there aren’t necessarily disjoint open
neighborhoods — is that in the category of schemes, the topological space X × X is
not in general the product of the topological space X with itself, see §10.1.2.)

11.2.3. Proposition. — Separated morphisms form a “reasonable” class in the sense of 8.1
— they are by preserved by composition, base change, and the notion is local on the target.
316 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Proof. The class of closed embeddings is “reasonable” in this sense (Exercise 10.4.A),
so by Theorem 11.1.2 the same is true of separated morphisms. □

11.2.C. E XERCISE . Show that monomorphisms are separated. Hence locally closed
embeddings (and in particular open and closed embeddings) are separated.

11.2.D. E XERCISE . Show that separated morphisms are quasiseparated. (Hint:

closed embeddings are affine, hence quasicompact.)

11.2.4. Proposition. — Affine morphisms are separated. In particular, finite morphisms

are separated.

Proof. By Proposition 11.2.1(a), every morphism of affine schemes is separated. The

notion of separatedness is local on the target, so we are done. □
Many of the following exercises will drive home the importance of the Cancel-
lation Theorem 11.1.1.

11.2.E. E XERCISE . Show that an A-scheme is separated (over A) if and only if it is

separated over Z. In particular, a complex scheme is separated over C if and only if
it is separated over Z, so complex geometers and arithmetic geometers can discuss
separated schemes with each other without confusion.
π / ρ
/Z.
11.2.F. E ASY E XERCISE . Suppose we have morphisms X Y
(a) Show that if ρ ◦ π is a locally closed embedding (resp., locally of finite type,
separated), then so is π.
(b) If ρ ◦ π is quasicompact, and Y is Noetherian, show that π is quasicompact. Hint:
Exercise 8.3.B(a).
(c) If ρ ◦ π is quasiseparated, show that π is quasiseparated. Hint: Exercise 8.3.B(b).

11.2.G. E XERCISE . Suppose µ : Z → X is a morphism, and σ : X → Z is a section of

µ, i.e., µ ◦ σ is the identity on X.
ZV
µ σ

X
Show that σ is a locally closed embedding. Show that if µ is separated, then σ is a
closed embedding. Show that if µ is quasiseparated, then σ is quasicompact. Give
an example to show that σ need not be a closed embedding if µ is not separated.

11.2.5. The graph morphism. Suppose π : X → Y is a morphism of Z-schemes. The

morphism γπ : X → X ×Z Y given by γπ = (id, π) is called the graph morphism
(see Figure 11.2). (It secretly appeared in Exercise 11.1.A.) Then π factors as pr2 ◦ γπ ,
where pr2 is the second projection. The diagram of Figure 11.2 is often called
the graph of a morphism π. (We will discuss graphs of rational maps in §11.3.4.)
Exercise 11.2.G applied to X ×Z Y → X yields the following.

11.2.6. Proposition. — The graph morphism γπ is always a locally closed embedding. If

Y is a separated Z-scheme (i.e., Y → Z is separated), then γπ is a closed embedding. If Y is
a quasiseparated Z-scheme, then γπ is quasicompact.
 © 2024 Ravi Vakil. Published by Princeton University Press. 317

F IGURE 11.2. The graph morphism

The image of γπ (a locally closed subscheme of X ×Z Y) is often also called the

graph of π; we might denote it Γπ .

11.2.H. E XERCISE WE WON ’ T USE . Suppose P is a class of morphisms such that

closed embeddings are in P, and P is preserved by fibered product and composition.
Recall the Definition 10.7.8 of the “reduction” of a morphism. Show that if π : X → Y
is in P then πred : Xred → Y red is in P. (Two examples are the classes of separated
morphisms and quasiseparated morphisms.) Possible hint:
πred / Y red
Xred
∈P ∈Pδ
 "/ 
X Y
π∈P

11.2.7. Important examples.

11.2.I. E XERCISE . Show that the line with doubled origin (Example 4.4.5) is not
separated, by verifying that the image of the diagonal morphism is not closed.
(Another argument is given below, in Exercise 11.3.C. A fancy argument is given in
Exercise 13.7.C.)
We next come to our first example of something separated but not affine.
The following single calculation will imply that all quasiprojective A-schemes
are separated (once we know that the composition of separated morphisms are
separated, Proposition 11.2.3).

A → Spec A is separated.
11.2.8. Proposition. — The morphism Pn
We give two proofs. The first is by direct calculation. The second requires
no calculation, and just requires that you remember some classical constructions
described earlier.
Proof 1: Direct calculation. We cover Pn n
A ×A PA with open sets of the form Ui ×A Uj ,
where U0 , . . . , Un form the “usual” affine open cover. The case i = j was taken care
of before, in the proof of Proposition 11.2.1(a). If i ̸= j then
Ui ×A Uj = ∼ Spec A[x0/i , . . . , xn/i , y0/j , . . . , yn/j ]/(xi/i − 1, yj/j − 1).
318 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Now δ−1 (pr−1 1 (Ui )) is Ui (as the diagonal morphism composed with projection to
the first factor is the identity), and similarly Uj = δ−1 (pr−1
2 (Uj )). Thus δ
−1
(Ui ×A
Uj ) = Ui ∩ Uj , so the diagonal morphism over Ui ×A Uj is Ui ∩ Uj → Ui ×A Uj .
This is a closed embedding, as the corresponding map of rings

A[x0/i , . . . , xn/i , y0/j , . . . , yn/j ] / A[x0/i , . . . , xn/i , x−1 ]/(xi/i − 1)

j/i

(given by xk/i 7→ xk/i , yk/j 7→ xk/i /xj/i ) is clearly a surjection (as each generator
of the ring on the right is clearly in the image — note that x−1 j/i is the image of yi/j ).
□
Proof 2: Classical geometry. Note that the diagonal morphism δ : Pn A → P A ×A
n
n2 +2n
PA followed by the Segre embedding S : PA ×A PA → P
n n n
(§10.6, a closed
embedding) can also be factored as the second Veronese embedding ν2 : Pn A →
n+2 n+2
(
P 2 ) −1 (
(§9.3.6) followed by a linear map L : P 2 ) −1
→Pn 2
+2n
, both of which
are closed embeddings (the linear map L by Exercise 9.3.C).
Pn n
A: ×A PA
δ S
l. (?) cl. emb. cl. emb. %
2
Pn
A Pn9 +2n

cl. emb. cl. emb.

ν2 $ L
n+2
P( 2 )−1
Informally, in coordinates:
([x0 , x1 , . . . , xn ], [x0 , x1 , . . . , xn ])
@
S
δ
%
x20 ,
 
x 0 x1 , ··· x0 xn ,
 x1 x0 , x21 , ··· x1 xn , 
[x0 , x1 , . . . , xn ]
 
 .. .. .. .. 
 . . . . 
xn x0 , xn x 1 , ··· x2n
9
ν2

 L

[x20 , x0 x1 , . . . , xn−1 xn , x2n ]

2
A → PA
n +2n
The composed map Pn may be written as
[x0 , . . . , xn ] 7→ [x20 , x0 x1 , x0 x2 , . . . , x2n ],
where the subscripts on the right run over all ordered pairs (i, j) where 0 ≤ i, j ≤ n.)
This forces δ to send closed sets to closed sets (or else S◦δ won’t, but L◦ν2 does). □

11.2.9. Important Definition (finally!). A variety over a field k, or k-variety, is a

reduced, separated scheme of finite type over k. For example, a reduced finite type
affine k-scheme is a variety. (Do you see why separatedness holds?) The notion
of variety generalizes our earlier notion of affine variety (§5.3.7) and projective
 © 2024 Ravi Vakil. Published by Princeton University Press. 319

variety (§5.3.E, see Proposition 11.2.15). By Exercise 5.3.F, the closed points of a
variety are dense. (Notational caution: In some sources, the additional condition of
irreducibility is imposed. Also, it is often assumed that k is algebraically closed.)
Varieties over k form a category: morphisms of k-varieties are just morphisms
as k-schemes. (Of course, the category of varieties over an algebraically closed field
k long predates modern scheme theory.)

11.2.J. E ASY E XERCISE . Show that morphisms of k-varieties are finite type and
separated.
A subvariety of a variety X is a reduced locally closed subscheme of X (which
you can quickly check is a variety itself). An open subvariety of X is an open
subscheme of X. (Reducedness is automatic in this case.) A closed subvariety of X
is a reduced closed subscheme of X.

11.2.K. ⋆⋆ E XERCISE ( FOR THOSE WITH APPROPRIATE BACKGROUND : COMPLEX

ALGEBRAIC VARIETIES YIELD COMPLEX ANALYTIC VARIETIES ). Show that the
analytification (Exercise 7.3.P) of a complex algebraic variety is a complex analytic
variety.

11.2.10. Products of varieties.

11.2.11. Proposition. —
(a) If k is perfect, the product of k-varieties (the fibered product over Spec k) is a k-variety.
(b) If k is algebraically closed, the product of irreducible k-varieties is an irreducible k-
variety.
(c) If k is algebraically closed, the product of connected k-varieties is a connected k-variety.

Proof. (a) The finite type and separated statements are straightforward, as both
properties are preserved by base change and composition. For reducedness, reduce
to the affine case, then use Corollary 10.5.23.
(b) It only remains to show irreducibility. Reduce to the affine case using
Exercise 10.5.N (as in the proof of Proposition 10.5.15). Then use Exercise 10.5.M,
which requires separable closure. (As an exercise, you can avoid the complications
of §10.5 by instead using Exercise 10.4.H.)

11.2.L. E XERCISE . Prove (c).

□

11.2.12. Back to separatedness. Here is a very handy consequence of separated-

ness.

11.2.13. Proposition. — Suppose X → Spec A is a separated morphism to an affine

scheme, and U and V are affine open subsets of X. Then U ∩ V is an affine open subset of X.

Before proving this, we give an unexpected consequence. Taking X = Spec A,

we see that the intersection of any two affine open subsets of Spec A is also an
affine open subset. This is certainly not obvious! We know the intersection of two
distinguished affine open subsets is affine (from D(f) ∩ D(g) = D(fg)), but we have
little handle on affine open subsets in general.
320 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Warning: this property does not characterize separatedness. (The correct

property is “affine-diagonal”, which we will not discuss, but which is useful in the
theory of stacks.) For example, if A is a field k and X is the line with doubled origin
over k, then X also has this property.
In the proof of Proposition 11.2.13, we will use the following nifty result.

11.2.M. E ASY E XERCISE . Suppose U and V are open subsets of an A-scheme X.

Show that ∆ ∩ (U ×A V) =∼ U ∩ V. (The intersection of two locally closed subschemes
was defined in Exercise 10.2.C.) Hints: (i) this was basically proved in the course
of Proof 1 of Proposition 11.2.8. (ii) See Figure 11.3. (iii) You can interpret this
statement using the Diagonal-Base-Change diagram, Exercise 1.2.S:

U∩V / U ×A V

 
X
δ / X ×A X.

∼ U ∩ V (Exercise 11.2.M)
F IGURE 11.3. ∆ ∩ (U × V) =

∼ U∩V, where
11.2.14. Proof of Proposition 11.2.13. By Exercise 11.2.M, (U×A V)∩∆ =
∆ is the diagonal. But U ×A V is affine (the fibered product of two affine schemes
over an affine scheme is affine, Step 1 of our construction of fibered products,
Theorem 10.1.1), and (U ×A V) ∩ ∆ is a closed subscheme of the affine scheme
U ×A V, and hence U ∩ V is affine. □

11.2.15. Corollary. — Every quasiprojective A-scheme is separated over A. In particular,

every reduced quasiprojective k-scheme is a k-variety.

Proof. Suppose X → Spec A is a quasiprojective A-scheme. The structure morphism

can be factored into an open embedding composed with a closed embedding
followed by Pn A → Spec A. Open embeddings and closed embeddings are separated
A → Spec A is separated (Proposition 11.2.8). Compositions
(Exercise 11.2.C), and Pn
of separated morphisms are separated (Proposition 11.2.3), so we are done. □
Here is an exercise we won’t use, but you may like. It includes a converse to
Proposition 11.2.13.
 © 2024 Ravi Vakil. Published by Princeton University Press. 321

11.2.N. E XERCISE . Show that X → Spec A is separated if and only if, for all affine
open subsets U and V of X, (i) the intersection U ∩ V is affine, and (ii) the map
O(U) ⊗A O(V) → O(U ∩ V) is surjective. Show that it is enough to check that this
holds as U and V range over the sets in any affine cover X = ∪Ui . (Hint: we largely
did this in Proof 1 of Proposition 11.2.8.)

11.2.16. ⋆⋆ Universally injective morphisms and the diagonal.

11.2.O. E XERCISE . Show that π : X → Y is universally injective if and only if the

diagonal morphism δπ : X → X ×Y X is surjective. (Recall that δπ is always injective,
by Proposition 11.2.1(b).)
Because surjective morphisms form a “reasonable” class (Exercise 10.4.G), we
see that universally injective morphisms also form a “reasonable” class.

11.2.P. E ASY E XERCISE . If π : X → Y and ρ : Y → Z are morphisms, and ρ ◦ π is

universally injective, show that π is universally injective.

11.2.Q. E XERCISE .
(a) Show that universally injective morphisms are separated.
(b) Show that a map between finite type schemes over an algebraically closed field
k is universally injective if and only if it is injective on closed points.

11.3 The locus where two morphisms from X to Y agree, and the
“Reduced-to-Separated” Theorem

When we introduced rational maps in §7.5, we promised that in good circum-

stances, a rational map has a “largest domain of definition”. We are now ready
to make precise what “good circumstances” means, in the Reduced-to-Separated
Theorem 11.3.2. We first introduce an important result making sense of locus where
two morphisms with the same source and target “agree”.

11.3.A. U SEFUL EXERCISE : T HE LOCUS WHERE TWO MORPHISMS AGREE . Suppose

π : X → Y and π ′ : X → Y are two morphisms over some scheme Z.

π
X 3+ Y
π′

 
Z

We can now give meaning to the phrase ’the locus where π and π ′ agree’, and
that in particular there is a largest locally closed subscheme where they agree —
which is closed if Y is separated over Z. Suppose µ : W → X is some morphism
(not assumed to be a locally closed embedding). We say that π and π ′ agree on µ if
π ◦ µ = π ′ ◦ µ. Show that there is a locally closed subscheme i : V ,→ X on which π
and π ′ agree, such that any morphism µ : W → X on which π and π ′ agree factors
uniquely through i, i.e., there is a unique j : W → V such that µ = i ◦ j. Show further
that if Y → Z is separated, then i : V ,→ X is a closed embedding. Hint: define V to
322 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

be the following fibered product:

V /Y

δ
 (π,π ′ ) 
X / Y ×Z Y.

As δ is a locally closed embedding, V → X is too. Then if µ : W → X is any

morphism such that π ◦ µ = π ′ ◦ µ, then µ factors through V.
The fact that the locus where two maps agree can be nonreduced should not
come as a surprise: consider two maps from A1k to itself, π(x) = 0 and π ′ (x) = x2 .
They agree when x = 0, but the situation is (epsilonically) better than that — they
should agree even on Spec k[x]/(x2 ).

11.3.1. Minor Remarks.

(i) In the previous exercise, we describe V ,→ X by a universal property. Taking
this as the definition, it is not a priori clear that V is a locally closed subscheme of
X, or even that it exists.
(ii) Warning: consider two maps from Spec C to itself over Spec R, the identity
and complex conjugation. These are both maps from a point to a point, yet they
do not agree despite agreeing as maps of sets. (If you are not convinced that they
disagree as morphisms, this might help: after base change Spec C → Spec R, they
do not agree even as maps of sets.)
(iii) More generally, the (set-theoretic) locus where π and π ′ agree can be
interpreted as follows: π and π ′ agree at p if π(p) = π ′ (p) and the two maps
of residue fields are the same. (You may enjoying thinking this through as an
exercise.)

11.3.B. E XERCISE : MAPS OF k- VARIETIES ARE DETERMINED BY THE MAPS ON

CLOSED POINTS . Suppose π : X → Y and π ′ : X → Y are two morphisms of k-
varieties that are the same at the level of closed points (i.e., for each closed point
p ∈ X, π(p) = π ′ (p)). Show that π = π ′ . (This implies that the functor from
the category of “classical varieties over k”, which we won’t define here, to the
category of k-schemes, is fully faithful. Can you generalize this appropriately to
non-algebraically closed fields?)

11.3.C. L ESS IMPORTANT EXERCISE . Show that the line with doubled origin X
(Example 4.4.5) is not separated, by finding two morphisms π : W → X, π ′ : W → X
whose domain of agreement is not a closed subscheme (cf. Proposition 11.2.1(b)).
(Another argument was given above, in Exercise 11.2.I. A fancy argument will be
given in Exercise 13.7.C.)
We now come to the central result of this section.

11.3.2. Reduced-to-Separated Theorem. — Two S-morphisms π : U → Z, π ′ : U → Z

from a reduced scheme to a separated S-scheme agreeing on a dense open subset of U are the
same.

Proof. Let V be the locus where π and π ′ agree. It is a closed subscheme of U (by
Exercise 11.3.A), which contains a dense open set. But the only closed subscheme
of a reduced scheme U whose underlying set is dense is all of U. (Do you see
 © 2024 Ravi Vakil. Published by Princeton University Press. 323

why this last statement is true? Can you give a counterexample if the reducedness
hypothesis is removed?) □

11.3.3. Consequence 1. Hence (as X is reduced and Y is separated) if we have two

morphisms from open subsets of X to Y, say π : U → Y and π ′ : V → Y, and they
agree on a dense open subset Z of U ∩ V, then they necessarily agree on U ∩ V.
Consequence 2. A rational map has a largest domain of definition on which
π : U 99K Y is a morphism, which is the union of all the domains of definition.
In particular, a rational function on a reduced scheme has a largest domain of
definition. For example, the domain of definition of A2k 99K P1k given by (x, y) 7→
[x, y] has domain of definition A2k \ {(0, 0)} (cf. §7.5.3). This partially extends the
definition of the domain of a rational function on a locally Noetherian scheme
(Definition 6.6.36). The complement of the domain of definition is called the
indeterminacy locus (a phrase used earlier in §6.6.36), and its points are sometimes
called fundamental points of the rational map, although we won’t use these
phrases. (We will see in Exercise 22.4.L that a rational map to a projective scheme
can be upgraded to an honest morphism by “blowing up” a scheme-theoretic
version of the locus of indeterminacy.)

11.3.D. E XERCISE . Show that the Reduced-to-Separated Theorem 11.3.2 is false

if we give up reducedness of the source or separatedness of the target. Here are
some possibilities. For the first, consider the two maps from Spec k[x, y]/(y2 , xy)
to Spec k[t], where we take π given by t 7→ x and π ′ given by t 7→ x + y; π and π ′
agree on the distinguished open set D(x). For the second, consider the two maps
from Spec k[t] to the line with the doubled origin, one of which maps to the “upper
half”, and one of which maps to the “lower half”. These two morphisms agree on
the dense open set D(t), see Figure 11.4.

F IGURE 11.4. Two different maps to a nonseparated scheme agree-

ing on a dense open set (see Exercise 11.3.D)

11.3.4. Graphs of rational maps. (Graphs of morphisms were defined in §11.2.5.) If X

is reduced and Y is separated, define the graph Γπ of a rational map π : X 99K Y as
follows. Let (U, π ′ ) be any representative of this rational map (so π ′ : U → Y is a
morphism). Let Γπ be the scheme-theoretic closure of Γπ ′ ,→ U × Y ,→ X × Y, where
the first map is a closed embedding (Proposition 11.2.6), and the second is an open
embedding. The product here should be taken in the category you are working in.
For example, if you are working with k-schemes, the fibered product should be
taken over k.
324 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

11.3.E. E XERCISE . Show that the graph of a rational map π : X 99K Y is independent
of the choice of representative of π. Hint: Suppose ξ ′ : U → Y and ξ : V → Y are
two representatives of π. Reduce to the case where V is the domain of definition of
π (§11.3.3), and ξ ′ = ξ|U . Reduce to the case V = X. Show an isomorphism Γπ = ∼ X,
∼
and Γξ|U = U. (Remark: the separatedness of Y is not necessary.)
In analogy with graphs of morphisms, the following diagram of a graph of a
rational map π can be useful (cf. Figure 11.2).
cl. emb./
ΓπO  X×Y
pr2
γπ
pr1
| "
X
π / Y.

11.3.F. E XERCISE ( THE BLOW- UP OF THE AFFINE PLANE AS THE GRAPH OF A RATIO -
NAL MAP ). Consider the rational map A2 k 99K Pk given by (x, y) 7→ [x, y]. Show
1

that this rational map cannot be extended over the origin. (A similar argument
arises in Exercise 7.5.J on the Cremona transformation.) Show that the graph of the
rational map is the morphism (the blow-up) described in Exercise 10.3.F. (When
we define blow-ups in general, we will see that they are often graphs of rational
maps, see Exercise 22.4.M.)

11.3.5. Variations.
Variations of the short proof of Theorem 11.3.2 yield other useful results. Exer-
cise 11.3.B is one example. Here is another.

11.3.G. E XERCISE ( MAPS TO A SEPARATED SCHEME CAN BE EXTENDED OVER AN

EFFECTIVE C ARTIER DIVISOR IN AT MOST ONE WAY ). Suppose X is a Z-scheme
(not necessarily reduced!), and Y is a separated Z-scheme. Suppose further that D
is an effective Cartier divisor on X. Show that any Z-morphism X \ D → Y can be
extended in at most one way to a Z-morphism X → Y. (Hint: reduce to the case
where X = Spec A, and D is the vanishing scheme of t ∈ A. Reduce to showing
that the scheme-theoretic image of D(t) in X is all of X. Show this by showing that
A → At is an inclusion.)
As noted in §7.5.2, rational maps can be defined from any X that has associated
points to any Y. The Reduced-to-Separated Theorem 11.3.2 can be extended to this
setting, as follows.

11.3.H. E XERCISE ( THE “A SSOCIATED - TO -S EPARATED T HEOREM ”). Prove that two
S-morphisms π and π ′ from a locally Noetherian scheme U to a separated S-scheme
Z, agreeing on an open subset containing the associated points of U, are the same.

11.4 Proper morphisms

A map of topological spaces (also known as a continuous map!) is said to be
proper if the preimage of any compact set is compact. Properness of morphisms is an
analogous property. For example, a variety over C will be proper if it is compact
in the classical topology (see [Se3, §7]). Alternatively, we will see that projective
 © 2024 Ravi Vakil. Published by Princeton University Press. 325

A-schemes are proper over A — so this is a nice property satisfied by projective

schemes, which also is convenient to work with.
Recall (§8.3.5) that a (continuous) map of topological spaces π : X → Y is closed
if for each closed subset S ⊂ X, π(S) is also closed. A morphism of schemes is
closed if the underlying continuous map is closed. We say that a morphism of
schemes π : X → Y is universally closed if for every morphism Z → Y, the induced
morphism Z ×Y X → Z is closed. In other words, a morphism is universally closed
if it remains closed under any base change. By now you will automatically expect
the following exercise.

11.4.A. E ASY E XERCISE . Show that the class of universally closed morphisms is
“reasonable” in the sense of §8.1.
To motivate the definition of properness for schemes, we remark that a continu-
ous map π : X → Y of locally compact Hausdorff spaces which have countable bases
for their topologies is universally closed if and only if it is proper (i.e., preimages of
compact subsets are compact). You are welcome to prove this as an exercise.

11.4.1. Definition. A morphism π : X → Y is proper if it is separated, finite type, and

universally closed. A scheme X is often said to be proper if some implicit structure
morphism is proper. For example, a k-scheme X is often described as proper if
X → Spec k is proper. (A k-scheme is often said to be complete if it is proper. We
will not use this terminology.) If A is a ring, one often says that an A-scheme is
proper over A if it is proper over Spec A.
Let’s try this idea out in practice. We expect that A1C → Spec C is not proper,
because the complex manifold corresponding to A1C is not compact. However, note
that this map is separated (it is a map of affine schemes), finite type, and (trivially)
closed. So the “universally” is what matters here.

11.4.B. E ASY E XERCISE . Show that π : A1C → Spec C is a closed map, but not
universally closed. Hence π is not proper. Possible hint: Consider a well-chosen
map such as A1C ×C P1C → P1C . (See Figure 19.1 for another finite type, separated,
closed morphism that is not proper. Showing that morphism is not proper requires
more creativity.)

11.4.2. Example. As a first example: closed embeddings are proper. They are
clearly separated, as affine morphisms are separated, Proposition 11.2.4. They are
finite type. After base change, they remain closed embeddings (§10.2.2), and closed
embeddings are always closed. This easily extends further as follows.

11.4.3. Proposition. — Finite morphisms are proper.

Proof. Finite morphisms are separated (as they are affine by definition, and affine
morphisms are separated, Proposition 11.2.4), and finite type (basically because
finite modules over a ring are automatically finitely generated). To show that finite
morphisms are closed after any base change, we note that they remain finite after
any base change (finiteness is preserved by base change, Exercise 10.4.D(d)), and
finite morphisms are closed (Exercise 8.3.L). □

11.4.4. Proposition. —
326 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

(a) The notion of “proper morphism” is stable under base change.

(b) The notion of “proper morphism” is local on the target (i.e., π : X → Y is proper
if and only if for any affine open cover Ui → Y, π−1 (Ui ) → Ui is proper). Note
that the “only if” direction follows from (a) — consider base change by Ui ,→ Y.
(c) The notion of “proper morphism” is preserved by composition.
(d) The product of two proper morphisms is proper: if π : X → Y and π ′ : X ′ → Y ′ are
proper, where all morphisms are morphisms of Z-schemes, then π×π ′ : X×Z X ′ →
Y ×Z Y ′ is proper.
(e) Suppose

(11.4.4.1) X
π /Y

τ ρ
 
Z
is a commutative diagram, and τ is proper, and ρ is separated. Then π is proper.
A sample application of (e): a morphism (over Spec k) from a proper k-scheme
to a separated k-scheme is always proper.

Proof. Parts (a) through (c) say that proper morphisms form a “reasonable” class
of morphisms in the sense of §8.1. So they follow because separated morphisms,
finite type morphisms, and universal closed morphisms are all reasonable classes.
Then (d) follows from (a) and (c) by Exercise 8.1.A. Closed embeddings are proper
(Example 11.4.2), so (e) follows from the Cancellation Theorem 11.1.1 for proper
morphisms. □

11.4.C. E XERCISE (“ IMAGES OF PROPER SCHEMES ARE PROPER ”). Suppose in the
diagram
X
π /Y

τ ρ
 
Z
that τ is proper and ρ is separated and finite type. Show that the scheme-theoretic
image of X under π is a proper Z-scheme. (Thus properness in algebraic geometry
behaves like properness in topology.)
We now come to the most important example of proper morphisms.

11.4.5. Theorem. — Projective A-schemes are proper over A.

(As finite morphisms to Spec A are projective A-schemes, Exercise 8.3.I, The-
orem 11.4.5 can be used to give a second proof that finite morphisms are proper,
Proposition 11.4.3.)

Proof. The structure morphism of a projective A-scheme X → Spec A factors as

a closed embedding followed by Pn A → Spec A. Closed embeddings are proper
(Example 11.4.2), and compositions of proper morphisms are proper (Proposi-
A → Spec A is proper. We have already
tion 11.4.4), so it suffices to show that Pn
seen that this morphism is finite type (Easy Exercise 5.3.D) and separated (Propo-
A → Spec A is universally closed. As
sition 11.2.8), so it suffices to show that Pn
 © 2024 Ravi Vakil. Published by Princeton University Press. 327

PnA = PZ ×Z Spec A, it suffices to show that PX := PZ ×Z X → X is closed for

n n n

any scheme X. But the property of being closed is local on the target on X, so
B → Spec B is
by covering X with affine open subsets, it suffices to show that Pn
closed for all rings B. This is the Fundamental Theorem of Elimination Theory
(Theorem 8.4.10). □

11.4.6. Remark: “Reasonable” proper schemes are projective. It is not easy to come up
with an example of an A-scheme that is proper but not projective! Over a field,
all proper curves are projective (see Remark 19.2.I), and all smooth surfaces are
projective (see [Ba, Thm. 1.28] for a proof of this theorem of Zariski; smoothness of
course is not yet defined). We will meet a first example of a proper but not projective
variety (a singular threefold) in §15.2.11. We will later see an example of a proper
nonprojective surface in §19.11.11, and a simpler one in Exercise 20.2.G. Once we
know about flatness, we will see Hironaka’s example of a proper nonprojective
irreducible smooth threefold over C (§24.7.6).

11.4.7. Functions on connected reduced proper k-schemes must be constant.

As an enlightening application of these ideas, we show that if X is a connected
reduced proper k-scheme where k = k, then Γ (X, OX ) = k. The analogous fact
for connected compact complex manifolds uses the maximum principle. We saw
this in the special case X = Pn in Exercise 4.4.E; it will be vastly generalized by
Grothendieck’s Coherence Theorem 18.9.1.
Suppose f ∈ Γ (X, OX ) (f is a function on X). This is the same as a map π : X →
A1k (Exercise 7.3.G, discussed further in §7.6.1). Let π ′ be the composition of π
with the open embedding A1 ,→ P1 . By Proposition 11.4.4(e), π ′ is proper, and
in particular closed. As X is connected, the image of π ′ is as well. Thus the set-
theoretic image of π ′ must be either a closed point, or all of P1 . But the set-theoretic
image of π ′ lies in A1 , so it must be a closed point p (which we identify with an
element of k).
By Corollary 9.4.5, the support of the scheme-theoretic image of π is the closed
point p. By Exercise 9.4.A, the scheme-theoretic image is precisely p (with the
reduced structure). Thus π can be interpreted as the structure map to Spec k,
followed by a closed embedding to A1 identifying Spec k with p. You should be
able to verify that this is the map to A1 corresponding to the constant function
f = p.
(What are counterexamples if different hypotheses are relaxed? The condition
k = k cannot be ignored — if k = R, then X = Spec C is a connected reduced proper
projective k-scheme, with Γ (X, OX ) = k2 . Interesting fact: x2 = 0 in P2k shows that
even nonreduced proper k-schemes can have only constants as functions; see §5.2.3
and Exercise 18.6.U.)

11.4.8. Facts (not yet proved) that may help you correctly think about finiteness.
The following facts may shed some light on the notion of finiteness. We will
prove them later.
A morphism of locally Noetherian schemes is finite if and only if it is proper
and affine, if and only if it is proper and quasifinite (Theorem 28.5.2). We have
proved parts of this statement, but we will only finish the proof once we know
Zariski’s Main Theorem, cf. §8.3.12).
328 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

As an application: quasifinite morphisms from proper schemes to separated

schemes are finite. Here is why: suppose π : X → Y is a quasifinite morphism over
Z, where X is proper over Z. Then by the Cancellation Theorem 11.1.1 for proper
morphisms, X → Y is proper. Hence as π is quasifinite and proper, π is finite.
As an explicit example, consider the map π : P1k → P1k given by [x, y] 7→
[f(x, y), g(x, y)], where f and g are homogeneous polynomials of the same de-
gree with no common roots in P1 . The fibers are finite, and π is proper (from the
Cancellation Theorem 11.1.1 for proper morphisms, as discussed after the statement
of Theorem 11.4.4), so π is finite. This could be checked directly as well, but now
we can save ourselves the annoyance.

11.4.9. ⋆⋆ Group varieties. The rest of §11.4 is double-starred, and we will

also throughout work in the category of varieties over a field k. We briefly discuss
group varieties, mainly because we can, not because we have anything profound
to say. We discuss them right now only because properness gives an unexpected
consequence, that proper group varieties are abelian (thanks to the surprisingly
simple Rigidity Lemma 11.4.12).
As discussed in §11.2.9, now that we have the category of varieties over k, we
immediately have the notion of a group variety (thanks to §7.6.4 on group objects
in any category). An algebraic group (over k) is a smooth group variety. Examples
include GLn (Exercise 7.6.N(b)), which includes Gm as a special case, and SLn
(Exercise 7.6.Q).

11.4.10. Side Remarks (that we won’t prove or use). Group varieties are automatically
smooth (hence algebraic groups) in characteristic 0. Group varieties are not neces-
sarily smooth in positive characteristic. (Let k be an imperfect field of characteristic
p, and choose t ∈ k without a pth root in k. Consider the “closed sub-group scheme”
xp = typ of the “additive group” G2a = Spec k[x, y]. This group scheme is reduced,
but not geometrically reduced; and not smooth, or even regular.) Algebraic groups
are automatically quasiprojective. An algebraic group G is affine (as a scheme)
if and only if it admits a closed immersion into GLn for some n ≥ 0, such that
G → GLn is a homomorphism of group schemes (Exercise 7.6.N(a)). For further
discussion on these and other issues, see [P2, §5].

11.4.11. Definition: Abelian varieties. We can now define one of the most important
classes of varieties: abelian varieties. The bad news is that we have no real example
yet. We will later see elliptic curves as an example (§19.10), but we will not meet
anything more exotic than that. Still, it is pleasant to know that we can make the
definition this early.
An abelian variety over a field k is an algebraic group that is geometrically
integral and projective (over k). It turns out that the analytification of any abelian
variety over C is a complex torus, i.e., Cn /Λ, where Λ is a lattice (of rank 2n).
(The key idea: connected compact complex Lie groups are commutative, see Re-
mark 11.4.14; the universal cover is a simply connected commutative complex Lie
group, and thus Cn . See [BL, Lem. 1.1.1].)
We now show that abelian varieties are abelian, i.e., abelian varieties are com-
mutative algebraic group varieties. (This is less tautological than it sounds! The
adjective “abelian” is used in two different senses; what they have in common is that
they are derived from the name of Niels Henrik Abel.) As algebraic groups need
 © 2024 Ravi Vakil. Published by Princeton University Press. 329

not be commutative (witness GLn ), it is somehow surprising that commutativity

could be forced by “compactness”.
The key result is the following fact, beautiful in its own right.

11.4.12. Rigidity Lemma. — Let X, Y, and Z be varieties, where X is proper and

geometrically integral, with a k-valued point p, and Y is irreducible. Let pr : X × Y → Y
be the projection, and let α : X × Y → Z be any morphism. Suppose q ∈ Y is such that α is
constant on X × {q}; say α(X × {q}) = r ∈ Z.
(a) Then there is a morphism ψ : Y → Z such that α = ψ ◦ pr. (In particular, for every
q ′ ∈ Y, α is constant on X × {q ′ }.)
(b) If α is also constant on {p} × Y, then α is a constant function on all of X × Y.

< ZN

X × {q} / X×Y
V
ψ
pr σp
 
q /Y

The hypotheses can be relaxed in a number of ways, but this version suffices
for our purposes. If you have not read §10.5.1, where geometric integrality is
discussed, you need not worry too much; this is automatic if k = k and X is integral
(Hard Fact 10.5.3). Even if you have read §10.5.1, it is helpful to first read this proof
under the assumption that k = k, to avoid being distracted from the main idea by
geometric points. But of course all of §11.4.9 is double-starred, so you shouldn’t be
reading this anyway.

11.4.D. E XERCISE . Show that the properness hypothesis on X is necessary. (Can

you make this accord with your intuition?)
Proof of the Rigidity Lemma 11.4.12. Define β : X × Y → Z by β(x, y) = α(p, y). (More
precisely, if σp is the section of X × Y → Y pulled back from Spec k 7→ p, then
β = α ◦ σp ◦ pr.) We will show that β(x, y) = α(x, y). This will imply (a) (take
ψ = α ◦ σp ), which in turn immediately implies (b).
Proposition 10.5.19 (geometrically reduced times reduced is reduced) implies
that X × Y is reduced. Exercise 10.5.O (geometrically irreducible times irreducible is
irreducible) implies that X × Y is irreducible. As Z is separated, by Exercise 11.3.A,
it suffices to show that α = β on a nonempty (hence dense) open subset of X × Y.
Let U ⊂ Z be an affine open neighborhood of r. Then α−1 (Z \ U) is a closed sub-
set of X × Y. As the projection pr is proper (using properness of X, and preservation
of properness under base change, Proposition 11.4.4(a)), we have pr(α−1 (Z \ U)) is
closed. Its complement is open, and contains q; let V ⊂ Y be an open neighborhood
of q disjoint from pr(α−1 (Z \ U)).

11.4.E. E XERCISE . Suppose γ : Spec K → V is a geometric point of V. Show that

the fiber XK over Spec K (or more precisely, over γ) is mapped to a point in Z by
330 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

the restriction αK of α to XK . Hint: review §11.4.7.

55 Z
J
αK

βK
XK / X×V α|V

pr
 
Spec K
γ
/V

By Exercise 11.3.B (maps of varieties over an algebraically closed field are

determined by the map of closed points), αK = βK .

11.4.F. E XERCISE . Use Exercise 11.3.A (the fact that the locus where α = β is
a closed subscheme — and represents a certain functor) to show that α = β on
pr−1 (V).
□

11.4.13. Corollary. — Suppose A is an abelian variety, G is a group variety, and

ϕ : A → G is any morphism of varieties. Then ϕ is the composition of a translation and a
homomorphism.

Proof. Let eA and eG be the identity points of A and G respectively. Composing

ϕ with a translation, we may assume that ϕ(eA ) = eG . Consider the morphism
α : A × A → G given by α(a1 , a2 ) = ϕ(a1 a2 )ϕ(a1 )−1 ϕ(a2 )−1 . Then α({eA } × A) =
α(A × {eA }) = {eB }, so by the Rigidity Lemma 11.4.12, α is a constant, and sends
A × A to eB . Thus ϕ(a1 a2 ) = ϕ(a1 )ϕ(a2 ), so ϕ is a homomorphism. □

11.4.G. E XERCISE ( ABELIAN VARIETIES ARE ABELIAN ). Show that an abelian

variety is an abelian group variety. Hint: apply Corollary 11.4.13 to the inversion
morphism i : A → A.

11.4.14. Remark. A similar idea is used to show that connected compact complex
Lie groups are abelian, see for example [BL, Lem. 1.1.1].
 Part IV

“Geometric” properties of schemes

 CHAPTER 12

Dimension

Everyone knows what a curve is, until he has studied enough mathematics to become
confused ...
— F. Klein, [RS, p. 90]
At this point, you know a fair bit about schemes, but there are some fundamen-
tal notions you cannot yet define. In particular, you cannot use the phrase “smooth
surface”, as it involves the notion of dimension and of smoothness. You may be
surprised that we have gotten so far without using these ideas. You may also be
disturbed to find that these notions can be subtle, but you should keep in mind that
they are subtle in all parts of mathematics.
In this chapter, we will address the first notion, that of dimension of schemes.
This should agree with, and generalize, our geometric intuition. Although we
think of dimension as a basic notion in geometry, it is a slippery concept, as it is
throughout mathematics. Even in linear algebra, the definition of dimension of a
vector space is surprising the first time you see it, even though it quickly becomes
second nature. The definition of dimension for manifolds is equally nontrivial.
For example, how do we know that there isn’t an isomorphism between some 2-
dimensional manifold and some 3-dimensional manifold? Your answer will likely
use topology, and hence you should not be surprised that the notion of dimension
is often quite topological in nature.
A caution for those thinking over the complex numbers: our dimensions will
be algebraic, and hence half that of the “real” picture. For example, we will see
very shortly that A1C , which you may picture as the complex numbers (plus one
generic point), has dimension 1.

12.1 Dimension and codimension

12.1.1. Definition(s): dimension. Surprisingly, the right definition is purely topologi-

cal — it just depends on the topological space, and not on the structure sheaf. We
define the dimension of a topological space X (denoted dim X) as the supremum
of lengths of chains of closed irreducible sets, starting the indexing of the closed
irreducible sets with 0. (The dimension may be infinite.) Scholars of the empty
set can take the dimension of the empty set to be −∞. (An analogy from linear
algebra: the dimension of a vector space is the supremum of the length of chains

This material is published by Princeton University Press as The Rising Sea: Foundations of Algebraic
Geometry by Ravi Vakil. This version is free to view and download for personal use only. Not for
redistribution, resale, or use in derivative works. © 2024 Ravi Vakil.

333
334 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

of subspaces.) Define the dimension of a ring as the supremum of the lengths

of the chains of nested prime ideals (where indexing starts at zero). These two
definitions of dimension are sometimes called Krull dimension. (You might think
a Noetherian ring has finite dimension because all chains of prime ideals are finite,
but this isn’t necessarily true — see Exercise 12.1.M.)

12.1.A. E ASY E XERCISE . Show that dim Spec A = dim A. (Hint: Exercise 3.7.F
gives a bijection between irreducible closed subsets of Spec A and prime ideals of
A. It is “inclusion-reversing”.)
The homeomorphism between Spec A and Spec A/N(A) (§3.4.7: the Zariski
topology disregards nilpotents) implies that dim Spec A = dim Spec A/N(A).

12.1.2. Examples. We have identified all the prime ideals of k[t] (they are 0, and
(f(t)) for irreducible polynomials f(t)), Z ((0) and (p)), k (only (0)), and k[x]/(x2 )
(only (x)), so we can quickly check that dim A1k = dim Spec Z = 1, dim Spec k = 0,
dim Spec k[x]/(x2 ) = 0.

12.1.3. We must be careful with the notion of dimension for reducible spaces. If Z is
the union of two closed subsets X and Y, then dim Z = max(dim X, dim Y). Thus
dimension is not a “local” characteristic of a space. This sometimes bothers us, so
we try to only talk about dimensions of irreducible topological spaces. We say a
topological space is pure dimensional or equidimensional (resp., pure dimension
n or equidimensional of dimension n) if each of its irreducible components has
the same dimension (resp., they are all of dimension n). A pure dimension 1 (resp.,
2, n) topological space is said to be a curve (resp., surface, n-fold).

12.1.B. U SEFUL E XERCISE . Show that a scheme has dimension n if and only if it
admits an open cover by affine open subsets of dimension at most n, where equality
is achieved for some affine open subset. Hint: You may find it helpful, here and
later, to show the following. For any topological space X and open subset U ⊂ X,
there is a bijection between irreducible closed subsets of U, and irreducible closed
subsets of X that meet U.

12.1.C. E ASY E XERCISE . Show that a Noetherian scheme of dimension 0 has a finite
number of points.

12.1.D. E ASY E XERCISE . Show that a surjection of integral domains of the same
dimension must be an isomorphism.

12.1.E. E XERCISE ( FIBERS OF INTEGRAL MORPHISMS , PROMISED IN §8.3.8). Sup-

pose π : X → Y is an integral morphism. Show that every (nonempty) fiber of π
has dimension 0. Hint: As integral morphisms are preserved by base change (Exer-
cise 10.4.D(d)), we assume that Y = Spec k. Hence we must show that if ϕ : k → A
is an integral extension, then dim A = 0. Outline of proof: Suppose p ⊂ m are two
prime ideals of A. Mod out by p, so we can assume that A is an integral domain.
I claim that any nonzero element is invertible: Say x ∈ A, and x ̸= 0. Then the
minimal monic polynomial for x has nonzero constant term. But then x is invertible
— recall the coefficients are in a field.
 © 2024 Ravi Vakil. Published by Princeton University Press. 335

12.1.F. I MPORTANT E XERCISE . Show that if π : Spec A → Spec B corresponds to

an integral extension of rings, then dim Spec A = dim Spec B. Hint: show that a
chain of prime ideals downstairs gives a chain upstairs of the same length, by
the Going-Up Theorem (Exercise 8.2.F). Conversely, a chain upstairs gives a chain
downstairs. Use Exercise 12.1.E to show that no two elements of the chain upstairs
go to the same element [q] ∈ Spec B of the chain downstairs.
e → X is the normalization of a scheme (possibly
12.1.G. E XERCISE . Show that if ν : X
in a finite extension of the function field of X), then dim Xe = dim X. Feel free to
assume that X is integral for convenience.

12.1.H. U SEFUL E XERCISE . Suppose X is an affine k-scheme, and K/k is an algebraic

field extension.
(a) Suppose X has pure dimension n. Show that XK := X ×k K has pure dimension
n. (See Exercise 24.5.F for a generalization, which for example removes the affine
hypothesis. Also, see Exercise 12.2.M and Remark 12.2.17 for the fate of possible
generalizations to arbitrary field extensions.) Hint: If X = Spec A, reduce to the case
where A is an integral domain. An irreducible component of XK corresponds to a
minimal prime p of AK := A ⊗k K. Suppose a ∈ ker(A → AK /p). Show that a = 0,
using the fact that a lies in a minimal prime p of AK (and is hence a zerodivisor,
by §6.6.12 combined with §6.6.22), and AK is a free A-module (so multiplication
in AK by a ∈ A is injective if a is nonzero). Thus A → AK /p is injective. Then use
Exercise 12.1.F.
(b) Prove the converse to (a): show that if XK has pure dimension n, then X has
pure dimension n.

12.1.I. E XERCISE . Show that dim Z[x] = 2. (Hint: The prime ideals of Z[x] were
implicitly determined in Exercise 3.2.Q.)

12.1.4. Codimension. Because dimension behaves oddly for disjoint unions, we

need some care when defining codimension, and in using the phrase. For example,
if Y is a closed subset of X, we might define the codimension to be dim X − dim Y,
but this behaves badly. For example, if X is the disjoint union of a point Y and a
curve Z, then dim X − dim Y = 1, but this has nothing to do with the local behavior
of X near Y.
A better definition is as follows. In order to avoid excessive pathology, we de-
fine the codimension of Y in X only when Y is irreducible. (Use extreme caution when
using this word in any other setting.) Define the codimension of an irreducible
subset Y ⊂ X of a topological space as the supremum of lengths of increasing chains
of irreducible closed subsets starting with Y (where indexing starts at 0 — recall
that the closure of an irreducible set is irreducible, Exercise 3.6.B(b)). In particular,
the codimension of a point is the codimension of its closure. The codimension of Y
in X is denoted by codimX Y.
We say that a prime ideal p in a ring has codimension equal to the supremum
of lengths of the chains of decreasing prime ideals starting at p, with indexing
starting at 0. Thus in an integral domain, the ideal (0) has codimension 0; and in Z,
the ideal (23) has codimension 1. Note that the codimension of the prime ideal p in
A is dim Ap (see §3.2.9). (This notion is often called height.) Thus the codimension
of p in A is the codimension of [p] in Spec A.
336 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

(Continuing an analogy with linear algebra: the codimension of a vector sub-

space Y ⊂ X is the supremum of lengths of increasing chains of subspaces starting
with Y. This is a better definition than dim X − dim Y, because it works even when
dim X = ∞. You might prefer to define codimX Y as dim(X/Y); that is analogous
to defining the codimension of p in A as the dimension of Ap — see the previous
paragraph.)

12.1.J. E XERCISE . Show that if Y is an irreducible closed subset of a scheme X, and

η is the generic point of Y, then the codimension of Y is the dimension of the local
ring OX,η .

12.1.5. Notice that Y is codimension 0 in X if it is an irreducible component of X.

Similarly, Y is codimension 1 if it is not an irreducible component, and for every
irreducible component Y ′ it is contained in, there is no irreducible closed subset
strictly between Y and Y ′ . (See Figure 12.1 for examples.) A closed subset all of
whose irreducible components are codimension 1 in some ambient space X is said
to be a hypersurface in X. (Be careful: the word “hypersurface” can be used to
mean slightly different things in algebraic geometry, for entirely reasonable reasons,
see for example Definition 9.3.2.)

F IGURE 12.1. Behavior of codimension

12.1.K. E ASY E XERCISE . If Y is an irreducible closed subset of a scheme X, show

that

(12.1.5.1) codimX Y + dim Y ≤ dim X.

We will soon see that equality always holds if X and Y are irreducible varieties
(Theorem 12.2.9), but equality doesn’t hold in general (§12.3.13).
Warning. The notion of codimension still can behave slightly oddly. For exam-
ple, consider Figure 12.1. (You should think of this as an intuitive sketch.) Here the
total space X has dimension 2, but point p is dimension 0, and codimension 1. We
also have an example of a codimension 2 subset q contained in a codimension 0
subset C with no codimension 1 subset “in between”.
Worse things can happen; we will soon see an example of a closed point in an
irreducible surface that is nonetheless codimension 1, not 2, in §12.3.13. However, for
irreducible varieties this can’t happen, and inequality (12.1.5.1) must be an equality
(Theorem 12.2.9).
 © 2024 Ravi Vakil. Published by Princeton University Press. 337

12.1.6. In unique factorization domains, codimension 1 prime ideals are principal.

We make a short observation, for future reference.

12.1.7. Lemma. — In a unique factorization domain A, all codimension 1 prime ideals are
principal.
This is a first glimpse of the fact that codimension one is rather special — this
theme will continue in §12.3. We will see that the converse of Lemma 12.1.7 holds
as well (when A is a Noetherian integral domain, Proposition 12.3.7).

Proof. Suppose p is a codimension 1 prime. Choose any f ̸= 0 in p, and let g be any

irreducible/prime factor of f that is in p (there is at least one). Then (g) is a nonzero
prime ideal contained in p, so (0) ⊂ (g) ⊂ p. As p is codimension 1, we must have
p = (g), and thus p is principal. □

12.1.8. Local rings of dimension zero.

Dimension zero objects aren’t just points; they have further information, which
might be thought of as some sort of infinitesimal “shape”. We caught a glimpse of
this earlier in §6.5 and Exercise 8.3.G.

12.1.9. Useful Proposition. — Suppose (A, m) is a Noetherian local ring. Then the
following are equivalent.

(i) dim A = 0;
(ii) m = N(A);
(iii) for some n > 1, mn = 0;
(iv) A has finite length;
(v) every descending sequence of ideals of A is eventually constant.

Such a ring is called an Artinian or finite length local ring (Definition 6.5.7).

12.1.L. E XERCISE . Prove Proposition 12.1.9.

12.1.10. A fun but unimportant counterexample. We end this introductory section

with a fun pathology. As a Noetherian ring has no infinite chain of prime ideals,
you may think that Noetherian rings must have finite dimension. Nagata, the
master of counterexamples, shows you otherwise with the following example.

12.1.M. ⋆⋆ E XERCISE : AN INFINITE - DIMENSIONAL N OETHERIAN RING . Let A =

k[x1 , x2 , . . . ]. Choose an increasing sequence of positive integers m1 , m2 , . . . whose
differences are also increasing (mi+1 −mi > mi −mi−1 ). Let pi = (xmi +1 , . . . , xmi+1 )
and S = A \ ∪i pi .
(a) Show that S is a multiplicative set.
(b) Show that S−1 pi in S−1 A is the largest prime ideal in a chain of prime ideals of
length mi+1 − mi . Hence conclude that dim S−1 A = ∞.
(c) Suppose B is a ring such that (i) for every maximal ideal m, Bm is Noetherian, and
(ii) every nonzero b ∈ B is contained in finitely many maximal ideals. Show that B
is Noetherian. (One possible approach: show that for any x1 , x2 , . . . , (x1 , x2 , . . . ) is
finitely generated.)
(d) Use (c) to show that S−1 A is Noetherian.
338 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

12.1.11. Remark: Noetherian local rings have finite dimension. However, we shall see
in Exercise 12.3.K(a) that Noetherian local rings always have finite dimension. (This
requires a surprisingly hard fact, Krull’s Height Theorem 12.3.10.) Thus points of
locally Noetherian schemes always have finite codimension.

12.2 Dimension, transcendence degree, and Noether normalization

We now give a powerful alternative interpretation for dimension for irreducible

varieties, in terms of transcendence degree. The proof will involve a classical
construction, Noether normalization, which will be useful in other ways as well. In
case you haven’t seen transcendence theory, here is a lightning introduction.

12.2.A. E XERCISE / DEFINITION . Recall that an element of a field extension E/F

is algebraic over F if it is integral over F. Recall also that a field extension E/F is an
algebraic extension if it is an integral extension (if all elements of E are algebraic
over F, cf. §8.2). The composition of two algebraic extensions is algebraic, by
Exercise 8.2.C. If E/F is a field extension, and F ′ and F ′′ are two intermediate field
extensions, then we write F ′ ∼ F ′′ if F ′ F ′′ is algebraic over both F ′ and F ′′ . Here
F ′ F ′′ is the compositum of F ′ and F ′′ , the smallest field extension in E containing F ′
and F ′′ .
(a) Show that ∼ is an equivalence relation on subextensions of E/F. A transcendence
basis of E/F is a set of elements {xi } that are algebraically independent over F (there
is no nontrivial polynomial relation among the xi with coefficients in F) such that
F({xi }) ∼ E.
(b) Show that if E/F has two transcendence bases, and one has cardinality n, then
both have cardinality n. (Hint: show that you can substitute elements from the one
basis into the other one at a time.) The size of any transcendence basis is called
the transcendence degree (which may be ∞), and is denoted trdeg. Any finitely
generated field extension necessarily has finite transcendence degree. (Remark: See
Theorem 10.5.17 for a related result.)

12.2.1. Theorem (dimension = transcendence degree). — Suppose A is a finitely

generated domain over a field k (i.e., a finitely generated k-algebra that is an integral
domain). Then dim Spec A = trdeg K(A)/k. Hence if X is an irreducible k-variety, then
dim X = trdeg K(X)/k.
We will prove Theorem 12.2.1 shortly (§12.2.7). We first show that it is useful by
giving some immediate consequences. We seem to have immediately dim An k = n.
However, our proof of Theorem 12.2.1 will go through this fact, so it isn’t really a
consequence.
A more substantive consequence is the following. If X is an irreducible k-
variety, then dim X is the transcendence degree of the function field (the stalk at the
generic point) OX,η over k. Thus (as the generic point lies in all nonempty open
sets) the dimension can be computed in any open set of X. (Warning: this is false
without the finite type hypothesis, even in quite reasonable circumstances: let X
be the two-point space Spec k[x](x) , and let U consist of only the generic point, see
Exercise 3.4.K.)
 © 2024 Ravi Vakil. Published by Princeton University Press. 339

Another consequence is a second proof of the Nullstellensatz 3.2.6.

12.2.B. E XERCISE : THE N ULLSTELLENSATZ FROM DIMENSION THEORY. Suppose

A = k[x1 , . . . , xn ]/I. Show that the residue field of any maximal ideal of A is a finite
extension of k. (Hint: the maximal ideals correspond to dimension 0 points, which
correspond to transcendence degree 0 finitely generated extensions of k, i.e., finite
extensions of k.)
Yet another consequence is geometrically believable.

12.2.C. E XERCISE . If π : X → Y is a dominant morphism of irreducible k-varieties,

then dim X ≥ dim Y. (This is false more generally: consider the inclusion of the
generic point into an irreducible curve.)

12.2.D. E XERCISE . Suppose f1 (x1 , . . . , xn ), . . . , fm (x1 , . . . , xn ) ∈ k[x1 , . . . , xn ] are

m polynomials in n variables over a field, with m > n. Show that the polynomials
f1 , . . . , fm are algebraically dependent. (Remark: if m = n, how can you tell if the fi
are algebraically dependent? There is a remarkably simple answer in characteristic
0: they are algebraically dependent if and only if the determinant of the Jacobian
matrix is the zero polynomial, see Exercise 21.6.C! This is false in characteristic p:
if fi (x1 , . . . , xn ) = xp
i , then the fi are algebraically independent, yet the Jacobian
matrix is the zero matrix.)

12.2.E. E XERCISE ( PRACTICE WITH THE CONCEPT ). Show that the three equations

wz − xy = 0, wy − x2 = 0, xz − y2 = 0

cut out an integral surface S in A4k . (You may recognize these equations from
Exercises 3.6.F and 9.3.A.) You might expect S to be a curve, because it is cut out by
three equations in four-space. One of many ways to proceed: cut S into pieces. For
example, show that D(w) = ∼ Spec k[x, w]w . (You may recognize S as the affine cone
over the twisted cubic. The twisted cubic was defined in Exercise 9.3.A.) It turns
out that you need three equations to cut out this surface. The first equation cuts out
a threefold in A4k (by Krull’s Principal Ideal Theorem 12.3.3, which we will meet
soon). The second equation cuts out a surface: our surface, along with another
surface. The third equation cuts out our surface, and removes the “extraneous
component”. One last aside: notice once again that the cone over the quadric
surface k[w, x, y, z]/(wz − xy) makes an appearance.

12.2.2. Definition: degree of a rational map of irreducible varieties. If π : X 99K Y is

a dominant rational map of irreducible (hence integral) k-varieties of the same
dimension, the degree of the field extension is called the degree of the rational
map. This readily extends if X is reducible: we add up the degrees on each of the
components of X. If π is a rational map of integral affine k-varieties of the same
dimension that is not dominant we say the degree is 0. We will interpret this degree
in terms of counting preimages of general points of Y in §24.5.14. Note that degree
is multiplicative under composition: if ρ : Y 99K Z is a rational map of integral
k-varieties of the same dimension, then deg(ρ ◦ π) = deg(ρ) deg(π), as degrees of
field extensions are multiplicative in towers.

12.2.3. Noether normalization.

340 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Our proof of Theorem 12.2.1 will use another important classical notion, Noether
normalization.

12.2.4. Noether Normalization Lemma. — Suppose A is an integral domain, finitely

generated over a field k. If trdeg K(A)/k = n, then there are elements x1 , . . . , xn ∈ A,
algebraically independent over k, such that A is a finite (hence integral by Corollary 8.2.2)
extension of k[x1 , . . . , xn ].
The geometric content behind this result is that given any integral finite type
affine k-scheme X, we can find a surjective finite morphism X → An k , where n is
the transcendence degree of the function field of X (over k). Surjectivity follows
from the Lying Over Theorem 8.2.5, in particular Exercise 12.1.F. This interpretation
is sometimes called geometric Noether normalization. (See Remark 29.3.7 for a
projective version of Noether normalization.)

12.2.5. Nagata’s proof of the Noether Normalization Lemma 12.2.4. Suppose we can
write A = k[y1 , . . . , ym ]/p, i.e., that A can be chosen to have m generators. Note
that m ≥ n. We show the result by induction on m. The base case m = n is
immediate.
Assume now that m > n, and that we have proved the result for smaller m. We
will find m−1 elements z1 , . . . , zm−1 of A such that A is finite over k[z1 , . . . , zm−1 ]/q
(i.e., the subring of A generated by z1 , . . . , zm−1 , where the zi satisfy the relations
given by the ideal q). Then by the inductive hypothesis, k[z1 , . . . , zm−1 ]/q is finite
over some k[x1 , . . . , xn ], and A is finite over k[z1 , . . . , zm−1 ], so by Exercise 8.3.H,
A is finite over k[x1 , . . . , xn ].

A = k[y1 , . . . , ym ]/p
finite

k[z1 , . . . , zm−1 ]/q

finite

k[x1 , . . . , xn ]

As y1 , . . . , ym are algebraically dependent in A, there is some nonzero algebraic

relation f(y1 , . . . , ym ) = 0 among them (where f is a polynomial in m variables
over k).
Let z1 = y1 − yrm1 , z2 = y2 − yrm2 , . . . , zm−1 = ym−1 − yrmm−1 , where r1 , . . . ,
rm−1 are positive integers to be chosen shortly. Then

f(z1 + yrm1 , z2 + yrm2 , . . . , zm−1 + yrmm−1 , ym ) = 0.

Then upon expanding this out, each monomial in f (as a polynomial in m variables)
will yield a single term that is a constant times a power of ym (with no zi factors).
By choosing the ri so that 0 ≪ r1 ≪ r2 ≪ · · · ≪ rm−1 , we can ensure that the
powers of ym appearing are all distinct, and so that in particular there is a leading
term yN m , and all other terms (including those with factors of zi ) are of smaller
degree in ym . Thus we have described an integral dependence of ym on z1 , . . . ,
zm−1 as desired. □
 © 2024 Ravi Vakil. Published by Princeton University Press. 341

12.2.6. The geometry behind Nagata’s proof. Here is the geometric intuition behind
Nagata’s argument. Suppose we have an n-dimensional variety in Am k with n < m,
for example xy = 1 in A2 . One approach is to hope the projection to a hyperplane
is a finite morphism. In the case of xy = 1, if we projected to the x-axis, it wouldn’t
be finite, roughly speaking because the asymptote x = 0 prevents the map from
being closed (cf. Exercise 8.3.K). If we instead projected to a random line, we
might hope that we would get rid of this problem, and indeed we usually can:
this problem arises for only a finite number of directions. But we might have a
problem if the field were finite: perhaps the finite number of directions in which
to project each have a problem. (You can show that if k is an infinite field, then
the substitution in the above proof zi = yi − yrmi can be replaced by the linear
substitution zi = yi − ai ym where ai ∈ k, and that for a nonempty Zariski-open
choice of ai , we indeed obtain a finite morphism.) Nagata’s trick in general is to
“jiggle” the variables in a nonlinear way, and this jiggling kills the nonfiniteness of
the map.

12.2.F. E XERCISE ( DIMENSION IS ADDITIVE FOR PRODUCTS OF FINITE TYPE k-

SCHEMES ). If X and Y are irreducible k-varieties, show that dim X ×k Y =
dim X + dim Y. Hint: If we had surjective finite morphisms X → Adim k
X
and
Y → Ak , we could construct a surjective finite morphism X ×k Y → Ak
dim Y dim X+dim Y
.
(Warning: this can be false without the finite type assumption, see Remark 12.2.17.)
In particular, we have dim A[x] = dim A + 1 when A is finitely generated over
a field. (This is true more generally when A is Noetherian, although we will not
prove this.)

12.2.G. L ESS IMPORTANT ( BUT FUN ) E XERCISE .

(a) Suppose X ⊂ Pn is an irreducible projective k-variety. Show that the affine cone
and projective cone over X both have dimension dim X + 1.
(b) Show that a prime ideal of k[x0 , . . . , xn ] which is minimal over a homogeneous
ideal is also homogeneous.
(c) Suppose p ⊂ q are two homogeneous prime ideals of k[x0 , . . . , xn ] such that
there is no homogeneous prime ideal strictly between the two. Show that there is
no prime ideal (not necessarily homogeneous) strictly between the two.

12.2.7. Proof of Theorem 12.2.1 on dimension and transcendence degree. Suppose

X is an integral affine k-scheme of finite type. We show that dim X equals the
transcendence degree n of its function field, by induction on n. (The idea is that we
reduce from X to An to a hypersurface in An to An−1 .) Assume the result is known
for all transcendence degrees less than n.
By the Noether Normalization Lemma 12.2.4, there exists a surjective finite
morphism X → An n
k . By Exercise 12.1.F, dim X = dim Ak . If n = 0, we are done, as
0
dim Ak = 0.
We now show that dim An n
k = n for n > 0, by induction. Clearly dim Ak ≥ n,
as we can describe a chain of irreducible subsets of length n: if x1 , . . . , xn are
coordinates on An , consider the chain of ideals
(0) ⊂ (x1 ) ⊂ · · · ⊂ (x1 , . . . , xn )
in k[x1 , . . . , xn ]. Suppose we have a chain of prime ideals of length at least n + 1:
(0) = p0 ⊂ · · · ⊂ pm .
342 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Choose any nonzero element g of p1 , and let f be any irreducible factor of g lying
in p1 . Then replace p1 by (f). (Of course, p1 may have been (f) to begin with.) Then
K(k[x1 , . . . , xn ]/(f(x1 , . . . , xn ))) has transcendence degree n − 1, so by induction,

dim k[x1 , . . . , xn ]/(f) = n − 1,

so dim k[x1 , . . . , xn ] ≤ n, completing the proof. □

12.2.8. Codimension is the difference of dimensions for irreducible varieties.

Noether normalization will help us show that codimension is the difference of
dimensions for irreducible varieties, i.e., that the inequality (12.1.5.1) is always an
equality.

12.2.9. Theorem. — Suppose X is a pure-dimensional k-scheme locally of finite type

(for example, an irreducible k-variety), Y is an irreducible closed subset, and η is the
generic point of Y. Then dim Y + dim OX,η = dim X. Hence by Exercise 12.1.J, dim Y +
codimX Y = dim X — inequality (12.1.5.1) is always an equality.
Proving this will give us an excuse to introduce some useful notions, such
as the Going-Down Theorem for finite extensions of integrally closed domains
(Theorem 12.2.12). Before we begin the proof, we give an algebraic translation.

12.2.H. E XERCISE . A ring A is called catenary if for every nested pair of prime
ideals p ⊂ q ⊂ A, every strictly increasing chain of prime ideals “from p to q” is
contained in a maximal such chain, and all maximal chains have the same (finite)
length. (We will not use this term in any serious way later.) Show that if A is a
localization of a finitely generated ring over a field k, then A is catenary.

12.2.10. Remark. Most rings arising naturally in algebraic geometry are catenary.
Important examples include: localizations of finitely generated Z-algebras; com-
plete Noetherian local rings; Dedekind domains (§13.5.11); and Cohen-Macaulay
rings (see §26.2.13). It is hard to give an example of a noncatenary ring; see for
example [Stacks, tag 02JE] or [He].

12.2.11. Proof of Theorem 12.2.9.

F IGURE 12.2. Key step in proof of Theorem 12.2.9

 © 2024 Ravi Vakil. Published by Princeton University Press. 343

12.2.I. E XERCISE . Reduce the proof of Theorem 12.2.9 to the following problem.
If X is an irreducible affine k-variety and Z is a closed irreducible subset maximal
among those smaller than X (the only larger closed irreducible subset is X), then
dim Z = dim X − 1.
Let d = dim X for convenience. By Noether normalization 12.2.4, we have a
finite morphism π : X → Ad corresponding to a finite extension of rings. Then π(Z)
is an irreducible closed subset of Ad (finite morphisms are closed, Exercise 8.3.L).

12.2.J. E XERCISE . Show that it suffices to show that π(Z) is a hypersurface. (Hint:
the dimension of any hypersurface is d − 1 by Theorem 12.2.1 on dimension and
transcendence degree. Exercise 12.1.F implies that dim π−1 (π(Z)) = dim π(Z). But
be careful: Z is not π−1 (π(Z)) in general.)
Now if π(Z) is not a hypersurface, then it is properly contained in an irreducible
hypersurface H, so by the Going-Down Theorem 12.2.12 for finite extensions of inte-
grally closed domains (which we shall now prove), there is some closed irreducible
subset Z ′ of X properly containing Z, contradicting the maximality of Z. □

12.2.12. Theorem (Going-Down Theorem for finite extensions of integrally closed

domains). — Suppose ϕ : B ,→ A is a finite extension of rings (so A is a finite B-module),
B is an integrally closed domain, and A is an integral domain. Then given nested prime
ideals q ⊂ q ′ of B, and a prime p ′ of A lying over q ′ (i.e., p ′ ∩ B = q ′ ), there exists a prime
p of A contained in p ′ , lying over q.
As usual, you should sketch a geometric picture of the statement. This theorem
is usually stated about extending a chain of ideals, in the same way as the Going-
Up Theorem (Exercise 8.2.F), and you may want to think this through. (Another
Going-Down Theorem, for flat morphisms, will be given in Exercise 24.5.E.)
This theorem is true more generally with “finite” replaced by “integral”; see [E,
p. 291] (“Completion of the proof of 13.9”) for the extension of Theorem 12.2.12, or
else see [AtM, Thm. 5.16] or [Mat1, Thm. 5(v)] for completely different proofs. See
[E, Fig. 10.4] for an example (in the form of a picture) of why the “integrally closed”
hypothesis on B cannot be removed.
In the course of the proof, we will use the following easy fact. We could prove
it now, but we leave it until Exercise 12.3.E, because the proof uses a trick arising in
Exercise 12.3.D.

12.2.13. Proposition (prime avoidance). — Suppose p1 , . . . , pn are prime ideals of a

ring A, and I is another ideal of A not contained in any pi . Then I is not contained in ∪pi :
there is an element f ∈ I not in any of the pi .
(Can you give a geometric interpretation of this result? Can you figure out why
it is called “prime avoidance”?)

12.2.14. Proof of Theorem 12.2.12 (Going-Down Theorem for finite extensions of integrally
closed domains). The proof uses Galois theory. Let L be the normal closure of
K(A)/K(B) (the smallest subfield of K(B) containing K(A), and that is mapped to
itself by any automorphism over K(B)/K(B)). Let C be the integral closure of B in
L (discussed in Exercise 10.7.I). Because A ,→ C is an integral extension, there is a
prime Q of C lying over q ⊂ B (by the Lying Over Theorem 8.2.5), and a prime Q ′ of
344 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

C containing Q lying over q ′ (by the Going-Up Theorem, Exercise 8.2.F). Similarly,
there is a prime P ′ of C lying over p ′ ⊂ A (and thus over q ′ ⊂ B). We would
be done if P ′ = Q ′ (just take p = Q ∩ A), but this needn’t be the case. However,
Lemma 12.2.15 below shows there is an automorphism σ of C over B, that sends
Q ′ to P ′ , and then the image of σ(Q) in A will do the trick, completing the proof.
The following diagram, in geometric terms, may help.

Spec C/P ′  x
O
σ
 *
Spec C/Q ′  / Spec C/Q   / Spec C LO

   ?
Spec A/p ′  / ?  / Spec A K(A)
O

      ?
Spec B/q ′  / Spec B/q   / Spec B K(B)

12.2.15. Lemma. — Suppose B is an integrally closed domain, L/K(B) is a finite normal

field extension, and C is the integral closure of B in L. If q ′ is a prime ideal of B, then
automorphisms of L/K(B) act transitively on the prime ideals of C lying over q ′ .
This result is often first seen in number theory, with B = Z and L a Galois
extension of Q.

Proof. Let P and Q1 be two prime ideals of C lying over q ′ , and let Q2 , . . . , Qn be
the prime ideals of C conjugate to Q1 (the image of Q1 under Aut(L/K(B))). If P
is not one of the Qi , then P is not contained in any of the Qi . (Do you see why?
Hint: Exercise 12.1.E or 12.1.F.) Hence by prime avoidance (Proposition 12.2.13),
P is not contained in their union, so there is some a ∈ P not contained in any Qi .
Thus no conjugate of a can be contained in Q1 , so the norm NL/K(B) (a) ∈ B is
not contained in Q1 ∩ B = q ′ . (Recall: If L/K(B) is separable, the norm is just the
product of the conjugates. But even if L/K(B) is not separable, the norm is the
product of conjugates to the appropriate power, because we can factor L/K(B) as a
separable extension followed by a purely inseparable extension, or because we can
take an explicit basis for the extension and calculate the norm of a. See [Lan, §VI.5]
for more on norms.) But since a ∈ P, its norm lies in P, but also in B, and hence in
P ∩ B = q ′ , yielding a contradiction. □
We end with some exercises for your practice and enlightenment.

12.2.K. E XERCISE / DEFINITION . Suppose X is a Noetherian scheme. Define a

function dimX (·) : X → Z≥0 from the underlying set of X, by taking dimX p to be
the dimension of the largest irreducible component of X containing p. We call this
the dimension of X at p. Show that dimX (·) is an upper semicontinuous function.
(Recall that a function f from a topological space X to R is upper semicontinuous if
for each x ∈ R, f−1 ((−∞, x)) is open. Informally: the function can jump “up” upon
taking limits. Our upper semicontinuous functions will map to Z, so informally
 © 2024 Ravi Vakil. Published by Princeton University Press. 345

functions jump “up” on closed subsets. Similarly, f is lower semicontinuous if

x ∈ R, f−1 ((x, ∞)) is open.

12.2.16. Recurring Theme: Semicontinuities. Semicontinuity is an important notion in

algebraic geometry. Here are some examples.
(i) the rank of a matrix of functions (because rank drops on closed subsets,
where various discriminants vanish)
(ii) the dimension of a variety at a point (Exercise 12.2.K above)
(iii) fiber dimension (Theorem 12.4.3)
(iv) the rank of a finite type quasicoherent sheaf (Exercise 14.3.J)
(v) degree of a finite morphism, as a function of the target (§14.3.4)
(vi) dimension of tangent space at closed points of a variety over an alge-
braically closed field (Exercise 21.2.G)
(vii) rank of cohomology groups of coherent sheaves, in proper flat families
(the Semicontinuity Theorem 25.1.1)
All but (i) are upper semicontinuous; (i) is a lower semicontinuous function.

12.2.L. E XERCISE . Suppose p is a closed point of a locally finite type k-scheme X.

Show that the following three integers are the same: (i) dimX p (Exercise 12.2.K);
(ii) dim OX,p ; and (iii) codimX p. (We define the Zariski topology on the closed
points of a finite type k-scheme in the obvious way: closed subsets are cut out by
equations, as in §8.4.8.)

12.2.M. E XERCISE ( THE DIMENSION OF A LOCALLY FINITE TYPE k- SCHEME IS PRE -

SERVED BY any FIELD EXTENSION , CF. E XERCISE 12.1.H( A )). Suppose X is a locally
finite type k-scheme of pure dimension n, and K/k is a field extension (not necessar-
ily algebraic). Show that XK has pure dimension n. Hint: Reduce to the case where
X is affine, so say X = Spec A. Reduce to the case where A is an integral domain.
Show (using the axiom of choice) that K/k can be written as an algebraic extension
of a purely transcendental extension. Hence by Exercise 12.1.H(a), it suffices to
deal with the case where K/k is purely transcendental, say with transcendence
basis {ei }i∈I (possibly infinite). Show that AK := A ⊗k K is an integral domain,
by interpreting it as a certain localization of the domain A[{ei }]. If t1 , . . . , td is a
transcendence basis for K(A)/k, show that {ei } ∪ {tj } is a transcendence basis for
K(AK )/k. Show that {tj } is a transcendence basis for K(AK )/K.
Exercise 12.2.M is conceptually important. For example, if X is described by
some equations with Q-coefficients, the dimension of X doesn’t depend on whether
we consider it as a Q-scheme or as a C-scheme.

12.2.17. Remark. Unlike Exercise 12.1.H, Exercise 12.2.M has finite type hypotheses
on X. It is not true that if X is an arbitrary k-scheme of pure dimension n, and K/k
is an arbitrary extension, then XK necessarily has pure dimension n. For example,
Exercise 10.2.L shows that if k is a field, then dim k(x) ⊗k k(y) = 1.

12.2.18. ⋆ Lines on hypersurfaces, part 1. Notice: although dimension theory is

not central to the following statement, it is essential to the proof.

12.2.N. E NLIGHTENING STRENUOUS EXERCISE : M OST SURFACES IN THREE - SPACE

OF DEGREE d > 3 HAVE NO LINES . In this exercise, we work over an algebraically
346 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

closed field k. For any d > 3, show that most degree d surfaces in P3 contain no lines.
Here, “most” means “all closed points of a Zariski-open subset of the parameter
space for degree d homogeneous polynomials in 4 variables, up to scalars”. As
there are d+3


3 such monomials, the degree d hypersurfaces are parametrized by
d+3
(
P 3 ) −1
(see Remark 4.5.3). Hint: Construct an “incidence correspondence”
d+3
X = {(ℓ, H) : [ℓ] ∈ G(1, 3), [H] ∈ P( 3 )−1 , ℓ ⊂ H},

parametrizing lines in P3 contained in a hypersurface: define a closed subscheme

d+3
X of P( 3 )−1 × G(1, 3) that makes this notion precise. (Recall that G(1, 3) is a
d+3
Grassmannian.) Show that X is a P( 3 )−1−(d+1) -bundle over G(1, 3). (Possible
hint for this: how many degree d hypersurfaces contain the line x = y = 0?)
Show that dim G(1, 3) = 4 (see §7.7: G(1, 3) has an open cover by A4 ’s). Show that
d+3
dim X = d+3 −1−(d+1)+4. Show that the image of the projection X → P( 3 )−1


3
must lie in a proper closed subset. The following diagram may help.
d+3


dim 3 − 1 − (d + 1) + 4

X
d+3
P( )−1−(d+1)
3

d+3
u )
P( 3 )−1 G(1, 3) dim 4

(The argument readily generalizes to show that if d > 2n − 3, then “most” degree d
hypersurfaces in Pn have no lines. The case n = 1 and n = 2 are trivial but worth
thinking through.)

12.2.19. Side Remark. If you do the previous Exercise, your dimension count will
suggest the true facts that degree 1 hypersurfaces — i.e., hyperplanes — have
2-dimensional families of lines, and that most degree 2 hypersurfaces have 1-
dimensional families (rulings) of lines, as shown in Exercise 9.3.L. They will also
suggest that most degree 3 hypersurfaces contain a finite number of lines, which
reflects the celebrated fact that regular cubic surfaces over an algebraically closed
field always contain 27 lines (Theorem 27.0.1), and we will use this “incidence
correspondence” or “incidence variety” to prove it (§27.3). The statement about
quartic surfaces generalizes to the Noether-Lefschetz Theorem implying that a very
general surface of degree d at least 4 contains no curves that are not the intersection
of the surface with a hypersurface, see [Lef, GH2]. “Very general” means that in the
parameter space (in this case, the projective space parametrizing surfaces of degree
d), the statement is true away from a countable union of proper Zariski-closed
subsets. Like “general” (which was defined in §10.3.7), “very general” is a weaker
version of the phrase “almost every”.

12.3 Krull’s Theorems

 © 2024 Ravi Vakil. Published by Princeton University Press. 347

We next introduce a number of results that we will repeatedly use, relating

functions and the codimension one points where they vanish. The key result is
Krull’s Principal Ideal Theorem, which will be geometrically believable, and whose
proof is short yet somehow nonintuitive and algebraic. This section will require
Noetherian hypotheses.

12.3.1. Krull’s Principal Ideal Theorem. In a vector space, a single linear equation
always cuts out a subspace of codimension 0 or 1 (and codimension 0 occurs only
when the equation is 0). The Principal Ideal Theorem, due to W. Krull, generalizes
this linear algebra fact.

12.3.2. Krull’s Principal Ideal Theorem (geometric version). — Suppose X is a

locally Noetherian scheme, and f is a function. The irreducible components of V(f) are
codimension 0 or 1.
This is clearly a consequence of the following algebraic statement.

12.3.3. Krull’s Principal Ideal Theorem (algebraic version). — Suppose A is a

Noetherian ring, and f ∈ A. Then every prime p minimal among those containing f has
codimension at most 1. If furthermore f is not a zerodivisor, then every such prime p
containing f has codimension precisely 1.
For example, locally principal closed subschemes have “codimension 0 or 1”,
and effective Cartier divisors have “pure codimension 1”. Here is another example
(which you could certainly prove directly, without the Principal Ideal Theorem).

12.3.A. E XERCISE . Show that an irreducible homogeneous polynomial in n + 1

variables over a field k describes an integral scheme of dimension n − 1 in Pn
k.

You should just note how short the proof of Theorem 12.3.3 is, and skip ahead
to §12.3.5 until you have used the theorem enough that you very much appreciate
its value.

12.3.4. ⋆⋆ Proof of Krull’s Principal Ideal Theorem 12.3.3. We dispose of the last
sentence first: if p is codimension zero, then it is an associated prime (Exercise 6.6.M),
and hence any element of p (including f) is a zerodivisor (Proposition 6.6.13).
We next wish to show that codim p ≤ 1. We reduce to the case of a local ring:
replacing A by Ap (and p by pAp ) does not change the problem so we may assume
that p is a maximal ideal.
Suppose otherwise that q is a strictly smaller prime ideal of A (so f ∈ / q by the
minimality of p among prime ideals containing f). We wish to show that q has
codimension zero.
The key idea of the proof is the following clever construction.

12.3.B. E XERCISE . Let ϕ : A → Aq be the localization, and define q(n) :=

−1 n
ϕ (q Aq ). (This is called the nth symbolic power of q, but we will not use
this terminology.) Note that q(1) ⊃ q(2) ⊃ · · · is a descending sequence of ideals
of A. If ag ∈ q(n) and g ∈ / q, show that a ∈ q(n) . (Translation: q(n) is q-primary,
although we won’t use this terminology either.)
p
Now (f) ⊂ A is the intersection of the minimal prime ideals containing f
(Exercise 3.4.F), which is precisely the maximal ideal p. Thus in the local ring A/(f),
348 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

the nilradical ideal is the maximal ideal, so by Proposition 12.1.9 (on dimension
0 Noetherian local rings), any descending chain of ideals of A/(f) eventually sta-
bilizes. In particular, the image of q(1) ⊃ q(2) ⊃ · · · in A/(f) eventually stabilizes,
so

(12.3.4.1) q(n) ≡ q(n+1) (mod f) for n ≥ n0 sufficiently large.

We next claim that in fact q(n) = q(n+1) + fq(n) for n ≥ n0 . Reason: suppose
bn ∈ q(n) . Then by (12.3.4.1), bn = bn+1 + af for bn+1 ∈ q(n+1) ⊂ q(n) and a ∈ A.
Hence af ∈ q(n) ; but f ∈
/ q. By Exercise 12.3.B, a ∈ q(n) , so q(n) = q(n+1) + fq(n) as
claimed.
Thus by Nakayama’s Lemma 8.2.9 (using A, M = q(n) /q(n+1) , I = (f)), q(n) =
(n+1)
q , from which qn Aq = qn+1 Aq . By Nakayama’s Lemma 8.2.9 (using Aq ,
M = qn Aq , I = qAq ), qn Aq = 0, so by Proposition 12.1.9, Aq has dimension zero,
so q has codimension zero in A as desired. □

12.3.5. We now use Krull’s Principal Ideal Theorem to prove some geometric facts.
We start with a fun fact in preparation for Exercise 12.3.D(a), but that will also be
very useful in other circumstances.

12.3.C. I MPORTANT E XERCISE ( TO BE USED REPEATEDLY ). Suppose k is a field

(possibly finite!). Let {p1 , . . . , pm } ∈ Pn
k be a collection of points, not necessarily
closed, none of which are the generic point of Pn k . Show that for N ≫ 0, there is a
nonempty hypersurface of degree N not containing any of the pi . Hint: Show this
by induction on m. To get from m to m+1: take a hypersurface not vanishing on p1 ,
. . . , pm . If it doesn’t vanish on pm+1 , we are done. Otherwise, call this hypersurface
fm+1 . Do something similar with m + 1 replaced by i for each 1 ≤ i ≤ m. Then
P
consider i f1 · · · f^i · · · fm+1 .

12.3.D. I MPORTANT E XERCISE ( TO BE USED REPEATEDLY ). This is a cool argument.

(a) (Hypersurfaces meet everything of dimension at least 1 in projective space, unlike in
affine space.) Suppose X is a closed subset of Pn k of dimension at least 1, and H is a
nonempty hypersurface in Pn k . Show that H meets X. (Hint: note that the affine cone
over H contains the origin in An+1 k . Apply Krull’s Principal Ideal Theorem 12.3.3 to
the cone over X.)
(b) Suppose X ,→ Pn k is a closed subset of dimension r. Show that any codimension r
linear space meets X. Hint: Refine your argument in (a). (Exercise 12.3.G generalizes
this to show that any two things in projective space that you would expect to meet
for dimensional reasons do in fact meet.)
(c) Show further that there is an intersection of r + 1 nonempty hypersurfaces
missing X. (The key step: there is a hypersurface of sufficiently high degree that
doesn’t contain any generic point of X, by Exercise 12.3.C.) If k is infinite, show that
there is a codimension r + 1 linear subspace missing X. (The key step: show that
there is a hyperplane not containing any generic point of a component of X.)
(d) If k is an infinite field, show that there is an intersection of r hyperplanes meeting
X in a finite number of points. (We will see in Exercise 13.4.B that if k = k, for “most”
choices of these r hyperplanes, this intersection is reduced, and in Exercise 18.6.M
that the number of points is the “degree” of X. But first of course we must define
“degree”.)
 © 2024 Ravi Vakil. Published by Princeton University Press. 349

The proof of prime avoidance (the following exercise) is similar to that of

Exercise 12.3.D(c), and you may wish to solve it first. We will find it useful shortly.

12.3.E. E XERCISE . Prove Proposition 12.2.13 (prime avoidance). Hint: by induction

on n. Don’t look in the literature — you might find a much longer argument.

12.3.F. E XERCISE ( BOUND ON CODIMENSION OF INTERSECTIONS IN An k ). Let k = k

be an algebraically closed field. Suppose X and Y are irreducible closed subvarieties
(possibly singular) of codimension m and n respectively in Ad k . Show that every
d
component of X ∩ Y has codimension at most m + n in Ak as follows. Show that
the diagonal Ad ∼ d d
k = ∆ ⊂ Ak ×k Ak is a regular embedding of codimension d. (You
will quickly guess the d equations for ∆.) Exercise 11.2.M identifies the intersection
of X and Y in Ad with the intersection of X × Y with ∆ in Ad ×k Ad k . Then show that
locally, X ∩ Y is cut out in X × Y by d equations. Use Krull’s Principal Ideal Theo-
rem 12.3.3. You will also need Exercise 12.2.F. (The hypothesis of irreducibility can
be replaced by pure dimensionality by appropriately generalizing Exercise 12.2.F.
The hypothesis k = k can be removed by suitably invoking Exercise 12.1.H. See
Exercise 13.2.N for a generalization.)

12.3.G. F UN E XERCISE ( GENERALIZING E XERCISE 12.3.D( B )). Suppose X and Y are

pure dimensional closed subvarieties of Pn of codimensions d and e respectively,
and d + e ≤ n. Show that X and Y intersect. Hint: apply Exercise 12.3.F to the affine
cones of X and Y. Recall the argument you used in Exercise 12.3.D(a) or (b).

12.3.H. U SEFUL E XERCISE . Suppose f is an element of a Noetherian ring A,

contained in no codimension zero or one prime ideals. Show that f is invertible.
(Hint: if a function vanishes nowhere, it is invertible, by Exercise 4.3.F(b).)

12.3.6. A useful characterization of unique factorization domains.

We can use Krull’s Principal Ideal Theorem to prove one of the four useful
criteria for unique factorization domains, promised in §5.4.5.

12.3.7. Proposition. — Suppose that A is a Noetherian integral domain. Then A is a

unique factorization domain if and only if all codimension 1 prime ideals are principal.
This contains Lemma 12.1.7 and (in some sense) its converse.

Proof. One direction is Lemma 12.1.7: if A is a unique factorization domain, then

all codimension 1 prime ideals are principal. So we assume conversely that all
codimension 1 prime ideals of A are principal, and we will show that A is a unique
factorization domain. Now each codimension one prime ideal, being principal,
is generated by a prime element, which is unique up to multiplication by a unit.
All prime elements of an integral domain are irreducible, by the usual argument.
(Suppose p ∈ A is prime, and p = ab, with a, b ∈ A \ {0}. Then by primality of the
ideal (p), either a or b is in (p), say without loss of generality a = kp. Then p = kpb,
from which 1 = bk, from which b is unit. Thus p is irreducible.) Conversely, if
a ∈ A is irreducible, then by Krull’s Principal Ideal Theorem 12.3.3, V(a) contains
some codimension one point [(p)], so a ∈ (p), say a = kp. But by irreducibility of
a, as p is not a unit, k must be a unit, from which (a) = (p), so a is prime. Thus we
have shown that the prime elements are the irreducible elements (and incidentally,
350 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

that up to multiplicative units, they correspond to the codimension one prime

ideals).
Finally, A is a unique factorization domain by the usual argument (e.g., [DF,
§8.3]): each non-unit nonzero element can be factored into a finite number of
irreducibles, and the uniqueness of such a factorization can be shown using the fact
that those irreducibles are also prime. □

12.3.8. Using Krull’s Principal Ideal Theorem inductively. The following result
will allow us to use Krull’s Principal Ideal Theorem inductively.

12.3.9. Lemma. — Suppose p0 ⊊ p1 ⊊ · · · ⊊ pn is a strict chain of prime ideals in a

Noetherian ring A, and f ∈ pn \ p0 . Then there exists a strict chain of prime ideals
p0 = q0 ⊊ q1 ⊊ · · · ⊊ qn−1 ⊊ qn = pn .
in A, with f ∈ q1 \ q0 .
You should feel compelled to immediately sketch of a picture of this lemma.
We are given Spec A, and a nested sequence of irreducible closed subsets, as well
as a “hypersurface” containing the smallest closed subset. We are going to create a
new sequence of irreducible closed subsets, with the same start and finish, which
are all (after the first) contained in the hypersurface.

Proof. Our plan is to create the ideals as follows:

 / p1   p / · · ·   q / pn−2  q  / pn−1  q 
p0   p
̸= ̸= ̸= ̸= ̸=
/ pn

̸= ̸= ̸= ̸= ̸=
 " # "
f∈ q1  / · · ·  / qn−2   / qn−1   / qn
̸= ̸= ̸= ̸=

Unsurprisingly, we do this by descending induction. We prove the following.

Suppose pj−2 ⊂ pj−1 ⊂ pj with f ∈ pj \ pj−2 . We seek a prime ideal qj−1 containing
f, with pj−2 ⊊ qj−1 ⊊ pj . Take qj−1 to be a prime containing (f, pj−2 ), and minimal
among those contained in pj . Now qj−1 is codimension 1 in A/pj−2 by Krull’s
Principal Ideal theorem, and pj is codimension at least 2 (we have a chain pj−2 ⊂
pj−1 ⊂ pj of length 2). Thus qj−1 ⊊ pj . □

12.3.I. E XERCISE . Suppose (A, m) is a Noetherian local ring, and f ∈ m. Show that
dim A/(f) ≥ dim A − 1.
The following generalization of Krull’s Principal Ideal Theorem looks like it
might follow by induction from Krull, but it is more subtle.

12.3.10. Krull’s Height Theorem. — Suppose X = Spec A where A is Noetherian,

and Z is an irreducible component of V(f1 , . . . , fr ), where f1 , . . . , fr ∈ A. Then the
codimension of Z is at most r.
(As mentioned in §12.1.4, height is another common word for codimension.)

12.3.J. E XERCISE . Prove Krull’s Height Theorem 12.3.10 by reducing it to the case
of a Noetherian local ring (A, m), where Z = V(m), and then using Lemma 12.3.9
for the f1 , . . . , fr in order.
 © 2024 Ravi Vakil. Published by Princeton University Press. 351

12.3.K. I MPORTANT E XERCISE . Suppose (A, m) is a Noetherian local ring.

(a) (Noetherian local rings have finite dimension, promised in Remark 12.1.11) Use Krull’s
Height Theorem 12.3.10 to prove that if there are g1 , . . . , gℓ such that V(g1 , . . . , gℓ ) =
{[m]}, then dim A ≤ ℓ. Hence show that A has finite dimension. (For comparison,
Noetherian rings in general may have infinite dimension, see Exercise 12.1.M.)
(b) Let d = dim A. Show that there exist g1 , . . . , gd ∈ A such that V(g1 , . . . , gd ) =
{[m]}.

12.3.11. Definition. The geometric translation of the above exercise is the following.
Given a d-dimensional “germ of a reasonable space” around a point p, p can be cut
out set-theoretically by d equations, and we always need at least d equations. These
d elements of A are called a system of parameters at p. Essentially equivalently,
if (A, m) is a Noetherian localpring of dimension d, then f1 , . . . , fd is a system of
parameters for (A, m) if m = (f1 , . . . , fd ).
The following result will be central in the next chapter.

12.3.12. Theorem. — For any Noetherian local ring (A, m, k), dim A ≤ dimk m/m2 .

Proof. Now m is finitely generated, so m/m2 is a finitely generated (A/m = k)-

module, hence finite-dimensional. Say dimk m/m2 = n. Choose a basis of m/m2 ,
and lift it to elements f1 , . . . , fn of m. By Nakayama’s Lemma (version 4, Exer-
cise 8.2.H), (f1 , . . . , fn ) = m. By Exercise 12.3.K, dim A ≤ n. □
We conclude with two starred sections.

12.3.13. ⋆ Pathologies of the notion of “codimension”. We can use Krull’s

Principal Ideal Theorem to produce the example of pathology in the notion of
codimension promised earlier this chapter. Let A = k[x](x) [t]. In other words,
elements of A are polynomials in t, whose coefficients are quotients of polynomials
in x, where no factors of x appear in the denominator. (Warning: A is not k[x, t](x) .)
Clearly, A is an integral domain, so xt − 1 is not a zero divisor. You can verify that
A/(xt − 1) = ∼ k[x](x) [1/x] =
∼ k(x) — “in k[x](x) , we may divide by everything but x,
and now we are allowed to divide by x as well” — so A/(xt − 1) is a field. Thus
(xt − 1) is not just prime but also maximal. By Krull’s Principal Ideal Theorem,
(xt − 1) is codimension 1. Thus (0) ⊂ (xt − 1) is a maximal chain. However, A
has dimension at least 2: (0) ⊂ (t) ⊂ (x, t) is a chain of prime ideals of length 2.
(In fact, A has dimension precisely 2, although we don’t need this fact in order to
observe the pathology.) Thus we have a codimension 1 prime in a dimension 2 ring
whose quotient is dimension 0, not 1. Here is a picture of this partially ordered set
of ideals.

(x, t)

(t) (xt − 1)

(0)
352 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

This example comes from geometry, and it is enlightening to draw a picture,

see Figure 12.3. Spec k[x](x) corresponds to a “germ” of A1k near the origin, and
Spec k[x](x) [t] corresponds to “this × the affine line”. You may be able to see from
the picture some motivation for this pathology — V(xt − 1) doesn’t meet V(x), so
it can’t have any specialization on V(x), and there is nowhere else for V(xt − 1)
to specialize. It is disturbing that this misbehavior turns up even in a relatively
benign-looking ring.

F IGURE 12.3. Dimension and codimension behave oddly on the

surface Spec k[x](x) [t]

12.3.14. ⋆ Lines on hypersurfaces, part 2. (Part 1 was §12.2.18.) We now give a

geometric application of Krull’s Principal Ideal Theorem 12.3.3, applied through
Exercise 12.3.D(a). Throughout, we work over an algebraically closed field k.

12.3.L. E XERCISE .
(a) Suppose
f(x0 , . . . , xn ) = fd (x1 , . . . , xn ) + x0 fd−1 (x1 , . . . , xn ) + · · · + xd−1
0 f1 (x1 , . . . , xn )
is a homogeneous degree d polynomial (so deg fi = i) cutting out a hypersur-
face X in Pn containing p := [1, 0, . . . , 0]. Show that there is a line through p
contained in X if and only if f1 = f2 = · · · = fd = 0 has a common zero in
Pn−1 = Proj k[x1 , . . . , xn ]. (Hint: given a common zero [a1 , . . . , an ] ∈ Pn−1 , show
that line joining p to [0, a1 , . . . , an ] is contained in X.)
(b) If d ≤ n − 1, show that through any point p ∈ X, there is a line contained in X.
Hint: Exercise 12.3.D(a).
(c) If d ≥ n, show that for “most hypersurfaces” X of degree d in Pn (for all hy-
n+d
persurfaces whose corresponding point in the parameter space P( d )−1 — cf.
Remark 4.5.3 and Exercise 9.3.J — lies in some nonempty Zariski-open subset),
“most points p ∈ X” (all points in a nonempty dense Zariski-open subset of X) have
no lines in X passing through them. (Hint: first show that there is a single p in a
single X contained in no line. Chevalley’s Theorem 8.4.2 may help.)

12.3.15. Remark. A projective (or proper) k-variety X is uniruled if every point

p ∈ X is contained in some P1 ⊂ X. (We won’t use this word beyond this remark.)
Part (b) shows that all hypersurfaces of degree at most n − 1 are uniruled. One
 © 2024 Ravi Vakil. Published by Princeton University Press. 353

can show (using methods beyond what we know now, see [Ko1, Cor. IV.1.11])
that if char k = 0, then every smooth hypersurface of degree at least n + 1 in Pn k
is not uniruled (thus making the open set in (c) explicit). Furthermore, smooth
hypersurfaces of degree n are uniruled, but covered by conics rather than lines.
Thus there is a strong difference in how hypersurfaces behave depending on how
the degree relates to n + 1. This is true in many other ways as well. Smooth
hypersurfaces of degree less than n + 1 are examples of Fano varieties; smooth
hypersurfaces of degree n + 1 are examples of Calabi-Yau varieties (with the possible
exception of n = 1, depending on the definition); and smooth hypersurfaces of
degree greater than n + 1 are examples of general type varieties. We define these
terms in §21.5.5.

12.4 Dimensions of fibers of morphisms of varieties

In this section, we show that the dimensions of fibers of morphisms of varieties

behave in a way you might expect from our geometric intuition. The reason we
have waited until now to discuss this is because we will use Theorem 12.2.9 (for
varieties, codimension is the difference of dimensions).
Before we begin, let’s make sure we are on the same page with respect to
our intuition. Elimination theory (Theorem 8.4.10) tells us that the projection
A → Spec A is closed. We can interpret this as follows. A closed subset X
π : Pn
of PnA is cut out by a bunch of homogeneous equations in n + 1 variables (over
A). The image of X is the subset of Spec A where these equations have a common
nontrivial solution. If we try hard enough, we can describe this by saying that the
existence of a nontrivial solution (or the existence of a preimage of a point under
π : X → Spec A) is an “upper semicontinuous” fact. More generally, your intuition
might tell you that the locus where a number of homogeneous polynomials in n + 1
variables over A have a solution space (in Pn A ) of dimension at least d should be a
closed subset of Spec A. (As a special case, consider linear equations. The condition
for m linear equations in n + 1 variables to have a solution space of dimension at
least d + 1 is a closed condition on the coefficients — do you see why, using linear
algebra?) This intuition will be correct, and will use properness in a fundamental
way (Theorem 12.4.3(b)). We will also make sense of upper semicontinuity in fiber
dimension on the source (Theorem 12.4.3(a)). A useful example to think through
is the map from the xy-plane to the xz-plane (Spec k[x, y] → Spec k[x, z]), given by
(x, z) 7→ (x, xy). (This example also came up in §8.4.1.)
We begin our substantive discussion with an inequality that holds more gener-
ally in the locally Noetherian setting.

12.4.A. K EY E XERCISE ( CODIMENSION BEHAVES AS YOU MIGHT EXPECT FOR A

MORPHISM , OR “ FIBER DIMENSIONS CAN NEVER BE LOWER THAN EXPECTED ”).
Suppose π : X → Y is a morphism of locally Noetherian schemes, and p ∈ X and
q ∈ Y are points such that q = π(p). Show that
codimX p ≤ codimY q + codimπ−1 (q) p
(see Figure 12.4). (This is similar to Exercise 12.1.K, but it is not the same!) Hint: take
a system of parameters (Definition 12.3.11) for q “in Y”, and a system of parameters
354 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

for p “in π−1 (q)”, and use them to find codimY q + codimπ−1 (q) p elements of OX,p
cutting out p = {[m]} in Spec OX,p . Use Exercise 12.3.K.

F IGURE 12.4. Exercise 12.4.A: the codimension of a point in the

total space is bounded by the sum of the codimension of the point
in the fiber plus the codimension of the image in the target

Does Exercise 12.4.A agree with your geometric intuition? You should be
able to come up with enlightening examples where equality holds, and where
equality fails. We will see that equality always holds for sufficiently nice — flat —
morphisms, see Proposition 24.5.6.
We now show that the inequality of Exercise 12.4.A is actually an equality over
“most of Y”.

12.4.1. Theorem. — Suppose π : X → Y is a finite type dominant morphism of integral

schemes such that the finitely generated field extension K(X)/K(Y) has transcendence
degree r. Then there exists a nonempty open subset U ⊂ Y such that for all q ∈ U, the fiber
over q is nonempty and has pure dimension r.
An interesting fact: we will use the Noether Normalization Lemma 12.2.4 in
the proof, even though the statement doesn’t involve varieties! If the generality of
the statement bothers you, we have the following special case for varieties.

12.4.2. Corollary. — Suppose π : X → Y is a (necessarily finite type) morphism of

irreducible k-varieties, with dim X = m and dim Y = n. Then there exists a nonempty
open subset U ⊂ Y such that for all q ∈ U, the fiber over q has pure dimension m − n, or
is empty.
(By convention, the empty set technically has pure dimension r — every irre-
ducible component of the empty set has dimension r. This makes the statement of
Corollary 12.4.2 a bit cleaner.)

12.4.B. E XERCISE . Verify that Theorem 12.4.1 implies Corollary 12.4.2.

Proof of Theorem 12.4.1. We begin with two quick reductions. (i) By shrinking Y if
necessary, we may assume that Y is affine, say Spec B. (ii) We may also assume that
 © 2024 Ravi Vakil. Published by Princeton University Press. 355

X is affine, say Spec A. (Reason: cover X with a finite number of affine open subsets
X1 , . . . , Xa , and take the intersection of the U’s for each of the π|Xi .)
In order to motivate the rest of the argument, we describe our goal. We will
produce a nonempty distinguished open subset U of Spec B so that π−1 (U) → U
factors through ArU via a finite surjective morphism:

(12.4.2.1) X= Spec A o
open emb.
? _ π−1 (U)

finite surj.

π ArU

 
Y= Spec B o
open emb.
? _U

12.4.C. E XERCISE . Show that this suffices to prove Theorem 12.4.1. Hint: Suppose
p ∈ U is a point, with residue field κ. Show that any irreducible variety mapping
finitely to Arκ has dimension at most r. This will show that the dimension of
any component of π−1 (p) has dimension at most r. To show equality, apply the
Going-Down Theorem 12.2.12 appropriately to Spec A → Spec B[x1 , . . . , xr ].
So we now work to build (12.4.2.1). We begin by noting that we have inclusions
of B into both A and K(B), and from both A and K(B) into K(A). The maps from A
and K(B) into K(A) both factor through A ⊗B K(B) (whose Spec is the generic fiber
of π), so the maps from both A and K(B) to A ⊗B K(B) must be inclusions.

(12.4.2.2) 3 9 K(A)
B

% / A ⊗B K(B)
AO
∴,→ O
∴,→
?  1
B / K(B)

Clearly K(A) ⊗B K(B) = K(A) (as A ⊗B K(B) can be interpreted as taking A

and inverting those nonzero elements of B), and A ⊗B K(B) is a finitely generated
algebra over the field K(B).
By Noether normalization 12.2.4, we can find elements t1 , . . . , tr ∈ A ⊗B
K(B), algebraically independent over K(B), such that A ⊗B K(B) is integral over
K(B)[t1 , . . . , tr ].
Now, we can think of the elements ti ∈ A⊗B K(B) as fractions, with numerators
in A and (nonzero) denominators in B. If f is the product of the denominators ap-
pearing for each ti , then by replacing B by Bf (replacing Spec B by its distinguished
open subset D(f)), we may assume that the ti are all in A. Thus (after sloppily
356 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

renaming Bf as B, and Af as A) we can trim and extend (12.4.2.2) to the following.


AO  / A ⊗B K(B)
O
integral
?  ?
B[t1 , . . . , tr ]  / K(B)[t1 , . . . , tr ]
O O

?  ?
B / K(B)

Now A is finitely generated over B, and hence over B[t1 , . . . , tr ]. But we cannot
yet be sure that A is finite over B[t1 , . . . , tr ], so we are not done. We will have to
localize B further.
Suppose A is generated over B by u1 , . . . , uq . Our Noether normalization
argument implies that each ui satisfies some monic equation fi (ui ) = 0, where
fi ∈ K(B)[t1 , . . . , tr ][t] (i.e., monic considered as a polynomial in t with coefficients
in K(B)[t1 , . . . , tr ]). The coefficients of fi (considered as polynomials in t1 , . . . , tr , t)
are a priori fractions in B, but by multiplying by all those denominators, we can
assume each fi ∈ B[t1 , . . . , tr ][t], at the cost of losing monicity of the fi (this time
considered as polynomials in t). Let b ∈ B be the product of the leading coefficients
(considered as polynomials in t) of all the fi . If U = D(b) (the locus where b is
invertible), then over U, the fi (can be taken to) have leading coefficient 1, so the ui
(in Ab ) are integral over Bb [t1 , . . . , tr ]. Thus Spec Ab → Spec Bb [t1 , . . . , tr ] is finite
and surjective (the latter by the Lying Over Theorem 8.2.5).
We have now constructed (12.4.2.1), as desired. □
There are a couple of things worth pointing out about the proof. First, although
the result was originally motivated by classical varieties over a field k, the proof
uses the theory of varieties over another field, the function field K(B). This is an
example of how the introduction of generic points to algebraic geometry is useful
even for considering more “classical” questions.
Second, the idea of the main part of the argument is that we have a result over
the generic point (Spec A ⊗B K(B) is finite and surjective over affine space over
K(B)), and we want to “spread it out” to an open neighborhood of the generic point
of Spec B. We do this by realizing that “finitely many denominators” appear when
correctly describing the problem, and inverting those functions. This “spreading
out” idea is a recurring theme.

12.4.D. E XERCISE ( USEFUL CRITERION FOR IRREDUCIBILITY ). Suppose π : X → Y is

a proper morphism to an irreducible variety, and all the fibers of π are nonempty,
and irreducible of the same dimension. Show that X is irreducible.

12.4.3. Theorem (upper semicontinuity of fiber dimension). — Suppose π : X → Y

is a morphism of finite type k-schemes.
(a) (upper semicontinuity on the source) The dimension of the fiber of π at p ∈ X (the
dimension of the largest component of π−1 (π(p)) containing p) is an upper semicontinuous
function in p (i.e., on X).
(b) (upper semicontinuity on the target) If furthermore π is closed (e.g., if π is proper), then
the dimension of the fiber of π over q ∈ Y is an upper semicontinuous function in q (i.e.,
on Y).
 © 2024 Ravi Vakil. Published by Princeton University Press. 357

You should be able to immediately construct a counterexample to part (b) if

the closedness hypothesis is dropped. (We also remark that Theorem 12.4.3(b) for
projective morphisms is done, in a simple way, in Exercise 18.1.B.)

Proof. (a) Let Fn be the subset of X consisting of points where the fiber dimension
is at least n. We wish to show that Fn is a closed subset for all n. We argue by
induction on dim Y. The base case dim Y = 0 is trivial. So we fix Y, and assume the
result for all smaller-dimensional targets.

12.4.E. E XERCISE . Show that it suffices to prove the result when X and Y are
integral, and π is dominant.
Let r = dim X−dim Y be the “relative dimension” of π. If n ≤ r, then Fn = X by
Exercise 12.4.A (combined with Theorem 12.2.9, that codimension is the difference
of dimensions for varieties).
If n > r, then let U ⊂ Y be the dense open subset of Theorem 12.4.1, where
“the fiber dimension is exactly r”. Then Fn does not meet the preimage of U.
By replacing Y with Y \ U (and X by X \ π−1 (U)), we are done by the inductive
hypothesis.

12.4.F. E ASY EXERCISE . Prove (b) (using (a)).

□

12.4.4. Proposition (“generically finite implies generally finite”, cf. Exercise 10.3.G).
— Suppose π : X → Y is a generically finite morphism of irreducible k-varieties of dimension
n. Then there is a dense open subset V ⊂ Y above which π is finite.
(If you wish, you can later relax the irreducibility hypothesis to simply requiring
X and Y to be simply of pure dimension n.)

Proof. As in the proof of Theorem 12.4.1, we may assume that Y is affine, and that π
is dominant. Suppose that X = ∪n i=1 Ui , where the Ui are affine open subschemes
of X. By Exercise 10.3.G, there are dense open subsets Vi ⊂ Y over which π|Ui is
finite. By replacing Y by an affine open subset of ∩Vi , we may assume that π|Ui is
finite.

12.4.G. E XERCISE . Show that π is closed. Hint: you will just use that π|Ui is closed,
and that there are a finite number of Ui .
Then X \ U1 is a closed subset, so π(X \ U1 ) is closed.

12.4.H. E XERCISE . Show that this closed subset is not all of Y.

Define V := Y \ π(X \ U1 ). Then π is finite above V: it is the restriction of the
finite morphism π|U1 : U1 → Y to the open subset V of the target Y. □
 CHAPTER 13

Regularity and smoothness

One natural notion we expect to see for geometric spaces is the notion of when
an object is “smooth”. In algebraic geometry, this notion, called regularity, is easy to
define (Definition 13.2.1) but a bit subtle in practice. This will lead us to a different
related notion of when a variety is smooth (Definition 13.2.4).
This chapter has many moving parts, of which §13.1–13.6 are the important
ones. In §13.1, the Zariski tangent space is motivated and defined. In §13.2, we
define regularity and smoothness over a field, the central topics of this chapter, and
discuss some of their important properties. In §13.3, we give a number of important
examples, mostly in the form of exercises. In §13.4, we discuss Bertini’s Theorem, a
fundamental classical result. In §13.5, we give many characterizations of discrete
valuation rings, which play a central role in algebraic geometry. Having seen clues
that “smoothness” is a “relative” notion rather than an “absolute” one, in §13.6 we
define smooth morphisms (and in particular, étale morphisms), and give some of their
properties. (We will revisit this definition in §21.2.28, once we know more.)
The remaining sections are less central. In §13.7, we discuss the valuative
criteria for separatedness and properness. In §13.8, we mention some more sophis-
ticated facts about regular local rings (the local rings at regular points). In §13.9, we
prove the Artin-Rees Lemma, because it will have been invoked in §13.5 (and will
be used later as well).

13.1 The Zariski tangent space

We begin by defining the tangent space of a scheme at a point. It behaves like
the tangent space you know and love at “smooth” points, but also makes sense at
other points. In other words, geometric intuition at the “smooth” points guides the
definition, and then the definition guides the algebra at all points, which in turn
lets us refine our geometric intuition.
The definition is short but surprising. The main difficulty is convincing yourself
that it deserves to be called the tangent space. This is tricky to explain, because
we want to show that it agrees with our intuition, but our intuition is worse than
we realize. So you should just accept this definition for now, and later convince
yourself that it is reasonable.

13.1.1. Definition. The Zariski cotangent space of a local ring (A, m) is defined to
be m/m2 ; it is a vector space over the residue field A/m. The dual vector space is

This material is published by Princeton University Press as The Rising Sea: Foundations of Algebraic
Geometry by Ravi Vakil. This version is free to view and download for personal use only. Not for
redistribution, resale, or use in derivative works. © 2024 Ravi Vakil.
359
360 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

∨
the Zariski tangent space. If X is a scheme, the Zariski cotangent space TX,p at
a point p ∈ X is defined to be the Zariski cotangent space of the local ring OX,p ,
and similarly for the Zariski tangent space TX,p . (Note that for any maximal ideal
m of any ring A, we have canonically m/m2 = mAm /m2 Am — make sure you see
why.) Elements of the Zariski cotangent space are called cotangent vectors or
differentials; elements of the tangent space are called tangent vectors.
The cotangent space is more algebraically natural than the tangent space, in
that the definition is shorter. There is a moral reason for this: the cotangent space is
more naturally determined in terms of functions on a space, and we are very much
thinking about schemes in terms of “functions on them”. This will come up later.
Here are two plausibility arguments that this is a reasonable definition. Hope-
fully one will catch your fancy.
In differential geometry, the tangent space at a point is sometimes defined as
the vector space of derivations at that point. A derivation at a point p of a manifold
is an R-linear operation that takes in functions f near p (i.e., elements of Op ), and
outputs elements f ′ (p) of R, and satisfies the Leibniz rule
(fg) ′ = f ′ g + g ′ f.
(We will later define derivations in a more general setting, §21.2.17.) A derivation is
the same as a map m → R, where m is the maximal ideal of Op . (The map Op → R
extends this, via the map Op → m given by f − f(p).) But m2 maps to 0, as if
f(p) = g(p) = 0, then
(fg) ′ (p) = f ′ (p)g(p) + g ′ (p)f(p) = 0.
Thus a derivation induces a map m/m2 → R, i.e., an element of (m/m2 )∨ .

13.1.A. E XERCISE . Check that this is reversible, i.e., that any linear map m/m2 → R
gives a derivation. In other words, verify that the Leibniz rule holds.
Here is a second, vaguer, motivation that this definition is plausible for the
cotangent space of the origin of An . (I prefer this one, as it is more primitive and
elementary.) Functions on An should restrict to a linear function on the tangent
space. What (linear) function does x2 + xy + x + y restrict to “near the origin”?
You will naturally answer: x + y. Thus we “pick off the linear terms”. Hence m/m2
are the linear functionals on the tangent space, so m/m2 is the cotangent space. In
particular, you should picture functions vanishing at a point (i.e., lying in m) as
giving functions on the tangent space in this obvious way.

13.1.2. Old-fashioned example. Computing the Zariski tangent space is actually

quite hands-on, because you can compute it just as you did when you learned
multivariable calculus. In A3 , we have a curve cut out by x + y + z2 + xyz = 0
and x − 2y + z + x2 y2 z3 = 0. (You can use Krull’s Principal Ideal Theorem 12.3.3
to check that this is a curve, but it is not important to do so.) What is the tangent
line near the origin? (Is it even smooth there?) Answer: the first surface looks like
x + y = 0 and the second surface looks like x − 2y + z = 0. The curve has tangent
line cut out by x + y = 0 and x − 2y + z = 0. It is smooth (in the traditional sense).
In multivariable calculus, the students work hard to get the answer, because we
aren’t allowed to tell them to just pick out the linear terms.
Let’s make explicit the fact that we are using. If A is a ring, m is a maximal
ideal, and f ∈ m is a function vanishing at the point [m] ∈ Spec A, then the Zariski
 © 2024 Ravi Vakil. Published by Princeton University Press. 361

tangent space of Spec A/(f) at m is cut out in the Zariski tangent space of Spec A
(at m) by the single linear equation f (mod m2 ). The next exercise will force you to
think this through.

13.1.B. I MPORTANT E XERCISE (“K RULL’ S P RINCIPAL I DEAL T HEOREM FOR TAN -
GENT SPACES ” — BUT MUCH EASIER THAN K RULL’ S P RINCIPAL I DEAL T HEO -
REM 12.3.3!). Suppose A is a ring, and m a maximal ideal. If f ∈ m, show that
the Zariski tangent space of A/f is cut out in the Zariski tangent space of A by f
(mod m2 ). (Note: we can quotient by f and localize at m in either order, as quotient
and localization commute, (4.3.7.1).) Hence the dimension of the Zariski tangent
space of Spec A/(f) at [m] is the dimension of the Zariski tangent space of Spec A at
[m], or one less. (That last sentence should be suitably interpreted if the dimension
is infinite, although it is less interesting in this case.)
Here is another example to see this principle in action, extending Exam-
ple 13.1.2: x + y + z2 = 0 and x + y + x2 + y4 + z5 = 0 cuts out a curve, which
obviously passes through the origin. If I asked my multivariable calculus students
to calculate the tangent line to the curve at the origin, they would do calculations
which would boil down (without them realizing it) to picking off the linear terms.
They would end up with the equations x + y = 0 and x + y = 0, which cut out a
plane, not a line. They would be disturbed, and I would explain that this is because
the curve isn’t smooth at a point, and their techniques don’t work. We on the other
hand bravely declare that the tangent space is cut out by x + y = 0, and (will soon)
use this pathology as definition of what makes a point singular (or non-regular).
(Intuitively, the curve near the origin is very close to lying in the plane x + y = 0.)
Notice: the cotangent space jumped up in dimension from what it was “supposed
to be”, not down. (This is not a coincidence: see Theorem 12.3.12.)

13.1.C. E XERCISE . Suppose Y and Z are closed subschemes of X, both containing

the point p ∈ X.
(a) Show that TZ,p is naturally a sub-κ(p)-vector space of TX,p .
(b) Show that TY∩Z,p = TY,p ∩ TZ,p . (On the left, ∩ means scheme-theoretic intersec-
tion. On the right, ∩ means intersection of subspaces of the vector space TX,p .)
(c) Show that TY∪Z,p contains the span of TY,p and TZ,p , where ∪ as usual in this
context is scheme-theoretic union.
(d) Show that TY∪Z,p can be strictly larger than the span of TY,p and TZ,p . (Hint:
Figure 4.5.)
Here is a pleasant consequence of the notion of Zariski tangent space.

13.1.3. Problem. Consider the ring A = k[x, y, z]/(xy − z2 ). Show that (x, z) is not
a principal ideal.
As dim A = 2 (why?), and A/(x, z) = ∼ k[y] has dimension 1, we see that this
ideal is codimension 1 (as codimension is the difference of dimensions for irre-
ducible varieties, Theorem 12.2.9). Our geometric picture is that Spec A is a cone
(we can diagonalize the quadric as xy − z2 = ((x + y)/2)2 − ((x − y)/2)2 − z2 , at
least if char k ̸= 2 — see Exercise 5.4.J), and that (x, z) is a line on the cone. (See
Figure 13.1 for a sketch.) This suggests that we look at the cone point.
362 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

F IGURE 13.1. V(x, z) ⊂ Spec k[x, y, z]/(xy − z2 ) is a line on a cone

13.1.4. Solution. Let m = (x, y, z) be the maximal ideal corresponding to the

origin. Then Spec A has Zariski tangent space of dimension 3 at the origin, and
Spec A/(x, z) has Zariski tangent space of dimension 1 at the origin. But Spec A/(f)
must have Zariski tangent space of dimension at least 2 at the origin by Exer-
cise 13.1.B. (As an added bonus, we have shown that A is not a Unique Factorization
Domain, by Lemma 12.1.7.))

13.1.5. ⋆ Remark. Another approach to solving the problem, not requiring the
definition of the Zariski tangent space, is to use the fact that the ring is graded
(where x, y, and z each have degree 1), and the ideal (x, z) is a homogeneous ideal.
(You may enjoy thinking this through.) The advantage of using the tangent space is
that it applies to more general situations where there is no grading. For example,
(a) (x, z) is not a principal ideal of k[x, y, z]/(xy − z2 − z3 ). As a different example,
(b) (x, z) is not a principal ideal of the local ring (k[x, y, z]/(xy − z2 ))(x,y,z) (the
“germ of the cone”). However, we remark that the graded case is still very useful.
The construction of replacing a filtered ring by its “associated graded” ring can
turn more general rings into graded rings (and can be used to turn example (a)
into the graded case). The construction of completion can turn local rings into
graded local rings (and can be used to turn example (b) into, essentially, the graded
case). Filtered rings will come up in §13.9, the associated graded construction will
implicitly come up in our discussions of the blow-up in §22.3, and many aspects of
completions will be described in Chapter 28.

13.1.D. E XERCISE . Show that (x, z) ⊂ k[w, x, y, z]/(wz − xy) is a codimension 1

ideal that is not principal, using the method of Solution 13.1.4. (See Figure 13.2
for the projectivization of this situation — a line on a smooth quadric surface.)
This example was promised just after Exercise 5.4.D. An improvement is given in
Exercise 15.5.N: this closed subset x = z = 0 is not even set-theoretically cut out by
one equation.

13.1.E. E XERCISE . Let A = k[w, x, y, z]/(wz−xy). Show that Spec A is not factorial.
(Exercise 5.4.L shows that A is not a unique factorization domain, but this is not
enough — why is the localization of A at the prime (w, x, y, z) not factorial? One
possibility is to do this “directly”, by trying to imitate the solution to Exercise 5.4.L,
but this is hard. Instead, use the intermediate result that in a unique factorization
domain, any codimension 1 prime is principal, Lemma 12.1.7, and considering
Exercise 13.1.D.) As A is integrally closed if char k ̸= 2 (Exercise 5.4.I(c)), this
 © 2024 Ravi Vakil. Published by Princeton University Press. 363

F IGURE 13.2. The line V(x, z) on the smooth quadric surface

V(wz − xy) ⊂ P3 .

yields an example of a scheme that is normal but not factorial, as promised in

Exercise 5.4.F. A slight generalization will be given in 22.4.N.

13.1.F. L ESS IMPORTANT EXERCISE (“ HIGHER - ORDER DATA”). (This exercise is fun,
but won’t be used.)
(a) In Exercise 3.7.B, you computed the equations cutting out the (union of the)
three coordinate axes of A3k . (Call this scheme X.) Your ideal should have had three
generators. Show that the ideal cannot be generated by fewer than three elements.
(Hint: working modulo m = (x, y, z) won’t give any useful information, so work
modulo a higher power of m.)
(b) Show that the coordinate axes in A3k are not regularly embedded in A3k . (This
was promised at the end of §9.5.)

13.1.6. Morphisms and tangent spaces. Suppose π : X → Y, and π(p) = q. Then if we

were in the category of differentiable manifolds, we would expect a tangent map,
from the tangent space of p to the tangent space at q. Indeed that is the case; we
have a map of stalks OY,q → OX,p , which sends the maximal ideal of the former n
to the maximal ideal of the latter m (we have checked that this is a “local morphism”
when we briefly discussed locally ringed spaces, see §7.3.1). Thus n2 maps to m2 ,
from which we see that n/n2 maps to m/m2 . If (OX,p , m) and (OY,q , n) have the
same residue field κ, so n/n2 → m/m2 is a linear map of κ-vector spaces, we have
a natural map (m/m2 )∨ → (n/n2 )∨ . This is the map from the tangent space of
p to the tangent space at q that we sought. (Aside: note that the cotangent map
always exists, without requiring p and q to have the same residue field — a sign
that cotangent spaces are more natural than tangent spaces in algebraic geometry.)

13.1.G. E XERCISE . Suppose X is a k-scheme. Describe a natural bijection from

Mork (Spec k[ϵ]/(ϵ2 ), X) to the data of a point p with residue field k (necessarily
a closed point) and a tangent vector at p. (This is important, for example in
deformation theory. We won’t use it, so you can treat it as a curiosity.)

13.1.H. E XERCISE ( FUN ARITHMETIC PRACTICE ). Find the dimension of the Zariski
∼ Z[x]/(x2 + 4). Find the dimension
tangent space at the point [(2, 2i)] of Z[2i] = √
of the Zariski tangent space at the point [(2, x)] of Z[ −2] = ∼ Z[x]/(x2 + 2). (If
364 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

you prefer geometric versions of the same examples, replace Z by C, and 2 by y:

consider C[x, y]/(x2 + y2 ) and C[x, y]/(x2 + y).)

13.1.7. Important interpretation in the finite type case: The Jacobian corank.

13.1.I. I MPORTANT E XERCISE ( THE J ACOBIAN COMPUTES THE Z ARISKI COTANGENT

SPACE AT A RATIONAL POINT ). Suppose X is a finite type k-scheme. Then locally it
is of the form Spec k[x1 , . . . , xn ]/(f1 , . . . , fr ). Show that the Zariski cotangent space
at a k-valued point (a closed point with residue field k) is given by the cokernel of
the “Jacobian map” kr → kn given by the Jacobian matrix
 ∂f1 ∂fr
∂x1 (p) · · ·

∂x1 (p)
(13.1.7.1) J= .. .. ..
.
 
. . .
∂f1 ∂fr
∂xn (p) ··· ∂xn (p)

(This makes precise our example of a curve in A3 cut out by a couple of equations,
where we picked off the linear terms, see Example 13.1.2.) You might be alarmed:
∂f
what does ∂x 1
mean? Do you need deltas and epsilons? No! Just define derivatives
formally, e.g.,
∂
(x2 + x1 x2 + x22 ) = 2x1 + x2 .
∂x1 1
Hint: Do this first when p is the origin, and consider linear terms, just as in
Example 13.1.2 and Exercise 13.1.B. For the general case, “translate p to the origin”.

13.1.8. Remark. This result can be extended to closed points of X whose residue
field is separable over k (and in particular, to all closed points if char k = 0 or if k is
finite), see Remark 21.2.34. The fact that we wait until Chapter 21 to show this does
not mean that it needs to be so complicated; a more elementary proof is possible.

13.1.9. Warning. It is more common in mathematics (but not universal) to define

the Jacobian matrix as the transpose of (13.1.7.1).

13.1.J. E XERCISE ( THE CORANK OF THE J ACOBIAN IS INDEPENDENT OF THE PRE -

SENTATION ). Suppose A is a finitely generated k-algebra, generated by x1 , . . . , xn ,
with ideal of relations I generated by f1 , . . . , fr . Let p be a point of Spec A.
(a) Suppose g ∈ I. Show that appending the column of partials of g to the Jacobian
matrix (13.1.7.1) does not change the corank (the dimension of the cokernel) at p.
Hence show that the corank of the Jacobian matrix at p does not depend on the
choice of generators of I.
(b) Suppose q(x1 , . . . , xn ) ∈ k[x1 , . . . , xn ]. Let h be the polynomial y−q(x1 , . . . , xn ) ∈
k[x1 , . . . , xn , y]. Show that the Jacobian matrix of (f1 , . . . , fr , h) with respect to
the variables (x1 , . . . , xn , y) has the same corank at p as the Jacobian matrix of
(f1 , . . . , fr ) with respect to (x1 , . . . , xn ). Hence show that the corank of the Jacobian
matrix at p is independent of the choice of generators for A.
Thus every finite type k-scheme X comes with an intrinsic “Jacobian corank
function” from (the underling set of) X to Z≥0 , whose value at k-valued points p
of X is the dimk TX,p . We temporarily call this function JC : X → Z≥0 , although
we will later discover that it is really the rank of the coherent sheaf of differentials
ΩX/k (Exercise 21.2.E).
 © 2024 Ravi Vakil. Published by Princeton University Press. 365

13.1.K. E ASY E XERCISE . Show that JC is an upper semicontinuous function.

13.1.L. E XERCISE . Show that JC is preserved by field extension of k. More precisely,

if X is a finite type k-scheme, and ℓ/k is a field extension, and under the base change
map X ×k ℓ → X, p 7→ q, show that JC(p) = JC(q).

13.1.M. E XERCISE . Prove that for any p ∈ X, JC(p) ≥ dimX p. (Necessarily X is

finite type over k, but p need not be a closed point of X.) The dimension of X at p,
dimX p, was defined in Exercise 12.2.K. Hint: Deal first with the case k = k. Here
use the closed points as leverage. Then prove the result for general k by showing
that both sides of the inequality are preserved by field extension (Exercises 13.1.L
and 12.2.M).

13.1.N. E XERCISE ( COMPUTING THE J ACOBIAN CORANK FUNCTION FOR PROJEC -

TIVE k- SCHEMES ). Suppose X ⊂ Pn k is cut out by homogeneous polynomials
fi ∈ k[x0 , . . . , xn ] (1 ≤ i ≤ r). Show that the Jacobian corank function of X is one
less than the corank of the “projective Jacobian matrix”:
 ∂f1 ∂fr
∂x0 (p) · · ·

∂x0 (p)
(13.1.9.1) JC(p) = corank  .. .. ..
 − 1.
 
. . .
∂f1 ∂fr
∂xn (p) · · · ∂xn (p)

You will have to make sense of the right side, as the projective Jacobian matrix has
entries which are not functions; what does the corank even mean?

13.2 Regularity, and smoothness over a field

The key idea in the definition of regularity is contained in Theorem 12.3.12 (the
dimension of the Zariski tangent space is at least the dimension of the local ring).

13.2.1. Definition. If equality holds in Theorem 12.3.12, we say that A is a regular

local ring. In particular, regular local rings are Noetherian by definition. (If a Noetherian
ring A is regular at all of its prime ideals, i.e., if Ap is a regular local ring for all
prime ideals p of A, then A is said to be a regular ring. We basically won’t use this
terminology. Fact 13.8.2 reassuringly implies that regular local rings are indeed
regular rings.) A locally Noetherian scheme X is regular at a point p if the local ring
OX,p is regular. (The word nonsingular is often used as well, notably in the case
when X is finite type over a field, but for the sake of consistency we will use “regular”
throughout.) It is singular at the point otherwise, and we say that the point is a
singularity. (The word nonregular is also used, but for the sake of consistency, we
will use “singular”, despite the fact that this choice is inconsistent with our choice of
“regular” over “nonsingular”. A possible motivation for the inconsistency: “regular
local ring” and “singularity” are both standard terminology, so at some point we
are forced to make a choice.) A scheme is regular (or nonsingular) if it is regular at
all points. It is singular otherwise (i.e., if it is singular at at least one point — if it has
a singularity).

13.2.A. E XERCISE . Show that a dimension 0 Noetherian local ring is regular if and
only if it is a field.
366 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

You will hopefully gradually become convinced that this is the right notion of
“smoothness” of schemes. Remarkably, Krull introduced the notion of a regular
local ring for purely algebraic reasons, some time before Zariski realized that it was
a fundamental notion in geometry in 1947.

13.2.B. E XERCISE . Suppose (A, m) is a regular local ring of dimension n > 0,

and f ∈ A. Show that A/(f) is a regular local ring of dimension n − 1 if and
only if f ∈ m \ m2 . (Hint: Krull’s Principal Ideal Theorem for tangent spaces,
Exercise 13.1.B.)

13.2.C. E XERCISE ( THE SLICING CRITERION FOR REGULARITY ). Suppose X is

a Noetherian scheme (such as a variety), D is an effective Cartier divisor on X
(Definition 9.5.1), and p ∈ D. Show that if p is a regular point of D then p is
a regular point of X. (Hint: Krull’s Principal Ideal Theorem for tangent spaces,
Exercise 13.1.B.)

13.2.D. E ASY E XERCISE ( PRACTICE WITH THE CONCEPT ). Suppose (A, m) is a

regular local ring of dimension d, and f1 , . . . , fd ∈ m are such that their images in
the vector space m/m2 form a basis. Show that they are a system of parameters
(Definition 12.3.11) for (A, m). (They are “particularly good” parameters.)

13.2.2. The Jacobian criterion for regularity.

A finite type k-scheme is locally of the form Spec k[x1 , . . . , xn ]/(f1 , . . . , fr ). The
Jacobian criterion for regularity (Exercise 13.2.E) gives a hands-on method for
checking for singularity at closed points, using the equations f1 , . . . , fr , if k = k.

13.2.E. I MPORTANT E XERCISE ( THE J ACOBIAN CRITERION FOR REGULARITY —

EASY, GIVEN E XERCISE 13.1.I). Suppose X = Spec k[x1 , . . . , xn ]/(f1 , . . . , fr ) has
pure dimension d. Show that a k-valued point p ∈ X is regular if and only if the
Jacobian corank function at p is d.

13.2.F. E ASY E XERCISE . Suppose k = k. Show that the singular closed points of the
hypersurface f(x1 , . . . , xn ) = 0 in An
k are given by the equations
∂f ∂f
f= = ··· = = 0.
∂x1 ∂xn
(Translation: the singular points of f = 0 are where the gradient of f vanishes. This
is not shocking.)

13.2.3. Smoothness over a field.

There seem to be two serious drawbacks with the Jacobian criterion. For finite
type schemes over k, the criterion gives a necessary condition for regularity, but
it is not obviously sufficient, as we need to check regularity at non-closed points
as well. We can prove sufficiency by working hard to show Fact 13.8.2, which
implies that the non-closed points must be regular as well. A second failing is that
the criterion requires k to be algebraically closed. These problems suggest that
old-fashioned ideas of using derivatives and Jacobians are ill-suited to the fancy
modern notion of regularity. But that is wrong — the fault is with the concept of
regularity. There is a better notion of smoothness over a field. Better yet, this idea
generalizes to the notion of a smooth morphism of schemes (to be discussed in §13.6,
and again in Chapter 24.8), which behaves well in all possible ways (including
 © 2024 Ravi Vakil. Published by Princeton University Press. 367

in the sense of §8.1). This is another sign that some properties we think of as of
objects (“absolute notions”) should really be thought of as properties of morphisms
(“relative notions”). We know enough to imperfectly (but correctly) define what it
means for a scheme to be k-smooth, or smooth over k.

13.2.4. Definition. A k-scheme is k-smooth of dimension d, or smooth of dimen-

sion d over k, if it is of pure dimension d, and there exists a cover by affine open
sets Spec k[x1 , . . . , xn ]/(f1 , . . . , fr ) where the Jacobian matrix has corank d at all
points (not just closed points). (In particular, smooth schemes are implicitly locally
of finite type.) A k-scheme is smooth over k if it is smooth of some dimension.
The k is often omitted when it is clear from context. By the theory of the Jacobian
corank function developed in §13.1.7, for any pure dimension d variety, this can be
checked on any affine open cover.

13.2.G. E XERCISE ( FIRST EXAMPLES ).

(a) Show that Ank is smooth for any n and k. For which characteristics is the curve
y2 z = x3 − xz2 in P2k smooth (cf. Exercise 13.3.C)?
(b) Suppose f ∈ k[x1 , . . . , xn ] is a polynomial such that the system of equations
∂f ∂f
f= = ··· = =0
∂x1 ∂xn
has no solutions in k. Show that the hypersurface f = 0 in An
k is smooth. (Compare
this to Exercise 13.2.F, which has the additional hypothesis k = k.)

13.2.H. E XERCISE ( SMOOTHNESS OF VARIETIES IS “ INDEPENDENT OF EXTENSION

OF BASE FIELD ”). Suppose X is a locally finite type k-scheme, and k ⊂ ℓ is a field
extension. Show that X is smooth over k if and only if X ×Spec k Spec ℓ is smooth
over ℓ.
The next exercise shows that we need only check closed points, thereby making
a connection to classical geometry.

13.2.I. E XERCISE . Show that if the Jacobian matrix for X = Spec k[x1 , . . . , xn ]/(f1 , . . . fr )
has corank d at all closed points, then it has corank d at all points. (Hint: the locus
where the Jacobian matrix has corank d can be described in terms of vanishing and
nonvanishing of determinants of certain explicit matrices.)

13.2.5. In defense of regularity. Having made a spirited case for smoothness, we

should be clear that regularity is still very useful. For example, it is the only concept
which makes sense in mixed characteristic. In particular, Z is regular at its points (it
is a regular ring), and more generally, discrete valuation rings are incredibly useful
examples of regular rings.

13.2.6. Regularity vs. smoothness.

13.2.J. E XERCISE . Suppose X is a locally finite type scheme of pure dimension d

over an algebraically closed field k = k. Show that X is regular at its closed points
if and only if it is smooth. (We will soon learn that for locally finite type k-schemes,
regularity at closed points is the same as regularity everywhere, Theorem 13.8.3.)
Hint to show regularity implies smoothness: use the Jacobian criterion to show that
the corank of the Jacobian is d at the closed points of X. Then use Exercise 13.2.I.
368 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

More generally, if k is perfect (e.g., if char k = 0 or k is a finite field), then

smoothness is the same as regularity at closed points (see Exercise 13.2.R). More
generally still, we will later prove the following fact. We mention it now because it
will make a number of statements cleaner long before we finally prove it. (There
will be no circularity.)

13.2.7. Smoothness-Regularity Comparison Theorem. —

(a) If k is perfect, every regular finite type k-scheme is smooth over k.
(b) Every smooth k-scheme is regular (with no hypotheses on perfection).
Part (a) will be proved in Exercise 13.2.R. Part (b) will be proved in §24.8.9. The
fact that Theorem 13.2.7(b) will be proved so far in the future does not mean that
we truly need to wait that long. But we deliberately postpone the proof until we
have machinery that will do much of the work for us. If you wish some insight
right away, here is the outline of the argument for (b), which you can even try
to implement after reading this chapter. We will soon show (in Exercise 13.3.P)
that Ank is regular for all fields k. Once we know the definition of étale morphism,
we will realize that every smooth variety locally admits an étale morphism to An k
(Observation 13.6.4). You can then show that if π : X → Y is an étale morphism of
locally Noetherian schemes, and p ∈ X, then p is regular if π(p) is regular. This
will be shown in Exercise 24.8.E, but you could reasonably do this after reading the
definition of étaleness.

13.2.8. Caution: Regularity does not imply smoothness. If k is not perfect, then regu-
larity does not imply smoothness, as demonstrated by the following example. Let
k = Fp (u), and consider the hypersurface X = Spec k[x]/(xp −u). Now k[x]/(xp −u)
df
is a field, hence regular. But if f(x) = xp − u, then f(u1/p ) = dx (u1/p ) = 0, so
the Jacobian criterion fails — X is not smooth over k. (Never forget that smooth-
ness requires a choice of field — it is a “relative” notion, and we will later define
smoothness over an arbitrary scheme, in §13.6.)

13.2.9. In case the previous example is too “small” to be enlightening (because

the scheme in question is smooth over a different field, namely k[x]/(xp − u)),
here is another. Let k = Fp (u) as before, with p > 2, and consider the curve
Spec k[x, y]/(y2 − xp + u). Then the closed point (y, xp − u) is regular but not
smooth.
Thus you should not use “regular” and “smooth” interchangeably.

13.2.10. Regular local rings are integral domains.

You might expect from geometric intuition that a scheme is “locally irreducible”
at a “smooth” point. Put algebraically:

13.2.11. Theorem. — Suppose (A, m, k) is a regular local ring of dimension n. Then A

is an integral domain.
Before proving it (in §13.2.14), we give some consequences.

13.2.K. E XERCISE . Suppose p is a regular point of a locally Noetherian scheme X.

Show that only one irreducible component of X passes through p.
 © 2024 Ravi Vakil. Published by Princeton University Press. 369

13.2.L. E ASY E XERCISE . Show that a nonempty regular Noetherian scheme is

irreducible if and only if it is connected.

13.2.M. I MPORTANT E XERCISE ( REGULAR SCHEMES IN REGULAR SCHEMES ARE

REGULAR EMBEDDINGS ).
(a) Suppose (A, m, k) is a regular local ring of dimension n, and I ⊂ A is an ideal of
A cutting out a regular local ring of dimension d. Let r = n − d. Show that Spec A/I
is a regular embedding in Spec A. Hint: show that there are elements f1 , . . . , fr of
I giving a basis of the k-vector space I/(I ∩ m2 ). Show that the quotient of A by
both (f1 , . . . , fr ) and I yields dimension d regular local rings. By Exercise 12.1.D, a
surjection of integral domains of the same dimension must be an isomorphism.
(b) Suppose π : X → Y is a closed embedding of regular schemes. Show that π is a
regular embedding.
Exercise 13.2.M has the following striking geometric consequence.

13.2.N. E XERCISE ( GENERALIZING E XERCISE 12.3.F).

(a) Suppose W is a smooth variety of pure dimension d over an algebraically closed
field k, and X and Y are pure dimensional subvarieties (possibly singular) of W
codimension m and n respectively. Show that every component of X ∩ Y has
codimension at most m + n in W as follows. Show that the diagonal W = ∼ ∆⊂
W ×k W is a regular embedding of codimension d. Then follow the rest of the hint
to Exercise 12.3.F. (Why is there a hypothesis of smoothness rather than regularity
on W?)
(b) Remove the requirement that k be algebraically closed from part (a).

13.2.12. Remark. The following example shows that the smoothness hypotheses in
Exercise 13.2.N cannot be (completely) dropped. Let W = Spec k[w, x, y, z]/(wz −
xy) be the cone over the smooth quadric surface, which is an integral threefold.
Let X be the surface w = x = 0 and Y the surface y = z = 0; both lie in W. Then
X ∩ Y is just the origin, so we have two codimension 1 subvarieties meeting in a
codimension 3 subvariety. (It is no coincidence that X and Y are the affine cones
over two lines in the same ruling, see Exercise 9.3.L. This example will arise again
in Exercise 22.4.N.)

13.2.13. Proposition. — Smooth k-schemes are reduced.

This argument is surprisingly serpentine!

Proof. Suppose X is a smooth k-scheme, and let Xk := X ×Spec k Spec k as usual. Then
Xk is smooth (Exercise 13.2.H), so Xk is regular at its closed points (Exercise 13.2.J),
so the local rings of Xk at closed points are integral domains (Theorem 13.2.11)
and thus reduced, so Xk is reduced at its closed points and hence reduced (the
nonreduced locus is closed, §9.4.10), so X is reduced (Do you see why?). □

13.2.O. E XERCISE . Suppose p is a closed point of a smooth k-scheme X of dimension

n. Show that OX,p is an integral domain (necessarily of dimension n).

13.2.14. Proof of Theorem 13.2.11 (following [Liu, Prop. 2.11]). We prove the result by
induction on n. The case n = 0 is clear (dimension 0 regular local rings are fields,
Exercise 13.2.A). So assume n > 0, and that the result holds for smaller dimension.
370 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Suppose p is prime, and dim A/p = n. We wish to show that p = (0). Now
A/p is a Noetherian local ring of dimension n, and its Zariski cotangent space
is a quotient of that of A cut out by the “equations of p” (Exercise 13.1.B). But
dim m/m2 = n, so p ⊂ m2 . Hence A/p is a regular local ring of dimension n.
Choose any f ∈ m \ m2 (so f ∈/ p). By Exercise 12.3.I (applied to both A/(f) and
A/(p, f)), dim A/(f) ≥ n − 1, and dim A/(f, p) ≥ n − 1, but both rings have Zariski
tangent space of dimension n − 1, so both are regular local rings of dimension
n − 1, and by the inductive hypothesis are both integral domains. The only way
one integral domain can be the quotient of another integral domain of the same
dimension is if the quotient is an isomorphism (Exercise 12.1.D). Thus (p, f) = (f),
from which p ⊂ (f), from which p ⊂ fp, from which p = fp. Then by Nakayama’s
Lemma 8.2.9, p = (0) as desired. □

13.2.15. ⋆⋆ Checking regularity of k-schemes at closed points by base changing to

k.
(We revisit these ideas using a different approach in 21.2.Y, so you should read
this only if you are particularly curious.) The Jacobian criterion is a great criterion
for checking regularity of finite type k-schemes at k-valued points. The following
result extends its applicability to more general closed points.
Suppose X is a finite type k-scheme of pure dimension n, and p ∈ X is a closed
point with residue field k ′ . By the Nullstellensatz 3.2.6, k ′ /k is a finite extension
of fields; suppose that it is separable. Define π : Xk := X ×k k → X by base change
from Spec k → Spec k.

13.2.P. E XERCISE .
(a) Suppose f(x) ∈ k[x] is a separable polynomial (i.e., f has distinct roots in k), and
irreducible, so k ′′ := k[x]/(f(x)) is a field extension of k. Show that k ′′ ⊗k k is, as a
ring, k × · · · × k, where there are deg f = deg k ′′ /k factors.
(b) Show that π−1 (p) consists of deg(k ′ /k) reduced points.

13.2.Q. E XERCISE . Suppose p is a closed point of X, with residue field k ′ that is

separable over k of degree d. Show that Xk is regular at all the preimages p1 , . . . ,
pd of p if and only if X is regular at p as follows.
(a) Reduce to the case X = Spec A.
(b) Let m ⊂ A be the maximal ideal corresponding to p. By tensoring the exact
sequence

0 /m /A / k′ /0

with k (field extensions preserve exactness of sequences of vector spaces), we have

the exact sequence

0 / m ⊗k k / A ⊗k k / k ′ ⊗k k / 0.

Show that m ⊗k k ⊂ A ⊗k k is the ideal corresponding to the pullback of p to

Spec A ⊗k k. Verify that (m ⊗k k)2 = m2 ⊗k k.
(c) By tensoring the short exact sequence of k-vector spaces

0 / m2 /m / m/m2 /0
 © 2024 Ravi Vakil. Published by Princeton University Press. 371

with k, show that

X
d
dimk TXk ,pi = d dimk ′ TX,p .
i=1

(d) Use Exercise 12.1.H(a) and the inequalities dimk TXk ,pi ≥ dim Xk and dimk ′ TX,p ≥
dim X (Theorem 12.3.12) to conclude.

13.2.16. Remark. In fact, regularity at a single pi is enough to conclude regularity

at p. You can show this by following up on Exercise 13.2.Q; first deal with the
case when k ′ /k is Galois, and obtain some transitive group action of Gal(k ′ /k) on
{p1 , . . . , pd }. Another approach is given in Exercise 21.2.Y.

13.2.R. E XERCISE . Suppose k is perfect, and X is a finite type k-scheme. Show

that X is smooth if and only if X is regular at all closed points. In particular,
Theorem 13.2.7(a) holds. Hint: let Xk := X ×k k. Explain why X is smooth if and
only if Xk is smooth if and only if Xk is regular at all closed points if and only if X
is regular at all closed points. See Exercise 13.2.Q for the last step.

13.3 Examples

We now explore regularity and smoothness in practice through a series of

examples and exercises.

13.3.1. Geometric examples.

13.3.A. E ASY E XERCISE . Suppose k is a field. Show that A1k and A2k are regular, by
directly checking the regularity of all points. Show that P1k and P2k are regular.
The generalization to arbitrary dimension is harder, so we leave it to Exer-
cise 13.3.P. By contrast, it is basically immediate that An n
k and Pk are smooth.

13.3.B. E XERCISE ( THE J ACOBIAN CRITERION FOR REGULARITY FOR PROJECTIVE

HYPERSURFACES ).
(a) Show that the non-smooth points of the hypersurface f = 0 in Pn
k correspond to
the locus
∂f ∂f
f= = ··· = = 0.
∂x0 ∂xn
If the degree of the hypersurface is not divisible by char k (e.g., if char k = 0), show
∂f ∂f
P ∂f
that it suffices to check ∂x 0
= · · · = ∂xn
= 0. Hint: show that (deg f)f = i xi ∂x i
.
(b) If furthermore k = k, show that the singular closed points are given by the same
locus. (In fact, this gives all the singular points, not just the singular closed points,
cf. §13.2.3. We won’t use this, so we won’t prove it.)

13.3.C. E XERCISE . Suppose char k ̸= 2. Show that y2 z = x3 − xz2 in P2k is

an irreducible smooth curve. (Eisenstein’s Criterion gives one way of showing
irreducibility. Warning: we didn’t specify char k ̸= 3, so be careful when using the
Jacobian criterion.)
372 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

13.3.D. E XERCISE . Suppose char k ̸= 2. Show that a quadric hypersurface in Pn is

smooth if and only if it is maximal rank (defined in Exercise 5.4.J).

13.3.E. E XERCISE . Suppose k = k has characteristic 0. Show that there exists a

regular (projective) plane curve of degree d. Hint: try a “Fermat curve” xd + yd +
zd = 0. (Feel free to weaken the hypotheses. Bertini’s Theorem 13.4.2 will give
another means of showing existence.)

13.3.F. E XERCISE ( SEE F IGURE 13.3). Assume k = k and char k = 0 to avoid

distractions. Find the singular closed points (which will be the non-smooth points)
of the following plane curves.
(a) y2 = x2 + x3 . This is an example of a node, which we saw earlier in
Exercise 7.5.H.
(b) y2 = x3 . This is called a cusp; we met it earlier in Exercise 10.7.F.
(c) y2 = x4 . This is called a tacnode; we met it earlier in Exercise 10.7.G.

F IGURE 13.3. Plane curve singularities

13.3.2. Remark. There are different possible definitions of node, cusp, and tacnode.
One reasonable definition in the case k = k is the following. Suppose C is pure
dimension 1, and p ∈ C is a closed point. Let (A, m) be the local ring OC,p , so
A/m = k. We say p is a node of C if dimk m/m2 = 2, the kernel of

Sym2 (m/m2 ) / m2 /m3

has dimension one, and the kernel is not spanned by the square of an element
in m/m2 . We won’t need this definition, so you needn’t memorize it. But can
you figure out why it is reasonable? Can you make a reasonable guess for a
definition of a cusp? Another definition of node, cusp, and more will be given in
Definitions 28.2.1 and 28.2.2.

13.3.G. E XERCISE . Show that the twisted cubic Proj k[w, x, y, z]/(wz − xy, wy −
x2 , xz − y2 ) is smooth. (You could do this using the fact that it is isomorphic to P1 .
But do this with the explicit equations, for the sake of practice. The twisted cubic
was defined in Exercise 9.3.A.)

13.3.3. Tangent planes and tangent lines.

Suppose a scheme X ⊂ An is cut out by equations f1 , . . . , fr , and X is smooth
of (pure) dimension d at the k-valued point a = (a1 , . . . , an ). Then the tangent
d-plane to X at a (sometimes denoted Ta X) is given by the r equations
   
∂fi ∂fi
(a) (x1 − a1 ) + · · · + (a) (xn − an ) = 0,
∂x1 ∂xn
 © 2024 Ravi Vakil. Published by Princeton University Press. 373

∂fi ∂fi
where (as in (13.1.7.1)) ∂x1 (a) is the evaluation of ∂x1 at a = (a1 , . . . , an ).

13.3.H. E XERCISE . Why is this independent of the choice of defining equations f1 ,

. . . , fr of X?
The Jacobian criterion for regularity (Exercise 13.2.E) ensures that these r equa-
tions indeed cut out a d-plane. (If d = 1, this is called the tangent line.) This can be
readily shown to be notion of tangent plane that we see in multivariable calculus,
but note that here this is the definition, and thus don’t have to worry about δ’s and
ϵ’s. Instead we will have to just be careful that it behaves the way we want it to.

13.3.I. E XERCISE . Compute the tangent line to the curve of Exercise 13.3.F(b) at
(1, 1).

13.3.J. E XERCISE . Compute the (projective) tangent line to the twisted cubic curve
of Exercise 13.3.G at [1, 1, 1, 1].

13.3.K. E XERCISE . Suppose X ⊂ Pn k (k as usual a field) is cut out by homogeneous

equations f1 , . . . , fr , and p ∈ X is a k-valued point that is smooth of (pure) dimen-
sion d. Define the (projective) tangent d-plane to X at p. (Definition 9.3.3 gives the
definition of a d-plane in Pn k , but you shouldn’t need to refer there.)

13.3.4. Side Remark to help you think cleanly. We would want the definition of tangent
k-plane to be natural in the sense that for any automorphism σ of An k (or, in the case
of the previous Exercise, Pn k ), σ(Tp X) = Tσ(p) σ(X). You could verify this by hand,
but you can also say this in a cleaner way, by interpreting the equations cutting out
the tangent space in a coordinate free manner. Informally speaking, we are using
the canonical identification of n-space with the tangent space to n-space at p, and
using the fact that the Jacobian “linear transformation” cuts out Tp X in Tp An in a
way independent of choice of coordinates on An or defining equations of X. Your
solution to Exercise 13.3.H will help you start to think in this way.

13.3.L. E XERCISE . Suppose X ⊂ Pn k is a degree d hypersurface cut out by f = 0, and

L is a line not contained in X. Exercise 9.3.D (a case of Bézout’s Theorem) showed
that X and L meet at d points, counted “with multiplicity”. Suppose L meets X
“with multiplicity at least 2” at a k-valued point p ∈ L ∩ X, and that p is a regular
point of X. Show that L is contained in the tangent plane to X at p. (Do you have a
picture of this in your mind?)

13.3.5. Arithmetic examples.

13.3.M. E ASY E XERCISE . Show that Spec Z is a regular curve.

13.3.N. E XERCISE . (This tricky exercise is for those who know about the primes of
the Gaussian integers Z[i].) There are several ways of showing that Z[i] is dimension
1. (For example: (i) it is a principal ideal domain; (ii) it is the normalization of Z in
the field extension Q(i)/Q; (iii) using Krull’s Principal Ideal Theorem 12.3.3 and
the fact that dim Z[x] = 2 by Exercise 12.1.I.) Show that Spec Z[i] is a regular curve.
(There are several ways to proceed. You could use Exercise 13.1.B. As an example
to work through first, consider the prime (1 + i), which is cut out by the equation
374 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

1 + x in Spec Z[x]/(x2 + 1).) We will later (§13.5.9) have a simpler approach once
we discuss discrete valuation rings.

13.3.O. E XERCISE . Show that [(5, 5i)] is the unique singular point of Spec Z[5i].
∼ Z[5i]5 . Use the previous exercise.)
(Hint: Z[i]5 =

13.3.6. Back to geometry: An

k is regular.
The key step to showing that Ank is regular (where k is a field) is the following.

13.3.7. Proposition. — Suppose (B, n, k) is a regular local ring of dimension d. Let

ϕ : B → B[x]. Suppose p is a prime ideal of A := B[x] such that nB[x] ⊂ p. Then Ap is a
regular local ring.

Proof. Geometrically: we have a morphism π : X = Spec B[x] → Y = Spec B, and

π([p]) = [n]. The fiber π−1 ([n]) is Spec(B[x]/nB[x]) = Spec k[x]. Thus either (i) [p] is
the fiber A1k above [n], or (ii) [p] is a closed point of the fiber A1k . (See Figure 13.4.)

F IGURE 13.4. The geometry behind the proof of Proposition 13.3.7

Before considering these two cases, we make two remarks. As (B, n) is a

regular local ring of dimension d, n is generated by d elements of B, say f1 , . . . , fd
(Exercise 13.2.D), and there is a chain of prime ideals
(13.3.7.1) 0 = q0 ⊂ q1 ⊂ · · · ⊂ qd = n.
Case (i): [p] is the fiber A1k . In this case, p = nB[x]. Thus by Krull’s Height
Theorem 12.3.10, the codimension of p = nB[x] is at most d, because p is generated
by the d elements f1 , . . . , fd . But p has codimension at least d: given the chain
(13.3.7.1) ending with n, we have a corresponding chain of prime ideals of B[x]
ending with nB[x]. Hence the codimension of p is precisely d, and pAp is generated
by f1 , . . . , fd , implying that it is a regular local ring.
Case (ii): [p] is a closed point of the fiber A1k . The closed point p of k[x] corresponds
to some monic irreducible polynomial g(x) ∈ k[x]. Arbitrarily lift the coefficients of
g to B; we sloppily denote the resulting polynomial in B[x] by g(x) as well. Then
p = (f1 , . . . , fd , g), so by Krull’s Height Theorem 12.3.10, the codimension of p
is at most d + 1. But p has codimension at least d + 1: given the chain (13.3.7.1)
ending with n, we have a corresponding chain in B[x] ending with nB[x], and we
 © 2024 Ravi Vakil. Published by Princeton University Press. 375

can extend it by appending p. Hence the codimension of p is precisely d + 1, and

pAp is generated by f1 , . . . , fd , g, implying that it is a regular local ring. □

13.3.P. E XERCISE . Use Proposition 13.3.7 to show that if X is a regular (locally

Noetherian) scheme, then so is X × A1 . In particular, show that An
k is regular. (Hint:
think about fibers.)

13.4 Bertini’s Theorem

We now discuss Bertini’s Theorem, a fundamental classical result.

13.4.1. Definition: dual projective space. The dual (or dual projective space) to Pn k
(with projective coordinates x0 , . . . , xn ), is informally the space of hyperplanes in
Pn n
k . Somewhat more precisely, it is a projective space Pk with projective coordinates
n∨
a0 , . . . , an (which we denote Pk in the futile hope of preventing confusion),
along with the data of the “incidence variety” or “incidence correspondence” I ⊂
Pn × Pn∨ cut out by the equation a0 x0 + · · · + an xn = 0. (This is often called the
universal hyperplane.) Note that the k-valued points of Pn∨ indeed correspond to
hyperplanes in Pn defined over k, and this is also clearly a duality relation (there
is a symmetry in the definition between the x-variables and the a-variables). This
definition is concrete enough to use in practice, and extends over an arbitrary base
(notably Spec Z). (But if you have a delicate and refined sensibility, you may want
to come up with a coordinate-free definition.)

13.4.2. Bertini’s Theorem. — Suppose X is a smooth subvariety of Pn k of (pure) dimension

d. Then there is a nonempty (=dense) open subset U of dual projective space Pn∨
k such that
for every point p = [H] ∈ U, H doesn’t contain any component of X, and the scheme H ∩ X
is smooth over κ(p) of (pure) dimension d − 1.
(You should perhaps think first of the case where κ(p) = k, i.e., H is a hy-
perplane cut out by a linear form with coefficients in k.) Any theorem of this
flavor is often called a “Bertini Theorem”. One example is the Kleiman-Bertini
Theorem 21.6.10, which was not proved jointly by Kleiman and Bertini.
As an application of Bertini’s Theorem 13.4.2, a general degree d > 0 hyper-
surface in Pnk intersects X in a smooth subvariety of codimension 1 in X: replace
X ,→ Pn with the composition

X / Pn   νd / PN

where νd is the dth Veronese embedding (9.3.8). Here “general” has its usual mean-
ing in algebraic geometry, see §10.3.7. (See Exercise 13.4.C for a generalization.) A
useful consequence of this, taking X = Pn : we immediately see that there exists a
smooth degree d hypersurface in Pn (over any infinite field k), without any messing
around with specific equations (as in the special case Exercise 13.3.E).
Exercise 21.6.I gives a useful improvement of Bertini’s Theorem in characteristic
0 (see Exercise 21.6.J).

Proof. The central idea of the proof is quite naive. We will describe the hyperplanes
that are “bad”, and show that they form a closed subset of dimension at most n − 1
376 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

∨
of Pn
k , and hence that the complement is a dense open subset. Somewhat more
∨
precisely, we will define a projective variety Z ⊂ X × Pn
k that can informally be
described as:
(13.4.2.1) Z = {(p ∈ X, H ⊂ Pn
k ) : p ∈ H,
and either p is a singular point of H ∩ X, or X ⊂ H} .
We will see that the projection π : Z → X has fibers at closed points that are projective
spaces of dimension n − 1 − dim X, and hence that dim Z ≤ n − 1. Thus the image
∨
of Z in Pn
k will be a closed subset (Theorem 8.4.10), of dimension of at most n − 1,
so its complement will be open and nonempty. We now put this strategy into
action.

13.4.3. We first define Z more precisely, in terms of equations on Pn × Pn ∨ , where

the coordinates on Pn are x0 , . . . xn , and the dual coordinates on Pn ∨ are a0 , . . . ,
an . Suppose X is cut out by f1 , . . . , fr . Then we take these equations as the first
of the defining equations of Z. (So far we have defined the subscheme X × Pn ∨ .)
We also add the equation a0 x0 + · · · + an xn = 0. (So far we have described the
subscheme of Pn × Pn ∨ corresponding to points (p, H) where p ∈ X and p ∈ H.)
Note that the projective Jacobian matrix (13.1.9.1)
 ∂f1 ∂fr
∂x0 (p) · · ·

∂x0 (p)
 .. .. .. 
 . . . 
∂f1 ∂fr
∂xn (p) ··· ∂xn (p)

has corank precisely d + 1 at all points of X by the projective Jacobian corank

criterion (Exercise 13.1.N). We then require that the projective Jacobian matrix with
a new column  
a0
 .. 
 . 
an
appended has corank ≥ d + 1 (hence = d + 1). This is precisely the notion that we
wished to capture — do you see why? This is cut out by equations (the determinants
of certain minors). (Remark 13.4.4 works through an example, which may help
clarify how this works.)
We next show that dim Z ≤ n − 1. For each closed point p ∈ X, let Wp be the
locus of hyperplanes containing p, such that H ∩ X is singular at p, or else contains
all of X; what is the dimension of Wp ? Suppose dim X = d. Then the restrictions
on the hyperplanes in the definition of Wp correspond to d + 1 linear conditions.
(Do you see why?) This means that Wp is a codimension d + 1, or dimension
n − d − 1, projective space. Thus the fiber of π : Z → X over each closed point has
pure dimension n − d − 1. By Key Exercise 12.4.A, this implies that dim Z ≤ n − 1.
(If you wish, you can use Exercise 12.4.D to show that dim Z = n − 1, and you can
later show that Z is a projective bundle over X, once you know what a projective
bundle is, Definition 17.2.4. But we don’t need this for the proof.) □

13.4.4. Remark. Here is an example that may help convince you that the algebra of
paragraph 13.4.3 is describing the geometry we desire. Consider the plane conic
x20 − x21 − x22 = 0 over a field of characteristic not 2, which you might picture as the
 © 2024 Ravi Vakil. Published by Princeton University Press. 377

circle x2 + y2 = 1 from the real picture in the chart U0 . Consider the point [1, 1, 0],
corresponding to (1, 0) on the circle. We expect the tangent line in the affine plane
to be x = 1, which corresponds to x0 − x1 = 0. Let’s see what the algebra gives us.
The projective Jacobian matrix is
   
2x0 2
 −2x1  =  −2  ,
−2x2 0
which indeed has rank 1 as expected. Our recipe asks that the matrix
 
2 a0
 −2 a1 
0 a2
have rank 1 (i.e., a0 = −a1 and a2 = 0), and also that a0 x0 + a1 x1 + a2 x2 = 0,
which (you should check) is precisely what we wanted.

13.4.A. E ASY E XERCISE .

(a) Prove Bertini’s Theorem with the weaker hypothesis that X has finitely many
non-smooth points.
(b) Prove Bertini’s Theorem with the weaker hypothesis that X → Pn k is a locally
closed embedding (but with X still smooth).

13.4.B. E XERCISE . Show that if X is a projective variety of dimension n in Pm ,

then the intersection of X with n general hyperplanes consists of a finite number
of reduced points. More precisely: if Pm∨ is the dual projective space, then there
is a nonempty Zariski-open subset U ⊂ (Pm∨ )n such that for each closed point
(H1 , . . . , Hn ) of U, the scheme-theoretic intersection H1 ∩ · · · ∩ Hn ∩ X consists of a
finite number of reduced points. (The number of such points, counted correctly, is
called the degree of the variety, see Exercise 18.6.M.)

13.4.C. E XERCISE ( MANY- HYPERSURFACE B ERTINI ). Suppose X is a smooth sub-

variety of Pn
k of (pure) dimension d ≥ r. Suppose e1 , . . . , er are positive integers,
so
e1 +n er +n
(13.4.4.1) P( n )−1 × · · · × P( n )−1
parametrizes r-tuples of hypersurfaces (H1 , . . . , Hr ) of degrees e1 , . . . , er respec-
tively. Show that there is a dense open subset U of (13.4.4.1) such that for every
point p = [(H1 , . . . , Hr )] ∈ U, H1 ∩ · · · ∩ Hr doesn’t contain any component of X,
and H1 ∩ · · · ∩ Hr ∩ X is smooth over κ(p) of (pure) dimension d − r. Informally:
for a general choice of hypersurfaces H1 , . . . , Hr of degrees e1 , . . . , er respectively,
H1 ∩ · · · ∩ Hr ∩ X is smooth of (pure) dimension d − r. (Hint: Don’t apply Bertini’s
Theorem 13.4.2. Instead, try to give the “same” proof.) The lesson of this exercise
isn’t that you should memorize this specific result. It is that you can prove any
result of this flavor that you need.

13.4.5. Dual varieties. Our discussion of Bertini gives us an excuse to mention

a classical construction. The image of Z (see (13.4.2.1)) in Pn∨ is called the dual
variety of X. As dim Z = n − 1, we “expect” the dual of X to be a hypersurface of
Pn∨ . It is a nonobvious fact that this in fact is a duality: the dual of the dual of X is
378 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

X itself (see [H2, Thm. 15.24]). The following exercise will give you some sense of
the dual variety.

13.4.D. E XERCISE . Show that the dual of a hyperplane in Pn is the corresponding

point of the dual space Pn∨ . In this way, the duality between Pn and Pn∨ is a
special case of duality between projective varieties.

13.4.E. E XERCISE . Suppose C ⊂ P2 is a smooth conic over an algebraically closed

field of characteristic not 2. Show that the dual variety to C is also a smooth conic.
Thus for example, through a general point in the plane (if k = k), there are two
tangents to C. (The points on a line in the dual plane corresponds to those lines
through a point of the original plane.)

13.4.F. ⋆ E XERCISE ( THERE IS ONE SMOOTH CONIC TANGENT TO FIVE GENERAL

LINES , AND GENERALIZATIONS ). Continuing the notation of the previous problem,
show that the number of smooth conics C containing i generally chosen points
and tangent to 5 − i generally chosen lines is 1, 2, 4, 4, 2, 1 respectively for i =
0, 1, 2, 3, 4, 5. You might interpret the symmetry of the sequence in terms of the
duality between the conic and the dual conic. This fact was likely known in the
paleolithic era.

13.5 Discrete valuation rings, and Algebraic Hartogs’s Lemma

The case of (co)dimension 1 is important, because if you understand how prime
ideals behave that are separated by dimension 1, then you can use induction to
prove facts in arbitrary dimension. This is one reason why Krull’s Principal Ideal
Theorem 12.3.3 is so useful.
A dimension 1 regular local ring can be thought of as a “germ of a smooth curve”
(see Figure 13.5). Two examples to keep in mind are k[x](x) = {f(x)/g(x) : x ̸ | g(x)}
and Z(5) = {a/b : 5 ̸ | b}. The first example is “geometric” and the second is
“arithmetic”, but hopefully it is clear that they have something fundamental in
common.

F IGURE 13.5. A germ of a curve

The purpose of this section is to give a long series of equivalent definitions of

these rings. Before beginning, we quickly sketch these seven definitions. There
are a number of ways a Noetherian local ring can be “nice”. It can be regular, or a
principal ideal domain, or a unique factorization domain, or normal. In dimension
1, these are the same. Also equivalent are nice properties of ideals: if m is principal;
or if all ideals are either powers of the maximal ideal, or 0. Finally, the ring can have
a discrete valuation, a measure of “size” of elements that behaves particularly well.

13.5.1. Theorem. — Suppose (A, m) is a Noetherian local ring of dimension 1. Then the
following are equivalent.
 © 2024 Ravi Vakil. Published by Princeton University Press. 379

(a) (A, m) is regular.

(b) m is principal.

Proof. Here is why (a) implies (b). If A is regular, then m/m2 is one-dimensional.
Choose any element t ∈ m \ m2 . Then t generates m/m2 , so generates m by
Nakayama’s Lemma 8.2.H (m is finitely generated by the Noetherian hypothesis).
We call such an element a uniformizer.
Conversely, if m is generated by one element t over A, then m/m2 is generated
by one element t over A/m = k. Since dimk m/m2 ≥ 1 by Theorem 12.3.12, we
have dimk m/m2 = 1, so (A, m) is regular. □
We will soon use a useful fact, which is geometrically motivated, and is a
special case of an important result, the Artin-Rees Lemma 13.9.3. We will prove it
in §13.9.

13.5.2. Proposition. — If (A, m) is a Noetherian local ring, then ∩i mi = 0.

13.5.3. The geometric intuition for this is that any function that is analytically zero
at a point (vanishes to all orders) actually vanishes in an open neighborhood of
that point. (Exercise 13.9.B will make this precise.) The geometric intuition also
2
suggests an example showing that Noetherianness is necessary: consider e−1/x
in the ring of germs of C∞ -functions on R at the origin. (In particular, this implies
that the ring of C∞ -functions on R, localized at the origin, is not Noetherian!)
It is tempting to argue that
(13.5.3.1) m(∩i mi ) = ∩i mi ,
and then to use Nakayama’s Lemma 8.2.9 to argue that ∩i mi = 0. Unfortunately, it
is not obvious that this first equality is true: product does not commute with infinite
descending intersections in general. (Aside: product also doesn’t commute with
finite intersections in general, as for example in k[x, y, z]/(xz − yz), (z)((x) ∩ (y)) ̸=
(xz) ∩ (yz).) We will establish Proposition 13.5.2 in Exercise 13.9.A(b). (We could
do it directly right now without too much effort.)

13.5.4. Theorem. — Suppose (A, m) is a Noetherian local ring of dimension 1. Then (a)
and (b) are equivalent to:
(c) A is an integral domain, and all ideals are of the form mn (for n ≥ 0) or (0).

Proof. Assume (a): suppose (A, m, k) is a regular local ring of dimension 1. Then I
claim that mn ̸= mn+1 for any n. Otherwise, by Nakayama’s Lemma, mn = 0, from
which tn = 0. But A is an integral domain (by Theorem 13.2.11), so t = 0, from
which A = A/m is a field, which doesn’t have dimension 1, contradiction.
I next claim that mn /mn+1 is dimension 1. Reason: mn = (tn ). So mn is
generated as an a A-module by one element, and mn /(mmn ) is generated as a
(A/m = k)-module by one element (nonzero by the previous paragraph), so it is a
one-dimensional vector space.
So we have a chain of ideals A ⊃ m ⊃ m2 ⊃ m3 ⊃ · · · with ∩mi = (0)
(Proposition 13.5.2). We want to say that there is no room for any ideal besides
these, because each pair is “separated by dimension 1”, and there is “no room at
the end”. Proof: suppose I ⊂ A is an ideal. If I ̸= (0), then there is some n such that
I ⊂ mn but I ̸⊂ mn+1 . Choose some u ∈ I \ mn+1 . Then (u) ⊂ I. But u generates
380 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

mn /mn+1 , hence by Nakayama’s Lemma it generates mn , so we have mn ⊂ I ⊂ mn ,

so we are done: (c) holds.
We now show that (c) implies (a). Assume (a) does not hold: suppose we
have a dimension 1 Noetherian local integral domain that is not regular, so m/m2
has dimension at least 2. Choose any u ∈ m − m2 . Then (u, m2 ) is an ideal, but
m2 ⊊ (u, m2 ) ⊊ m. □

13.5.A. E ASY E XERCISE . Suppose (A, m) is a Noetherian dimension 1 local ring.

Show that (a)–(c) above are equivalent to:
(d) A is a principal ideal domain.

13.5.5. Discrete valuation rings. We next define the notion of a discrete valuation
ring. Suppose K is a field. A discrete valuation on K is a surjective homomorphism
v : K× → Z (in particular, v(xy) = v(x) + v(y)) satisfying

v(x + y) ≥ min(v(x), v(y))

except if x + y = 0 (in which case the left side is undefined). (Such a valuation is
called non-archimedean, although we will not use that term.) It is often convenient
to say v(0) = ∞. More generally, a valuation is a surjective homomorphism
v : K× → G to a totally ordered abelian group G, although this isn’t so important to
us. The symbol “val” will denote valuations.
Here are three key examples.
(i) (the 5-adic valuation) K = Q, v(r) is the “power of 5 appearing in r”, e.g.,
v(35/2) = 1, v(27/125) = −3.
(ii) K = k(x), v(f) is the “power of x appearing in f.”
(iii) K = k(x), v(f) is the negative of the degree. This is really the same as (ii),
with x replaced by 1/x.
Then 0 ∪ {x ∈ K× : v(x) ≥ 0} is a ring, which we denote Ov . It is called the
valuation ring of v. (Not every valuation is discrete. Consider the ring of Puiseux
series over a field k, K = ∪n≥1 k((x1/n )), with v : K× → Q given by v(xq ) = q.)

13.5.B. E XERCISE . Describe the valuation rings in the three examples (i)–(iii) above.
(You will notice that they are familiar-looking dimension 1 Noetherian local rings.
What a coincidence!)

13.5.C. E XERCISE . Show that {0} ∪ {x ∈ K× : v(x) > 0} is the unique maximal ideal
of the valuation ring. (Hint: show that everything in the complement is invertible.)
Thus the valuation ring is a local ring.
An integral domain A is called a discrete valuation ring (or DVR) if there exists
a discrete valuation v on its fraction field K = K(A) for which Ov = A. Similarly, A
is a valuation ring if there exists a valuation v on K for which Ov = A.
Now if A is a regular local ring of dimension 1, and t is a uniformizer (a
generator of m as an ideal, or equivalently of m/m2 as a k-vector space) then any
nonzero element r of A lies in some mn \ mn+1 , so r = tn u where u is invertible
(as tn generates mn by Nakayama, and so does r), so K(A) = At = A[1/t]. So any
element of K(A)× can be written uniquely as utn where u is invertible and n ∈ Z.
Thus we can define a valuation v by v(utn ) = n.
 © 2024 Ravi Vakil. Published by Princeton University Press. 381

13.5.D. E XERCISE . Show that v is a discrete valuation.

13.5.E. E XERCISE . Conversely, suppose (A, m) is a discrete valuation ring. Show

that (A, m) is a regular local ring of dimension 1. (Hint: Show that the ideals are all
of the form (0) or In = {r ∈ A : v(r) ≥ n}, and (0) and I1 are the only prime ideals.
Thus we have Noetherianness, and dimension 1. Show that I1 /I2 is generated by
the image of any element of I1 − I2 .)
Hence we have proved:

13.5.6. Theorem. — An integral domain A is a Noetherian local ring of dimension 1

satisfying (a)–(d) if and only if
(e) A is a discrete valuation ring.

13.5.F. E XERCISE . Show that there is only one discrete valuation on a discrete
valuation ring.

13.5.7. Definition. Thus any regular local ring of dimension 1 comes with a unique
valuation on its fraction field. If the valuation of an element is n > 0, we say that
the element has a zero of order n. If the valuation is −n < 0, we say that the
element has a pole of order n. We will come back to this shortly, after dealing with
(f) and (g).

13.5.8. Theorem. — Suppose (A, m) is a Noetherian local ring of dimension 1. Then

(a)–(e) are equivalent to:
(f) A is a unique factorization domain, and
(g) A is an integral domain, integrally closed (in its fraction field).

Proof. (a)–(e) clearly imply (f), because we have the following stupid unique factor-
ization: each nonzero element of A can be written uniquely as utn where n ∈ Z≥0
and u is invertible.
Now (f) implies (g), because unique factorization domains are integrally closed
in their fraction fields (Exercise 5.4.F).
It remains to check that (g) implies (a)–(e). We will show that (g) implies (b).
Suppose (A, m) is a Noetherian local integral domain of dimension 1, integrally
closed in its fraction field K(A). Choose any nonzero r ∈ m. Then S = A/(r)
is a Noetherian local ring of dimension 0 — its only prime is the image of m,
which we denote n to avoid confusion. Then nN = 0, where N is sufficiently large
(Proposition 12.1.9). Hence there is some n such that nn = 0 but nn−1 ̸= 0.
Now comes the crux of the argument. Thus in A, mn ⊂ (r) but mn−1 ̸⊂ (r).
Choose s ∈ mn−1 − (r). Consider s/r ∈ K(A). As s ∈ / (r), s/r ∈
/ A, so as A is
integrally closed, s/r is not integral over A.
Now (inside K(A)) sr m ̸⊂ m (or else sr m ⊂ m would imply that m is a faithful
A[ r ]-module, contradicting Exercise 8.2.I). But sm ⊂ mn ⊂ rA, so sr m ⊂ A. Thus
s
s r
r m = A, from which m = s A, so m is principal. □

13.5.9. Example: Spec Z[i] is regular. We now have a simpler explanation of why
Spec Z[i] is regular (Exercise 13.3.N). First, Z[i] is a unique factorization domain
(§5.4.5(0)), hence normal (§5.4.6). It is dimension 1, so its closed (codimension 1)
382 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

points are regular by Theorem 13.5.8. Its generic point is also regular, as Z[i] is an
integral domain.

13.5.10. Example: “curves are regular almost everywhere”. By Theorem 13.5.8 and
Exercise 10.7.O, every one-dimensional variety is regular away from finitely many
closed points.

13.5.11. Aside: Dedekind domains. A Dedekind domain is a Noetherian integral

domain of dimension at most one that is normal (integrally closed in its fraction
field). The localization of a Dedekind domain at any prime but (0) (i.e., a codimen-
sion one prime) is hence a discrete valuation ring. This is an important notion,
but we won’t use it much. Rings of integers of number fields are examples, see
§10.7.1.
√ In particular, if n is a square-free integer congruent to 3 (mod 4), then
Z[ n] is a Dedekind domain, by Exercise 5.4.I(a). If you wish you can prove unique
factorization of ideals in a Dedekind domain: any nonzero ideal in a Dedekind domain
can be uniquely factored into prime ideals. (Hint: the Jordan-Hölder package, §6.5.)
By the same argument as in §13.5.9, Dedekind domains are regular.

13.5.12. Poles and zeroes of functions on Noetherian schemes. We now generalize

the sense in which (x−2)2 /(x−3)4 on A1C has a double zero at x = 2 and a quadruple
pole at x = 3.

13.5.13. Definition: Regular in codimension one (R1 ). We say that a Noetherian ring A
is regular in codimension 1 (or R1 for short) if its localization at all codimension
one primes are regular local rings. We say a locally Noetherian scheme A is
regular in codimension 1 (or R1 ) if it is regular at its codimension one points. By
Theorem 13.5.8, Noetherian normal schemes are regular in codimension 1.

13.5.14. Definition: Valuation at an R1 point. Suppose X is a locally Noetherian

scheme. Then for any regular codimension 1 point p (i.e., any point p where OX,p
is a regular local ring of dimension 1), we have a discrete valuation valp . If f is any
nonzero element of the fraction field of OX,p (e.g., if X is integral, and f is a nonzero
element of the function field of X), then if valp (f) > 0, we say that f has a zero of
order valp (f) at p, and if valp (f) < 0, we say that f has a pole of order − valp (f)
at p. (We are not yet allowed to discuss order of vanishing at a point that is not
regular and codimension 1. One can make a definition, but it doesn’t behave as
well as it does when have you have a discrete valuation.) Can you see why 75/34
has a double zero at 5, and a single pole at 2? Can you describe the zeros and poles
of x3 (x + y)/(x2 + y)3 on A2k ?

13.5.G. E XERCISE ( FINITENESS OF ZEROS AND POLES ON N OETHERIAN SCHEMES ).

Suppose X is an integral Noetherian scheme, and f ∈ K(X)× is a nonzero element of
its function field. Show that f has a finite number of zeros and poles. (Hint: reduce
to X = Spec A. If f = f1 /f2 , where fi ∈ A, prove the result for fi .)

13.5.15. Remark. A (Noetherian) scheme can be singular in codimension 2 and

still be normal. For example, you have shown that the cone x2 + y2 = z2 in A3 in
characteristic not 2 is normal (Exercise 5.4.I(b)), but it is singular at the origin (the
Zariski tangent space is visibly three-dimensional).
 © 2024 Ravi Vakil. Published by Princeton University Press. 383

But singularities of normal schemes are not so bad in some ways: we will
very shortly (§13.5.18) discuss Algebraic Hartogs’s Lemma 13.5.19 for Noetherian
normal schemes, which states that you can extend functions over codimension 2
sets.

13.5.16. Remark. We know that for Noetherian rings we have implications

unique factorization domain =⇒ integrally closed =⇒ regular in codimension 1.
Hence for locally Noetherian schemes, we have similar implications:
factorial =⇒ normal =⇒ regular in codimension 1.
Here are two examples to show you that these inclusions are strict.

13.5.H. E XERCISE ( THE pinched plane). Let A be the subring k[x3 , x2 , xy, y] of
k[x, y]. (Informally, we allow all polynomials that don’t include a nonzero multiple
of the monomial x.) Show that Spec k[x, y] → Spec A is a normalization. Show that
A is not integrally closed. Show that Spec A is regular in codimension 1. (Hint for
the last part: show it is dimension 2, and when you throw out “the origin” you get
something regular, by inverting x2 and y respectively, and considering Ax2 and
Ay .)

13.5.17. Example (the quadric surface once again). Suppose k is algebraically closed
of characteristic not 2. Then k[w, x, y, z]/(wz − xy) is integrally closed, but not a
unique factorization domain, see Exercise 5.4.L (and Exercise 13.1.E).

13.5.18. Algebraic Hartogs’s Lemma for Noetherian normal schemes.

Hartogs’s Lemma in several complex variables states (informally) that a holo-
morphic function defined away from a codimension two set can be extended over
that. We now describe an algebraic analog, for Noetherian normal schemes. (It may
also be profitably compared to the Second Riemann Extension Theorem.) We will
use this repeatedly and relentlessly when connecting line bundles and divisors in
§15.4.

13.5.19. Algebraic Hartogs’s Lemma. — Suppose A is a integrally closed Noetherian

integral domain. Then
A = ∩p codimension 1 Ap .
The equality takes place in K(A); recall that any localization of an integral
domain A is naturally a subset of K(A) (Exercise 1.2.C). Warning: few people call
this Algebraic Hartogs’s Lemma. I call it this because it parallels the statement in
complex geometry.
One might say that if f ∈ K(A) does not lie in Ap where p has codimension 1,
then f has a pole at [p], and if f ∈ K(A) lies in pAp where p has codimension 1, then
f has a zero at [p].

13.5.20. It is worth geometrically interpreting Algebraic Hartogs’s Lemma as saying

that a rational function on a Noetherian normal scheme with no poles is in fact regular (an
element of A). Informally: “Noetherian normal schemes have the Hartogs property.”
One can state Algebraic Hartogs’s Lemma more generally in the case that
Spec A is a Noetherian normal scheme, meaning that A is a product of Noetherian
normal integral domains; the reader may wish to do so.
384 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Another generalization (and something closer to the “right” statement) is that

if A is a subring of a field K, then the integral closure of A in K is the intersection
of all valuation rings of K containing A; see [AtM, Cor. 5.22] for explanation and
proof.

13.5.21. ⋆ Proof of Algebraic Hartogs’s Lemma 13.5.19.

The key clever construction in the proof is the following. If A is an integral
domain, and r ∈ K(A), then r ∈ A if and only if r ∈ Ap for all primes p of A, i.e., if
r is regular at all points of Spec A. For this reason, it makes sense to ask at what
points r ∈ K(A) is not regular. Our “ideal of denominators” construction (from the
proof Proposition 5.4.2) shows that it is a closed subset: if I = {a ∈ A : ar ∈ A}, then
the points [p] of V(I) = Supp A/I are precisely the points where r is not regular.
We get a more refined understanding of this closed subset from an easy-to-define
module Ma/b .

13.5.22. Suppose r = a/b ∈ / A, where a, b ∈ A. Let a be the image of a in A/(b).

The fact that r ∈/ A means that a ̸= 0 in A/(b). Let Ma/b be the A-submodule of
a/b
A/(b) generated by a. We have constructed Ma/b so that Mq = 0 if and only if
a/b ∈ Aq . (Translation: a is not a multiple of b in Ap if and only if a ̸= 0 in Ap /(b).)
We wish to show Algebraic Hartogs’s Lemma, which says that if A is integrally
closed, then it suffices to instead check only the height one prime ideals. We begin
with an approximation of this fact, which works without the “integrally closed”
hypothesis.

13.5.23. Lemma. — Suppose A is a Noetherian integral domain, and r ∈ K(A). Then

r ∈ A if and only if for every prime p ∈ Ass(A/(g)) for every nonzero g ∈ A, we have
r ∈ Ap .

Proof. Clearly if r ∈ A, then r ∈ Ap for all p. So assume that r = a/b ∈ / A. We

will find such a p so that r ∈
/ Ap as described in the statement of the Lemma. We
take p to be any associated prime ideal of the nonzero module Ma/b constructed in
§13.5.22. As Ma/b is a submodule of A/(b), p is an associated prime of A/(b) as
well (Exercise 6.6.G(a)). □

13.5.24. Proposition. — If A is a Noetherian integrally closed integral domain, then

(i) A is regular in codimension 1 (Definition 13.5.13), and

(ii) for any b ∈ A, the A-module A/(b) has no embedded points.

Proof. (i) Integral closure is preserved by localization (Exercise 5.4.A), so for each
codimension 1 prime p, Ap is integrally closed. Thus by Theorem 13.5.8, Ap is
regular.
(ii) Let p be an associated prime of A/(b). We wish to show that p is codimen-
sion 1 in A. The truth of the statement is preserved by replacing A by its localization
Ap , so it suffices to prove the following: “Suppose (A, m) is a Noetherian local ring,
an integrally closed integral domain, and b ∈ m \ {0}. Suppose m ∈ Ass(A/(b)).
Then m is principal (and thus codimension at most 1).”
We now prove this statement. Choose a ∈ A/(b) such that Ann(a) = m. Then
a ̸= 0. Lift a to a ∈ A. Then a/b ∈ K(A) \ A, and (a/b)m ⊂ A.
 © 2024 Ravi Vakil. Published by Princeton University Press. 385

We show by contradiction that (a/b)m ̸⊂ m; assume otherwise. Recall Exer-

cise 8.2.I: “Suppose A is a subring of K, r ∈ K, and N is a faithful A[r]-module that
is finitely generated as an A-module. Then r is integral over A.” We apply this
here with r = a/b and N = m. Then a/b is integral over A, implying (from integral
closure of A) that a/b ∈ A, contradiction.
Hence (a/b)m = A, from which m = (b/a)A, from which m is principal as
desired. □

13.5.25. Theorem. — Suppose A is a Noetherian integral domain, regular in codimension

1. Then (a) A is integrally closed if and only if (b) A = ∩p codimension 1 Ap .
This implies Algebraic Hartogs’s Lemma 13.5.19.

Proof. (a) implies (b). We apply Lemma 13.5.23. From Proposition 13.5.24(ii), the only
primes we need to check are codimension one.
(b) implies (a). The Ap are dimension one Noetherian domains that are regular,
and hence integrally closed by Theorem 13.5.8. Suppose r ∈ K(A) is integral
over A. Then r ∈ K(A) is integral over Ap for all codimension 1 p, so r ∈ Ap , so
r ∈ ∩p codimension 1 Ap = A. □
We conclude with a fact that we will later (§26.3), along with Proposition 13.5.24,
recognize as Serre’s R1 + S2 criterion for normality.

13.5.26. Proposition. — Suppose A is a Noetherian integral domain satisfying conditions

(i) and (ii) of Proposition 13.5.24. Then A is integrally closed.

Proof. Now A = ∩p Ap as p runs over the primes described in Lemma 13.5.23. From
condition (ii) of Proposition 13.5.24, these are all codimension one primes. The result
then follows from Theorem 13.5.25, using condition (i) of Proposition 13.5.24. □

13.5.27. Consequence of Algebraic Hartogs’s Lemma 13.5.19 we will use.

13.5.I. U SEFUL E XERCISE . If f is a nonzero rational function on a locally Noetherian

normal scheme, and f has no poles, show that f is regular.

13.6 Smooth (and étale) morphisms: First definition

In §13.2.3, we defined smoothness over a field. We now know enough to define

smooth morphisms (and, as a special case, étale morphisms) in general. Our definition
will be imperfect in a number of ways. For example, it will look a little surprising.
As another sign, it will not be obvious that it actually generalizes smoothness over a
field. A third flaw is that because it is of the form “there exist open covers satisfying
some property”, it is poorly designed to be used to show that some morphism is
not smooth. But it has a major advantage that we can give the definition right now,
and the basic properties of smooth morphisms will be straightforward to show.

13.6.1. Differential geometric motivation (see Figure 24.6). The notion of a smooth
morphism is motivated by the following idea from differential geometry (see,
e.g., [Tu] or [Var, §1.1–1.2] for the differential geometry). For the purposes of
386 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

this discussion, we say a map π : X → Y of manifolds (or real analytic spaces,

or any variant thereof) is smooth of relative dimension n if locally (on X) it looks
like Y ′ × Rn → Y ′ . (In differential geometry, this is usually called a submersion.)
Translation: it is “locally on the source a smooth fibration”. (If you have not heard
the word fibration before, don’t worry. Informally speaking, a fibration in some
“category of geometric objects” is a map π : X → Y that locally on the source is a
product V × W → V; in some sort of category of manifolds, a fibration is a “locally
on the source a smooth fibration” if the W in the local description can be taken to
be Ra for some a.)
In particular, “smooth of relative dimension 0” is the same as the important
notion of “local isomorphism” (“isomorphism locally on the source” — not quite the
same as “covering space”). Carrying this idea too naively into algebraic geometry
leads to the “wrong” definition (see Exercise 13.6.F). We give a better definition
now, and some equivalent definitions in §24.8.

13.6.2. Definition. A morphism π : X → Y is smooth of relative dimension n if

there exist open covers {Ui } of X and {Vi } of Y, with π(Ui ) ⊂ Vi , such that for every
i we have a commutative diagram

Ui o
∼ / W   open / Spec B[x1 , . . . , xn+r ]/(f1 , . . . , fr )
ρ
π|Ui ρ|W
 ∼
 u
Vi o / Spec B

where ρ is induced by the obvious map of rings in the opposite direction, and W is
an open subscheme of Spec B[x1 , . . . , xn+r ]/(f1 , . . . , fr ), such that the determinant
of the Jacobian matrix of the fi ’s with respect to the first r xi ’s
 
∂fj
(13.6.2.1) det
∂xi i,j≤r
is an invertible (= nowhere zero) function on W.
Étale means smooth of relative dimension 0.

13.6.3. Quick observations. Be sure to notice that if B is a field, then we have just the
right number of equations (r of them) to hope to cut out something of dimension
n in An+rk . More generally, intuitively and imprecisely, we have just the “right
number of equations” to cut out W in An+r B , with no redundant ones to spare.
From the definition, smooth morphisms are locally of finite presentation (hence
locally of finite type). Also, AnB → Spec B is immediately seen to be smooth of
relative dimension n. From the definition, the locus on X where π is smooth of
relative dimension n is open. In particular, the locus where π is étale is open.
As easy examples, open embeddings are étale, and the projection An × Y → Y is
smooth of relative dimension n. You may already have some sense of what makes
this definition imperfect. For example, how do you know a morphism is smooth if
it isn’t given to you in a form where the “right” variables xi and “right” equations
are clear?

13.6.A. M OTIVATING E XERCISE ( FOR THOSE WITH DIFFERENTIAL - GEOMETRIC

BACKGROUND ). Show how Definition 13.6.2 gives the definition in differen-
tial geometry, as described in §13.6.1. (Your argument might use the implicit
 © 2024 Ravi Vakil. Published by Princeton University Press. 387

function theorem, which “doesn’t work” in algebraic geometry, see Exercise 13.6.F.)
Show that for complex varieties, analytification turns étale morphisms into “local
biholomorphisms”.

13.6.B. E XERCISE . Show that the notion of smoothness of relative dimension n

(and in particular, étaleness) is local on both the source and target.
We can thus make sense of the phrase “π : X → Y is smooth of relative dimen-
sion n at p ∈ X”: it means that there is an open neighborhood U of p such that π|U
is smooth of relative dimension n.
The phrase smooth morphism (without reference to relative dimension n)
often informally means “smooth morphism of some relative dimension n”, but
sometimes can mean “smooth of some relative dimension in the neighborhood of
every point”.

13.6.C. E ASY E XERCISE . Show that the notion of smoothness of relative dimension
n (and hence the notion of étaleness) is preserved by base change.

13.6.D. E XERCISE . Suppose π : X → Y is smooth of relative dimension m and

ρ : Y → Z is smooth of relative dimension n. Show that ρ ◦ π is smooth of relative
dimension m + n. (In particular, the composition of étale morphisms is étale.
Furthermore, this implies that smooth morphisms are closed under product, by
Exercise 8.1.A.)
Exercises 13.6.B–13.6.D imply that smooth morphisms (and étale morphisms)
form a “reasonable” class in the sense of §8.1.

13.6.4. Observation. Suppose π : X → Y is smooth of relative dimension n. Then

locally on X, π can be described as an étale cover of An
Y . More precisely, for every
p ∈ X, there is an open neighborhood U of p, such that π|U can be factored into

U
α / An β
/Y
Y

where α is étale and β is the obvious projection.

13.6.E. E XERCISE . Show that Spec k → Spec k[ϵ]/(ϵ2 ) is not an étale morphism.
(Does your mental picture of étale morphisms already capture this fact?) Possible
approach: if it were étale, show that it would induce an “isomorphism on tangent
spaces”.

13.6.5. More sophisticated approaches to Exercise 13.6.E may lead you to consider
the following exercises: Classify the étale maps to Spec k. Classify the étale maps
to Spec k[ϵ]/(ϵ2 ). If A = k[x1 , . . . , xn ]/I is such that Spec A has only one point (an
example of an Artinian local scheme, Definition 6.5.9, and A is an Artinian local
ring, §12.1.8), classify the étale maps to Spec A. These questions will lead you in
different important directions, to deformation theory, Witt vectors, and more.

13.6.F. E XERCISE . Suppose k is a field of characteristic not 2. Let Y = Spec k[t], and
X = Spec k[u, 1/u]. Show that the morphism π : X → Y induced by t 7→ u2 is étale.
(Sketch π!) Show that there is no nonempty open subset U of X on which π is an
isomorphism. (This shows that étale cannot be defined as “local isomorphism”,
i.e., “isomorphism locally on the source”. In particular, the naive extension of the
388 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

differential intuition of §13.6.1 does not give the right definition of étaleness in
algebraic geometry.)
We now show that our new Definition 13.6.2 specializes to Definition 13.2.4
when the target is a field.

13.6.6. Theorem. — Suppose X is a k-scheme. Then the following are equivalent.

(i) X is smooth of relative dimension n over Spec k (Definition 13.6.2).
(ii) X has pure dimension n, and is smooth over k (in the sense of Definition 13.2.4).

This allows us to use the phrases “smooth over k” and “smooth over Spec k”
interchangeably, as we would expect. More generally, “smooth over a ring A”
means “smooth over Spec A”.

Proof. (i) implies (ii). Suppose X is smooth of relative dimension n over Spec k
(Definition 13.6.2). We will show that X has pure dimension n, and is smooth over
k (Definition 13.2.4). Our goal is a Zariski-local statement, so we may assume that
X is an open subset

W ⊂ Spec k[x1 , . . . , xn+r ]/(f1 , . . . , fr )

as described in Definition 13.6.2.

The central issue is showing that X has “pure dimension n”, as the latter half
of the statement would then follow because the corank of the Jacobian matrix is
then n.
Let Xk := X ×k k. By Exercise 12.1.H, Xk is pure dimension n if and only if X
has pure dimension n. Let p be a closed point of the (finite type scheme) Xk . Then
Xk has dimension at least n at p, by Krull’s Height Theorem 12.3.10, combined with
Theorem 12.2.9 (codimension is the difference of dimensions for varieties). But the
Zariski tangent space at a closed point of a finite type scheme over k is cut out
by the Jacobian matrix (Exercise 13.1.I), and thus has dimension exactly n (by the
Jacobian condition in the definition of k-smooth). As the dimension of the Zariski
tangent space bounds the dimension of the scheme (Theorem 12.3.12), Xk has pure
dimension n as desired.
(ii) implies (i). It suffices to prove the result in a neighborhood of any given
closed point p of X. (Here we use that any union of open neighborhoods of all the
closed points is all of X, by Exercise 5.3.F.) We thus consider the following problem.
Suppose we are given X := Spec k[x1 , . . . , xn+r ]/(f1 , . . . , fN ), and a point p ∈ X,
and an open neighborhood U of p in X such that U is pure dimension n, and the
Jacobian matrix of the fj ’s with respect to the xi ’s has corank n everywhere on U.
We wish to show that the structure morphism π : X → Spec k is smooth of relative
dimension n at p (Definition 13.6.2). By permuting the fi (and shrinking U), we
may assume that the Jacobian matrix of the first f1 , . . . , fr already has corank n
(i.e., rank r). Then Y := Spec k[x1 , . . . , xn+r ]/(f1 , . . . , fr ) is smooth at p of relative
dimension n over Spec k (Definition 13.6.2), so the local ring OY,p is an integral
domain by Exercise 13.2.O, of pure dimension n by the proof that (i) implies (ii)
above.
For the same reasons, OX,p is also an integral domain of pure dimension n. We
have a surjection OY,p ↠ OY,p /(fr+1 , . . . , fN ) = OX,p of integral domains of the
same dimension, so by Easy Exercise 12.1.D, OY,p = OX,p , i.e., fr+1 = · · · = fN = 0
 © 2024 Ravi Vakil. Published by Princeton University Press. 389

in OY,p . Thus near p, X is isomorphic to Y, so X is indeed smooth over k of relative

dimension n (Definition 13.6.2) as desired. □

13.7 ⋆ Valuative criteria for separatedness and properness

(The only reason this section is placed here is that we need the theory of discrete
valuation rings.)
In reasonable circumstances, it is possible to verify separatedness by checking
only maps from spectra of valuations rings. There are four reasons you might like
this (even if you never use it). First, it gives useful intuition for what separated
morphisms look like. Second, given that we understand schemes by maps to them
(the Yoneda philosophy), we might expect to understand morphisms by mapping
certain maps of schemes to them, and this is how you can interpret the diagram
appearing in the valuative criterion. And the third concrete reason is that one of
the two directions in the statement is much easier (a special case of the Reduced-
to-Separated Theorem 11.3.2, see Exercise 13.7.A), and this is the direction we will
repeatedly use. Finally, the criterion is very useful!
Similarly, there is a valuative criterion for properness.
In this section, we will meet the valuative criteria, but aside from outlining the
proof of one result (the DVR version of the valuative criterion of separatedness), we
will not give proofs, and satisfy ourselves with references. There are two reasons
for this controversial decision. First, the proofs require the development of some
commutative algebra involving valuation rings that we will not otherwise need.
Second, we will not use these results in any substantial or necessary way later in
this book.
We begin with a valuative criterion for separatedness that applies in a case
that will suffice for the interests of most people, that of finite type morphisms of
Noetherian schemes. We will then give a more general version for more general
readers.

13.7.1. Theorem (valuative criterion for separatedness, DVR version; see Fig-
ure 13.6). — Suppose π : X → Y is a morphism of finite type of locally Noetherian schemes.
Then π is separated if and only if the following condition holds: for any discrete valuation
ring A, and any diagram of the form

(13.7.1.1) Spec K(A) /X

_
open emb. π
 
Spec A /Y

(where the vertical morphism on the left corresponds to the inclusion A ,→ K(A)), there is
at most one morphism Spec A → X such that the diagram

(13.7.1.2) Spec K(A) /; X

_
open emb. ≤1 π
 
Spec A /Y
390 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

commutes.

F IGURE 13.6. The valuative criterion for separatedness

The idea behind the proof is explained in §13.7.3. We can show one direction
right away, in the next exercise.

13.7.A. E XERCISE ( THE EASY DIRECTION ). Use the Reduced-to-Separated Theo-

rem 11.3.2 to prove one direction of the theorem: that if π is separated, then the
valuative criterion holds.

13.7.B. E XERCISE . Suppose X is an integral Noetherian separated curve. If p ∈ X is

a regular closed point, then OX,p is a discrete valuation ring, so each regular point
yields a discrete valuation on K(X). Use the previous exercise to show that distinct
points yield distinct discrete valuations.
Here is the intuition behind the valuative criterion (see Figure 13.6). We think
of Spec of a discrete valuation ring A as a “germ of a curve”, and Spec K(A) as
the “germ minus the origin” (even though it is just a point!). Then the valuative
criterion says that if we have a map from a germ of a curve to Y, and have a lift of
the map away from the origin to X, then there is at most one way to lift the map
from the entire germ. In the case where Y is the spectrum of a field, you can think
of this as saying that limits of one-parameter families are unique (if they exist).
For example, this captures the idea of what is wrong with the map of the line
with the doubled origin over k (Figure 13.7): we take Spec A to be the germ of the
affine line at the origin, and consider the map of the germ minus the origin to the
line with doubled origin. Then we have two choices for how the map can extend
over the origin.

13.7.C. E XERCISE . Make this precise: show that map of the line with doubled origin
over k to Spec k fails the valuative criterion for separatedness. (Earlier arguments
were given in Exercises 11.2.I and 11.3.C.)

13.7.2. ⋆⋆ Remark for experts: moduli spaces and the valuative criterion for separatedness.
If Y = Spec k, and X is a (fine) moduli space (a term we won’t define here) of
some type of object, then the question of the separatedness of X (over Spec k) has a
natural interpretation: given a family of your objects parametrized by a “punctured
discrete valuation ring”, is there always at most one way of extending it over the
closed point?
 © 2024 Ravi Vakil. Published by Princeton University Press. 391

F IGURE 13.7. The line with the doubled origin fails the valuative
criterion for separatedness. (You may notice Figure 11.4 embedded
in this diagram.)

13.7.3. Idea behind the proof of Theorem 13.7.1 (the valuative criterion for separatedness,
DVR version) in the case of varieties. (One direction was done in Exercise 13.7.A.) If π
is not separated, our goal is to produce a diagram (13.7.1.1) that can be completed
to (13.7.1.2) in more than one way. If π is not separated, then δ : X → X ×Y X is a
locally closed embedding that is not a closed embedding.

13.7.D. E XERCISE . Show that you can find points p not in the diagonal ∆ of X ×Y X
and q in ∆ such that p ∈ q, and there are no points “between p and q” (no points r
distinct from p and q with p ∈ r and r ∈ q). (Exercise 8.4.C may shed some light.)
Let Q be the scheme obtained by giving the induced reduced subscheme
structure to q. Let B = OQ,p be the local ring of Q at p.

13.7.E. E XERCISE . Show that B is a Noetherian local integral domain of dimension

1.
If B were regular, then we would be done: composing the inclusion morphism
Q → X ×Y X with the two projections induces the same morphism q → X (i.e.,
Spec κ(q) → X) but different extensions to B precisely because p is not in the diago-
nal. To complete the proof, one shows that the normalization of B is Noetherian;
this is where need the hypothesis that we are working with varieties, to invoke
Exercise 10.7.M. Then localizing at any prime above p (there is one by the Lying
Over Theorem 8.2.5) yields the desired discrete valuation ring A.
For an actual proof, see [Stacks, tag 0207] or [Gr-EGA, II.72.3].
With a more powerful invocation of commutative algebra, we can prove a
valuative criterion with much less restrictive hypotheses.

13.7.4. Theorem (valuative criterion for separatedness). — Suppose π : X → Y is a

quasiseparated morphism. Then π is separated if and only if for any valuation ring A, and
any diagram of the form (13.7.1.1), there is at most one morphism Spec A → X such that
the diagram (13.7.1.2) commutes.
Because I have already failed to completely prove the DVR version, I feel no
urge to prove this harder fact. The proof of one direction, that π separated implies
that the criterion holds, follows from the identical argument as in Exercise 13.7.A.
For a complete proof, see [Stacks, tags 01KY and 01KZ] or [Gr-EGA, II.7.2.3].
392 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

13.7.5. Valuative criteria for (universal closedness and) properness.

There is a valuative criterion for properness too. It is philosophically useful,
and sometimes directly useful, although we won’t need it. It naturally comes from
the valuative criterion for separatedness combined with a valuative criterion for
universal closedness.

13.7.6. Theorem (valuative criterion for universal closedness and properness,

DVR version). — Suppose π : X → Y is a morphism of finite type of locally Noetherian
schemes. Then π is universally closed (resp., proper) if and only if for any discrete valuation
ring A, and any diagram (13.7.1.1), there is at least one (resp., exactly one) morphism
Spec A → X such that the diagram (13.7.1.2) commutes.
See [Gr-EGA, II.7.3.8], [Ha1, Thm. II.4.7], or [Gr-EGA’, I.5.5] for proofs. A
comparison with Theorem 13.7.1 will convince you that these three criteria naturally
form a family.
In the case where Y is a field, you can think of the valuative criterion of proper-
ness as saying that limits of one-parameter families in proper varieties always exist,
and are unique. This is a useful intuition for the notion of properness.

13.7.F. E ASY E XERCISE . Use the valuative criterion for properness to prove that
PnA → Spec A is proper if A is Noetherian. (Don’t be fooled: Because this requires
the valuative criterion, this is a difficult way to prove a fact that we already showed
in Theorem 11.4.5.)

13.7.G. E XERCISE ( CF. E XERCISE 13.7.B). Suppose X is an irreducible regular

(Noetherian) curve, proper over a field k (respectively, over Z). Describe a bijection
between the discrete valuations on K(X) for which the elements of k (respectively
Z) have valuation 0, and the closed points of X.

13.7.7. Remarks for experts. There is a moduli-theoretic interpretation similar to

that for separatedness (Remark 13.7.2): X is proper if and only if there is always
precisely one way of filling in a family over a punctured spectrum of a discrete
valuation ring.

13.7.8. Finally, here is a fancier version of the valuative criterion for universal
closedness and properness.

13.7.9. Theorem (valuative criterion for universal closedness and properness). —

Suppose π : X → Y is a quasiseparated, finite type (hence quasicompact) morphism. Then π
is universally closed (resp., proper) if and only if the following condition holds. For any
valuation ring A and any diagram of the form (13.7.1.1), there is at least one (resp., exactly
one) morphism Spec A → X such that the diagram (13.7.1.2) commutes.
Clearly the valuative criterion for properness is a consequence of the valuative
criterion for separatedness (Theorem 13.7.4) and the valuative criterion for universal
closedness. For proofs, see [Stacks, tag 01KF] or [Gr-EGA, II.7.3.8].

13.7.10. On the importance of valuation rings in general. Although we have only

discussed discrete valuation rings in depth, general valuation rings should not be
considered an afterthought. Serre makes the case to Grothendieck in a letter, [GrS,
p. 125]:
 © 2024 Ravi Vakil. Published by Princeton University Press. 393

Tu es bien sévère pour les Valuations! Je persiste pourtant à les garder, pour plusieurs
raisons: d’abord une pratique: n rédacteurs ont sué dessus, il n’y a rien à reprocher au
fourbi, on ne doit pas le vider sans raisons très sérieuses (et tu n’en as pas). . . . Même un
noethérien impénitent a besoin des valuations discrètes, et de leurs extension; en fait, Tate,
Dwork, et tous les gen p-adiques te diront qu’on ne peut se limiter au cas discret et que le
cas de rang 1 est indispensable; à ce moment, les méthodes noethériennes deviennent un vrai
carcan, et on comprend beaucoup mieux en rédigeant le cas général que uniquement le cas
de rang 1. . . . Il n’y a pas lieu d’en faire un plat, bien sûr, et c’est pourquoi j’avias vivement
combattu le plan Weil initial qui en faisait le théorème central de l’Algèbre Commutative,
mais d’autre part il faut le garder.
You are very harsh on Valuations! I persist nonetheless in keeping them, for several
reasons, of which the first is practical: n people have sweated over them, there is nothing
wrong with the result, and it should not be thrown out without very serious reasons (which
you do not have). . . . Even an unrepentant Noetherian needs discrete valuations and their
extensions; in fact, Tate, Dwork and all the p-adic people will tell you that one cannot
restrict oneself to the discrete case and the rank 1 case is indispensable; Noetherian methods
then become a burden, and one understands much better if one considers the general case
and not only the rank 1 case. . . . It is not worth making a mountain out of it, of course,
which is why I energetically fought Weil’s original plan to make it the central theorem of
Commutative Algebra, but on the other hand it must be kept.

13.8 ⋆ More sophisticated facts about regular local rings

Regular local rings have essentially every good property you could want, but
some of them require hard work. We now discuss a few fancier facts that may help
you sleep well at night.

13.8.1. Localizations of regular local rings are regular local rings.

13.8.2. Fact ([E, Cor. 19.14], [Mat2, Thm. 19.3]). — If (A, m) is a regular local ring,
then any localization of A at a prime is also a regular local ring.
(We will not need this, and hence will not prove it.) This major theorem was
an open problem in commutative algebra for a long time until settled by Serre and
Auslander-Buchsbaum using homological methods.
Hence to check if Spec A is regular (A Noetherian), it suffices to check at closed
points (at maximal ideals). Assuming Fact 13.8.2 (and using Exercise 5.1.E), you
can check regularity of a Noetherian scheme by checking at closed points.
We will prove two important cases of Fact 13.8.2. The first you can do right
now.

13.8.A. E XERCISE . Suppose X is a Noetherian dimension 1 scheme that is regular at

its closed points. Show that X is reduced. Hence show (without invoking Fact 13.8.2)
that X is regular.
The second important case will be proved in §21.6.2:

13.8.3. Theorem. — If X is a finite type scheme over a perfect field k that is regular at its
closed points, then X is regular.
394 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

More generally, Exercise 21.6.B will show that Fact 13.8.2 holds if A is the
localization of a finite type algebra over a perfect field.

13.8.B. E XERCISE ( GENERALIZING E XERCISE 13.3.E). Suppose k is an algebraically

closed field of characteristic 0. Assuming Theorem 13.8.3, show that there exists a
regular hypersurface of every positive degree d in Pn . (As in Exercise 13.3.E, feel
free to weaken the hypotheses.)

13.8.4. Regular local rings are unique factorization domains, integrally closed,
and Cohen-Macaulay.

13.8.5. Fact (Auslander-Buchsbaum Theorem). — Regular local rings are unique

factorization domains.
(This is a hard theorem, so we will not prove it, and will therefore not use it.
For a proof, see [E, Thm. 19.19], [Mat2, Thm. 20.3] or [Gr-EGA, IV4 .21.11.1], or
[Mu7, §III.7] in the special case of varieties.) Thus regular schemes are factorial,
and hence normal by Exercise 5.4.F.

13.8.6. Remark: Factoriality is weaker than regularity. The implication “regular implies
factorial” is strict. Here is an example showing this. Suppose k is an algebraically
closed field of characteristic not 2. Let A = k[x1 , . . . , xn ]/(x21 + · · · + x2n ) (cf. Exer-
cise 5.4.I). Note that Spec A is clearly singular at the origin. In Exercise 15.5.R, we
will show that A is a unique factorization domain when n ≥ 5, so Spec A is factorial.
In particular, A(x1 ,...,xn ) is a Noetherian local ring that is a unique factorization
domain, but not regular. (More generally, it is a consequence of Grothendieck’s
proof of a conjecture of P. Samuel that a Noetherian local ring that is a complete
intersection — in particular a hypersurface — that is factorial in codimension at
most 3 must be factorial, [SGA2, Exp. XI, Cor. 3.14]. For a shorter and somewhat
more elementary proof, see [CL].)

13.8.7. Regular local rings are integrally closed. The Auslander-Buchsbaum Theorem
(Fact 13.8.5) implies that regular local rings are integrally closed (by Exercise 5.4.F).
We will prove this (without appealing to Fact 13.8.5) in §26.3.5.

13.8.8. Regular local rings are Cohen-Macaulay. In §26.2.5, we will show that regular
local rings are Cohen-Macaulay (a notion defined in Chapter 26).

13.9 ⋆ Filtered rings and modules, and the Artin-Rees Lemma

We conclude Chapter 13 by discussing the Artin-Rees Lemma 13.9.3, which was

used to prove Proposition 13.5.2. The Artin-Rees Lemma generalizes the intuition
behind Proposition 13.5.2, that any function that is analytically zero at a point
actually vanishes in an open neighborhood of that point (§13.5.3). Because we will
use it later, and because it is useful to recognize it in other contexts, we discuss it in
some detail.
 © 2024 Ravi Vakil. Published by Princeton University Press. 395

13.9.1. Definitions. Suppose I is an ideal of a ring A. A descending filtration of an

A-module M
(13.9.1.1) M = M0 ⊃ M1 ⊃ M 2 ⊃ · · ·
is called an I-filtration if Id Mn ⊂ Mn+d for all d, n ≥ 0. An example is the I-adic
filtration where Mk = Ik M. We say an I-filtration is I-stable if for some s and all
d ≥ 0, Id Ms = Md+s . For example, the I-adic filtration is I-stable.
Let A(I)• be the graded ring ⊕n≥0 In (with the convention I0 = A). This
is called the Rees algebra of the ideal I in A, although we will not need this
terminology. Define M(I)• := ⊕Mn . It is naturally a graded module over A(I)• .

13.9.2. Proposition. — If A is Noetherian, M is a finitely generated A-module, and

(13.9.1.1) is an I-filtration, then M(I)• is a finitely generated A(I)• -module if and only if
the filtration (13.9.1.1) is I-stable.

Proof. Note that A(I)• is Noetherian (by Exercise 4.5.G(b), as A is Noetherian, and I
is a finitely generated A-module).
Assume first that M(I)• is finitely generated over the Noetherian ring A(I)• ,
and hence Noetherian. Consider the increasing chain of A(I)• -submodules whose
kth element Lk is
M ⊕ M1 ⊕ M2 ⊕ · · · ⊕ Mk ⊕ IMk ⊕ I2 Mk ⊕ · · ·
(which agrees with M(I)• up until Mk , and then “I-stabilizes”). This chain must
stabilize by Noetherianness. But ∪Lk = M(I)• , so for some s ∈ Z≥0 , Ls = M(I)• ,
so Id Ms = Ms+d for all d ≥ 0 — (13.9.1.1) is I-stable.
For the other direction, assume that Md+s = Id Ms for a fixed s and all d ≥ 0.
Then M(I)• is generated over A(I)• by M ⊕ M1 ⊕ · · · ⊕ Ms . But each Mj is finitely
generated, so M(I)• is indeed a finitely generated A(I)• -module. □

13.9.3. Artin-Rees Lemma. — Suppose A is a Noetherian ring, and (13.9.1.1) is an

I-stable filtration of a finitely generated A-module M. Suppose that L ⊂ M is a submodule,
and let Ln := L ∩ Mn . Then
L = L0 ⊃ L1 ⊃ L2 ⊃ · · ·
is an I-stable filtration of L.

Proof. Note that L• is an I-filtration, as ILn ⊂ IL ∩ IMn ⊂ L ∩ Mn+1 = Ln+1 . Also,

L(I)• is an A(I)• -submodule of the finitely generated A(I)• -module M(I)• , and
hence finitely generated by Exercise 3.6.Y (as A(I)• is Noetherian, see the proof of
Proposition 13.9.2). □
An important special case is the following.

13.9.4. Corollary. — Suppose I ⊂ A is an ideal of a Noetherian ring, and M is a finitely

generated A-module, and L is a submodule. Then for some integer s, Id (L ∩ Is M) =
L ∩ Id+s M for all d ≥ 0.
Warning: it need not be true that Id L = L ∩ Id M for all d. (Can you think of a
counterexample to this statement?)

Proof. Apply the Artin-Rees Lemma 13.9.3 to the filtration Mn = In M. □

396 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

13.9.A. E XERCISE (K RULL I NTERSECTION T HEOREM ).

(a) Suppose I is an ideal of a Noetherian ring A, and M is a finitely generated
A-module. Show that there is some a ≡ 1 (mod I) such that a ∩∞ j
j=1 I M = 0. Hint:
Apply the Artin-Rees Lemma 13.9.3 with L = ∩∞ j n
j=1 I M and Mn = I M. Show that
L = IL, and apply the first version of Nakayama (Lemma 8.2.8).
(b) Show that if A is a Noetherian integral domain or a Noetherian local ring,
and I is a proper ideal, then ∩∞ j
j=1 I = 0. In particular, you will have proved
Proposition 13.5.2: if (A, m) is a Noetherian local ring, then ∩i mi = 0.

13.9.B. E XERCISE . Make the following precise, and prove it (thereby justifying
the intuition in §13.5.3): if X is a locally Noetherian scheme, and f is a function
on X that is analytically zero at a point p ∈ X, then f vanishes in a (Zariski) open
neighborhood of p.
 Part V

Quasicoherent sheaves on schemes,

and their uses
 CHAPTER 14

More on quasicoherent and coherent sheaves

In Chapter 6, we discussed how algebraically, quasicoherent and coherent

sheaves generalized modules over a ring. In this chapter, we explore how geometri-
cally, quasicoherent and coherent sheaves generalize vector bundles. Informally, a
vector bundle V on a geometric space X (such as a manifold) is a family of vector
spaces continuously parametrized by points of X. In other words, for each point p
of X, there is a vector space, and these vector spaces are glued into a space V so that,
as p varies, the vector space above p varies continuously. Nontrivial examples to
keep in mind are the tangent bundle to a differentiable manifold, and the Möbius
strip over a circle (interpreted as a line bundle). We will make this somewhat
more precise in §14.1, but if you have not seen this idea before, you shouldn’t be
concerned; the main thing we will use this notion for is motivation.

14.0.1. (Locally) free sheaves. A free sheaf on a ringed space X is an OX -module

isomorphic to OX⊕I where the sum is over some index set I. The rank of the free
sheaf is the cardinality of I. (This agrees with its rank as an O-module at a point,
see Definition 4.3.8.) A locally free sheaf on a ringed space X is an OX -module
locally isomorphic to a free sheaf. This corresponds to the notion of a vector bundle
(§14.1). Quasicoherent sheaves form a convenient abelian category containing the
locally free sheaves (i.e., vector bundles) that is much smaller than the category of
O-modules. Quasicoherent sheaves generalize locally free sheaves in much the way
that modules generalize free modules. Coherent sheaves are, roughly speaking, a
finite rank version of quasicoherent sheaves, which form a well-behaved abelian
category containing finite rank locally free sheaves (or equivalently, finite rank
vector bundles). Just as the notion of free modules leads us to the notion of
modules in general, and finitely generated modules, the notion of free sheaves will
lead us inevitably to the notion of quasicoherent sheaves and coherent sheaves.
(There is a slight fib in comparing finitely generated modules to coherent sheaves,
see §6.4.)

14.1 Vector bundles “=” locally free sheaves

We recall somewhat more precisely the notion of vector bundles on manifolds.
Arithmetically-minded readers shouldn’t tune out: for example, fractional ideals of
the ring of integers in a number field (defined in §10.7.1) turn out to be an example
of a “line bundle on a smooth curve” (Exercise 14.1.K).

This material is published by Princeton University Press as The Rising Sea: Foundations of Algebraic
Geometry by Ravi Vakil. This version is free to view and download for personal use only. Not for
redistribution, resale, or use in derivative works. © 2024 Ravi Vakil.
399
400 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

A rank n vector bundle on a manifold M is a map π : V → M with the structure

of an n-dimensional real vector space on π−1 (p) for each point p ∈ M, such that
for every p ∈ M, there is an open neighborhood U and a homeomorphism

ϕ : U × Rn → π−1 (U)
over U (so that the diagram
∼
π−1 (U) o / U × Rn
=
(14.1.0.1)

π|π−1 (U) projection to first factor

# {
U
commutes) that is an isomorphism of vector spaces over each p ∈ U. An isomor-
phism (14.1.0.1) is called a trivialization over U. If U = M, we say that V is a trivial
bundle.
We call n the rank of the vector bundle. A rank 1 vector bundle is called a
line bundle. (It can also be convenient to be agnostic about the rank of the vector
bundle, so it can have different ranks on different connected components. It is also
sometimes convenient to consider infinite-rank vector bundles.)

14.1.1. Transition functions. Given trivializations over U1 and U2 , over their

intersection, the two trivializations must be related by an element T12 of GLn with
entries consisting of functions on U1 ∩ U2 . (There is an issue of convention as to
−1
whether T12 should be replaced by T12 . Our choice: We want this to be a matrix
multiplying on the left, sending U1 -coordinates to U2 -coordinates.) If {Ui } is a
cover of M, and we are given trivializations over each Ui , then the {Tij } must satisfy
the cocycle condition:
(14.1.1.1) Tjk |Ui ∩Uj ∩Uk ◦ Tij |Ui ∩Uj ∩Uk = Tik |Ui ∩Uj ∩UK .
−1
(This implies Tij = Tji .) The data of the Tij are called transition functions (or
transition matrices) for the trivialization.
This is reversible: given the data of a cover {Ui } and transition functions Tij ,
we can recover the vector bundle (up to unique isomorphism) by “gluing together
the various Ui × Rn along Ui ∩ Uj using Tij ”.

14.1.2. The sheaf of sections. Fix a rank n vector bundle V → M. The sheaf of
sections F of V (Exercise 2.2.G) is an OM -module — given any open set U, we can
multiply a section over U by a function on U and get another section.
Moreover, given a trivialization over U, the sections over U are naturally
identified with n-tuples of functions of U.
n
U×R
V
π n-tuple of functions

U
Thus given a trivialization, over each open set Ui , we have an isomorphism
∼ ⊕n
F |Ui ←→ OU i
. As mentioned in §14.0.1, we say that such an F is a locally free
sheaf of rank n.
 © 2024 Ravi Vakil. Published by Princeton University Press. 401

14.1.3. Transition functions for the sheaf of sections. Suppose we have a vector
bundle on M, along with a trivialization over an open cover Ui . Suppose we have
a section of the vector bundle over M. (This discussion will apply with M replaced
by any open subset.) Then over each Ui , the section corresponds to an n-tuple of
functions over Ui , say ⃗si .

14.1.A. E XERCISE . Show that over Ui ∩ Uj , the vector-valued function ⃗si is related
to ⃗sj by the (same) transition functions: Tij⃗si = ⃗sj . (Don’t do this too quickly —
make sure your i’s and j’s are on the correct side.)
Given a locally free sheaf F with rank n, and a trivializing open neighborhood
of F (an open cover {Ui } such that over each Ui , F |Ui = ∼ O ⊕n as O-modules),
Ui
we have transition functions Tij ∈ GLn (O(Ui ∩ Uj )) satisfying the cocycle con-
dition (14.1.1.1). Thus the data of a locally free sheaf of rank n is equivalent to
the data of a vector bundle of rank n. This change of perspective is useful, and
is similar to an earlier change of perspective when we introduced ringed spaces:
understanding spaces is the same as understanding (sheaves of) functions on the
spaces, and understanding vector bundles (a type of “space over M”) is the same
as understanding (sheaves of) sections.

14.1.4. Definition. A rank 1 locally free sheaf is called an invertible sheaf. (Unim-
portant aside: “invertible sheaf” is a heinous term for something that is essentially
a line bundle. The motivation is that if X is a locally ringed space, and F and G are
OX -modules with F ⊗OX G = ∼ OX , then F and G are invertible sheaves [MO33489].
Thus in the monoid of OX -modules under tensor product, invertible sheaves are
the invertible elements. We will never use this fact. People often informally use the
phrase “line bundle” when they mean “invertible sheaf”. The phrase line sheaf has
been proposed but has not caught on.)

14.1.5. Notation. Recall (Definition 6.2.6) that given a section s of F and a section t
of G , we have a section s ⊗ t of F ⊗ G . If F is an invertible sheaf, this section is
often denoted st.

14.1.6. Locally free sheaves more generally.

We can generalize the notion of locally free sheaves to schemes (or more gener-
ally, ringed spaces) without change. A locally free sheaf of rank n on a scheme
X is defined as an OX -module F that is locally a free sheaf of rank n. Precisely,
there is an open cover {Ui } of X such that for each Ui , F |Ui =∼ O ⊕n . This open
Ui
cover determines transition functions — the data of a cover {Ui } of X, and functions
Tij ∈ GLn (O(Ui ∩ Uj )) satisfying the cocycle condition (14.1.1.1) — which in turn
determine the locally free sheaf. As before, given these data, we can find the sec-
tions over any open set U. Informally, they are sections of the free sheaves over
each U ∩ Ui that agree on overlaps. More formally, for each i, they are

si1
 

⃗si =  ...  ∈ Γ (U ∩ Ui , OX )⊕n ,

 

sin

satisfying Tij⃗si = ⃗sj on U ∩ Ui ∩ Uj .

402 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

You should think of these as vector bundles, but just keep in mind that they are
not the “same”, just equivalent notions. We will later (Definition 17.1.5) define the
“total space” of the vector bundle V → X (a scheme over X) in terms of the sheaf
version of Spec. But the locally free sheaf perspective will prove to be more useful.
As one example: the definition of a locally free sheaf is much shorter than that of a
vector bundle.
As in our motivating discussion, it is sometimes convenient to let the rank vary
among connected components, or to consider infinite rank locally free sheaves.

14.1.7. Pulling back vector bundles.

14.1.B. E SSENTIAL EXERCISE : PULLING BACK VECTOR BUNDLES . Suppose π : X →

Y is a morphism of ringed spaces. Recall the definition of π∗ : ModOY → ModOX ,
pulling back O-modules. (Now is a good time to do Exercise 7.2.D, if you haven’t
done it before.) Suppose n ∈ Z+ .
(a) Show that (canonically) π∗ (OY⊕n ) = ∼ O ⊕n .
X
(b) If G is a locally free sheaf on Y of rank n, show that π∗ G is a locally free sheaf
on X of rank n.
(c) If G is a locally free sheaf on Y of rank n, and {Ui } are trivializing neighbor-
hoods for G on Y, with transition matrices Tij , show that {π−1 Ui } are trivializing
neighborhoods for π∗ G on X, with transition matrices π∗ Tij . (Do you see what that
means?)

14.1.8. The problem with locally free sheaves / vector bundles (as opposed to
vector spaces).
Recall that OX -modules form an abelian category: we can talk about kernels,
cokernels, and so forth, and we can do homological algebra. Similarly, vector
spaces form an abelian category. But locally free sheaves (i.e., vector bundles),
along with reasonably natural maps between them (those that arise as maps of OX -
modules), don’t form an abelian category. As a motivating example in the category
of manifolds, consider the map of the trivial line bundle on R (with coordinate t) to
itself, corresponding to multiplying by the coordinate t. Then this map jumps rank,
and if you try to define a kernel or cokernel you will get confused.
This problem is resolved by enlarging our notion of nice OX -modules in a
natural way, to quasicoherent sheaves.

OX -modules ⊃ quasicoherent sheaves ⊃ locally free sheaves

(abelian category) (abelian category) (not an abelian category)

You can turn this into a definition of quasicoherent sheaves, equivalent to those
we gave in §6.1. We want a notion that is local on X of course. So we ask for the
smallest abelian subcategory of ModOX that is “local” and includes vector bundles.
It turns out that the main obstruction to vector bundles to be an abelian category
is the failure of cokernels of maps of locally free sheaves — as OX -modules — to
be locally free; we could define quasicoherent sheaves to be those OX -modules
that are locally cokernels, yielding a description that works more generally on
ringed spaces, as described in Exercise 6.3.B. You may wish to later check that our
definitions are equivalent to these.
Similarly, in the locally Noetherian setting, finite rank locally free sheaves will
sit in a nice smaller abelian category, that of coherent sheaves.
 © 2024 Ravi Vakil. Published by Princeton University Press. 403

quasicoherent sheaves ⊃ coherent sheaves ⊃ finite rank locally free sheaves

(abelian category) (abelian category) (not an abelian category)

14.1.9. Useful constructions on locally free sheaves: Hom , dual, ⊗.

We now give some useful constructions in the form of a series of exercises
about locally free sheaves on a scheme. They are useful, important, and surprisingly
nontrivial! Two hints: Exercises 14.1.C–14.1.I will apply for ringed spaces in general,
so you shouldn’t use special properties of schemes. Furthermore, they are all local
on X, so you can reduce to the case where the locally free sheaves in question are
actually free.

14.1.C. E XERCISE . Suppose F and G are locally free sheaves on X of rank m and n
respectively. Show that Hom OX (F , G ) is a locally free sheaf of rank mn.

14.1.D. E XERCISE . If E is a locally free sheaf on X of (finite) rank n, Exercise 14.1.C

implies that E ∨ := Hom (E , OX ) is also a locally free sheaf of rank n. This is called
the dual of E (cf. §2.3.4). Given transition functions for E , describe transition
functions for E ∨ . (Note that if E is rank 1, i.e., invertible, the transition functions
of the dual are the inverse of the transition functions of the original.) Show that
E =∼ E ∨∨ . (Caution: your argument showing that there is a canonical isomorphism
∼ ∼
F ←→ (F ∨ )∨ better not also show that there is an isomorphism F ←→ F ∨ ! We
will see an example in §15.1 of a locally free F that is not isomorphic to its dual:
the invertible sheaf O(1) on Pn .)

14.1.E. E XERCISE . If F and G are locally free sheaves, show that F ⊗ G is a locally
free sheaf. (Here ⊗ is tensor product as OX -modules, defined in Exercise 2.6.K.) If
F is an invertible sheaf, show that F ⊗ F ∨ = ∼ OX .

14.1.F. E XERCISE . Recall that tensor products tend to be only right-exact in general
(Exercise 1.2.H). Show that tensoring by a locally free sheaf is exact. More precisely,
if F is a locally free sheaf, and G ′ → G → G ′′ is an exact sequence of OX -modules,
then so is G ′ ⊗ F → G ⊗ F → G ′′ ⊗ F . (Possible hint: it may help to check
exactness by checking exactness at stalks. Recall that the tensor product of stalks
can be identified with the stalk of the tensor product, so for example there is a
∼
“natural” isomorphism (G ⊗OX F )p ←→ Gp ⊗OX,p Fp , Exercise 2.6.K(b).)

14.1.G. E XERCISE . If E is a locally free sheaf of finite rank, and F and G are
OX -modules, show that

∼ Hom (F ⊗ E ∨ , G ) =
Hom (F , G ⊗ E ) = ∼ Hom (F , G ) ⊗ E .

∼ D∨ ⊗ G .
In particular, if D is a locally free sheaf of finite rank, Hom (D, G ) =
(Possible hint: first consider the case where E is free.)

14.1.10. The Picard group.

14.1.H. E XERCISE AND IMPORTANT DEFINITION . Show that the invertible sheaves
on X, up to isomorphism, form an abelian group under tensor product. This is
called the Picard group of X, and is denoted Pic X.
404 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

14.1.I. E ASY E XERCISE ( CF. E XERCISE 14.1.B). Show that Pic is a contravariant
functor from the category of ringed spaces to the category Ab of abelian groups.

14.1.J. ⋆ E XERCISE ( FOR THOSE WITH SUFFICIENT COMPLEX - ANALYTIC BACK -

GROUND ). Recall the analytification functor (Exercises 7.3.P, 7.3.Q, and 11.2.K),
that takes a complex finite type reduced scheme and produces a complex analytic
space.
(a) If L is an invertible sheaf on a complex (algebraic) variety X, define (up to
unique isomorphism) the corresponding invertible sheaf on the complex variety
Xan .
(b) Show that the induced map Pic X → Pic Xan is a group homomorphism.
(c) Show that this construction is functorial: if π : X → Y is a morphism of complex
varieties, the following diagram commutes:

π∗ / Pic X
Pic Y

 π∗ 
Pic Yan
an
/ Pic Xan

where the vertical maps are the ones you have defined.

14.1.K. ⋆ E XERCISE ( FOR THOSE WITH SUFFICIENT ARITHMETIC BACKGROUND ; SEE

ALSO P ROPOSITION 15.4.15 AND §15.5.5). Recall the definition of the ring of
integers OK in a number field K, Remark 10.7.1. A fractional ideal a of OK is a
nonzero OK -submodule of K such that there is a nonzero a ∈ OK such that aa ⊂ OK .
Products of fractional ideals are defined analogously to products of ideals in a
ring (defined in Exercise 3.4.C): ab consists of (finite) OK -linear combinations of
products of elements of a and elements of b. Thus fractional ideals form an abelian
semigroup under multiplication, with OK as the identity. In fact fractional ideals of
OK form a group.
(a) Explain how a fractional ideal on a ring of integers in a number field yields
an invertible sheaf. (Although we won’t need this, it is worth noting that
a fractional ideal is the same as an invertible sheaf with a trivialization at
the generic point.)
(b) A fractional ideal is principal if it is of the form rOK for some r ∈ K× .
Show that any two fractional ideals that differ by a principal fractional
ideal yield the same invertible sheaf.
(c) Show that two fractional ideals that yield the same invertible sheaf differ
by a principal ideal.
(d) The class group is defined to be the group of fractional ideals modulo the
principal ideals (i.e., modulo K× ). Give an isomorphism of the class group
with the Picard group of OK .
(This discussion applies to any Dedekind domain. See Exercise 15.5.O for a follow-
up.)

14.2 Locally free sheaves on schemes in particular

 © 2024 Ravi Vakil. Published by Princeton University Press. 405

We now discuss aspects of locally free sheaves more specific to schemes. Based
on your intuition for line bundles on manifolds, you might hope that every point
has a “small” open neighborhood on which all invertible sheaves (or locally free
sheaves) are trivial. Sadly, this is not the case. We will eventually see (§19.11.1)
that for the curve y2 − x3 − x = 0 in A2C , every nonempty open set has nontrivial
invertible sheaves. (This will use the fact that it is an open subset of an elliptic curve.)
This is a persistent problem in algebraic geometry: the Zariski topology is too blunt.

14.2.A. E ASY E XERCISE . Prove that locally free sheaves on a scheme are quasico-
herent.
Similarly, finite rank locally free sheaves are always finite type, and if OX is
coherent, finite rank locally free sheaves on X are coherent. (Remember: If OX is
not coherent, then coherence is a pretty useless notion on X.)

14.2.B. U NIMPORTANT EXERCISE . Use the example of Exercise 6.1.B(b) to show

that not every quasicoherent sheaf is locally free.

14.2.C. E XERCISE . Show that every (finite) rank n vector bundle on A1k is trivial of
rank n. Hint: finitely generated modules over a principal ideal domain are finite
direct sums of cyclic modules, as mentioned in Remark 6.4.4. (Be careful: Before
invoking this, you need to be sure that the module in question is finitely generated!)
See the aside in §15.4.11 for the difficult generalization to An
k.

14.2.1. Definition. We define rational (and regular) sections of a locally free sheaf
on a scheme X just as we did rational (and regular) functions (see for example §6.6
and §7.5).

14.2.D. E XERCISE . Suppose s is a section of a locally free sheaf F on a scheme X.

Define the notion of the closed subscheme cut out by s = 0, denoted (for obvious
reasons) V(s). Be sure to check that your definition is independent of choices! (This
exercise gives a new solution to Exercise 4.5.S.) Hint: given a trivialization over
an open set U, s corresponds to a number of functions f1 , . . . on U; on U, take the
scheme cut out by these functions. Alternate hint that avoids coordinates: figure
out how to define it as the largest closed subscheme on which s restricts to 0.

14.2.E. E ASY E XERCISE . Show that locally free sheaves on locally Noetherian
normal schemes satisfy “Algebraic Hartogs’s Lemma”: sections defined away from
a set of codimension at least 2 extend over that set. (Algebraic Hartogs’s Lemma
for Noetherian normal schemes is Theorem 13.5.19.)

14.2.F. E ASY E XERCISE . Suppose s is a nonzero rational section of an invertible

sheaf on a locally Noetherian normal scheme. Show that if s has no poles, then s is
regular. (Hint: Exercise 13.5.I.)

14.2.2. Locally free sheaves in short exact sequences.

14.2.G. E XERCISE ( POSSIBLE HELP FOR LATER PROBLEMS ).

(a) Suppose

(14.2.2.1) 0 / F′ /F / F ′′ /0
406 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

is a short exact sequence of quasicoherent sheaves on X. Suppose U = Spec A is

an affine open set where F ′ , F ′′ are free, say F ′ |Spec A = A
e ⊕a , F ′′ |Spec A = A
e ⊕b .
(Here a and b are assumed to be finite for convenience, but this is not necessary,
so feel free to generalize to the infinite rank case.) Show that F is also free on
Spec A, and that 0 → F ′ → F → F ′′ → 0 can be interpreted as coming from the
tautological exact sequence 0 → A⊕a → A⊕(a+b) → A⊕b → 0. (As a consequence,
given an exact sequence of quasicoherent sheaves (14.2.2.1) where F ′ and F ′′ are
locally free, F must also be locally free.)
(b) In the finite rank case, show that given an open covering by trivializing affine
open sets (of the form described in (a)), the transition functions (really, matrices) of
F may be interpreted as block upper triangular matrices, where the top left a × a
blocks are transition functions for F ′ , and the bottom b × b blocks are transition
functions for F ′′ .

14.2.H. E XERCISE . Suppose (14.2.2.1) is a short exact sequence of quasicoherent

sheaves on X. By Exercise 14.2.G(a), if F ′ and F ′′ are locally free, then F is too.
(a) If F and F ′′ are locally free of finite rank, show that F ′ is too. Hint: Reduce
to the case X = Spec A and F and F ′′ free. Interpret the map ϕ : F → F ′′ as
an n × m matrix M with values in A, with m the rank of F and n the rank of
F ′′ . For each point p of X, show that there exist n columns {c1 , . . . , cn } of M that
are linearly independent at p and hence near p (as linear independence is given
by nonvanishing of the appropriate n × n determinant). Thus X can be covered
by distinguished open subsets in bijection with the choices of n columns of M.
Restricting to one subset and renaming columns, reduce to the case where the
determinant of the first n columns of M is invertible. Then change coordinates on
A⊕m = F (Spec A) so that M with respect to the new coordinates is the identity
matrix in the first n columns, and 0 thereafter. Finally, in this case interpret F ′ as
^ .
A⊕(m−n)
(b) If F ′ and F are both locally free, show that F ′′ need not be. Hint: consider
(14.2.3.1), which we can interpret as the closed subscheme exact sequence (9.1.2.1)
for a point on A1 . (If an injection of locally free sheaves α : F ′ ,→ F is such that
the cokernel F ′′ is also locally free, people sometimes say that α is an injection
of vector bundles, not just of locally free sheaves. Implicitly, they are considering
a category of vector bundles which is different from the category of locally free
sheaves; the cokernel of a map of vector bundles is then always a vector bundle.
See §14.2.4 for further discussion.)

14.2.I. E XERCISE . (a) Suppose 0 → F ′ → F → F ′′ → 0 is a short exact sequence

of finite rank locally free sheaves. Show that

0 / F ′′∨ / F∨ / F ′∨ /0

(obtained by applying Hom (·, OX ) to the original sequence) is also a short exact
sequence of finite rank locally free sheaves.
⋆ (b) Suppose · · · F2 → F1 → F0 → 0 is an exact sequence of finite rank locally
free sheaves. Show that the induced sequence

0 / F∨ / F∨ / F∨ / ···
0 1 2

is also an exact sequence of finite rank locally free sheaves.

 © 2024 Ravi Vakil. Published by Princeton University Press. 407

14.2.3. Vector bundle maps.

Maps of vector bundles (in the more conventional sense) are more restrictive
than morphisms of quasicoherent sheaves that happen to be locally free sheaves.
A locally free subsheaf of a locally free sheaf does not always yield a “subvector
bundle” of a vector bundle. The archetypal example is the exact sequence of
quasicoherent sheaves on A1k = Spec k[t] corresponding to the following exact
sequence of k[t]-modules:
(14.2.3.1) 0 / tk[t] / k[t] /k / 0.

The locally free sheaf tk[t]

g is a subsheaf of k[t],
g but it does not correspond to
a “subvector bundle”; the cokernel is not a vector bundle. (This example was
mentioned in §14.1.8.)
Because maps of locally free sheaves that are actually “maps of vector bundles”
have some particularly good and useful behavior, it seems worthwhile to give them
a name. There is no accepted name, so we will just choose one.

14.2.4. Definition. Let X be a scheme (or indeed more generally a ringed space), and
let E and F be two finite rank locally free sheaves on X, of rank a + b and a + c
respectively. We say that ϕ : E → F is a rank a map of vector bundles if for every
point p ∈ X, there is an open neighborhood U of p with a commutative diagram

E |U
ϕ
/ F |U
O O
∼ ∼
 
⊕(a+b) //O ⊕a   / O ⊕(a+c)
OU U U
⊕(a+b) ⊕a
where the map OU ↠ OU is projection to the first a summands, and the
⊕a ⊕(a+c)
map OU ,→ OU is inclusion as the first a summands. We say that ϕ is a
map of vector bundles if near every point p ∈ X, ϕ is a rank a map of vector
bundles for some a. If b = 0, we say ϕ is an injection of vector bundles (as in
Exercise 14.2.H(b)); we use this terminology sparingly.

14.2.5. Important but straightforward observations. (i) The notion “rank a map of vec-
tor bundles” commutes with any base change Y → X. (ii) The quasicoherent sheaves
ker ϕ, im ϕ, and coker ϕ are locally free (of finite rank b, a, and c respectively), and
their construction commutes with any base change Y → X.

0 / ker ϕ /E ϕ
/F / coker ϕ /0
O O O O
∼ ∼ ∼ ∼
   
0 / O ⊕b / O ⊕(a+b) / O ⊕(a+c) / O ⊕c /0
U U U U

(iii) If ϕ : E → F is a rank a map of vector bundles, then at any point p ∈ X, the

rank of ϕ|p : E |p → F |p is a.
In the next exercises we give three criteria for when a morphism ϕ of finite
rank locally free sheaves is a map of vector bundles.

14.2.J. E XERCISE ( USEFUL CHARACTERIZATION OF “ MAP OF VECTOR BUNDLES ”).

Suppose ϕ : E → F is a morphism of finite rank locally free sheaves, where F has
408 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

rank a + c. Show that ϕ is a rank a map of vector bundles if and only if coker ϕ
is locally free of rank c. (In particular, every surjection of finite rank locally free
sheaves is a map of vector bundles.)

14.2.K. E XERCISE . Suppose ϕ : E → F is a morphism of finite rank locally free

sheaves, and p ∈ X. Show that the natural map (ker ϕ)|p → ker(ϕ|p ) is surjective if
and only if ϕ is a map of vector bundles in some neighborhood of p.

14.2.6. ⋆ Tensor algebra constructions.

(These important ideas require more work, so you may wish to leave them
for later.) For the next exercises, recall the following. If M is an A-module, then
the tensor algebra T • (M) is a noncommutative algebra, graded by Z≥0 , defined
as follows. T 0 (M) = A, T n (M) = M ⊗A · · · ⊗A M (where n terms appear in the
product), and multiplication is what you expect.
The symmetric algebra Sym• M is a symmetric algebra, graded by Z≥0 , defined
as the quotient of T • (M) by the (two-sided) ideal generated by all elements of the
form x ⊗ y − y ⊗ x for all x, y ∈ M. Thus Symn M is the quotient of M ⊗ · · · ⊗ M
by the relations of the form m1 ⊗ · · · ⊗ mn − m1′ ⊗ · · · ⊗ mn′ where (m1′ , . . . , mn′ ) is
a rearrangement of (m1 , . . . , mn ).
The exterior algebra ∧• M is defined to be the quotient of T • M by the (two-
sided) ideal generated by all elements of the form x ⊗ x for all x ∈ M. Expanding
(a + b) ⊗ (a + b), we see that a ⊗ b = −b ⊗ a in ∧2 M. This implies that if 2 is
invertible in A (e.g., if A is a field of characteristic not 2), ∧n M is the quotient of
M⊗· · ·⊗M by the relations of the form m1 ⊗· · ·⊗mn −(−1)sgn(σ) mσ(1) ⊗· · ·⊗mσ(n)
where σ is a permutation of {1, . . . , n}. The exterior algebra is a “skew-commutative”
A-algebra.
Better: both Sym and ∧ can be defined by universal properties. For example,
any map of A-modules M⊗n → N that is symmetric in the n entries factors uniquely
through Symn A (M).
It is most correct to write TA •
(M), Sym•A (M), and ∧•A (M), but the “base ring”
A is usually omitted for convenience.

14.2.L. E XERCISE . If F is locally free of rank m, show that T n F , Symn F , and

∧n F are locally free, and find their ranks. We note that in this case, ∧rank F F is
denoted det F , and is called the determinant (line) bundle or (both better and
worse) the determinant locally free sheaf.
(Remark: These constructions can be defined for O-modules on an arbitrary
ringed space.)
You can also define the sheaf of noncommutative algebras T • F , the sheaf of
commutative algebras Sym• F , and the sheaf of skew-commutative algebras ∧• F .

14.2.M. E XERCISE . Suppose 0 → F ′ → F → F ′′ → 0 is an exact sequence of

locally free sheaves. Show that for any r, there is a filtration of Symr F

Symr F = G 0 ⊃ G 1 ⊃ · · · ⊃ G r ⊃ G r+1 = 0

with subquotients

∼ (Symp F ′ ) ⊗ (Symr−p F ′′ ).
G p /G p+1 =
 © 2024 Ravi Vakil. Published by Princeton University Press. 409

(Here are two different possible hints for this and Exercise 14.2.N: (1) Interpret
the transition matrices for F as block upper triangular, with two blocks, where
one diagonal block gives the transition matrices for F ′ , and the other gives the
transition matrices for F ′′ (cf. Exercise 14.2.2.1(b)). Then appropriately interpret
the transition matrices for Symr F as block upper triangular, with r + 1 blocks. (2)
It suffices to consider a small enough affine open set Spec A, where F ′ , F , F ′′ are
free, and to show that your construction behaves well with respect to localization
at an element f ∈ A. In such an open set, the sequence is 0 → A⊕p → A⊕(p+q) →
A⊕q → 0 by the Exercise 14.2.G. Let e1 , . . . , ep be the standard basis of A⊕p , and
f1 , . . . , fq be the the standard basis of A⊕q . Let e1′ , . . . , ep′ be denote the images of
e1 , . . . , ep in A⊕(p+q) . Let f1′ , . . . , fq′ be any lifts of f1 , . . . , fq to A⊕(p+q) . Note that
fi′ is well-defined modulo e1′ , . . . , ep′ . Note that
∼ ⊕r Symi F ′ |Spec A ⊗O
Symr F |Spec A = Symr−i F ′′ |Spec A .
i=0 Spec A

Show that G p := ⊕ri=p Symi F ′ |Spec A ⊗OSpec A Symr−i F ′′ |Spec A gives a well-defined
(locally free) subsheaf that is independent of the choices made, e.g., of the basis e1 ,
. . . , ep , f1 , . . . , fq , and the lifts f1′ , . . . , fq′ .)

14.2.N. U SEFUL E XERCISE . Suppose 0 → F ′ → F → F ′′ → 0 is an exact sequence

of locally free sheaves. Show that for any r, there is a filtration of ∧r F :
∧r F = G 0 ⊃ G 1 ⊃ · · · ⊃ G r ⊃ G r+1 = 0
with subquotients
G p /G p+1 =∼ (∧p F ′ ) ⊗ (∧r−p F ′′ )
for each p. In particular, if the sheaves have finite rank, then det F = (det F ′ ) ⊗
(det F ′′ ).

14.2.O. E XERCISE . Suppose F is locally free of rank n. Describe a map ∧r F ×

∼ ∨
∧n−r F → ∧n F that induces an isomorphism ∧r F −→ (∧n−r F ) ⊗ ∧n F . This
is called a perfect pairing of vector bundles. (If you know about perfect pairings
of vector spaces, do you see why this is a generalization?) You might use this later
in showing duality of Hodge numbers of regular varieties over algebraically closed
fields, Exercise 21.5.L.

14.2.P. E XERCISE ( DETERMINANT LINE BUNDLES BEHAVE WELL IN EXACT SE -

QUENCES ). Suppose 0 → F1 → · · · → Fn → 0 is an exact sequence of finite
rank locally free sheaves on X. Show that “the alternating product of determinant
bundles is trivial”:
det(F1 ) ⊗ det(F2 )∨ ⊗ det(F3 ) ⊗ det(F4 )∨ ⊗ · · · ⊗ det(Fn )(−1) =
n
∼ OX .
(Hint: break the exact sequence into short exact sequences. Use Exercise 14.2.H(a)
to show that they are short exact sequences of finite rank locally free sheaves. Then
use Exercise 14.2.N.)

14.2.Q. E XERCISE . Suppose F is a quasicoherent sheaf. Define the quasicoherent

sheaves T n F , Symn F , and ∧n F . (One possibility: describe them on each affine
open set, and use the characterization of Important Exercise 6.2.D.)
410 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

14.3 More pleasant properties of finite type and coherent sheaves

We now describe a number of important facts about quasicoherent sheaves

when we have some finiteness conditions.

14.3.A. E XERCISE (Hom BEHAVES REASONABLY IF THE SOURCE IS FINITELY PRE -

SENTED , CF. E XERCISE 1.5.H).
(a) (“Hom (fpr, qcoh) is qcoh”) Suppose F is a finitely presented (quasicoherent)
sheaf on X, and G is a quasicoherent sheaf on X. Show that Hom (F , G ) is a
quasicoherent sheaf. Hint: take a “finite presentation” of F locally: on U ⊂ X,
O ⊕m → O ⊕n → F → 0. Apply the left-exact functor (from OX -modules to OX -
modules, §2.6.4) Hom OX (·, G ). (Aside: For an example of quasicoherent sheaves F
and G on a scheme X such that Hom (F , G ) is not quasicoherent, let A be a discrete
valuation ring with uniformizer t, let X = Spec A, let F = M f and G = N e with
∞ ∞
M = ⊕i=1 A and N = A. Then Mt = ⊕i=1 At , and of course Nt = At . Consider the
homomorphism ϕ : Mt → Nt sending 1 in the ith factor of Mt to 1/ti . Then ϕ is
not the localization of any element of HomA (M, N).)
(b) (“Hom (fpr, coh) is coh”) If further G is coherent, show that Hom (F , G ) is also
coherent.
(c) Suppose F is a finitely presented sheaf on X, and G is a quasicoherent sheaf
on X. Show that Hom (F , ·) gives a left-exact covariant functor QCohX → QCohX
and CohX → CohX , and that Hom (·, G ) gives a left-exact contravariant functor from
finitely presented sheaves on X to QCohX (cf. Exercise 2.6.I).

14.3.1. Duals of coherent sheaves. From Exercise 14.3.A(b), if F is coherent, its

dual F ∨ := Hom (F , OX ) (see §2.3.4) is too. This generalizes the notion of duals
of vector bundles in Exercise 14.1.D. Your argument there generalizes to show
that there is always a natural “double dual” morphism F → (F ∨ )∨ . Unlike
in the vector bundle case, this is not always an isomorphism. (For an example,
let F be the coherent sheaf associated to k[t]/(t) on A1 = Spec k[t], and show
that F ∨ = 0.) Coherent sheaves for which the “double dual” morphism is an
isomorphism are called reflexive sheaves, but we won’t use this notion. The
canonical map F ⊗ F ∨ → OX is called the trace map — can you see why?

14.3.B. E XERCISE . Suppose

(14.3.1.1) 0 /F /G /H /0

is an exact sequence of quasicoherent sheaves on a scheme X, where H is a locally

free quasicoherent sheaf, and suppose E is a quasicoherent sheaf. By left-exactness
of Hom (Exercise 2.6.I),

0 / Hom (H , E ) / Hom (G , E ) / Hom (F , E ) /0

is exact except possibly on the right. Show that it is also exact on the right. (Hint:
this is local, so you can assume that X is affine, say Spec A, and H = A g ⊕I , so

(14.3.1.1) can be written as 0 → M → N → A → 0. Show that this exact sequence

⊕I

splits, so we can write N = M ⊕ A⊕I in a way that respects the exact sequence.)
In particular, if F , G , and H are coherent, and H is locally free, then we have an
 © 2024 Ravi Vakil. Published by Princeton University Press. 411

exact sequence of coherent sheaves

0 /H∨ / G∨ / F∨ / 0.

14.3.C. E XERCISE (“Supp(FTF ) IS CLOSED ”). Suppose F is a sheaf of abelian

groups. (a) Show that the support of a finite type quasicoherent sheaf on a scheme
X is a closed subset. (Hint: Exercise 6.6.B(b).) (b) Show that the support of a
quasicoherent sheaf need not be closed. (Hint: If A = C[t], then C[t]/(t − a) is an
A-module supported at a. Consider
Q ⊕a∈C C[t]/(t − a). Be careful: this example
won’t work if ⊕ is replaced by .)
We next come to a geometric interpretation of Nakayama’s Lemma, which is
why Nakayama’s Lemma should be considered a geometric fact (with an algebraic
proof). Informally: if some sections of a finite type quasicoherent sheaf generate
the sheaf “at p” (the fiber at p), then they generate “near p” (in an actual open
neighborhood of p).

14.3.D. U SEFUL E XERCISE : G EOMETRIC N AKAYAMA ( FOR FINITE TYPE SHEAVES ,

GENERATORS AT p ARE GENERATORS NEAR p). Suppose X is a scheme, and F
is a finite type quasicoherent sheaf. Show that if U ⊂ X is an open neighborhood
of p ∈ X and a1 , . . . , an ∈ F (U) so that their images a1 |p , . . . , an |p generate the
fiber F |p (defined as Fp ⊗ κ(p), §4.3.8), then there is an affine open neighborhood
p ∈ Spec A ⊂ U of p such that “a1 |Spec A , . . . , an |Spec A generate F |Spec A ” in the
following senses:
(i) a1 |Spec A , . . . , an |Spec A generate F (Spec A) as an A-module;
(ii) for any q ∈ Spec A, a1 , . . . , an generate the stalk Fq as an OX,q -module
(and hence for any q ∈ Spec A, the fibers a1 |q , . . . , an |q generate the fiber
F |q as a κ(q)-vector space).
In particular, if F |p = 0, then there exists an open neighborhood V of p such that
F |V = 0.

14.3.E. U SEFUL E XERCISE ( FOR FINITELY PRESENTED SHEAVES , LOCALLY FREE =

FREE STALKS ). Suppose F is a finitely presented sheaf on a scheme X. Show that
if Fp is a free OX,p -module for some p ∈ X, then F is locally free in some open
neighborhood of p. Hence F is locally free if and only if Fp is a free OX,p -module
for all p ∈ X. Translation: for a finitely presented sheaf, local freeness is a stalk-local
property, and free stalks imply local freeness nearby.
Extended hint: Find an open neighborhood U of p, and n elements of F (U)
that generate Fp . Using Geometric Nakayama, Exercise 14.3.D, show that the
sections generate Fq for all q in some open neighborhood Y of p in U. Thus you
have described a surjection OY⊕n → F |Y . Show that the kernel of this map is finite
type (using Lemma 6.4.6, “finitely presented implies always finitely presented”),
⊕n
and hence has closed support (say Z ⊂ Y), which does not contain p. Thus OY\Z →
F |Y\Z is an isomorphism.
This is enlightening in a number of ways. It shows that for finitely presented
sheaves, local freeness is a stalk-local condition. Furthermore, on an integral scheme,
any finitely presented sheaf F is automatically free over the generic point (do you
see why?), so every finitely presented sheaf on an integral scheme is locally free
over a dense open subset. And any finitely presented sheaf that is 0 at the generic
412 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

point of an irreducible scheme is necessarily 0 on a dense open subset. The last

two sentences show the utility of generic points; such statements would have been
more mysterious in classical algebraic geometry. An example showing the necessity
of finiteness hypotheses: the Z-module A ⊂ Q, where A consists of fractions with
squarefree denominator, see [Stacks, tag 065J].

14.3.F. E XERCISE . (Torsion-free and torsion sheaves were defined in §6.1.4.)

(a) Show that torsion-free coherent sheaves on a regular (hence implicitly locally
Noetherian) curve are locally free.
(b) Show that torsion coherent sheaves on a quasicompact regular integral curve
are supported at a finite number of closed points.
(c) Suppose F is a coherent sheaf on a quasicompact (for convenience) regular
curve. Describe a canonical short exact sequence 0 → Ftors → F → Flf → 0, where
Ftor is a torsion sheaf, and Flf is locally free.
To answer Exercise 14.3.F, use Useful Exercise 14.3.E (local freeness can be
checked at stalks) to reduce to the discrete valuation ring case, and recall Re-
mark 6.4.4, the structure theorem for finitely generated modules over a principal
ideal domain A: any such module can be written as the direct sum of principal
modules A/(a). Hence:

14.3.2. Proposition. — If M is a finitely generated module over a principal ideal domain,

then M is torsion-free if and only if M is free.
(Exercise 24.1.B is closely related.)
Proposition 14.3.2 is false without the finite generation hypothesis: consider
M = K(A) for a suitably general ring A. It is also false if we give up the principal
ideal domain hypothesis: consider (x, y) ⊂ C[x, y] and (x, y) ⊂ C[x, y]/(xy). (These
examples require some verification.) These examples show that Exercise 14.3.F(a) is
false if we give up the “dimension 1” or “regular” hypothesis.

14.3.3. The fiber and rank of a quasicoherent sheaf at a point.

We now apply the definitions of fiber and rank of an OX -module at a point
p ∈ X (§4.3.8) to a quasicoherent sheaf F on a scheme X. The fiber of F at p is the
vector space
F |p := Fp /mFp = Fp ⊗OX,p κ(p),
where m is the maximal ideal of OX,p , and κ(p) is as usual the residue field OX,p /m
at p. A section of F over an open set containing p can be said to take on a value at
p, which is an element of this vector space. The rank of a quasicoherent sheaf F at
a point p is dimκ(p) Fp /mFp (possibly infinite). More explicitly, on any affine set
Spec A where p = [p] and F (Spec A) = M, then the rank is dimK(A/p) Mp /pMp .
Note that this definition of rank is consistent with the notion of rank of a locally
free sheaf. In the locally free case, the rank is a (locally) constant function of the
point. The converse is sometimes true, see Exercise 14.3.K below.

14.3.G. E XERCISE . Consider the coherent sheaf F on A1k = Spec k[t] corresponding
to the module k[t]/(t). Find the rank of F at every point of A1 . Don’t forget the
generic point!

14.3.H. E XERCISE . Show that at any point, rank(F ⊕ G ) = rank(F ) + rank(G ) and
rank(F ⊗ G ) = rank F rank G . (Hint: Show that direct sums and tensor products
 © 2024 Ravi Vakil. Published by Princeton University Press. 413

∼ M/IM ⊕
commute with ring quotients and localizations, i.e., (M ⊕ N) ⊗R (R/I) =
∼ ∼
N/IN, (M ⊗R N) ⊗R (R/I) = (M ⊗R R/I) ⊗R/I (N ⊗R R/I) = M/IM ⊗R/I N/IM,
etc.)
If X is irreducible, and F is a quasicoherent (usually coherent) sheaf on X,
then rank F (with no mention of a point) by convention means at the generic
point. (For example, a rank 0 quasicoherent sheaf on an integral scheme is a torsion
quasicoherent sheaf, see Definition 6.1.5.)

14.3.I. E ASY E XERCISE . Show that the rank of coherent sheaves on an integral
scheme X is additive in exact sequences: if

0 /F /G /H /0

is an exact sequence of coherent sheaves, show that

rank F − rank G + rank H = 0.
Hint: localization is exact. (Caution: your argument will use the fact that the rank
is at the generic point; the example

0 / k[t]
g ×t
/ k[t]
g / k[t]/(t)
^ /0

on A1k shows that rank at a closed point is not additive in exact sequences.)
If F is a finite type quasicoherent sheaf, then the rank of F at p is finite, and
by Nakayama’s Lemma, the rank is the minimal number of generators of Mp as an
Ap -module.

14.3.J. I MPORTANT E XERCISE ( RANK IS UPPER SEMICONTINUOUS FOR FINITE TYPE

QUASICOHERENT SHEAVES ). If F is a finite type quasicoherent sheaf on X, show
that rank(F ) is an upper semicontinuous function on X. Hint: generators at a point
p are generators nearby by Geometric Nakayama’s Lemma, Exercise 14.3.D. (The
example in Exercise 14.3.C shows the necessity of the finite type hypothesis.)

14.3.K. I MPORTANT E XERCISE ( FOR FINITE TYPE QUASICOHERENT SHEAVES ON

REDUCED SCHEMES , CONSTANT RANK = VECTOR BUNDLE ).
(a) If X is reduced, F is a finite type quasicoherent sheaf on X, and the rank of F is
constant, show that F is locally free.
Hint: prove the result in an open neighborhood of an arbitrary point p as
follows. Suppose n = rankp F . Choose n generators of the fiber F |p (a basis
as a κ(p)-vector space). By Geometric Nakayama’s Lemma 14.3.D, we can find
an affine open neighborhood Spec A ⊂ X of p, with F |Spec A = M, f so that the
chosen generators F |p lift to generators m1 , . . . , mn of M. Let ϕ : A⊕n //M
P
((r1 , . . . , rn ) 7→ ri mi ) be the corresponding surjection. Suppose ker ϕ ̸= 0 — say
(r1 , . . . , rn ) ∈ ker ϕ, with ri ̸= 0. As X is reduced, there is some p where ri ∈ / p.
Then ri is invertible in Ap , so Mp has fewer than n generators, contradicting the
constancy of rank.
(b) Then use upper semicontinuity of rank (Exercise 14.3.J) to show that finite type
quasicoherent sheaves on a reduced scheme are locally free on a dense open set.
(c) Show that part (a) can be false without the condition of X being reduced. (Hint:
Consider A = k[x]/(x2 ), M = k.)
414 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

14.3.L. E XERCISE . Suppose X is reduced, and ϕ : E → F is a morphism of

finite rank locally free sheaves. Show that ϕ is a rank a map of vector bundles
(Definition 14.2.4) if and only if ϕ|p : E |p → F |p is rank a for all points p ∈ X.
(Hint: use the previous exercise, and Exercise 14.2.J.)
You can use the notion of rank to help visualize finite type quasicoherent
sheaves, or even quasicoherent sheaves. For example, finite type quasicoherent
sheaves generalize finite rank vector bundles as follows: to each point there is
an associated vector space, and although the ranks can jump, they fit together in
families as well as one might hope. You might try to visualize the example of
Example 14.3.G. Nonreducedness can fit into the picture as well — how would
you picture the coherent sheaf on Spec k[ϵ]/(ϵ2 ) corresponding to k[ϵ]/(ϵ) (Exer-
cise 14.3.K(c))? How about k[ϵ]/(ϵ2 ) ⊕ k[ϵ]/(ϵ)?

14.3.4. Degree of a finite morphism at a point. Suppose π : X → Y is a finite morphism.

Then π∗ OX is a finite type (quasicoherent) sheaf on Y, and the rank of this sheaf at
a point p is called the degree of the finite morphism at p. By Exercise 14.3.J, the
degree of π is a upper semicontinuous function on Y. The degree can jump: consider
the closed embedding of a point into a line corresponding to k[t] → k given by
t 7→ 0. It can also be constant in cases that you might initially find surprising — see
Exercise 10.3.4, where the degree is always 2, but the 2 is obtained in a number of
different ways.

14.3.M. E XERCISE . Suppose π : X → Y is a finite morphism. By unwinding the

definition, verify that the degree of π at p is the dimension of the space of functions
on the scheme-theoretic preimage of p, considered as a vector space over the residue
field κ(p). In particular, the degree is zero if and only if π−1 (p) is empty.

14.4 Pushforwards of quasicoherent sheaves

We conclude the chapter by discussing pushforward and pullbacks of quasico-

herent sheaves, their properties, and some applications.
Suppose B → A is a morphism of rings. Recall (from Exercise 1.4.E) that
(· ⊗B A, ·B ) is an adjoint pair between the categories of A-modules and B-modules:
we have a bijection
∼ HomB (N, MB )
HomA (N ⊗B A, M) =
functorial in both arguments. These constructions behave well with respect to
localization (in an appropriate sense), and hence work (often) in the category
of quasicoherent sheaves on schemes (and indeed always in the category of O-
modules on ringed spaces, see Definition 14.5.6, although we won’t particularly
care). The easier construction (M 7→ MB ) will turn into our old friend pushforward.
The other (N 7→ A ⊗B N) will be a relative of pullback, whom I’m reluctant to call
an “old friend”.
We begin with the pushforwards, for which we have already done much of the
work.
The main moral of this section is that in “reasonable” situations, the pushfor-
ward of a quasicoherent sheaf is quasicoherent, and that this can be understood
 © 2024 Ravi Vakil. Published by Princeton University Press. 415

in terms of one of the module constructions defined above. We begin with a

motivating example:

14.4.A. E XERCISE . Let π : Spec A → Spec B be a morphism of affine schemes, and

suppose M is an A-module, so M f is a quasicoherent sheaf on Spec A. Give an
∼ g
isomorphism π∗ M −→ MB . (Hint: There is only one reasonable way to proceed:
f
look at distinguished open sets.)
In particular, π∗ M
f is quasicoherent. Perhaps more important, this implies
that the pushforward of a quasicoherent sheaf under an affine morphism is also
quasicoherent.

14.4.B. E XERCISE . If π : X → Y is an affine morphism, show that π∗ is an exact

functor QCohX → QCohY .
The following result generalizes the fact that the pushforward of a quasicoher-
ent sheaf under an affine morphism is also quasicoherent.

14.4.C. E SSENTIAL E XERCISE ( COROLLARY TO E XERCISE 6.2.G: “π∗ : QCohX →

QCohY FOR QCQS π”). Suppose π : X → Y is a quasicompact quasiseparated
morphism, and F is a quasicoherent sheaf on X. Show that π∗ F is a quasicoherent
sheaf on Y. (For counterexamples without the quasicompact or quasiseparated
hypotheses, see [Stacks, tag 078C].)
Coherent sheaves do not always push forward to coherent sheaves. For exam-
ple, consider the structure morphism π : A1k → Spec k, corresponding to k → k[t].
Then π∗ OA1k is the quasicoherent sheaf corresponding to k[t], which is not a finitely
generated k-module. But in good situations, coherent sheaves do push forward.
For example:

14.4.D. E XERCISE . Suppose π : X → Y is a finite morphism of locally Noetherian

schemes. If F is a coherent sheaf on X, show that π∗ F is a coherent sheaf. Hint:
Show first that π∗ OX is finite type.
Once we define cohomology of quasicoherent sheaves, we will quickly prove
that if F is a coherent sheaf on Pn
k , then Γ (Pk , F ) is a finite-dimensional k-module,
n

and more generally if F is a coherent sheaf on Proj S• , then Γ (Proj S• , F ) is a

coherent A-module (where S0 = A). This is a special case of the fact that the
“pushforwards of coherent sheaves by projective morphisms are also coherent
sheaves” (Theorem 18.7.1(d)). (The notion of projective morphism, a relative version
of Proj S• → Spec A, will be defined in §17.3.)
More generally, given Noetherian hypotheses, pushforwards of coherent sheaves
by proper morphisms are also coherent sheaves (Grothendieck’s Coherence Theo-
rem 18.9.1).

14.4.1. Remark. Vector bundles do not always pushforward to vector bundles,

which is another reason to work in the larger category of (quasi)coherent sheaves.
(Can you think of an example? Hint: Spec k[t]/(t) → Spec k[t].) But in particularly
good circumstances they can, which is one of the purposes of the Cohomology and
Base Change Theorem 25.1.6.
416 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

14.5 Pullbacks of quasicoherent sheaves: three different

perspectives

We next discuss the pullback of a quasicoherent sheaf: if π : X → Y is a mor-

phism of schemes, π∗ is a covariant functor QCohY → QCohX . (In fact our discussion
will apply readily to pulling back O-modules on ringed spaces, see §14.5.5 and
§14.5.6.) The notion of the pullback of a quasicoherent sheaf can be confusing on
first (and second) glance. (For example, it is not the inverse image sheaf, although
we will see that it is related.)

14.5.1. Here are three contexts in which you have seen the pullback, or can under-
stand it quickly. It may be helpful to keep these in mind, to keep you anchored in
the long discussion that follows. Suppose G is a quasicoherent sheaf on a scheme Y.
(i) (restriction to open subsets) If i : U ,→ Y is an open embedding, then i∗ G is
G |U , the restriction of G to U (Example 2.2.8).
(ii) (restriction to points) If i : p ,→ Y is the “inclusion” of a point p in Y (for
example, the closed embedding of a closed point; see Exercise 7.3.L(b)), then i∗ G is
G |p , the fiber of G at p (Definition 4.3.8).
The similarity of the notation G |U and G |p is precisely because both are pull-
backs. Pullbacks (especially to locally closed subschemes or generic points) are
often called restriction. For this reason, if π : X → Y is some sort of inclusion (such
as a locally closed embedding, or an “inclusion of a generic point”) then π∗ G is
often written as G |X and called the restriction of G to X, when π can be interpreted
as some type of “inclusion”.
(iii) (pulling back vector bundles) Suppose G is a locally free sheaf on Y, and
π : X → Y is any morphism. Then π∗ G is the pullback of G as a vector bundle, as
defined in §14.1.7. (This will be established in §14.5.6.)

14.5.2. Strategy. We will see three different ways of thinking about the pullback.
Each has significant disadvantages, but together they give a good understanding.
(a) Because we are understanding quasicoherent sheaves in terms of affine open
sets, and modules over the corresponding rings, we begin with an interpretation
in this vein. This will be very useful for proving and understanding facts. The
disadvantage is that it is annoying to make a definition out of this (when the target
is not affine), because gluing arguments can be tedious.
(b) As we saw with fibered product, gluing arguments can be made simpler
using universal properties, so our second “definition” will be by universal property.
This is elegant, but has the disadvantage that it still needs a construction, and
because it works in the larger category of O-modules, it isn’t clear from the universal
property that it takes quasicoherent sheaves to quasicoherent sheaves. But if the
target is affine, our construction of (a) is easily seen to satisfy universal property.
Furthermore, the universal property is “local on the target”: if π : X → Y is any
morphism, i : U ,→ Y is an open embedding, and G is a quasicoherent sheaf on
Y, then if π∗ G exists, then its restriction to π−1 (U) is (canonically identified with)
(π|U )∗ (G |U ). Thus if the pullback exists in general (even as an O-module), affine-
locally on Y it looks like the construction of (a) (and thus is quasicoherent).
(c) The third definition is one that works on ringed spaces in general. It is
short, and is easily seen to satisfy the universal property. It doesn’t obviously
take quasicoherent sheaves to quasicoherent sheaves (at least in the way that we
 © 2024 Ravi Vakil. Published by Princeton University Press. 417

have defined quasicoherent sheaves) — a priori it takes quasicoherent sheaves to

O-modules. But thanks to the discussion at the end of (b) above, which used (a),
this shows that the pullback of a quasicoherent sheaf is indeed quasicoherent.

14.5.3. First attempt at describing the pullback, using affines. Suppose π : X → Y

is a morphism of schemes, and G is a quasicoherent sheaf on Y. We want to define
the pullback quasicoherent sheaf π∗ G on X in terms of affine open sets on X and
Y. Suppose Spec A ⊂ X, Spec B ⊂ Y are affine open sets, with π(Spec A) ⊂ Spec B.
Suppose G |Spec B =∼ N.
e Perhaps motivated by the fact that pullback should relate to
tensor product, we want
(14.5.3.1) Γ (Spec A, π∗ G ) = N ⊗B A.
More precisely, we would like Γ (Spec A, π∗ G ) and N ⊗B A to be identified. This
could mean that we use this to construct a definition of π∗ G , by “gluing all this
information together” (and showing it is well-defined). Or it could mean that we
define π∗ G in some other way, and then find a natural identification (14.5.3.1). The
first approach can be made to work (and §14.5.4 is the first step), but we will follow
the second.

F IGURE 14.1. The pullback of a quasicoherent sheaf (module-theoretically)

14.5.4. We begin this project by fixing an affine open subset Spec B ⊂ Y. To avoid
confusion, let ϕ = π|π−1 (Spec B) : π−1 (Spec B) → Spec B. We will define a quasico-
herent sheaf on π−1 (Spec B) that will turn out to be ϕ∗ (G |Spec B ) (and will also be
the restriction of π∗ G to π−1 (Spec B)).
If Spec Af ⊂ Spec A is a distinguished open set, then
Γ (Spec Af , ϕ∗ G ) = N ⊗B Af = (N ⊗B A)f = Γ (Spec A, ϕ∗ G )f
where “=” means “canonically isomorphic”. Define the restriction map Γ (Spec A, ϕ∗ G ) →
Γ (Spec Af , ϕ∗ G ),
(14.5.4.1) Γ (Spec A, ϕ∗ G ) / Γ (Spec A, ϕ∗ G ) ⊗A Af ,

by α 7→ α ⊗ 1 (of course). Thus ϕ∗ G is (or: extends to) a quasicoherent sheaf on

π−1 (Spec B) (by Exercise 6.2.D).
We have now defined a quasicoherent sheaf on π−1 (Spec B), for every affine
open subset Spec B ⊂ Y. We want to show that this construction, as Spec B varies,
glues into a single quasicoherent sheaf on X.
418 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

You are welcome to do this gluing appropriately, for example using the dis-
tinguished affine base of Y (§6.2.1). This works, but can be confusing, so we take
another approach.

14.5.5. Universal property definition of pullback. Suppose π : X → Y is a

morphism of ringed spaces, and G is an OY -module. (We are of course interested in
the case where π is a morphism of schemes, and G is quasicoherent. Even once we
specialize our discussion to schemes, much of our discussion will extend without
change to this more general situation.) We “define” the pullback π∗ G as an OX -
module, using the following adjointness universal property: for any OX -module F ,
∼
there is a bijection HomOX (π∗ G , F ) ←→ HomOY (G , π∗ F ), and these bijections are
functorial in F . By universal property nonsense, this determines π∗ G up to unique
isomorphism; we just need to make sure that it exists (which is why the word
“define” is in quotes). Notice that we avoid worrying about when the pushforward
of a quasicoherent sheaf F is quasicoherent by working in the larger category of
O-modules.

14.5.A. I MPORTANT E XERCISE . If Y is affine, say Y = Spec B, show that the

construction of the quasicoherent sheaf in §14.5.4 satisfies this universal property
of pullback of G . Thus calling this sheaf π∗ G is justified. Hint: Interpret both sides
of the alleged bijection explicitly. The adjointness in the ring/module case should
turn up.

14.5.B. I MPORTANT E XERCISE . Suppose i : U ,→ X is an open embedding of ringed

spaces, and F is an OX -module. Show that F |U satisfies the universal property of
i∗ F (and thus deserves to be called i∗ F ). In other words, for each OU -module E ,
describe a bijection

∼
HomOU (F |U , E ) o / HomO (F , i∗ E ),
X

functorial in E .
We next show that if π∗ G satisfies the universal property (for the morphism
π : X → Y), then if j : V ,→ Y is any open subset, and i : U = π−1 (V) ,→ X (see
(14.5.5.1)), then (π∗ G )|U satisfies the universal property for π|U : U → V. Thus
(π∗ G )|U deserves to be called π|∗U (G |V ). You will notice that we really need to work
with O-modules, not just with quasicoherent sheaves.


U /X
i
(14.5.5.1) π−1 (V)
π|U π
  
V /Y
j
 © 2024 Ravi Vakil. Published by Princeton University Press. 419

If F ′ is an OU -module, we have a series of bijections (using Important Exer-

cise 14.5.B and adjointness of pullback and pushforward):
∼
HomOU ((π∗ G )|U , F ′ ) ←→ HomOU (i∗ (π∗ G ), F ′ )
∼
←→ HomOX (π∗ G , i∗ F ′ )
∼
←→ HomOY (G , π∗ i∗ F ′ )
∼
←→ HomOY (G , j∗ (π|U )∗ F ′ )
∼
←→ HomOV (j∗ G , (π|U )∗ F ′ )
∼
←→ HomOV (G |V , (π|U )∗ F ′ ).
We have thus described a bijection
∼
HomOU ((π∗ G )|U , F ′ ) o / HomO (G |V , (π|U )∗ F ′ )
V

which is (by construction) functorial in F ′ . Hence the discussion in the first

paragraph of §14.5.5 is justified. For example, thanks to Important Exercise 14.5.A,
the pullback π∗ exists if Y is an open subset of an affine scheme.
At this point, we could show that the pullback exists, following the idea behind
the construction of the fibered product: we would start with the definition when Y
is affine, and “glue”. We will instead take another route.

14.5.6. Third definition: pullback of O-modules via explicit construction. Sup-

pose π : X → Y is a morphism of ringed spaces, and G is an OY -module. Of course,
our case of interest is if π is a morphism of schemes, and G is quasicoherent. Now
π−1 G is a π−1 OY -module. (Notice that we are using the ringed space (X, π−1 OY ),
not (X, OX ). The inverse image construction π−1 was discussed in §2.7.) Further-
more, OX is also a π−1 OY -module, via the map π−1 OY → OX that is part of the data
of the morphism π. Define the pullback of G by π as the OX -module
(14.5.6.1) π∗ G := π−1 G ⊗π−1 OY OX .
It is immediate that pullback is a covariant functor π∗ : ModOY → ModOX , and that
π∗ OY is canonically isomorphic to OX . We have now established §14.5.1(iii), that
pullback of quasicoherent sheaves subsumes our older definition of pullback of
locally free sheaves.

14.5.C. I MPORTANT E XERCISE . Show that this definition (14.5.6.1) of pullback

satisfies the universal property (§14.5.5). Thus the pullback exists, at least as a
functor ModOY → ModOX .

14.5.D. I MPORTANT E XERCISE ( PULLBACK IS A COVARIANT FUNCTOR ON QUASI -

COHERENT SHEAVES ). Show that if π : X → Y is a morphism of schemes, then π∗
gives a covariant functor QCohY → QCohX . (You might use §14.5.4, Exercise 14.5.B,
Exercise 14.5.A, and/or §14.5.5.)
If G is a quasicoherent sheaf on Y, note that we can interpret a global section of
Y as a map s : OY → G (the section of G is the image of the section 1 of OY ). Note
that a section of G is the data of a map OY → G . By Exercise 14.5.D, if s : OY → G is
a section of G , then there is a natural section π∗ s : OX → π∗ G of π∗ G . We call π∗ s
the pullback of the section s.
420 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

The following is immediate from the universal property, and Exercise 14.5.D.
The “quasicompact and quasiseparated” hypotheses are solely to ensure that π∗
indeed sends QCohX to QCohY (Exercise 14.4.C).

14.5.7. Proposition. — Suppose π : X → Y is a quasicompact, quasiseparated morphism.

Then (π∗ : QCohY → QCohX , π∗ : QCohX → QCohY ) are an adjoint pair: there is an
isomorphism
∼
(14.5.7.1) HomOX (π∗ G , F ) o / HomO (G , π∗ F ),
Y

functorial in both F ∈ QCohX and G ∈ QCohY .

14.5.E. I MPORTANT EXERCISE ( PULLBACK IS A right-exact FUNCTOR ). Suppose

π : X → Y is a morphism of schemes. Show that π∗ : QCohY → QCohX is a
right-exact functor. Possible hint: ⊗ is right-exact.
For an important class of morphisms, called flat morphisms, we will see that
pullback is an exact functor (Exercise 24.1.H).

14.5.F. E XERCISE ( PULLBACK PRESERVES FINITE TYPE AND FINITELY PRESENTED

QUASICOHERENT SHEAVES ). Suppose that π : X → Y is a morphism of schemes,
and G is a quasicoherent sheaf on Y. If G is finite type (resp., finitely presented),
show that π∗ G is as well. Hint: cover Y by open sets for which there are surjections
⊕n ⊕m ⊕n
OU → G |U → 0 (resp., OU → OU → G |U → 0). Apply the the right-exact
functor π . (Hence if X is locally Noetherian, and G is coherent, then so is π∗ G . It is
∗

not always true that the pullback of a coherent sheaf is coherent, and the interested
reader can think of a counterexample.)

14.5.G. E ASY EXERCISE . Suppose ξ : W → X and π : X → Y are morphisms of

schemes, and G is a quasicoherent sheaf on Y. Show that there is a canonical
∼
isomorphism ξ∗ π∗ G ←→ (π ◦ ξ)∗ G .

14.5.H. E XERCISE . Suppose π : X → Y is a morphism of schemes, with π(p) = q,

and G is a quasicoherent sheaf on Y.
(a) (pullback on stalks) Show that pullback induces an isomorphism
∼ / G q ⊗O
(14.5.7.2) (π∗ G )p Y,q
OX,p .

(b) (pullback on fibers) Show that pullback induces an isomorphism

∼ / G |q ⊗κ(q) κ(p) .
(π∗ G )|p

Possible hint: Given a ring map B → A with [m] 7→ [n], where m ⊂ A and n ⊂ B
∼ (N ⊗B (B/n)) ⊗B/n (A/m) by
are maximal ideals, show that (N ⊗B A) ⊗A (A/m) =
showing both sides are isomorphic to N ⊗B (A/m).

14.5.I. E XERCISE : PULLBACK PRESERVES TENSOR PRODUCT. Suppose π : X → Y is

a morphism of schemes, and G and G ′ are quasicoherent sheaves on Y. Show that
π∗ (G ⊗OY G ′ ) = π∗ G ⊗OX π∗ G ′ .

14.5.J. U NIMPORTANT E XERCISE . Verify that the following is an example showing

that pullback is not left-exact: consider the exact sequence of sheaves on A1k , where
 © 2024 Ravi Vakil. Published by Princeton University Press. 421

i : p ,→ A1k is the origin:

0 / OA1 (−p) / OA1 / i∗ O|p / 0.

k k

(This is the closed subscheme exact sequence (9.1.2.1) for p ∈ A1k , and corresponds
to the exact sequence of k[t]-modules 0 → tk[t] → k[t] → k → 0. Warning: here
O|p is not the stalk Op ; it is the structure sheaf of the scheme p.) Restrict to p.

14.5.K. E XERCISE ( THE PUSH - PULL MAP FOR QUASICOHERENT SHEAVES ). Suppose
that
ψ′
X′ /X

π′ π
 
Y′
ψ
/Y
is a commutative diagram of schemes, and π and π ′ are both quasicompact and qua-
siseparated (so pushforward by both sends quasicoherent sheaves to quasicoherent
sheaves). Describe a natural map (functorial in F ∈ QCohX ) of quasicoherent
sheaves on Y ′ ,
ψ∗ π ∗ F / π∗′ (ψ ′ )∗ F

following the Definition 2.7.4 of the push-pull map for sheaves in general, as you
may have done in Exercise 7.2.D(f) in the case of O-modules. (See Exercise 18.7.B
for a generalization to cohomology.)

14.5.8. By applying the above exercise in the special case where Y ′ is a point q of Y,
we see that there is a natural map from the fiber of the pushforward to the sections
over the fiber:
(14.5.8.1) π∗ F ⊗ κ(q) / Γ (π−1 (q), F |π−1 (q) ).

One might hope that (14.5.8.1) is an isomorphism, i.e., that π∗ F “glues together”
the fibers Γ (π−1 (q), F |π−1 (q) ). This is too much to ask, but at least (14.5.8.1) gives
a map. (In fact, under just the right circumstances, (14.5.8.1) is an isomorphism, see
§25.1.)

14.5.L. I MPORTANT E XERCISE ( THE PROJECTION FORMULA , TO BE GENERALIZED

IN E XERCISE 18.7.E). Suppose π : X → Y is quasicompact and quasiseparated, and
F and G are quasicoherent sheaves on X and Y respectively.
(a) Describe a natural morphism (π∗ F ) ⊗ G → π∗ (F ⊗ π∗ G ).
(b) If G is locally free, show that this natural morphism is an isomorphism. (Hint:
what if G is free?)
(c) If π is affine, then show that this natural morphism is an isomorphism.
(Sadly, the phrase “projection formula” is used for a number of facts, not all
related.)

14.5.9. Remark: Pulling back ideal sheaves. There is an important subtlety in

pulling back quasicoherent ideal sheaves. Suppose i : X ,→ Y is a closed embedding,
and µ : Y ′ → Y is an arbitrary morphism. Let X ′ := X ×Y Y ′ . As “closed embeddings
pull back to closed embeddings” (§10.2.2), the pulled back map i ′ : X ′ → Y ′ is a
422 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

closed embedding.
X ′ _ / X
_
′
i i
 
Y′
µ
/Y
∼ ∼
Now µ∗ induces canonical isomorphisms µ∗ OY ←→ OY ′ and µ∗ (i∗ OX ) ←→ (i∗′ OX ′ ).
(Why is the latter isomorphism true? Hint: check affine-locally.) But it is not nec-
essarily true that µ∗ IX/Y = IX ′ /Y ′ . (Exercise 14.5.J yields an example.) This is
because the application of µ∗ to the closed subscheme exact sequence

0 / IX/Y / OY / i∗ OX /0

yields something that is a priori only right-exact:

µ∗ IX/Y / OY ′ / i∗′ OX ′ / 0.

Thus, as IX ′ /Y ′ is the kernel of OY ′ → i∗′ OX ′ , we see that IX ′ /Y ′ is the image of

µ∗ IX/Y in OY ′ . We can also see this explicitly from Exercise 10.2.B: affine-locally,
the ideal of the pullback is generated by the pullback of the ideal.
Note also that if µ is an exact functor (which we will later see as a consequence
of flatness, in Exercise 24.1.H), then µ∗ IX/Y → IX ′ /Y ′ is an isomorphism.

14.6 The quasicoherent sheaf corresponding to a graded module

We now describe quasicoherent sheaves on a projective A-scheme. Recall that

a projective A-scheme is produced from the data of a Z≥0 -graded ring S• , with
S0 = A, and S+ is a finitely generated ideal (a “finitely generated graded ring over
A”, §4.5.6). The resulting scheme is denoted Proj S• .
Suppose M• is a graded S• -module, graded by Z. (While reading the next
section, you may wonder why we don’t grade by Z≥0 . You will see that it doesn’t
matter. A Z-grading will make things cleaner when we produce an M• from a
quasicoherent sheaf on Proj S• .) We define the quasicoherent sheaf M g• as follows.
(I will avoid calling it M, as this might cause confusion with the affine case; but
f
g• is not graded in any way.) For each homogeneous f of positive degree, we define
M
• (f) on the distinguished open D(f) = {p : f(p) ̸= 0}
a quasicoherent sheaf M ^
by ((M^ • )f )0 — note that ((M• )f )0 is an ((S• )f )0 -module, and recall that D(f) is
identified with Spec((S• )f )0 (Exercise 4.5.I). As in (4.5.7.1), the subscript 0 means
“the 0-graded piece”. We have obvious isomorphisms of the restriction of M ^ • (f) and

M^• (g) to D(fg), satisfying the cocycle conditions. (Think through this yourself, to
be sure you agree with the word “obvious”!) By Exercise 2.5.E, these sheaves glue
together to a single sheaf Mg• on Proj S• . We then discard the temporary notation
M^• (f).
The O-module M g• is clearly quasicoherent, because it is quasicoherent on each
D(f), and quasicoherence is local.
 © 2024 Ravi Vakil. Published by Princeton University Press. 423

14.6.A. E XERCISE . Give an isomorphism between the stalk of M g• at a point corre-

sponding to homogeneous prime p ⊂ S• and ((M• )p )0 . Here ((M• )p )0 means “the
degree 0 homogeneous fractions in (M• )p ”, and p should be a “relevant” homoge-
neous prime (i.e., [p] is a point of Proj S• ). (Remark: You can use this exercise to give
an alternate definition of M g• in terms of “compatible stalks”, cf. Exercise 4.5.P.)

Given a map of graded modules ϕ : M• → N• , we get an induced map of

sheaves M g• → N f• . Explicitly, over D(f), the map M• → N• induces (M• )f →
(N• )f , which induces ϕf : ((M• )f )0 → ((N• )f )0 ; and this behaves well with re-
spect to restriction to smaller distinguished open sets, i.e., the following diagram
commutes.
((M• )f )0
ϕf
/ ((N• )f )0

 ϕfg 
((M• )fg )0 / ((N• )fg )0 .

Thus ∼ is a functor from the category of graded S• -modules to the category of

quasicoherent sheaves on Proj S• .

14.6.B. E ASY E XERCISE . Show that ∼ is an exact functor. (Hint: everything in the
construction is exact.)

14.6.C. E XERCISE . Show that if M• and M•′ agree in high enough degrees, then
M
g• = ∼M
g′ . Then show that the map from graded S• -modules (up to isomorphism)
•
to quasicoherent sheaves on Proj S• (up to isomorphism) is not a bijection. (Really:
show this is not an equivalence of categories.)

14.6.D. E XERCISE . Describe a map of S0 -modules M0 → Γ (Proj S• , M

g• ). (This
foreshadows the “saturation map” of §15.7.5 that takes a graded module to its
saturation, see Exercise 15.7.C.)

14.6.1. Exercise 14.6.C shows that ∼ isn’t an isomorphism (or equivalence) of

categories, but it is close. The relationship is akin to that between presheaves
and sheaves, and the sheafification functor (see §15.7). One convenient way of
understanding this is by starting with the category f.g.ModS• of finitely generated
S• -modules, and creating a new category C with the same objects, but different
morphisms: the morphisms from M• to N• in this new category are
 
HomC (M• , N• ) := lim Homf.g.ModS (M•≥m , N•≥m ) .
m→∞ •

(Here M•≥m = ⊕n≥m M• as you might expect.) Then you can show that C is
equivalent to the category of finite type quasicoherent sheaves on Proj S• , and this
equivalence corresponds to the map M• 7→ M g• . You are welcome to prove these
facts, but we will not use them.

14.6.2. Example: Homogeneous ideals of S• give closed subschemes of Proj S• .

Recall that a homogeneous ideal I• ⊂ S• yields a closed subscheme Proj S• /I• ,→
∼ P3 . The ideal I• =
Proj S• . For example, suppose S• = k[w, x, y, z], so Proj S• =
2 2
(wz − xy, x − wy, y − xz) yields our old friend, the twisted cubic (defined in
Exercise 9.3.A)
424 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

14.6.E. E XERCISE . Show that if the functor ∼ is applied to the exact sequence of
graded S• -modules
0 / I• / S• / S• /I• /0

we obtain the closed subscheme exact sequence (9.1.2.1) for Proj S• /I• ,→ Proj S• .
All closed subschemes of Proj S• arise in this way (see Exercise 15.7.I).

14.6.3. Remark. If M• is finitely generated (resp., finitely presented, coherent),

then so is M
g• , [Roh, (4.4.11)]. We will not need this fact. (Quick proof of the first
two: if M• is finitely generated, resp., finitely presented, so is the quasicoherent
sheaf M
g• on the affine cone Spec M• . These two notions are preserved by pullback,
and the standard open subsets of Proj S• are closed subschemes of the affine cone
Spec S• .)
 CHAPTER 15

Line bundles, maps to projective space, and divisors

When learning algebraic geometry for the first time, it is surprising to hear that
line bundles (invertible sheaves) are of such vital importance. i) They are, of course,
the simplest vector bundles. ii) We can often classify them, using “codimension 1”
information. In particular, we can build them and manipulate them. iii) Knowing
line bundles on a variety X helps us classify X, and understand maps to projective
space, and to other varieties. iv) And this is only the beginning.
In this chapter, we next describe convenient and powerful ways of working
with and classifying line bundles. We begin with a fundamental example, the
line bundles O(m) on projective space, §15.1. We then discuss how line bundles
and some sections give maps to projective space, and indeed how all maps to
projective schemes arise in this way, §15.2. This is simultaneously elementary,
subtle, and powerful. In §15.3, we describe a concretization of the idea that “one-
parameter families to projective space have unique limits”. We then come to the
most challenging part of the the chapter in §15.4. We introduce Weil divisors, and
use them to determine all the line bundles on X in a number of circumstances. Our
hard work leads to many fun examples we can work out by hand in §15.5. We then
discuss a different means of easily describing some line bundles as sheaves of ideals
that happen to be invertible (effective Cartier divisors, §15.6). We conclude with a
brief section once again connecting quasicoherent sheaves on projective schemes to
graded modules, §15.7.

15.1 Some line bundles on projective space

We now describe an important family of invertible sheaves on projective space
over a field k.
As a warm-up, we begin with the invertible sheaf OP1k (1) on P1k = Proj k[x0 , x1 ].
The subscript P1k refers to the space on which the sheaf lives, and is often omitted
when it is clear from the context. We describe the invertible sheaf O(1) using
transition functions. It is trivial on the usual affine open sets U0 = D(x0 ) =
Spec k[x1/0 ] and U1 = D(x1 ) = Spec k[x0/1 ]. (We continue to use the convention
xi/j for describing coordinates on patches of projective space, see §4.4.9.) Thus the
data of a section over U0 is a polynomial in x1/0 . The transition function from U0
to U1 is multiplication by x0/1 = x−11/0 . The transition function from U1 to U0 is
hence multiplication by x1/0 = x−1
0/1 .

This material is published by Princeton University Press as The Rising Sea: Foundations of Algebraic
Geometry by Ravi Vakil. This version is free to view and download for personal use only. Not for
redistribution, resale, or use in derivative works. © 2024 Ravi Vakil.
425
426 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

This information is summarized below:

open cover U0 = Spec k[x1/0 ] U1 = Spec k[x0/1 ]

×x0/1 =x−1
1/0
-
trivialization and transition functions k[x1/0 ] m k[x0/1 ]
×x1/0 =x−1
0/1

To test our understanding, let’s compute the global sections of O(1). This will
∼ k (Example 4.4.6). A global
generalize our hands-on calculation that Γ (P1k , OP1k ) =
section is a polynomial f(x1/0 ) ∈ k[x1/0 ] and a polynomial g(x0/1 ) ∈ k[x0/1 ] such
that f(1/x0/1 )x0/1 = g(x0/1 ). A little thought will show that f must be linear:
f(x1/0 ) = ax1/0 + b, and hence g(x0/1 ) = a + bx0/1 . Thus
dim Γ (P1k , O(1)) = 2 ̸= 1 = dim Γ (P1k , O).
Thus O(1) is not isomorphic to O, and we have constructed our first (proved)
example of a nontrivial line bundle!
We next define more generally OP1k (m) on P1k . It is defined in the same way,
except that the transition functions are the mth powers of those for O(1).

open cover U0 = Spec k[x1/0 ] U1 = Spec k[x0/1 ]

×xm −m
0/1 =x1/0
-
trivialization and transition functions k[x1/0 ] m k[x0/1 ]
×xm −m
1/0 =x0/1

In particular, thanks to the explicit transition functions, we see that O(m) = O(1)⊗m
(with the obvious meaning if m is negative: (O(1)⊗(−m) )∨ ). Clearly also O(m) ⊗
O(m ′ ) = O(m + m ′ ).

15.1.A. I MPORTANT E XERCISE . Show that dim Γ (P1 , O(m)) = m + 1 if m ≥ 0, and

0 otherwise.

15.1.1. Example. Long ago (in Exercise 2.6.K(a)), we warned that sheafification
was necessary when tensoring OX -modules: if F and G are two OX -modules on a
∼ (F ⊗ G )(X).
ringed space, then it is not necessarily true that F (X) ⊗OX (X) G (X) =
We now have an example: let X = Pk , F = O(1), G = O(−1), and use the fact that
1

O(−1) has no nonzero global sections.

̸ ∼ O(m ′ ). Hence conclude that

15.1.B. E XERCISE . Show that if m ̸= m ′ , then O(m) =
we have an injection of groups Z ,→ Pic Pk given by m 7→ O(m).
1

It is useful to identify the global sections of O(m) with the homogeneous

polynomials of degree m in x0 and x1 , i.e., with the degree m part of k[x0 , x1 ]
(cf. §15.1.2 for the generalization to Pn ). Can you see this from your solution to
Exercise 15.1.A? We will see that this identification is natural in many ways. For
example, you can show that the definition of O(m) doesn’t depend on a choice of
affine cover, and this polynomial description is also independent of cover. (For this,
see Example 4.5.13; you can later compare this to Exercise 25.1.N.) As an immediate
check of the usefulness of this point of view, ask yourself: where does the section
 © 2024 Ravi Vakil. Published by Princeton University Press. 427

x30 − x0 x21 of O(3) vanish? The section x0 + x1 of O(1) can be multiplied by the
section x20 of O(2) to get a section of O(3). Which one? Where does the rational
section x40 (x1 + x0 )/x71 of O(−2) have zeros and poles, and to what order? (We saw
the notion of zeros and poles in Definition 13.5.7, and will meet them again in §15.4,
but you should intuitively answer these questions already.)
We now define the invertible sheaf OPnk (m) on the projective space Pn k . On the
usual affine open set Ui = Spec k[x0/i , . . . , xn/i ]/(xi/i − 1) = Spec Ai , it is trivial,
so sections (as an Ai -module) are isomorphic to Ai . The transition function from
−m
Ui to Uj is multiplication by xm i/j = xj/i .

Ui = Spec k[x0/i , . . . , xn/i ]/(xi/i − 1) Uj = Spec k[x0/j , . . . , xn/j ]/(xj/j − 1)

×xm −m
i/j =xj/i
.
k[x0/i , . . . , xn/i ]/(xi/i − 1) n k[x0/j , . . . , xn/j ]/(xj/j − 1)
×xm −m
j/i =xi/j

Note that these transition functions clearly satisfy the cocycle condition.

15.1.C. E SSENTIAL EXERCISE
( CF. E XERCISE 9.3.J, T HEOREM 18.1.2). Show that
k , OPk (m)) =
n+m
dimk Γ (Pn n
n .

15.1.2. As in the case of P1 , sections of O(m) on Pn k are naturally identified with

homogeneous degree m polynomials in our n + 1 variables. (The connection
to homogeneous polynomials will be put on a more precise footing in the next
Exercise.) Thus x + y + 2z is a section of O(1) on P2 . It isn’t a function, but we know
where this section vanishes — precisely
where x + y + 2z = 0.
Also, notice that for fixed n, n+m n is a polynomial in the variable m, of degree
n, but only when m ≥ 0 (or better: when m ≥ −n). This should be telling you
that this function “wants to be a polynomial,” but won’t succeed without assis-
tance. We will later define h0 (Pn k , O(m)) := dimk Γ (Pk , O(m)), and later still we
n

will define higher

P∞ cohomology groups, and we will define the Euler characteristic
χ(Pnk , O(m)) := i=0 (−1) h (Pk , O(m)) (cohomology will vanish in degree higher
i i n

than n). We will discover the moral that the Euler characteristic is better-behaved
than h0 , and so we should now suspect (and later prove, see Theorem 18.1.2,
Remark 18.3.5, and Exercise 18.3.C) that this polynomial is in fact the Euler char-
acteristic, and the reason that it agrees with h0 for m ≥ 0 is because all the other
cohomology groups should vanish.
We finally note that we can define O(m) on Pn A for any ring A: the above
definition applies without change, and Exercise 15.1.C immediately generalizes
O n+m


to show that Γ (Pn A , P n (m)) is a free A-module of rank
A n , with the same
interpretation in terms of homogeneous polynomials.

15.1.D. E XERCISE ( ENLIGHTENING EXAMPLE AND GOOD PRACTICE ). Suppose n >

0. Let π : An+1
A \ V(x0 , . . . , xn ) → Pn
A be the morphism (x0 , . . . , xn ) 7→ [x0 , . . . , xn ]
(described in Exercise 7.3.F and Example 7.3.12). For any m ∈ Z, describe an
isomorphism
∼
π∗ OPnA (m) o / O n+1
A \V(x0 ,...,xn ) .
A
428 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

This induces a map π∗ : Γ (Pn A , OPA (m)) → Γ (AA \ V(x0 , . . . , xn ), OAn+1 ). Show
n
n+1
A
(as in §4.4.1) that the pullback

/ Γ (An+1 \ V(x0 , . . . , xn ), O n+1 )

A , OAn+1 )
α : A[x0 , . . . , xn ] = Γ (An+1
A A A A

is an isomorphism, and explain how the composition

/ A[x0 , . . . , xn ]
A , OPA (m))
α−1 ◦ π∗ : Γ (Pn n

A , OPA (m)) as the homogeneous degree m polynomials in A[x0 , . . . , xn ].

interprets Γ (Pn n

15.1.3. These are the only line bundles on Pn k.

Suppose that k is a field. We will see in §15.4.14 that these O(m) are the only
invertible sheaves on Pnk . The next Exercise shows this when n = 1.

15.1.E. E XERCISE . Show that every invertible sheaf on P1k is of the form O(m) for
some m. Hint: use the classification of finitely generated modules over a principal
ideal domain (Remark 6.4.4) to show that all invertible sheaves on A1k are trivial (a
special case of Exercise 14.2.C). Reduce to determining possible transition functions
between the two open subsets in the standard cover of P1k .
Caution: there can exist invertible sheaves on P1A not of the form O(m). You
may later be able to think of examples. (Hints to find an example for when you
know more: what if Spec A is disconnected? Or if that leads to too silly an example,
what if Spec A has nontrivial invertible sheaves?)

15.1.4. Line bundles on projective A-schemes.

Suppose M• is a graded S• -module. Define the graded module M(m)• by
^• satisfies
M(m)d := Mm+d . Thus the quasicoherent sheaf M(m)

^• ) = ((M• )f )m
Γ (D+ (f), M(m)

where here the subscript means we take the mth graded piece. (These subscripts
are admittedly confusing!)

^ ∼
15.1.F. E XERCISE . If S• = A[x0 , . . . , xn ], so Proj S• = Pn A , show S(m)• = O(m)
using transition functions (cf. §15.1). (Recall from Exercise 15.1.D that the global
sections of O(m) should be identified with the homogeneous degree m polynomials
in x0 , . . . , xn . Can you see that in the context of this exercise?)

15.1.5. Definition. Motivated by this, if S• is a graded ring generated in degree 1,

we define OProj S (m) (or simply O(m), where S• is implicit) by S(m)
•
^• .

15.1.G. I MPORTANT E XERCISE . If S• is generated in degree 1, show that OProj S• (m)

is an invertible sheaf.
If S• is generated in degree 1, and F is a quasicoherent sheaf on Proj S• , define
F (m) := F ⊗ O(m). This is often called twisting F by O(m), or twisting F by
m. More generally, if L is an invertible sheaf, then F ⊗ L is often called twisting
F by L .
 © 2024 Ravi Vakil. Published by Princeton University Press. 429

15.1.H. E XERCISE . If S• is generated in degree 1, show that M g• (m) = ∼ M(m)

^• .
(Hereafter, we can be cavalier with the placement of the “dot” in such situations.)
∼ O(m1 ) ⊗
15.1.I. E XERCISE . If S• is generated in degree 1, show that O(m1 + m2 ) =
O(m2 ).

15.1.6. ⋆ Unimportant remark. Even if S• is not generated in degree 1, then (if

S• is finitely generated) by Exercise 7.4.G, Sd• is generated in degree 1 for some
d. In this case, we may define the invertible sheaves O(dm) for m ∈ Z. This
does not mean that we can’t define O(1); this depends on S• . For example, if
S• is the polynomial ring k[x, y] with the usual grading, except without linear
terms (so S• = k[x2 , xy, y2 , x3 , x2 y, xy2 , y3 ]), then S2• and S3• are both generated
in degree 1, meaning that we may define O(2) and O(3). There is good reason to
call their “difference” O(1). (This algebra may remind of you of the pinched plane,
Exercise 13.5.H.)

15.2 Line bundles and maps to projective space

Recall that affine space An had a short description as a representable functor:
maps (from X) to An are exactly the same data as a choice of n functions (on X),
Exercise 7.6.E(a). In this section, we accidentally re-interpret projective space in the
same way.
Suppose x0 , x1 , and x2 are projective coordinates on P2k . Do you see how
(15.2.0.1) [x30 , x20 x2 , x0 x1 x2 + x31 , x32 ]
gives a morphism P2k → P3k ? The entries of (15.2.0.1) aren’t functions, so we can’t
make sense of them individually. But for any “point of P2k ”, their ratios can make
sense nonetheless. For example, the point [1, 1, 1] is sent to [1, 1, 2, 1], and the same
point [2, 2, 2] is sent to [8, 8, 16, 8], so there is no contradiction — the entries of
(15.2.0.1) are all homogeneous of degree 3, so the quotient of one by another is a
rational function. On the other hand,
(15.2.0.2) [x30 , x20 x2 , x0 x1 x2 + x31 , x31 ]
doesn’t give a morphism P2k → P3k , because the point [0, 0, 1] “doesn’t know where
to go”. But you might be able to interpret (15.2.0.2) as giving a morphism P2k \
{[0, 0, 1]} → P3k , or a rational map P2k 99K P3k . This may even remind you of
Exercise 7.3.O. We now make this precise, and state it much more generally. In
particular, we won’t need to work over a field.

15.2.A. E ASY E XERCISE ( A VITALLY IMPORTANT CONSTRUCTION ). Suppose s0 ,

. . . , sn are n + 1 global sections of an invertible sheaf L on a scheme X, with no
common zero. Define a corresponding map to Pn :
[s0 ,...,sn ]
X / Pn
Hint: If U is an open subset on which L is trivial, choose a trivialization, then
translate the si into functions using this trivialization, and use Exercise 7.3.O(a) to
obtain a morphism U → Pn . Then show that all of these maps (for different U and
different trivializations) “agree”, using Exercise 7.3.O(b).
430 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

(In Theorem 15.2.2, we will see that this yields all maps to projective space.)
Note that this exercise as written “works over Z” (as all morphisms are “over” the
final object in the category of schemes), although many readers will just work over
a field k.

15.2.1. Definitions. If L is an invertible sheaf on X, then those points where all

global sections of L vanish are called the base points of L , and the set of base
points is called the base locus of L ; it is a closed subset of X. (We can refine
this to a closed subscheme: by taking the scheme-theoretic intersection of the
vanishing loci of the sections of L , we obtain the scheme-theoretic base locus.)
The complement of the base locus is the base-point-free locus. If L has no base
points, it is base-point-free. The base-point-free locus is an open subset of X
More generally, if we have a bunch of global sections of L (perhaps not all of
them), we can define the base points (and all the related notions, of the previous
paragraph) of these sections. Thus for example Exercise 15.2.A states that if s0 , . . . ,
sn are a base-point-free family of global sections of an invertible sheaf L on X, then
they define a map to Pn .
A linear series (or linear system) on a k-scheme X is a k-vector space V (usually
finite-dimensional), an invertible sheaf L , and a linear map λ : V → Γ (X, L ). Such
a linear series is often called “V”, with the rest of the data left implicit. If the map λ
is an isomorphism, it is called a complete linear series, and is often written |L |.
The language of base points (base-point-free, base locus, scheme-theoretic base
locus, ...) readily applies to linear series (base-point-free linear series, ...).
As a reality check, Exercise 15.2.A states that an (n + 1)-dimensional linear
series on a k-scheme X, with choice of basis, with base-point-free locus U, defines a
morphism U → Pn k.

15.2.B. E ASY E XERCISE . If L and M are base-point-free invertible sheaves, show

that L ⊗ M is as well.

15.2.C. E XERCISE . Suppose we have a morphism π : X → Pn k , corresponding to

the base-point-free linear series Γ (Pnk , O(1)) → Γ (X, L ) (so L = π∗ O(1)). If the
n
scheme-theoretic image of X in Pk lies in a hyperplane, we say that the linear series
(or X itself) is degenerate (and otherwise, nondegenerate). Show that a base-point-
free linear series V → Γ (X, L ) is nondegenerate if and only if the map V → Γ (X, L )
is an inclusion. In particular, a complete linear series is always nondegenerate.
Theorem 15.2.2, the converse or completion to Exercise 15.2.A, will give one
reason why line bundles are crucially important: they tell us about maps to projec-
tive space, and more generally, to quasiprojective A-schemes. Given that we have
had a hard time naming any non-quasiprojective schemes, they tell us about maps
to essentially all interesting schemes.

15.2.2. Important Theorem. — For a fixed scheme X, morphisms X → Pn are in bijection

with the data (L , s0 , . . . , sn ), where L is an invertible sheaf and s0 , . . . , sn are sections
of L with no common zeros (i.e., (s0 , . . . , sn ) is base-point-free), up to isomorphism of
these data.
(This works over Z, or indeed any base.) Informally: morphisms to Pn corre-
spond to n + 1 sections of a line bundle, not all vanishing at any point, modulo
 © 2024 Ravi Vakil. Published by Princeton University Press. 431

global sections of OX∗ , as multiplication by an invertible function gives an automor-

phism of L . This is one of those important theorems in algebraic geometry that is
easy to prove, but quite subtle in its effect on how one should think. It takes some
time to properly digest. A “coordinate-free” version is given in Exercise 15.2.G.

15.2.3. Theorem 15.2.2 describes all morphisms to projective space, and hence by
the Yoneda philosophy, this can be taken as the definition of projective space: it
defines projective space up to unique isomorphism. Projective space Pn (over Z)
is the moduli space of line bundles L along with n + 1 sections of L with no common
zeros. (Can you give an analogous definition of projective space over X, denoted
PnX ?) Or if you prefer, maps X → Pk of k-schemes correspond to (n + 1)-dimensional
n

base-point-free linear series on X, along with choice of basis. (How would you rephrase
this if k is replaced by another ring?)
Every time you see a map to projective space, you should immediately simulta-
neously keep in mind the invertible sheaf and sections (or, if you prefer, the linear
series).
Maps to projective schemes can be described similarly. For example, if Y ,→ P2k
is the curve x22 x0 = x31 − x1 x20 , then maps from a scheme X to Y are given by an
invertible sheaf on X along with three sections s0 , s1 , s2 , with no common zeros,
satisfying s22 s0 − s31 + s1 s20 = 0.

15.2.4. The central idea of the proof of Theorem 15.2.2 is straightforward: we need
to show a bijection, so we just figure out what the maps are from left to right, and
from right to left, and show that they are indeed inverse. The key will be thinking
about transition functions between the trivializing neighborhoods.
Here more precisely is the correspondence of Theorem 15.2.2. Any n+1 sections
of L with no common zeros determine a morphism to Pn , by Exercise 15.2.A.
Conversely, if you have a map to projective space π : X → Pn , then we have n + 1
sections of OPn (1), corresponding to the hyperplane sections, x0 , . . . , xn . Then π∗ x0 ,
. . . , π∗ xn are sections of π∗ OPn (1), and they have no common zero. (Reminder:
it is helpful to think of pulling back invertible sheaves in terms of pulling back
transition functions.)
So to prove Theorem 15.2.2, we just need to show that these two constructions
compose to give the identity in either direction.
Proof of Important Theorem 15.2.2. Suppose we are given n + 1 sections s0 , . . . , sn of
an invertible sheaf L , with no common zeros, which (via Exercise 15.2.A) induce
a morphism π : X → Pn . For each si , we get a trivialization of L on the open set
Xsi where si doesn’t vanish. The transition functions for L are precisely si /sj on
Xsi ∩ Xsj . As O(1) is trivial on the standard affine open sets D+ (xi ) of Pn , π∗ (O(1))
is trivial on Xsi = π−1 (D+ (xi )). Moreover, si /sj = π∗ (xi /xj ) (directly from the
∼
construction of π in Exercise 15.2.A). This gives an isomorphism L ←→ π∗ O(1) —
the two invertible sheaves have the same transition functions.

15.2.D. E XERCISE . Show that this isomorphism can be chosen so that

∼
(L , s0 , . . . , sn ) o / (π∗ O(1), π∗ x0 , . . . , π∗ xn ),

thereby completing one of the two implications of the theorem.

432 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

For the other direction, suppose we are given a map π : X → Pn . Let si =

π xi ∈ Γ (X, π∗ (O(1))). As the xi ’s have no common zeros on Pn , the si ’s have
∗

no common zeros on X. The map [s0 , . . . , sn ] is the same as the map π. We see
this as follows. The preimage of D(xi ) (by the morphism [s0 , . . . , sn ]) is D(si ) =
D(π∗ xi ) = π−1 D(xi ), so “the right open sets go to the right open sets”. To show the
two morphisms D(si ) → D(xi ) (induced from (s1 , . . . , sn ) and π) are the same, we
use the fact that maps to an affine scheme D(xi ) are determined by their maps of
global sections in the opposite direction (Essential Exercise 7.3.G). Both morphisms
D(si ) → D(xi ) corresponds to the ring map π♯ : xj/i = xj /xi 7→ sj /si . □

15.2.5. Remark: Extending Theorem 15.2.2 to rational maps. Suppose s0 , . . . , sn are

sections of an invertible sheaf L on a scheme X. Then Theorem 15.2.2 yields a
morphism X \ V(s0 , . . . , sn ) → Pn . In particular, if X is integral, and the si are not
all 0, these data yield a rational map X 99K Pn .

15.2.6. Examples and applications.

15.2.7. Example: the Veronese embedding is |OPnk (d)|. Consider the line bundle
OPnk (d) on Pn k . We have checked that the sections of this line bundle form a vector
space of dimension n+d


d , with a basis corresponding to homogeneous degree
d polynomials in the projective coordinates for Pn k (Exercises 15.1.C and 15.1.D).
Also, they have no common zeros (as for example the subset of sections xd d
0 , x1 ,
d
. . . , xn have no common zeros). Thus the complete linear series is base-point-free,
n+d
and determines a morphism νd : Pn → P( d )−1 . This is the Veronese embedding
(Definition 9.3.8). For example, if n = 2 and d = 2, we get a map P2k → P5k .
In §9.3.8, we saw that this is a closed embedding. The following is a more
general method of checking that maps to projective space are closed embeddings.

15.2.E. L ESS IMPORTANT EXERCISE . Suppose π : X → Pn A corresponds to an

invertible sheaf L on X, and sections s0 , . . . , sn . Show that π is a closed embedding
if and only if
(i) each open set Xsi is affine, and
(ii) for each i, the map of rings A[y0 , . . . , yn ] → Γ (Xsi , O) given by yj 7→ sj /si
is surjective.

15.2.8. Special case of Example 15.2.7. Recall that the image of the Veronese embed-
ding when n = 1 is called a rational normal curve of degree d (Exercise 9.3.I). Our map
is P1k → Pd
k given by [x, y] 7→ [x , x
d d−1
y, . . . , xyd−1 , yd ].

15.2.F. E XERCISE . Suppose we are given a map π : P1k → Pnk where the correspond-
ing invertible sheaf on P1k is O(d). (This can reasonably be called a degree d map,
cf. Exercises 18.4.E and 18.6.H.) Show that if d < n, then the image is degenerate
(defined in Exercise 15.2.C). Show that if d = n and the image is nondegenerate,
then the image is isomorphic (via an automorphism of projective space) to a rational
normal curve.

15.2.G. E XERCISE .
(a) Suppose X is a k-scheme (although this statement can readily generalize). Show
that a (finite-dimensional) base-point-free linear series V on X corresponding to L
 © 2024 Ravi Vakil. Published by Princeton University Press. 433

induces a morphism to projective space

ϕV : X / PV ∨ .

(This should be seen as a coordinate-free version of Theorem 15.2.2.)

(b) ⋆ Complete the sentence: “PV ∨ represents the functor...”. Explain the meaning
of the statements Γ (PV, O(1)) = V ∨ , and Γ (PV, O(2)) = Sym2 V ∨ .

15.2.H. E XERCISE . Define the graded rings R• = k[u, v, w]/(uw − v2 ) and

S• = k[x, y] (with all variables having degree 1). By Exercise 9.3.H, we have
∼
an isomorphism Proj R• ←→ Proj S• (via the Veronese embedding ν2 ). Show that
this isomorphism is not induced by a map of graded rings S• → R• .

15.2.9. Remark. You may be able to show that after “regrading”, the isomorphism
∼
Proj R• ←→ Proj S• does arise from a map of graded rings (S2• → R• ). Exer-
cise 19.11.B will give an example where it is not possible to patch the lack of maps
of graded rings by just regrading.

15.2.I. E XERCISE : AN EARLY LOOK AT INTERSECTION THEORY, RELATED TO B ÉZOUT ’ S

T HEOREM . A classical definition of the degree of a curve in projective space is
as follows: intersect it with a “general” hyperplane, and count the number of
points of intersection, with appropriate multiplicity. We interpret this in the case of
π : P1k → Pn n 1
k . Show that there is a hyperplane H of Pk not containing π(Pk ). Equiv-
alently, π H ∈ Γ (P , OP1 (d)) (interpreted as the pullback of a section of OPn (1))
∗ 1

is not 0. Show that the number of zeros of π∗ H is precisely d. (You will have to
define “appropriate multiplicity”. What does it mean geometrically if π is a closed
embedding, and π∗ H has a double zero? Aside: Can you make sense of this even if
π is not a closed embedding?) Thus this classical notion of degree agrees with the
notion of degree in Exercise 15.2.F. (See Exercise 9.3.D for another case of Bézout’s
Theorem. Here we intersect a degree d curve with a degree 1 hyperplane; there we
intersect a degree 1 curve with a degree d hypersurface. Exercise 18.6.J will give a
common generalization.)

15.2.10. Example: The Segre embedding revisited. The Segre embedding can also be
interpreted in this way. This is a useful excuse to define some notation. Suppose
F is a quasicoherent sheaf on a Z-scheme X, and G is a quasicoherent sheaf on
a Z-scheme Y. Let πX , πY be the projections from X ×Z Y to X and Y respectively.
Then F ⊠ G (pronounced “F box-times G ”) is defined to be π∗X F ⊗ π∗Y G . In
particular, OPm ×Pn (a, b) is defined to be OPm (a) ⊠ OPn (b) (over any base Z). The
Segre embedding Pm × Pn → Pmn+m+n corresponds to the complete linear series
for the invertible sheaf O(1, 1).
When we first saw the Segre embedding in §10.6, we saw (in different language)
that this complete linear series is base-point-free. We also checked by hand (§10.6.1)
that it is a closed embedding, essentially by Exercise 15.2.E.
Recall that if L and M are both base-point-free invertible sheaves on a scheme
X, then L ⊗ M is also base-point-free (Exercise 15.2.B). We may interpret this fact
using the Segre embedding (under reasonable hypotheses on X). If ϕL : X → PM
is a morphism corresponding to a (base-point-free) linear series based on L , and
ϕM : X → PN is a morphism corresponding to a linear series on M , then the
434 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Segre embedding yields a morphism X → PM × PN → P(M+1)(N+1)−1 , which

corresponds to a base-point-free series of sections of L ⊗ M .

15.2.11. ⋆⋆ A proper nonprojective k-variety — and gluing schemes along closed

subschemes.
(The rest of this section should be skipped on a first reading.) We conclude by
using what we have developed to describe an example of a scheme that is proper
but not projective (promised in Remark 11.4.6). We use a construction that looks
so fundamental that you may be surprised to find that we won’t use it in any
meaningful way later.
∼ P3 , Zi be a line in Xi ,
Fix an algebraically closed field k. For i = 1 and 2, let Xi = k
and Zi be a regular conic in Xi disjoint from Zi (both Zi and Zi′ isomorphic to P1k ).
′

The construction of §15.2.12 will allow us to glue X1 to X2 so that Z1 is identified

with Z2′ and Z1′ is identified with Z2 , see Figure 15.1. (You will be able to make this
precise after reading §15.2.12.) The result, call it X, is proper, by Exercise 15.2.M.

F IGURE 15.1. Building a proper nonprojective variety

Then X is not projective. For if it were, then it would be embedded in projective

space by some invertible sheaf L . If X is embedded, then X1 is too, so L must
restrict to an invertible sheaf on X1 of the form OX1 (n1 ), where n1 > 0. You can
check that the restriction of L to Z1 is OZ1 (n1 ), and the restriction of L to Z1′ is
OZ1′ (2n1 ). Symmetrically, the restriction of L to Z2 is OZ2 (n2 ) for some n2 > 0,
and the restriction of L to Z2′ is OZ2′ (2n2 ). But after gluing, Z1 = Z2′ , and Z1′ = Z2 ,
so we have n1 = 2n2 and 2n1 = n2 , which is impossible.

15.2.12. Gluing two schemes together along isomorphic closed subschemes.

It is straightforward to show that you can glue two schemes along isomorphic
open subschemes. (More precisely, if X1 and X2 are schemes, with open subschemes
∼
U1 and U2 respectively, and an isomorphism U1 ←→ U2 , you can make sense
∼
of gluing X1 and X2 along U1 ←→ U2 . You should think this through.) You
can similarly glue two schemes along isomorphic closed subschemes. We now
make this precise. Suppose Z1 ,→ X1 and Z2 ,→ X2 are closed embeddings, and
∼ / Z2 is an isomorphism. We will explain how to glue X1 to X2 along ϕ.
ϕ : Z1
`
The result will be called X1 ϕ X2 .

15.2.13. Motivating example. Our motivating example is if Xi = Spec Ai and

∼ / A1 /I1 . Then the result
Zi = Spec Ai /Ii , and ϕ corresponds to ϕ♯ : A2 /I2
will be Spec R, where R is the ring of consisting of ordered pairs (a1 , a2 ) ∈ A1 × A2
 © 2024 Ravi Vakil. Published by Princeton University Press. 435

that “agree via ϕ”. More precisely, this is a fibered product of rings:
R := A1 ×ϕ♯ : A1 /I1 →A2 /I2 A2 .

15.2.14. The general construction, as a locally ringed space. In our general situation,
we might wish to cover X1 and X2 by open charts of this form. We would then
have
`to worry about gluing and choices, so to avoid this, we instead first construct
X1 ϕ X2 as a locally ringed space. As a topological space, the definition is clear:
we glue the underlying sets together ∼
` along the underlying sets of Z1 = Z2 , and
topologize it so that a subset of X1 ϕ X2 is open if and only if its restrictions to X1
and X2 are`both open. For convenience, let Z be the image of Z1 (or equivalently
Z2 ) in X1 ϕ X2 . We next define the stalk of the structure sheaf at any point
` `
p ∈ X1 ϕ X2 . If p ∈ Xi \ Z = (X1 ϕ X2 ) \ X3−i (hopefully the meaning of this is
clear), we define the stalk as OXi ,p . If p ∈ X1 ∩ X2 , we define the stalk to consist of
elements (s1 , s2 ) ∈ OX1 ,p × OX2 ,p which agree in OZ1 ,p = ∼ OZ ,p . The meaning of
2
everything in this paragraph will be clear to you if you can do the following.

15.2.J. E XERCISE . Define the structure sheaf of OX1 `ϕ X2 in terms of compatible

germs. (What should it mean for germs to be compatible? Hint: for z ∈ Z, suppose
we have open subsets U1 of X1 and U2 of X2 , with ` U1 ∩ Z = U2 ∩ Z, so U1 and
U2 glue together to give an open subset U of X1 ϕ X2 . Suppose we also have
functions f1 on U1 and f2 on U2 that “agree on U ∩ Z” — what does that mean?
Then we declare that the germs of the “function on U obtained by gluing together
f1 and f2 ” are compatible.) Show that the resulting ringed space is a locally ringed
space.
`
We next want to show that the locally ringed space X1 ϕ X2 is a scheme.
Clearly it is a scheme away from Z. We first verify a special case.

15.2.K. E XERCISE . Show that in Example 15.2.13 the construction of §15.2.14 indeed
yields Spec(A1 ×ϕ♯ A2 ).

15.2.L. E XERCISE . In the general case, suppose x ∈ Z. Show that there is an

affine open subset Spec Ai ⊂ X`i such that Z ∩ Spec A1 = Z ∩ Spec A2 . Then use
Exercise 15.2.J to show that X1 ϕ X2 is a scheme in an open neighborhood of x,
and thus a scheme.

15.2.15. Remarks.
(a) As the notation suggests, this is a fibered coproduct in the category of schemes,
and indeed in the category of locally ringed spaces. We won’t need this fact, but
you can prove it if you wish; it isn’t hard. Unlike the situation for products, fibered
coproducts don’t exist in general in the category of schemes. Miraculously (and
for reasons that are specific to schemes), the resulting cofibered diagram is also a
fibered (i.e., Cartesian) diagram. This has pleasant ramifications. For example, this
construction “behaves well with respect to” (or “commutes with”) flat base change.
(b) You might hope that if you have a single scheme X with two disjoint closed
∼
subschemes W ′ and W ′′ , and an isomorphism W ′ −→ W ′′ , then you should be
able to glue X to itself along W → W . This construction doesn’t work, and indeed
′ ′′

it may not be possible. You can still make sense of the quotient as an algebraic space,
which we will not define here.
436 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

15.2.M. E XERCISE . We continue to use the notation Xi , ϕ, etc. Suppose we are

working in the category of A-schemes. `
(a) If X1 and X2 are universally closed, show that X1 ϕ X2 is as well.
`
(b) If X1 and X2 are separated, show that X1 ϕ X2 is as well.
`
(c) If X1 and X2 are finite type over a Noetherian ring A, show that X1 ϕ X2 is
as well. (Hint: Reduce to the “affine” case of the Motivating Example 15.2.13.
Choose generators x1 , . . . , xn of A1 , and y1 , . . . , yn , such that xi modulo I1
agrees with yi modulo I2 via ϕ. Choose generators g1 , . . . , gm of I2 — here
use Noetherianness of A. Show that (xi , yi ) and (0, gi ) generate R ⊂ A1 × A2 ,
as follows. Suppose (a1 , a2 ) ∈ R. Then there is some polynomial m such that
′
a1 = m(x1 , . . . , xn ). Hence (a1 , a2 ) − m((x Pm 1 , y1 ), . . . , (xn , yn )) = (0, a2 ) for some
′ ′
a
P2m 2∈ I . Then a 2 can be written as ℓ
i=1 i 1(y , . . . , y )g
n i . But then (0, a2′ ) =
i=1 ℓi ((x1 , y1 ), . . . , (xn , yn ))(0, gi ).)
`
Thus if X1 and X2 are proper, so is X1 ϕ X2 .
` other hand, we have just seen (§15.2.11) that if X1 and X2 are projective,
On the
then X1 ϕ X2 needn’t be projective.

15.2.16. Fun with more general “gluing” (coproducts) of schemes.

If you have read this far, you can see more interesting examples quite cheaply.
Suppose more generally that we have a closed embedding Z ,→ Y, and only an
affine morphism Z → X. We can still “glue” X to Y along Z to get a scheme W, by
following the same recipe.

15.2.N. E XTENDED E XERCISE . Given a closed embedding α : Z ,→ X and an affine

morphism β : Z → Y, show (following the strategy of §15.2.12) that there is a
pushout diagram of schemes

(15.2.16.1) Z _
β affine
/Y
α closed α′
 β′ 
X / W.

This diagram is necessarily unique up to unique isomorphism.

15.2.O. E XERCISE . Show that (15.2.16.1) is also a pullback diagram (i.e., a Cartesian
diagram), and that α ′ is a closed embedding, and β ′ is an affine morphism.

15.2.17. A scheme with no closed points.

In the course of solving Extended Exercise 15.2.N, you will investigate the
(much easier) affine case, which corresponds to the following diagram of rings.

β♯
(15.2.17.1) A/I o BO
OO

Ao A ×A/I B.

Thus the ring corresponding to W can be interpreted as

A ×A/I B = {(a, b) ∈ (A, B) : a ≡ α♯ (b) (mod I)}.

 © 2024 Ravi Vakil. Published by Princeton University Press. 437

We can use this construction to see some interesting behavior. For example,
we now describe a scheme X with no closed points. It will have countably many
points {p1 , p2 , . . . }, and its open sets will be of the form {p1 , . . . , pn }. (Clearly this
topological space has no closed points.) Each open set will be affine: {p1 , . . . , pn } =
Spec An . We inductively build An as follows. Let A1 = k(x1 , x2 , . . . ) (a countably
generated transcendental extension of a field k). We inductively build An+1 with
the following special case of (15.2.17.1):
k(xn , xn+1 , . . . ) o ? _ k(xn+1 , xn+2 , . . . )[xn ]xn
OO OO

An o ? _ An+1 .

The top row is just a discrete valuation ring mapping into its fraction field K. The
left column is a complicated local ring whose residue field is that same field K.
This may be easier to understand when focusing on {p1 , p2 , p3 }. Let F =
k(x3 , x4 , . . . ). Then A1 is a field, A2 is the discrete valuation ring F(x2 )[x1 ]x1 , and
A3 is the subring of A2 whose residue modulo x1 is in the discrete valuation ring
F[x2 ]x2 . (The ring A3 is an example of a rank 2 valuation ring; can you see why it
deserves this name?)

15.2.P. E XERCISE . Use the above to describe a scheme ∪n≥1 Spec An with no closed
points.

15.3 The Curve-to-Projective Extension Theorem

We now use the main theorem of the previous section, Theorem 15.2.2, to prove
something useful and concrete. (In fact, we could have proved this far earlier
— with a little cleverness you can replace the invocation of Theorem 15.2.2 by
Exercise 7.3.M, and prove this as soon as you know about discrete valuation rings.)
As motivation, we consider a variation of the example at the start of §15.2:
(15.2.0.2) was a failed attempt at a morphism P2k → P3k . There was a reason the
example used P2k rather than P1k as the domain. For example, at first it appears as
though the rational map P1k 99K P3k given by
[x7 , x3 y4 , x4 y3 + 4x5 y2 , x5 y2 + x7 ]
doesn’t give a morphism P1k → P3k , because it is not defined at [x, y] = [0, 1]. But
we can extend the map over [0, 1], using
(15.3.0.1) [x7 , x3 y4 , x4 y3 + 4x5 y2 , x5 y2 + x7 ] = [x4 , y4 , xy3 + 4x2 y2 , x2 y2 + x4 ]
— we can just divide through by x3 to make the morphism make sense. Similarly, it
appears as though [x2 , x−2 y4 , x−1 y3 + 4y2 , y2 + x2 ] also is not defined at [x, y] =
[0, 1], but you may see what to do (and why it can be extended to the same map as
(15.3.0.1)). This will generalize to the following.

15.3.1. The Curve-to-Projective Extension Theorem. — Suppose C is a pure di-

mension 1 Noetherian scheme over an affine base S = Spec A, and p ∈ C is a regular
closed point of it. Suppose Y is a projective S-scheme. Then any morphism C \ {p} → Y (of
S-schemes) extends to all of C.
438 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

To preempt any confusion: C is “absolute pure dimension 1” not relative pure

dimension 1.
In practice, we will use this theorem when S = Spec k, and C is a k-variety.
The only reason we assume S is affine is because we won’t know the meaning of
“projective S-scheme” until we know what a projective morphism is (§17.3). But the
proof below extends immediately to general S once we know the meaning of the
statement.
Note that if such an extension exists, then it is unique: the nonreduced locus of
C is a closed subset (Exercise 6.6.N). Hence by replacing C by an open neighborhood
of p that is reduced, we can use the Reduced-to-Separated Theorem 11.3.2 that
maps from reduced schemes to separated schemes are determined by their behavior
on a dense open set. Alternatively, maps to a separated scheme can be extended
over an effective Cartier divisor in at most one way (Exercise 11.3.G).
The following exercise shows that the hypotheses are necessary.

15.3.A. E XERCISE . In each of the following cases, prove that the morphism
C \ {p} → Y cannot be extended to a morphism C → Y.
(a) Projectivity of Y is necessary. Suppose C = A1k , p = 0, Y = A1k , and C \ {p} → Y is
given by “t 7→ 1/t”.
(b) One-dimensionality of C is necessary. Suppose C = A2k , p = (0, 0), Y = P1k , and
C \ {p} → Y is given by (x, y) 7→ [x, y].
(c) Non-singularity of C is necessary. Suppose C = Spec k[x, y]/(y2 − x3 ), p = 0,
Y = P1k , and C \ {p} → Y is given by (x, y) 7→ [x, y].
We remark that by combining this (easy) theorem with the (hard) valuative
criterion for properness (Theorem 13.7.6), one obtains a proof of the properness
of projective space bypassing the (tricky but not hard) Fundamental Theorem
of Elimination Theory 8.4.10 (see Exercise 13.7.F). Fancier remark: the valuative
criterion of properness can be used to show that Theorem 15.3.1 remains true if Y is
only required to be proper, but it requires some thought.

15.3.2. Central idea of proof. The central idea of the proof may be summarized
as “clear denominators”, and is illustrated by the following motivating example.
Suppose you have a morphism from A1 \ {0} to projective space, and you wanted
to extend it to A1 . Suppose the map was given by t 7→ [t4 + t−3 , t−2 + 4t]. Then
of course you would “clear the denominators”, and replace the map by t 7→
[t7 + 1, t + 4t4 ]. Similarly, if the map was given by t 7→ [t2 + t3 , t2 + t4 ], you
would divide by t2 , to obtain the map t 7→ [1 + t, 1 + t2 ].

15.3.3. Proof. Our plan is to maneuver ourselves into the situation where we can
apply the idea of §15.3.2. We begin with some quick reductions. The nonreduced
locus of C is closed (Exercise 6.6.N) and doesn’t contain p, so by replacing C by an
appropriate open neighborhood of p, we may assume that C is reduced and affine.
We next reduce to the case where Y = PnA . Choose a closed embedding Y → PA .
n

If the result holds for P , and we have a morphism C → P with C \ {p} mapping
n n

to Y, then C must map to Y as well. Reason: we can reduce to the case where
the source is an affine open subset, and the target is An n
A ⊂ PA (and hence affine).
n
Then the functions vanishing on Y ∩ AA pull back to functions that vanish at the
generic points of the irreducible components of C and hence vanish everywhere on
C (using reducedness of C), i.e., C maps to Y.
 © 2024 Ravi Vakil. Published by Princeton University Press. 439

Choose a uniformizer t ∈ m \ m2 in the local ring of C at p. As t is a function in

some neighborhood of p, we may assume that t is a function on C (by replacing
C by this neighborhood of p). Then V(t) contains p (as t ∈ m), but V(t) does not
contain the component of C containing p (as t ∈ / m2 ), so by replacing C by an affine
open neighborhood of p in (C \ V(t)) ∪ {p} — note that (C \ V(t)) ∪ {p} ⊂ C is open
because C (almost) has the cofinite topology — we may assume that p is the only
zero of the function t (and of course t vanishes at p with multiplicity 1).
We have a map C \ {p} → Pn A , which by Theorem 15.2.2 corresponds to a line
bundle L on C \ {p} and n + 1 sections of it with no common zeros in C \ {p}. Let
U be a nonempty open set of C \ {p} on which L = ∼ O. Then by replacing C by
U ∪ {p}, we interpret the map to P as n + 1 rational functions f0 , . . . , fn , defined
n

away away from p, with no common zeros away from p. Let N = mini (valp fi ).
Then t−N f0 , . . . , t−N fn are n + 1 functions with no common zeros. Thus they
determine a morphism C → Pn A extending C \ {p} → PA as desired.
n
□

15.3.B. E XERCISE ( USEFUL PRACTICE ). Suppose X is a Noetherian k-scheme, and

Z is an irreducible codimension 1 subvariety whose generic point is a regular
point of X (so the local ring OX,Z is a discrete valuation ring). Suppose π : X 99K Y
is a rational map to a projective k-scheme. Show that the domain of definition
of the rational map includes a dense open subset of Z. In other words, rational
maps from Noetherian k-schemes to projective k-schemes can be extended over
regular codimension 1 sets. (We have already seen this principle in action — see
Exercise 7.5.J on the Cremona transformation.)

15.3.4. ⋆ Extended example: tangents to regular plane curves are limits of secants.
We now discuss an extended example that may give insight on many different
things. When learning calculus, we’re sometimes taught that tangent lines to plane
curves are limits of secants. We can now see this in terms of the Curve-to-Projective
Extension Theorem 15.3.1, and deltas and epsilons are turned into easy algebra.
This discussion also foreshadows the importance of the diagonal morphism in
understanding differentials.
Let k be a field, which you should initially take to be algebraically closed. (You
can later think about how this makes sense over arbitrary fields.) Suppose C ⊂ A2k
is a plane curve, given by the equation f(x, y) = 0. Fix a point p = (x0 , y0 ) of C.
Given a point q = (x1 , y1 ) of C, with x1 ̸= x0 , the slope of the line joining them is
m = (y1 − y0 )/(x1 − x0 ), and the equation of the line joining them is
y1 − y0
(y − y0 ) = (x − x0 ).
x1 − x0
More generally (even if x1 = x0 , but still with y1 ̸= y0 ), the line joining them
has “slope” [x1 − x0 , y1 − y0 ] ∈ P1k , and the equation of the line joining them is
(x1 − x0 )(y − y0 ) = (y1 − y0 )(x − x0 ). We thus have a map

(15.3.4.1) C\p / P1
k

sending q to the “slope” of the secant line pq. The Curve-to-Projective Extension
Theorem 11.1.1 states that this will extend over p where p is a smooth point of C.
Let’s work out the algebra. Keep in mind that now x0 , y0 ∈ k, as p is a fixed point.
Define F(x1 , y1 ) := f(x1 , y1 ) − f(x0 , y0 ) ∈ k[x1 , y1 ].
440 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

15.3.C. E XERCISE . Show that F ∈ (x1 − x0 , y1 − y0 ).

Hence we can write F = g(x1 , y1 )(x1 − x0 ) + h(x1 , y1 )(y1 − y0 ) for some
(nonunique) choice of g, h ∈ k[x1 , y1 ]. Choose such a g and h.

15.3.D. E XERCISE . Show that if p = (x0 , y0 ) is a smooth point of C with nonvertical

tangent vector, then h(x0 , y0 ) ̸= 0.

15.3.E. E XERCISE . Show that if p = (x0 , y0 ) is a smooth point of C, then the slope of
the tangent line at p is [−h(x0 , y0 ), g(x0 , y0 )] ∈ P1k , and the equation of the tangent
line is
g(x0 , y0 )(x − x0 ) + h(x0 , y0 )(y − y0 ) = 0.

15.4 Hard but important: Line bundles and Weil divisors

The notion of Weil divisors gives a great way of understanding and classifying
line bundles, at least on Noetherian normal schemes. Some of what we discuss
will apply in more general circumstances, and the expert is invited to consider
generalizations by judiciously weakening hypotheses in various statements. Before
we get started, you should be warned: this is one of those topics in algebraic
geometry that is hard to digest — learning it changes the way in which you think
about line bundles. But once you become comfortable with the imperfect dictionary
to divisors, it becomes second nature.
For the rest of this section, we consider only Noetherian normal schemes. We
do this because we will use finite decomposition into irreducible components
(Exercise 5.3.B), and Algebraic Hartogs’s Lemma 13.5.19.
Define a Weil divisor as a formal Z-linear combination of codimension 1
irreducible closed subsets of X. In other words, a Weil divisor is defined to be an
object of the form
X
nY [Y]
Y ⊂ X codimension 1

where the nY are integers, all but a finite number of which are zero. Weil divisors
obviously form an abelian group, denoted Weil X. For example, if X is a curve, the
Weil divisors are linear combinations of closed points.
P
15.4.1. We say that [Y] is an irreducible (Weil) divisor. A Weil divisor D = nY [Y]
is said to be effective if nY ≥ 0 for all Y. In this case we say D ≥ 0, and by
D1 ≥ D2 we mean D1 − D2 ≥ 0. The support of a Weil divisor D, denoted Supp D,
is the subset ∪nY ̸=0PY. If U ⊂ P X is an open set, we define the restriction map
Weil X → Weil U by nY [Y] 7→ Y∩U̸=∅ nY [Y ∩ U].

15.4.A. E ASY EXERCISE . Show that the irreducible divisors on Pn k correspond

to irreducible homogeneous polynomials in k[x0 , . . . , xn ], up to multiplication by
nonzero scalars k× .

15.4.2. Suppose now that X is also regular in codimension 1 (R1 , Definition 13.5.13).
We add this hypothesis because we will use properties of discrete valuation rings.
Assume also that X is reduced. (This is only so we can talk about rational functions
 © 2024 Ravi Vakil. Published by Princeton University Press. 441

without worrying about them being defined at embedded points. Feel free to relax
this hypothesis.) Suppose that L is an invertible sheaf, and s a rational section not
vanishing everywhere on any irreducible component of X. (Rational sections are
given by a section over a dense open subset of X, with the obvious equivalence,
§14.2.1.)
We can make sense of the valuation of s along an irreducible Weil divisor
Y (denoted valY (s)) as follows. Take any open set U containing the generic point
of Y where L is trivializable, along with any trivialization over U; under this
trivialization, s is a nonzero rational function on U, which thus has a valuation
(§13.5.14). Any two such trivializations differ by an invertible function (transition
functions are invertible), so this valuation is well-defined. Note that valY (s) = 0 for
all but finitely many Y, by Exercise 13.5.G.
Thus s determines a Weil divisor
X
div(s) := valY (s)[Y]
Y

the summation runs over all irreducible divisors Y of X. We call div(s) the divisor
of zeros and poles of the rational section s (cf. Definition 13.5.7).

15.4.3. The important group of “line bundles with rational sections”. Now
consider the set {(L , s)} of pairs of line bundles L with rational sections s of L ,
not the zero section on any irreducible component of X, up to isomorphism. (An
isomorphism (L , s) = ∼ (L ′ , s ′ ) means an isomorphism of line bundles under which
′
s is sent to s .) This set (after taking quotient by isomorphism) forms an abelian
group under tensor product ⊗, with identity (OX , 1). (Tricky question: what is the
inverse of (L , s) in this group?)
It is important to notice that if t is an invertible function on X, then multiplica-
tion by t gives an isomorphism
∼
(L , s) ←→ (L , st).
Similarly, (L , s)/(L , u) = (O, s/u). Here s/u is a rational function.
The map div yields a group homomorphism

(15.4.3.1) div : {(L , s)}/isomorphism / Weil X.

15.4.B. E ASIER E XERCISE .

(a) (divisors of rational functions) Verify that on A1k , div(x3 /(x + 1)) = 3[(x)] − [(x + 1)]
(“= 3[0] − [−1]”).
(b) (divisor of rational sections of a nontrivial invertible sheaf) On P1k , there is a rational
section of O(1) “corresponding to” x2 /(x + y). Figure out what this means, and
calculate div(x2 /(x + y)).
The homomorphism (15.4.3.1) will be the key to determining all the line bun-
dles on many X. (Note that any invertible sheaf will have such a rational section.
For each irreducible component, take a nonempty open set not meeting any other
irreducible component; then shrink it so that L is trivial; choose a trivialization;
then take the union of all these open sets, and choose the section on this union cor-
responding to 1 under the trivialization.) We will see that in reasonable situations,
this map div will be injective, and often an isomorphism. Thus by forgetting the
442 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

rational section (i.e., taking an appropriate quotient), we will have described the
Picard group of all line bundles. Let’s put this strategy into action.

15.4.4. Proposition. — If X is normal and Noetherian then the map div is injective.

Proof. Suppose div(L , s) = 0. Then s has no poles. By Exercise 14.2.F, s is a regular

section. We now show that the morphism ×s : OX → L is in fact an isomorphism;
∼
this will prove the Proposition, as it will give an isomorphism (OX , 1) ←→ (L , s).
It suffices to show that ×s is an isomorphism on an open subset U of X where
L is trivial, as X is covered by trivializing open neighborhoods of L (as L is
∼
locally trivial). Choose an isomorphism i : L |U −→ OU . Composing ×s with i
yields a map ×s : OU → OU that is multiplication by a rational function s ′ = i(s)
′

that has no zeros and no poles. The rational function s ′ is regular because it has
no poles (Exercise 13.5.I), and 1/s ′ is regular for the same reason. Thus s ′ is an
invertible function on U, so ×s ′ is an isomorphism. Hence ×s is an isomorphism
over U. □
Motivated by this, we try to find an inverse to div, or at least to determine the
image of div.

15.4.5. Important Definition: OX (D). Assume now that X is normal — this will be a
standing assumption for the rest of this section.
Assume also that X is irreducible (purely to avoid making (15.4.5.1) look uglier
— but feel free to relax this, see Exercise 15.4.C). In case it helps: recall that a normal
scheme is a disjoint union of irreducible normal schemes (Exercise 5.4.B) — so this
assumption truly is harmless.
Suppose D is a Weil divisor. Define the sheaf OX (D) by

(15.4.5.1) Γ (U, OX (D)) := {t ∈ K(X)× : div |U t + D|U ≥ 0} ∪ {0}.

Here D|U is the restriction of D to U (defined in §15.4.1), and div |U t means the
divisor of t considered as a rational function on U (i.e., consider just the irreducible
divisors of U). The subscript X in OX (D) is omitted when it is clear from context.
The sections of OX (D) over U are the rational functions on U that have poles and
zeros “constrained by D”: a positive coefficient in D allows a pole of that order; a
negative coefficient demands a zero of that order. Away from the support of D, this
is (isomorphic to) the structure sheaf (by Algebraic Hartogs’s Lemma 13.5.19).

15.4.6. Important Remark. It will be crucial to note that OX (D) comes along with a
canonical nonzero “rational section” corresponding to 1 ∈ K(X)× . (It is a rational
section in the sense that it is a section over a dense open set, namely the complement
of Supp D.)

15.4.C. L ESS IMPORTANT EXERCISE . Generalize the definition of OX (D) to the case
when X is not necessarily irreducible. (This is just a question of language. Once
you have done this, feel free to drop this hypothesis in the rest of this section.)

15.4.D. E ASY E XERCISE . Verify that OX (D) is a quasicoherent sheaf. (Hint: the
“distinguished affine” criterion for quasicoherence of Exercise 6.2.D.)
 © 2024 Ravi Vakil. Published by Princeton University Press. 443

In good situations, OX (D) is an invertible sheaf. For example, let X = A1k .

Consider
OX (−2[(x)] + [(x − 1)] + [(x − 2)]) ,
often written O(−2[0]+[1]+[2]) for convenience. Then 3x3 /(x−1) is a global section;
it has the required two zeros at x = 0 (and even one to spare), and takes advantage
of the allowed pole at x = 1, and doesn’t have a pole at x = 2, even though one
is allowed. (Unimportant aside: the statement remains true in characteristic 2,
although the explanation requires editing.)

15.4.E. E ASY E XERCISE . (This is a consequence of later discussion as well, but you
should be able to do this by hand.)
(a) Show that any global section of OA1k (−2[(x)] + [(x − 1)] + [(x − 2)]) is a k[x]-
multiple of x2 /((x − 1)(x − 2)).
(b) Extend the argument of (a) to give an isomorphism
∼
OA1k (−2[(x)] + [(x − 1)] + [(x − 2)]) o / OA1 .
k

The next exercise is seemingly small, but is absolutely crucial to lesson confu-
sion as the complications add up in the next few pages.

15.4.F. E XERCISE . We continue the example of Easy Exercise 15.4.E, letting k = C

for concreteness. Let X = A1C , and D = −2[(x)] + [(x − 1)] + [(x − 2)].
(a) Consider the section of OX (D) corresponding to x2 (x−3)/((x−1)(x−2)) ∈ K(X)
(under the correspondence of (15.4.5.1)). This section vanishes at precisely one point.
What point is it?
(b) Consider the section of OX (D) corresponding to x2 (x − 1) ∈ K(X). This section
vanishes at precisely two points. What are those points, and to what orders does it
vanish at those points?
(c) Consider the rational section of OX (D) corresponding to 1 ∈ K(X) (never forget,
via the correspondence of (15.4.5.1)). Where are its zeros and poles? What are their
orders? What is the most common wrong answer to this problem?
We next show that in good circumstances, OX (D) is an invertible sheaf. (In
fact the OX (D) construction can occasionally be useful even if OX (D) is not an
invertible sheaf, but this won’t concern us here. An example of an OX (D) that is
not an invertible sheaf is given in Exercise 15.4.K.)

15.4.G. H ARD BUT I MPORTANT E XERCISE . Suppose L is an invertible sheaf, and s

is a nonzero rational section of L .
∼
(a) Describe an isomorphism O(div s) ←→ L . (You will use the normality hypoth-
esis!) Hint: show that those open subsets U for which O(div s)|U = ∼ OU form a
base for the Zariski topology. For each such U, define ϕU : O(div s)(U) → L (U)
sending a rational function t (with zeros and poles “constrained by div s”) to st.
Show that ϕU is an isomorphism (with the obvious inverse map, division by s).
∼
Argue that this map induces an isomorphism of sheaves ϕ : O(div s) −→ L .
(b) Let σ be the map from K(X) to the rational sections of L , where σ(t) is the ratio-
∼ L defined via (15.4.5.1) (as described in Remark 15.4.6).
nal section of OX (div s) =
Show that σ sends the canonical section “1” of OX (D) to s. More concisely: σ(1) = s.
(Hint: the map in part (a) sends 1 to s.)
444 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

In conclusion, we can identify which Weil divisors D are in the image of

div (15.4.3.1): they are the D for which OX (D) is a line bundle, and then we can
construct the image — (O(D), σ(1)) the unique (L , s) (up to isomorphism) such
that div(L , s) = D.

15.4.H. E XERCISE ( THE EXAMPLE OF §15.1). Suppose X = Pn k , L = O(1), s is the

section of O(1) corresponding to x0 , and D = div s. Verify that O(mD) = ∼ O(m),
and the canonical rational section of O(mD) is precisely s . (Watch out for possible
m

confusion: 1 has no pole along x0 = 0, but σ(1) = sm does have a zero if m > 0.)
For this reason, O(1) is sometimes called the hyperplane class in Pic X. (Of course,
x0 can be replaced by any linear form.)

15.4.I. E XERCISE ( THE P ICARD GROUP OF PROJECTIVE SPACE ). You have identified
Weil(Pn n
k ) in Exercise 15.4.A. Use this to deduce that the only line bundles on Pk are
n ∼
isomorphic to O(m) for some m, so Pic Pk = Z.

15.4.7. Definition. If D is a Weil divisor on (Noetherian normal irreducible) X such

that D = div f for some rational function f, we say that D is principal. Principal
divisors clearly form a subgroup of Weil X; denote this group of principal divisors
Prin X. Note that div induces a group homomorphism K(X)× → Prin X. If X can be
covered with open sets Ui such that on Ui , D is principal, we say that D is locally
principal. (Caution: we used the phrase “locally principal” in a slightly different
way, when referring to closed subschemes.) Locally principal divisors form a
subgroup of Weil X, which we denote LocPrin X. (This notation is not standard,
and we only use it in (15.4.10.1).)

15.4.8. Important Observation: “O(principal) = ∼ O, O(locally principal) = line bundle”.

As a consequence of Exercise 15.4.G(a) (taking L = O), if D is principal (and X
is normal, our standing hypothesis), then (i) O(D) = ∼ O. (Diagram (15.4.10.1) will
imply that the converse holds: if O(D) = ∼ O, then D is principal.) Thus (ii) if D is
locally principal, OX (D) is locally isomorphic to OX — OX (D) is an invertible sheaf.

15.4.J. I MPORTANT E XERCISE . Suppose OX (D) is an invertible sheaf. Show the

converse to Important Observation 15.4.8: show that D is locally principal.

15.4.9. Remark. In definition (15.4.5.1), it may seem cleaner to consider those s

such that div s ≥ D|U . The reason for the convention comes from our desire that
div σ(1) = D. (Taking the “opposite” convention would yield the dual bundle, in
the case where D is locally principal.)

15.4.K. L ESS IMPORTANT EXERCISE : A W EIL DIVISOR THAT IS NOT LOCALLY PRIN -
CIPAL . Let X = Spec k[x, y, z]/(xy − z2 ), a cone, and let D be the line z = x = 0
(see Figure 13.1).
(a) Show that D is not a locally principal divisor. (Hint: consider the stalk at the
origin. Use the Zariski tangent space, see Problem 13.1.3.) In particular OX (D) is
not an invertible sheaf. (Caution: we earlier saw that it was not a locally principal
closed subscheme in Problem 13.1.3. It is unfortunate that “locally principal” has
two different but very similar meanings.)
(b) Show that div(x) = 2D. This corresponds to the fact that the plane x = 0 is
tangent to the cone X along D. Hence 2D is a locally principal divisor.
 © 2024 Ravi Vakil. Published by Princeton University Press. 445

15.4.L. I MPORTANT E XERCISE . If X is Noetherian and factorial, show that for any
Weil divisor D, O(D) is an invertible sheaf. (Hint: For motivation, consider the case
where D is irreducible, say D = [Y], and to cover X by open sets so that on each
open set U there is a function whose divisor is [Y ∩ U]. One open set will be X − Y.
Next, we find an open set U containing an arbitrary p ∈ Y, and a function on U. As
OX,p is a unique factorization domain, the prime corresponding to Y is codimension
1 and hence principal by Lemma 12.1.7. Let f be a generator of this prime ideal,
interpreted as an element of K(X). It is regular at p, it has a finite number of zeros
and poles, and through p, [Y] is the “only zero” (the only component of the divisor
of zeros). Let U be X minus all the other zeros and poles.)

15.4.10. The class group. We can now get a handle on the Picard group of a
normal Noetherian scheme. Define the class group of X, Cl X, by Weil X/ Prin X.
By taking the quotient of the inclusion (15.4.3.1) by Prin X, we have the inclusion
Pic X ,→ Cl X. This is summarized in the convenient and enlightening diagram.

(O(D),σ(1))←[D
t 
(15.4.10.1) {(L , s)}/iso.
∼ / LocPrin X  = if X / Weil X
div factorial
/ Prin X / Prin X
  
Pic X {L }/iso.
∼ / LocPrin X/ Prin X   = if X / Cl X
j factorial
O(D)←[D

This diagram is very important, and although it is short to state, it takes time to
digest.
In particular, if A is a unique factorization domain, then all Weil divisors on Spec A
are principal by Lemma 12.1.7, so Cl(Spec A) = 0, and hence Pic(Spec A) = 0.
n
15.4.11. As k[x1 , . . . , xn ] has unique factorization, Cl(An
k ) = 0, so Pic(Ak ) = 0 .
n
Geometers might find this believable — “C is a contractible complex manifold, and
hence should have no nontrivial line bundles” — even if some caution is in order,
as the kinds of line bundles being considered are entirely different: holomorphic
vs. topological or C∞ . (Aside: for this reason, you might expect that An k also has
no nontrivial vector bundles. This is the Quillen-Suslin Theorem, formerly known
as Serre’s Conjecture, part of Quillen’s work leading to his 1978 Fields Medal. The
case n = 1 was Exercise 14.2.C. For a short proof by Vaserstein, see [Lan, p. 850].)
Removing a closed subset of X of codimension greater than 1 doesn’t change
the class group, as it doesn’t change the Weil divisor group or the principal divisors.
(Warning: it can affect the Picard group, see Exercise 15.5.M.)
Removing an irreducible closed subset of pure codimension 1 changes the Weil
divisor group in a controllable way. Suppose Z is an irreducible codimension 1
subset (an irreducible divisor) of X. Then we clearly have an exact sequence:

17→[Z]
0 /Z / Weil X / Weil(X − Z) / 0.

When we take the quotient by the subgroup of principal divisors, taking into
account the fact that we may lose exactness on the left, we get an excision exact
446 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

sequence for class groups:

17→[Z]
(15.4.11.1) Z / Cl X / Cl(X − Z) / 0.

(Do you see why?)

15.4.12. For example, if U is an open subscheme of X = An , Pic U = {0}.

15.4.13. As another application, let X = Pn

k , and Z be the hyperplane x0 = 0. We
have
Z / Cl(Pn ) / Cl(An ) /0
k k

from which Cl(Pn n

k ) is generated by the class [Z], and Pic(Pk ) is a subgroup of this.

15.4.14. By Exercise 15.4.H, [Z] 7→ O(1), and as O(m) is nontrivial for m ̸= 0 (Exer-
cise 15.1.B), [Z] is not torsion in Cl Pn n n
k . Hence Pic(Pk ) ,→ Cl(Pk ) is an isomorphism,
∼
and Pic(Pn k ) = Z , with generator O(1). The degree of an invertible sheaf on P is
n

defined using this: define deg O(d) to be d. (You will have already proved that
Pic(Pn ∼
k ) = Z if you did Exercise 15.4.I; but we will use the strategy here to great
effect in §15.5.4.)
We have gotten good mileage from the fact that the Picard group of the spec-
trum of a unique factorization domain is trivial. More generally, Exercise 15.4.L
gives us:

15.4.15. Proposition. — If X is Noetherian and factorial, then for any Weil divisor D,
O(D) is invertible, and hence the map Pic X → Cl X is an isomorphism.
This can be used to make the connection to the class group in number theory
precise, see Exercise 14.1.K; see also §15.5.5.

15.4.16. Mild but important generalization: twisting line bundles by divisors.

The above constructions can be extended, with OX replaced by an arbitrary invert-
ible sheaf, as follows. Let L be an invertible sheaf on a normal Noetherian scheme
X. Then define L (D) by OX (D) ⊗ L . If D is locally principal, then L (D) is a line
bundle. Notice that in this case there are two different ways of interpreting sections
of L (D) over an open set, each with different advantages: as a section of the new
line bundle L (D), and as rational sections of L with constraints on poles and
zeros given by the divisor D.

15.4.M. E ASY E XERCISE .

(a) Assume for convenience that X is irreducible. Show that sections of L (D) can
be interpreted as rational sections of L with zeros and poles constrained by D, just
as in (15.4.5.1):
Γ (U, L (D)) := {t nonzero rational section of L : div |U t + D|U ≥ 0} ∪ {0}.
(b) Suppose D1 and D2 are locally principal. Show that
∼ O(D1 + D2 ).
(O(D1 ))(D2 ) =

15.4.17. A variation of the Qcqs Lemma. The Qcqs Lemma 6.2.9, proved in
Exercise 6.2.G, has the following generalization.
 © 2024 Ravi Vakil. Published by Princeton University Press. 447

15.4.N. I MPORTANT E XERCISE ( TO BE USED REPEATEDLY ). Suppose X is a qua-

sicompact quasiseparated scheme, L is an invertible sheaf on X with section s,
and F is a quasicoherent sheaf on X. Generalizing Definition 6.2.8, let Xs be the
open subset of X where s doesn’t vanish. We interpret s as a degree 1 element of
the graded ring R(L )• := ⊕n≥0 Γ (X, L ⊗n ). Note that ⊕n≥0 Γ (X, F ⊗OX L ⊗n ) is a
graded R(L )• -module.
(a) Describe a natural map

(⊕n≥0 Γ (X, F ⊗OX L ⊗n ))s / Γ (Xs , F ).

0

(Possible hint: for quasicoherent sheaves, “tensor product has no need to be sheafi-
fied when restricted to affine subschemes”, Exercise 6.2.F.)
(b) Show that this map is an isomorphism. (Hint: show this map is an isomorphism
in the affine case.)
Translation: Any section of F over Xs can be extended to a section over X
after multiplying by some some appropriate power of s. And if we have two such
extensions, they become equal after multiplying by another appropriate power of s.

15.5 The pay-off: Many fun examples

15.5.1. Fun examples: Projective transformations.

The fact that Pic Pn ∼
k = Z has many wonderful and cheap consequences.

15.5.A. E XERCISE ( AUTOMORPHISMS OF PROJECTIVE SPACE ). Show that all the

automorphisms of projective space Pn k (fixing k) correspond to (n + 1) × (n + 1)
invertible matrices over k, modulo scalars (also known as PGLn+1 (k)). (Hint:
k → Pk is an automorphism. Show that this induces an isomorphism
Suppose π : Pn n
∼
π∗ O(1) −→ O(1). Show that π∗ : Γ (Pn , O(1)) → Γ (Pn , O(1)) is an isomorphism.)

15.5.2. Automorphisms of projective space are often called projective transforma-

tions. Because of Exercise 15.5.A, in their incarnation of matrices modulo scalars,
projective transformations are also called projective changes of coordinates. Ex-
ercise 15.5.A will be useful later, especially for the case n = 1. In this case, these
automorphisms are called fractional linear transformations. (For experts: why was
Exercise 15.5.A not stated over an arbitrary base ring A? Where does the argument
go wrong in that case? For what rings A does the result still work?)

15.5.B. E XERCISE . Show that Aut(P1k ) is strictly three-transitive on k-valued points,

i.e., given two triplets (p1 , p2 , p3 ) and (q1 , q2 , q3 ) each of distinct k-valued points
of P1 , there is precisely one automorphism of P1 sending pi to qi (i = 1, 2, 3).

15.5.C. E XERCISE . Solve these problems over an arbitrary field k.

(a) Find a linear fractional transformation f(t) ∈ PGL(2) that has order precisely 3
in PGL(2).
(b) Show that any two order 3 elements of PGL(2) are conjugate. (Possible hint: use
transitivity.)
448 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

15.5.D. E XERCISE . Suppose p0 , . . . , pn+1 are n + 2 distinct k-valued points of

Pnk , no n + 1 of which lie on a hyperplane. Show that there is a unique projective
transformation taking pi (0 ≤ i ≤ n) to [0, . . . , 0, 1, 0, . . . , 0] (where the 1 is in the
ith position), and taking pn+1 to [1, . . . , 1].

15.5.E. F UN E XERCISE . Suppose X is a quasiprojective k-scheme, and π : Pn k →X

is any morphism (over k). Show that either the image of π has dimension n, or π
contracts Pnk to a point. In particular, there are no nonconstant maps from projective
space to a smaller-dimensional quasi-projective variety. Hint: show that it suffices
to assume k is algebraically closed, and in particular, infinite. If X ⊂ PN , define
d by π∗ OPN (1) =∼ OPn (d). Try to show that d = 0. To do that, show that if m ≤ n
k
then m nonempty hypersurfaces in Pn have nonempty intersection. For this, use
the fact that any nonempty hypersurface in Pn k has nonempty intersection with any
subscheme of dimension at least 1 (Exercise 12.3.D(a)).

15.5.F. ⋆ E XERCISE ( FOR THOSE WHO HAVE READ §7.6.4, THE DOUBLE - STARRED
SECTION ON GROUP SCHEMES ). Explain how GLn acts (nontrivially!) on Pn−1
(over Z, or over a field of your choice). (The group scheme GLn was defined in
Exercise 7.6.N. The action of a group scheme appeared earlier in Exercise 7.6.S(a).)
Hint: this is much more easily done with the language of functors, §7.6, using our
functorial description of projective space (§15.2.3), than with our old description of
projective space in terms of patches. (A generalization to the Grassmannian will be
given in Exercise 16.4.L.)

15.5.3. Remark. Over an algebraically closed field k, GLn acts transitively on

the closed points of Pn
k
, and the stabilizer of the point [1, 0, · · · , 0] consists of the
subgroup P of matrices with 0’s in the first column below the first row. (Side
Remark: the point is better written as a column vector, so the GLn -action can be
interpreted as matrix multiplication in the usual way.) This suggests that Pkn−1 is
the quotient GLn /P. This is largely true; but we first would have to make sense of
the notion of group quotient.

15.5.4. More fun examples.

We can now actually calculate some Picard and class groups. First, a useful
observation: notice that you can restrict invertible sheaves on Y to any subscheme
X, and this can be a handy way of checking that an invertible sheaf is not trivial.
Effective Cartier divisors (§9.5.1) sometimes restrict too: if you have an effective
Cartier divisor on Y, then it restricts to a closed subscheme on X, locally cut out
scheme-theoretically by one equation. If you are fortunate and this equation doesn’t
vanish on any associated point of X (§6.6.2), then you get an effective Cartier divisor
on X. You can check that the restriction of effective Cartier divisors corresponds to
restriction of invertible sheaves (in the sense of Exercise 14.1.I).

15.5.G. E XERCISE : A TORSION P ICARD GROUP. Suppose that Y is a hypersurface in

Pn n ∼
k corresponding to an irreducible degree d polynomial. Show that Pic(Pk − Y) =
n ∼
Z/(d). (For differential geometers: this is related to the fact that π1 (Pk −Y) = Z/(d).)
Hint: (15.4.11.1).
The next two exercises explore consequences of Exercise 15.5.G, and provide
us with some examples promised in Exercise 5.4.N.
 © 2024 Ravi Vakil. Published by Princeton University Press. 449

15.5.H. E XERCISE ( GENERALIZING E XERCISE 5.4.N). Keeping the same notation,

assume d > 1 (so Pic(Pn − Y) ̸= 0), and let H0 , . . . , Hn be the n + 1 coordinate
hyperplanes on Pn . Show that Pn \ Y is affine, and Pn − Y − Hi is a distinguished
open subset of it. Show that the Pn − Y − Hi form an open cover of Pn − Y. Show
that Pic(Pn − Y − Hi ) = 0. Show that each Pn − Y − Hi is the Spec of a unique
factorization domain, but Pn − Y is not. Thus the property of being a unique
factorization domain is not an affine-local property — it satisfies only one of the
two hypotheses of the Affine Communication Lemma 5.3.2.

15.5.I. E XERCISE . Keeping the same notation as the previous exercise, show that
on Pn − Y, Hi (restricted to this open set) is an effective Cartier divisor that is not
cut out by a single equation. (Hint: Otherwise it would give a trivial element of the
class group.)

15.5.J. E XERCISE . Show that A := R[x, y]/(x2 + y2 − 1) is not a unique factorization

domain, but A ⊗R C is. Hint: Exercise 15.5.H. (Non-hint: doesn’t (1 − y)(1 + y) = x2
suggest that A ⊗R C isn’t a unique factorization domain?)

15.5.K. E XERCISE : P ICARD GROUP OF P1 × P1 . Consider

∼ Proj k[w, x, y, z]/(wz − xy),

X = P1k ×k P1k =

∼
a smooth quadric surface (see Figure 9.2, and Example 10.6.2). Show that Pic X =
Z ⊕ Z as follows: Show that if L = {∞} ×k P ⊂ X and M = P ×k {∞} ⊂ X, then
1 1

X−L−M = ∼ A2 . This will give you a surjection Z ⊕ Z ↠ Cl X. Show that O(L)

restricts to O on L and O(1) on M. Show that O(M) restricts to O on M and O(1)
on L. (This exercise takes some time, but is enlightening.)

15.5.L. E XERCISE . Show that irreducible smooth projective surfaces (over k) can be
birational but not isomorphic. Hint: show P2 is not isomorphic to P1 × P1 using
the Picard group. (Aside: we will see in Exercise 20.2.D that the Picard group of the
“blown up plane” is Z2 , but in Exercise 20.2.E we will see that the blown up plane
is not isomorphic to P1 × P1 , using a little more information in the Picard group.)
This is unlike the case for curves: birational irreducible smooth projective
curves (over k) must be isomorphic, as we will see in Theorem 16.3.3. Nonetheless,
any two surfaces are related in a simple way: if X and X ′ are projective, regular, and
birational, then X can be sequentially blown up at judiciously chosen points, and
X ′ can too, such that the two results are isomorphic (see [Ha1, Thm. V.5.5]; blowing
up will be discussed in Chapter 22).

15.5.M. E XERCISE : P ICARD GROUP OF THE CONE . Let X = Spec k[x, y, z]/(xy − z2 ),
a cone, where char k ̸= 2. (The characteristic hypothesis is not necessary for the
result, but is included so you can use Exercise 5.4.H to show normality of X.) Show
that Pic X = 0, and Cl X = ∼ Z/2. Hint: show that the class of Z = {x = z = 0} (the
“affine cone over a line”) generates Cl X by showing that its complement D(x) is
isomorphic to an open subset of A2k . Show that 2[Z] = div(x) and hence principal,
and that Z is not principal, Exercise 15.4.K. (Remark: You know enough to show
that X \ {(0, 0, 0)} is factorial. So although the class group is insensitive to removing
loci of codimension greater than 1, §15.4.11, this is not true of the Picard group.)
450 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

A Weil divisor (on a normal scheme) with a nonzero multiple corresponding

to a line bundle is called Q-Cartier. (We won’t use this terminology beyond the
next exercise.) Exercise 15.5.M gives an example of a Weil divisor that does not
correspond to a line bundle, but is nonetheless Q-Cartier. We now give an example
of a Weil divisor that is not Q-Cartier.

15.5.N. E XERCISE ( A NON -Q-C ARTIER DIVISOR ). On the cone over the smooth
quadric surface X = Spec k[w, x, y, z]/(wz − xy), let Z be the Weil divisor cut out by
w = x = 0. Exercise 13.1.D showed that Z is not cut out scheme-theoretically by a
single equation. Show more: that if n ̸= 0, then n[Z] is not locally principal. Hence
show that Z is not even cut out locally set-theoretically by a single equation. Hint:
show that the complement of an effective Cartier divisor on an affine scheme is also
affine, using Proposition 8.3.4. Then if some multiple of Z were locally principal,
then the closed subscheme of the complement of Z cut out by y = z = 0 would be
affine — any closed subscheme of an affine scheme is affine. But this is the scheme
y = z = 0 (also known as the wx-plane) minus the point w = x = 0, which we have
seen is non-affine, §4.4.1.

15.5.O. ⋆ E XERCISE ( FOR THOSE WITH SUFFICIENT ARITHMETIC BACKGROUND ).

Identify the (ideal) class group of the ring of integers OK in a number field K,
as defined in Exercise 14.1.K, with the class group of Spec OK , as defined in this
section. In particular, you will recover the common description of the class group as
formal sums of prime ideals, modulo an equivalence relation coming from principal
fractional ideals.

15.5.5. More on class groups and unique factorization.

As mentioned in §5.4.5, there are few commonly used means of checking that a
ring is a unique factorization domain. The next exercise is one of them, and it is
useful. For example, it implies the classical fact that for rings of integers in number
fields, the class group is the obstruction to unique factorization (see Exercise 14.1.K
and Proposition 15.4.15).

15.5.P. E XERCISE . Suppose that A is a Noetherian integral domain. Show that A is a

unique factorization domain if and only if A is integrally closed and Cl Spec A = 0.
(One direction is easy: we have already shown that unique factorization domains
are integrally closed in their fraction fields. Also, Lemma 12.1.7 shows that all
codimension 1 prime ideals of a unique factorization domain are principal, so
that implies that Cl Spec A = 0. It remains to show that if A is integrally closed
and Cl Spec A = 0, then all codimension 1 prime ideals are principal, as this char-
acterizes unique factorization domains (Proposition 12.3.7). Algebraic Hartogs’s
Lemma 13.5.19, may arise in your argument.) This is the third important characteri-
zation of unique factorization domains promised in §5.4.5.
My final favorite method of checking that a ring is a unique factorization
domain (§5.4.5) is Nagata’s Lemma. It is also the least useful.

15.5.Q. ⋆⋆ E XERCISE (N AGATA’ S L EMMA ). Suppose A is a Noetherian domain,

x ∈ A an element such that (x) is prime and Ax = A[1/x] is a unique factorization
domain. Then A is a unique factorization domain. (Hint: Exercise 15.5.P. Use the
 © 2024 Ravi Vakil. Published by Princeton University Press. 451

short exact sequence

Z[(x)] / Cl Spec A / Cl Spec Ax /0

(15.4.11.1) to show that Cl Spec A = 0. Prove that A[1/x] is integrally closed, then
show that A is integrally closed as follows. Suppose T n + an−1 T n−1 + · · · + a0 = 0,
where ai ∈ A, and T ∈ K(A). Then by integral closedness of Ax , we have that
T = r/xm , where if m > 0, then r ∈ / (x). Then we quickly get a contradiction if
m > 0.)
This leads to a fun algebra fact promised in Remark 13.8.6. Suppose k is an
algebraically closed field of characteristic not 2. Let A = k[x1 , . . . , xn ]/(x21 +· · ·+x2m )
where m ≤ n. When m ≤ 2, we get some special behavior. (If m = 0, we get affine
space; if m = 1, we get a nonreduced scheme; if m = 2, we get a reducible scheme
that is the union of two affine spaces.)
If m ≥ 3, we have verified that Spec A is normal, in Exercise 5.4.I(b). In
fact, if m ≥ 3, then A is a unique factorization domain unless m = 3 or m = 4
(Exercise 5.4.L; see also Exercise 13.1.E). For the case m = 3:
A = k[x, y, z, w1 , . . . , wn−3 ]/(x2 + y2 − z2 )
is not a unique factorization domain, as it is has nonzero class group (by essentially
the same argument as for Exercise 15.5.M).
The failure at 4 comes from the geometry of the quadric surface: we have
checked that in Spec k[w, x, y, z]/(wz − xy), there is a codimension 1 irreducible
subset — the cone over a line in a ruling — that is not principal.

15.5.R. E XERCISE ( THE CASE m ≥ 5). Suppose that k is algebraically closed of

characteristic not 2. Show that if ℓ ≥ 3, then
A = k[a, b, x1 , . . . , xn ]/(ab − x21 − · · · − x2ℓ )
is a unique factorization domain, by using Nagata’s Lemma with x = a.

15.6 Effective Cartier divisors “=” invertible ideal sheaves

We now give a different means of describing invertible sheaves on a scheme.
One advantage of this over Weil divisors is that it can give line bundles on every-
where nonreduced schemes (such a scheme can’t be regular at any codimension 1
prime).
Suppose D ,→ X is a closed subscheme such that the corresponding ideal sheaf
I is an invertible sheaf. Then I is locally trivial; suppose U is a trivializing affine
open set Spec A. Then the closed subscheme exact sequence (9.1.2.1)
0 /I / OX / OD /0

corresponds to
0 /I /A / A/I /0
∼ A as A-modules. Thus I is generated by a single element, say a, and this
with I =
exact sequence starts as
0 /A ×a
/A
452 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

As multiplication by a is injective, a is not a zerodivisor. We conclude that D is

locally cut out by a single equation, that is not a zerodivisor. This was the definition
of effective Cartier divisor given in §9.5.1. This argument is clearly reversible, so we
have a quick new definition of effective Cartier divisor (an ideal sheaf I that is an
invertible sheaf — or equivalently, the corresponding closed subscheme).

15.6.A. E ASY E XERCISE . Show that a is unique up to multiplication by an invertible

function.
In the case where X is locally Noetherian, we can use the language of associated
points (§6.6.2), so we can restate this definition as: D is locally cut out by a single
equation, not vanishing at any associated point of X.
We now define an invertible sheaf corresponding to D. The seemingly obvious
definition would be to take ID , but instead we do the “opposite”.

15.6.1. Definition: invertible sheaves corresponding to effective Cartier divisors. If D is

∨
an effective Cartier divisor, define the invertible sheaf O(D) to be ID . (The reason
for the dual is Exercise 15.6.C.) Define O(nD) to be O(D) . In particular, define
⊗n

O(−D) to be O(D)∨ , i.e., O(−D) = ID .

15.6.B. E XERCISE . If X is normal, check that this agrees with our earlier definition
of O(D), Important Definition 15.4.5. (In Definition 15.4.5, D is a Weil divisor, but
in this exercise, D is an effective Cartier divisor. So you will have to define the
Weil divisor corresponding to an effective Cartier divisor on a normal Noetherian
scheme X.)
Because the ideal sheaf ID is O(−D), the closed subscheme exact sequence
(9.1.2.1) may be written as

0 / O(−D) /O / OD / 0.

The invertible sheaf O(D) has a canonical section sD : tensoring 0 → I → O

with I ∨ gives us O → I ∨ . (Easy unimportant fact: instead of tensoring I → O
with I ∨ , we could have dualized I → O, and we would get the same section.)

15.6.C. I MPORTANT AND SURPRISINGLY TRICKY EXERCISE . Recall that a section of

a locally free sheaf on X cuts out a closed subscheme of X (Exercise 14.2.D). Show
that the section sD cuts out D. (Compare this to Remark 15.4.9.)
This construction is reversible:

15.6.D. E XERCISE . Suppose L is an invertible sheaf, and s is a section that is locally

not a zerodivisor. (Make sense of this! In particular, if X is locally Noetherian, this
means “s does not vanish at an associated point of X”, see §6.6.2.) Show that s = 0
cuts out an effective Cartier divisor D, and O(D) =∼ L.

15.6.E. E XERCISE . Suppose I and J are invertible ideal sheaves (hence corre-
sponding to effective Cartier divisors, say D and D ′ respectively). Show that I J
is an invertible ideal sheaf. (We define the product of two quasicoherent ideal
sheaves I J as you might expect: on each affine, we take the product of the two
corresponding ideals, as defined in Exercise 3.4.C. To make sure this is well-defined,
we need only check that if A is a ring, and f ∈ A, and I, J ⊂ A are two ideals, then
 © 2024 Ravi Vakil. Published by Princeton University Press. 453

(IJ)f = If Jf in Af .) We define the corresponding Cartier divisor to be D + D ′ . Verify

that O(D + D ′ ) =∼ O(D) ⊗ O(D ′ ).

We thus have an important correspondence between effective Cartier divisors

(closed subschemes whose ideal sheaves are invertible, or equivalently locally cut
out by one non-zerodivisor, or in the locally Noetherian case, locally cut out by one
equation not vanishing at an associated point) and ordered pairs (L , s) where L
is an invertible sheaf, and s is a section that is not locally a zerodivisor (or in the
locally Noetherian case, not vanishing at an associated point). The effective Cartier
divisors form an abelian semigroup. We have a map of semigroups, from effective
Cartier divisors to invertible sheaves with sections not locally zerodivisors (and
hence also to the Picard group of invertible sheaves).

15.6.2. Normal (line) bundles to effective Cartier divisors. If D is an effective Cartier

divisor on a scheme X, then the restriction of the invertible sheaf O(D) to D itself is
often called the normal (line) bundle to D (although perhaps it should be called
the normal invertible sheaf). We denote it ND/X := OX (D)|D . The motivation for
this language is that if X and D are smooth complex varieties, the normal bundle
in this algebro-geometric sense agrees with the normal bundle in the differential-
geometric sense. This interpretation breaks down in more “singular” situations,
but the intuition is useful even then. We will define the normal bundle (and normal
sheaf) in a more general setting in §21.2.15.

15.6.3. Unimportant Remark. We get lots of invertible sheaves, by taking differences

of two effective Cartier divisors. In fact we “usually get them all” — it is very hard
to describe an invertible sheaf on a finite type k-scheme that is not describable in
such a way. For example, there are none if the scheme is regular or even factorial
(basically by Proposition 15.4.15 for factoriality; and regular schemes are factorial
by the Auslander-Buchsbaum Theorem 13.8.5). Exercise 16.2.F will imply that there
are none if the scheme is projective. It holds in all other reasonable circumstances,
see [Gr-EGA, IV4 .21.3.4]. However, it does not always hold; the first and best
example is due to Kleiman, see [Kl3].

15.6.4. ⋆⋆ Cartier divisors. There is a related notion of Cartier divisor. This notion is
considered essential by many, so I will try to explain why we are not discussing
it (except briefly here). Cartier divisors are indeed important, but arguably not
at the level of discussion we are engaging in. Using Cartier divisors involves
technical difficulties, in return for suprisingly little simplification of perspective
and results. (The expert is welcome to develop a parallel discussion in the language
of Cartier divisors in the margins to test this out.) For completeness, here is a
definition. On a scheme X, we define an OX -module MX , whose sections on an
affine open set Spec A are the (elements of the) total fraction ring of A (§6.6.36),
with the “obvious” restriction maps. If X is integral, MX is the constant sheaf
corresponding to the function field K(X). We have an induced map OX ,→ MX . Let
MX∗ be the sheaf of groups corresponding (on the level of sections) to invertible
elements, inducing OX∗ ,→ MX∗ . Then the Cartier divisors of X are defined as
Γ (X, OX∗ /MX∗ ) (with its structure of an abelian group). Many seeming advantages
of Cartier divisors actually come from other structures. Cartier divisors easily
pull back under morphisms, but so do line bundles / invertible sheaves. Effective
Cartier divisors are also very pleasant and simple. Cartier divisors easily yield
454 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

invertible sheaves, but it is not always true that every invertible sheaf comes from a
Cartier divisor, [Kl3].

15.7 The graded module corresponding to a quasicoherent sheaf

(This section answers some fundamental questions, but it is surprisingly tricky.
You may wish to skip this section, or at least the proofs, on first reading, unless you
have a particular need for them.)
Throughout this section, S• is a finitely generated graded algebra generated
in degree 1, so in particular we have the invertible sheaf O(m) for all m by Exer-
cise 15.1.G. Also, throughout, M• is a graded S• -module, and F is a quasicoherent
sheaf on Proj S• .
We know how to get quasicoherent sheaves on Proj S• from graded S• -modules.
We will now see that we can get them all in this way. We will define a functor Γ•
from (the category of) quasicoherent sheaves on Proj S• to (the category of) graded
S• -modules that will attempt to reverse the ∼ construction. They are not quite
inverses, as ∼ can turn two different graded modules into the same quasicoherent
sheaf (see for example Exercise 14.6.C). But we will see a natural isomorphism
∼
• (F ) ←→ F . In fact Γ• (M• ) is a better (“saturated”) version of M• , and there is a
Γ^ g
saturation functor M• → Γ• (M g• ) that is akin to groupification and sheafification
— it is adjoint to the forgetful functor from saturated graded modules to graded
modules. And thus we come to the fundamental relationship between ∼ and Γ• :
they are an adjoint pair.
graded S• -modules
S
∼

u
QCohProj S• saturate forget
equivalence
Γ•
) 
saturated graded S• -modules
We now make some of this precise, but as little as possible to move forward. In
particular, we will show that every quasicoherent sheaf on a projective A-scheme
arises from a graded module (Corollary 15.7.4), and we will see that every closed
subscheme of Proj S• arises from a homogeneous ideal I• ⊂ S• (Exercise 15.7.I).

15.7.1. Definition of Γ• . When you do Essential Exercise 15.1.C (on global sections
of OPnk (m)), you will suspect that in good situations,
∼ Γ (Proj S• , M(m)
Mm = f • ).

Motivated by this, we define

Γm (F ) := Γ (Proj S• , F (m)).

15.7.A. E XERCISE . Describe a morphism of S0 -modules Mm → Γ (Proj S• , M(m)

^• ),
extending the m = 0 case of Exercise 14.6.D.
 © 2024 Ravi Vakil. Published by Princeton University Press. 455

15.7.B. E XERCISE . Show that Γ• (F ) is a graded S• -module. (Hint: consider

Sm → Γ (Proj S• , O(m)).)

15.7.C. E XERCISE . Show that the map M• → Γ• (M g• ) arising from the previous two
exercises is a map of S• -modules. We call this the saturation map.

15.7.D. E XERCISE . Show that the saturation map need not be injective, nor need it
be surjective. (Hint: S• = k[x], M• = k[x]/(x2 ) or M• = xk[x].)

15.7.E. E XERCISE . Show that Γ• is a functor from QCohProj S• to the category of

graded S• -modules. In other words, if F → G is a morphism of quasicoherent
sheaves on Proj S• , describe the natural map Γ• F → Γ• G , and show that such maps
respect the identity and composition. Thus the saturation map can be better called
the saturation functor.

15.7.2. The special case M• = S• . We have a saturation map S• → Γ• S f• , which

is a map of S• -modules. But Γ• S f• has the structure of a graded ring (basically
because we can multiply sections of O(m1 ) by sections of O(m2 ) to get sections of
O(m1 + m2 ), see Exercise 15.1.I).

15.7.F. E XERCISE .
(a) Show that the saturation map is in fact a ring morphism.
(b) Show that the saturation map is an isomorphism when S• is a polynomial ring
(with the usual grading), in at least one variable.
(c) Show that the saturation map is not an isomorphism for S• = k[x, y]/(xy), where
both variables have degree 1.

15.7.3. ⋆ The reverse map. Now that we have defined the saturation map M• →
• F → F . While subtler to define, it will have the
g• , we will describe a map Γg
Γ• M
advantage of being an isomorphism.

15.7.G. E XERCISE . Define the natural map Γg • F → F as follows. First describe the
map on sections over D+ (f). Note that sections of the left side are of the form z/fn
′ ′
where z ∈ Γn deg f (F ), and z/fn = z ′ /fn if there is some N with fN (fn z−fn z ′ ) = 0.
Sections on the right are implicitly described in Exercise 15.4.N. Show that your
map behaves well on overlaps D+ (f) ∩ D+ (g) = D+ (fg).

15.7.H. T RICKY E XERCISE . Show that the natural map Γg • F → F is an isomor-

phism, by showing that it is an isomorphism of sections over D+ (f) for any f. First
show surjectivity, using Exercise 15.4.N to show that any section of F over D+ (f)
is of the form z/fn where z ∈ Γn deg f (F ). Then verify that it is injective.

15.7.4. Corollary. — Every quasicoherent sheaf on a projective A-scheme arises from the
∼ construction.

15.7.I. E XERCISE ( CONVERSE TO E XERCISE 9.3.B). Show that each closed subscheme
of Proj S• arises from a homogeneous ideal I• ⊂ S• . (Hint: Suppose Z is a closed
subscheme of Proj S• . Consider the exact sequence 0 → IZ → OProj S• → OZ → 0.
Apply Γ• , and then ∼. Be careful: Γ• is left-exact, but not necessarily exact.) This
fulfills promises made in Exercises 9.3.B and 14.6.E.
456 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

For the first time, we see that every closed subscheme of a projective scheme is
cut out by homogeneous equations. This is the analog of the fact that every closed
subscheme of an affine scheme is cut out by equations. It is disturbing that it is so
hard to prove this fact. (For comparison, this was easy on the level of the Zariski
topology — see Exercise 4.5.K(c).)

15.7.J. ⋆⋆ E XERCISE (Γ• AND ∼ ARE ADJOINT FUNCTORS ). Describe a natural

bijection Hom(M• , Γ• F ) = ∼ Hom(Mg• , F ), as follows.
(a) Show that maps M• → Γ• F are the “same” as maps ((M• )f )0 → ((Γ• F )f )0
as f varies through S+ , that are “compatible” as f varies, i.e., if D+ (g) ⊂
D+ (f), there is a commutative diagram
((M• )f )0 / ((Γ• F )f )0

 
((M• )g )0 / ((Γ• F )g )0

More precisely, give a bijection between Hom(M• , Γ• F ) and the set of

compatible maps
 
Hom((M• )f )0 → ((Γ• F )f )0 .
f∈S+

(b) Describe a bijection between the set of compatible maps (Hom((M• )f )0 →

((Γ• F )f )0 )f∈S+ and the set of compatible maps Γ (D+ (f), M
g• ) → Γ (D+ (f), F ).

15.7.5. ⋆⋆ Saturated S• -modules. We end with a remark: different graded S• -

modules give the same quasicoherent sheaf on Proj S• , but the results of this section
show that there is a “best” (= saturated) graded module for each quasicoherent
sheaf, and there is a map from each graded module to its “best” version, M• →
g• . A module for which this is an isomorphism (a “best” module) is called
Γ• M
saturated. We won’t use this term later.
This “saturation” map M• → Γ• M g• is analogous to the sheafification map,
taking presheaves to sheaves. For example, the saturation of the saturation equals
the saturation.
See §14.6.1 for another description of the category of saturated S• -modules.
 CHAPTER 16

Maps to projective space, and properties of line bundles

Line bundles on a variety X are important in part because they can tell a lot
about maps from X to other varieties, because they tell us about maps from X to
projective space (§15.2). The crucial concept of “ampleness” of a line bundle should
be seen as some sort of “positivity” (§16.2), which will helps us know when the
line bundle has lots of sections and might even give a map to projective space. The
fundamental facts of ampleness will take some work, but will be powerfully useful
from then on. For example, in §16.3, we will see many applications to curves.

16.1 Globally generated quasicoherent sheaves

We begin with a result letting us “get at” any finite type quasicoherent sheaf on
a projective scheme in terms of line bundles. Throughout this section, S• will be
a finitely generated graded ring over A, generated in degree 1. We will prove the
following result.

16.1.1. Theorem. — Any finite type quasicoherent sheaf F on Proj S• can be presented
in the form
⊕finite O(−m) /F / 0.

Because we can work with the line bundles O(−m) in a hands-on way, this
result will give us great control over all coherent sheaves (and in particular, vector
bundles) on Proj S• . As just a first example, it will allow us to show that every
coherent sheaf on a projective k-scheme has a finite-dimensional space of global
sections (Corollary 18.1.4). (This fact will grow up to be the fact that the higher
pushforwards of coherent sheaves under proper morphisms are also coherent, see
Theorem 18.7.1(d) and Grothendieck’s Coherence Theorem 18.9.1.)
Rather than proceeding directly to a proof, we use this as an excuse to introduce
notions that are useful in wider circumstances (global generation, base-point-freeness,
ampleness), and their interrelationships. But first we use it as an excuse to mention
an important classical result.

16.1.2. The Hilbert Syzygy Theorem.

Given any coherent sheaf F on Pn k , Theorem 16.1.1 gives the existence of a
surjection ϕ : ⊕finite O(−m) → F → 0. The kernel of the surjection is also coherent,

This material is published by Princeton University Press as The Rising Sea: Foundations of Algebraic
Geometry by Ravi Vakil. This version is free to view and download for personal use only. Not for
redistribution, resale, or use in derivative works. © 2024 Ravi Vakil.

457
458 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

so iterating this construction, we can construct an infinite resolution of F by a

direct sum of line bundles:
· · · ⊕finite O(m2,j ) / ⊕finite O(m1,j ) / ⊕finite O(m0,j ) /F / 0.

16.1.3. The Hilbert Syzygy Theorem. — There exists a finite resolution, of length at
most n + 1.
See the comments after Theorem 3.6.17 for the original history of this result.
We won’t use the theorem, but a proof will be given in the optional section §24.4.

16.1.4. Globally generated sheaves. Suppose X is a scheme, and F is an O-

module. The most important definition of this section is the following: F is
globally generated (or generated by global sections) if it admits a surjection from
a free sheaf on X:
O ⊕I //F .
Here I is some index set. The global sections in question are the images of the |I|
sections corresponding to 1 in the various summands of OX⊕I ; those images generate
the stalks of F . We say F is finitely globally generated (or generated by a finite
number of global sections) if the index set I can be taken to be finite.
More definitions in more detail: we say that F is globally generated at a point
p (or sometimes generated by global sections at p) if we can find ϕ : O ⊕I → F
that is surjective on stalks at p:
ϕp
Op⊕I / / Fp .

(It would be more precise to say that the stalk of F at p is generated by global
sections of F .) The key insight is that global generation at p means that every germ
at p is a linear combination (over the local ring OX,p ) of germs of global sections.
Note that F is globally generated if it is globally generated at all points p. (Reason:
Exercise 2.4.D showed that isomorphisms can be checked on the level of stalks. An
easier version of the same argument shows that surjectivity can also be checked on
the level of stalks.) Notice that we can take a single index set for all of X, by taking
the union of all the index sets for each p.

16.1.A. E ASY E XERCISE . An invertible sheaf L on X is globally generated if and

only if for any point p ∈ X, there is a global section of L not vanishing at p, i.e., L
is base-point-free (Definition 15.2.1).

16.1.B. E ASY E XERCISE ( REALITY CHECK ). Show that every quasicoherent sheaf on
every affine scheme is globally generated. Show that every finite type quasicoherent
sheaf on every affine scheme is generated by a finite number of global sections.
(Hint for both: for any A-module M, there is a surjection onto M from a free
A-module.)
Clearly if F and G are globally generated, then so is F ⊕ G (and similarly
for finitely globally generated, generated at a point p, etc.). Similarly for tensor
product:

16.1.C. E ASY E XERCISE ( GLOBALLY GENERATED ⊗ GLOBALLY GENERATED IS GLOB -

ALLY GENERATED ). Show that if quasicoherent sheaves F and G are globally
 © 2024 Ravi Vakil. Published by Princeton University Press. 459

generated at a point p, then so is F ⊗ G . (If F and G are invertible sheaves, this is

Exercise 15.2.B.)

16.1.D. E ASY BUT I MPORTANT E XERCISE . Suppose F is a finite type quasicoherent

sheaf on X.
(a) Show that F is globally generated at p if and only if “the fiber of F is gener-
ated by global sections at p”, i.e., the global sections span the fiber Fp /mFp as a
vector space, where m is the maximal ideal of OX,p . (Hint: Geometric Nakayama,
Exercise 14.3.D.)
(b) Show that if F is globally generated at p, then “F is globally generated near p”:
there is an open neighborhood U of p such that F is globally generated at every
point of U.
(c) Suppose further that X is a quasicompact scheme. Show that if F is globally
generated at all closed points of X, then F is globally generated at all points of X.
(Note that nonempty quasicompact schemes have closed points, Exercise 5.1.E.)

16.1.E. E ASY E XERCISE . If F is a finite type quasicoherent sheaf on X, and X is

quasicompact, show that F is globally generated if and only if it is generated by a
finite number of global sections.
We are now able to state a celebrated result of Serre.

16.1.5. Serre’s Theorem A. — Suppose S• is generated in degree 1, and finitely generated

over A = S0 . Let F be any finite type quasicoherent sheaf on Proj S• . Then there exists
some m0 such that for all m ≥ m0 , F (m) is finitely globally generated.
We could now prove Serre’s Theorem A directly, but will continue to use this
as an excuse to introduce more ideas; it will be a consequence of Theorem 16.2.2.

16.1.6. Proof of Theorem 16.1.1 assuming Serre’s Theorem A (Theorem 16.1.5). Suppose
we have n global sections s1 , . . . , sn of F (m) that generate F (m). This gives a
map
O ⊕n / F (m)
Proj S•

given by (f1 , . . . , fn ) 7→ f1 s1 + · · · + fn sn on any open set. Because these global

sections generate F (m), this is a surjection. Tensoring with O(−m) (which is exact,
as tensoring with any locally free sheaf is exact, Exercise 14.1.F) gives the desired
result. □

16.2 Ample and very ample line bundles

We now introduce ampleness, a central “positivity” property of line bundles
that will prove important for a number of disparate reasons — geometric, algebraic,
topological, computational, and more.
Suppose π : X → Spec A is a morphism, and L is an invertible sheaf on X. (The
case when A is a field is the one of most immediate interest.) We say that L is very
ample over A or π-very ample, or relatively very ample if X = ∼ Proj S• where S•
is a finitely generated graded ring over A generated in degree 1 (Definition 4.5.6),
and L = ∼ OProj S (1). One often just says very ample if the structure morphism is
•
460 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

clear from the context. Note that the existence of a very ample line bundle implies
that X is projective.

16.2.A. E ASY BUT I MPORTANT E XERCISE ( EQUIVALENT DEFINITION OF VERY AMPLE

OVER A). Suppose π : X → Spec A is proper, and L is an invertible sheaf on X.
Show that L is very ample if and only if there exist a finite number of global
sections s0 , . . . sn of L , with no common zeros, such that the morphism
[s0 , . . . , sn ] : X / Pn
A

is a closed embedding.

16.2.B. E ASY EXERCISE ( VERY AMPLE IMPLIES BASE - POINT- FREE ). Show that a
very ample invertible sheaf L on a proper A-scheme must be base-point-free. In
other words, show that for any point p on the scheme, there is a section of L not
vanishing there.

16.2.C. E XERCISE ( VERY AMPLE ⊗ BASE - POINT- FREE IS VERY AMPLE , HENCE VERY
AMPLE ⊗ VERY AMPLE IS VERY AMPLE ). Suppose L and M are invertible sheaves
on a proper A-scheme X, and L is very ample over A and M is base-point-free,
then L ⊗ M is very ample. (Hint: L gives a closed embedding X ,→ Pm , and
M gives a morphism X → Pn . Show that the product map X → Pm × Pn is a
closed embedding, using the Cancellation Theorem 11.1.1 for closed embeddings on
X → Pm ×Pn → Pm . Finally, consider the composition X ,→ Pm ×Pn ,→ Pmn+m+n ,
where the last closed embedding is the Segre embedding.)

16.2.D. E XERCISE ( VERY AMPLE ⊠ VERY AMPLE IS VERY AMPLE , CF. E XAMPLE 15.2.10).
Suppose X and Y are proper A-schemes, and L (resp., M ) is a very ample invert-
ible sheaf on X (resp., Y). If πX : X ×A Y → X and πY : X ×A Y → Y are the usual
projections, show that π∗X L ⊗ π∗Y M (also known as L ⊠ M , see §15.2.10) is very
ample on X ×A Y.

16.2.1. Definition. We say an invertible sheaf L on a proper A-scheme X is ample

over A, or π-ample (where π : X → Spec A is the structure morphism), or relatively
ample if one of the following equivalent conditions holds.

16.2.2. Theorem. — Suppose π : X → Spec A is proper, and L is an invertible sheaf on

X. Then the following are equivalent.
Projective geometry:
(a) For some N > 0, L ⊗N is very ample over A.
(a’) For all n ≫ 0, L ⊗n is very ample over A.
Global generation:
(b) For all finite type quasicoherent sheaves F , there is an n0 such that for n ≥ n0 ,
F ⊗ L ⊗n is globally generated.
Topology:
(c) As f runs over all the global sections of L ⊗n (over all n > 0), the open subsets
Xf = {p ∈ X : f(p) ̸= 0} form a base for the topology of X.
(c’) As f runs over the global sections of L ⊗n (over all n > 0), those open subsets Xf
which are affine form a base for the topology of X.
(c”) As f runs over the global sections of L ⊗n (over all n > 0), those open subsets Xf
which are affine cover X.
 © 2024 Ravi Vakil. Published by Princeton University Press. 461

(Variants of Theorem 16.2.2 in the “absolute” and “relative” settings will be

given in Theorems 16.2.6 and 17.3.7 respectively.)
Note that (b) and (c)–(c”) make no reference to the structure morphism π. We
will later (Theorem 18.8.1) meet a cohomological criterion for ampleness (due to
Serre). The Kodaira Embedding Theorem (see, for example, [GH1, p. 181]) also
gives a criterion for ampleness in the complex category: if X is a compact complex
manifold, then an invertible sheaf L on X is ample if and only if it admits a
Hermitian metric with positive curvature everywhere.
The different flavor of these conditions gives some indication that ampleness
is better-behaved than very ampleness in a number of ways. We mention without
proof another property (“ampleness is an open condition in families”): if π : X →
T is a finitely presented proper morphism, and L is an invertible sheaf on X,
then those points on T where L is ample on the fiber form an open subset of T .
Furthermore, on this open subset, L is relatively ample over the base. We won’t
use these facts (proved in [Gr-EGA, IV3 .9.6.4]), but they are good to know.
Before getting to the proof of Theorem 16.2.2, we give some sample applications.
We begin by noting that the fact that (a) implies (b) gives Serre’s Theorem A
(Theorem 16.1.5).

16.2.E. I MPORTANT E XERCISE . Suppose L and M are invertible sheaves on a

proper A-scheme X, and L is ample. Show that L ⊗n ⊗ M is very ample for n ≫ 0.
(Hint: use both (a) and (b) of Theorem 16.2.2, and Exercise 16.2.C.)

16.2.F. I MPORTANT E XERCISE . Show that every line bundle on a projective A-

scheme X is the difference of two very ample line bundles. More precisely, for any
invertible sheaf L on X, we can find two very ample invertible sheaves M and N
such that L = ∼ M ⊗ N ∨ . (Hint: use the previous Exercise.)

16.2.G. I MPORTANT E XERCISE ( USED REPEATEDLY ). Suppose π : X → Y is a finite

morphism of proper A-schemes, and L is an ample line bundle on Y. Show that
π∗ L is ample on X. Hint: use the criterion of Theorem 16.2.2(c”). (Remark for those
who have read about ampleness in the absolute setting in §16.2.5: the result applies
in that situation, i.e., with “proper A-schemes” changed to “schemes”, without
change. The only additional thing to note is that ampleness of L on Y implies
that Y is quasicompact from the definition, and separated from Theorem 16.2.6(d).
A relative version of this result appears in §17.3.6. It can be generalized even
further, with “π finite” replaced by “π quasiaffine” — to be defined in §17.3.11 —
see [Gr-EGA, II.5.1.12].)

16.2.H. E XERCISE ( AMPLE ⊗ AMPLE IS AMPLE , AMPLE ⊗ BASE - POINT- FREE IS

AMPLE ). Suppose L and M are invertible sheaves on a proper A-scheme X, and
L is ample. Show that if M is ample or base-point-free, then L ⊗ M is ample.

16.2.I. L ESS IMPORTANT E XERCISE ( AMPLE ⊠ AMPLE IS AMPLE ). Solve Exer-

cise 16.2.D with “very ample” replaced by “ample”.

16.2.J. ⋆⋆ E XERCISE ( CF. E XERCISE 12.3.C). Suppose X is a projective k-scheme, L

is an ample line bundle on X, M is a line bundle on X, and q1 , . . . , qm are points
of X (not necessarily closed). Show that for every N ≫ 0, there is a section of
L ⊗N ⊗ M that does not vanish on any of the qi . (Hint: if q1 is not in the closure of
462 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

any of the other qi , use the criterion of Theorem 16.2.2(b) with a judiciously chosen
F to find a section of L ⊗N ⊗ M vanishing at all of the qi except nonvanishing
at q1 .) This exercise is double-starred because it is interesting mainly when k is a
finite field, and because it will not be used in an essential way later in the book. But
it can be useful, and in particular you might use it to solve starred Exercise 18.4.L,
by way of Exercise 16.2.K.

16.2.K. ⋆⋆ E XERCISE . Suppose C is a reduced projective projective curve over

a field k, L is a line bundle on C, and q1 , . . . , qP
m are closed points of C. Using
Exercise 16.2.J, show that we can write L = ∼ O( nj pj ) where the pj are regular
points distinct from the qi . Hint: Use that the nonregular points of C form a finite
set (§13.5.10). Write L as the difference of two very ample line bundles.

16.2.3. Proof of Theorem 16.2.2 in the case X is Noetherian. Noetherian hypotheses are
used at only one point in the proof, and we explain how to remove them.

F IGURE 16.1. Structure of the proof of Theorem 16.2.2, up to the

last step

We work roughly from easiest to hardest implications. Obviously, (a’) implies

(a), and (c’) implies both (c) and (c”).
We now show that (c) implies (c’). Suppose we have a point p in an open subset
U of X. We seek an affine Xf containing p and contained in U. By shrinking U, we
may assume that U is affine. From (c), U contains some Xf containing p. But this Xf
is affine, as it is the complement of the vanishing locus of a section of a line bundle
on an affine scheme (Exercise 8.3.E), so (c’) holds. Note for future reference that the
equivalence of (c) and (c’) did not require the hypothesis of properness.
We next show that (a) implies (c). We embed X in projective space by some
power of L . Given a closed subset Z ⊂ X, and a point p of the complement X \ Z,
we seek a section of some L ⊗N that vanishes on Z and not on p. The existence of
such a section follows from the fact that V(I(Z)) = Z (Exercise 4.5.K(c)): there is
some (homogeneous) element of I(Z) that does not vanish on p. (This also shows
that (a) implies (c’) as well, but we won’t need it.)
We next show that (b) implies (c). Suppose we have a point p in an open subset
U of X. We seek a section of L ⊗N that doesn’t vanish at p, but vanishes on X \ U.
Let I be the sheaf of ideals of functions vanishing on X \ U (the quasicoherent
sheaf of ideals cutting out X \ U, with reduced structure). As X is Noetherian, I is
finite type, so by (b), I ⊗ L ⊗N is generated by global sections for some N, so there
 © 2024 Ravi Vakil. Published by Princeton University Press. 463

F IGURE 16.2. Visual aid for the proof of Theorem 16.2.2

is some global section of it not vanishing at p. (Noetherian note: This is the only part
of the argument where we use Noetherian hypotheses. They can be removed as
follows. Show that for a quasicompact quasiseparated scheme, every ideal sheaf is
generated by its finite type subideal sheaves. Indeed, any quasicoherent sheaf on a
quasicompact quasiseparated scheme is the union of its finite type quasicoherent
subsheaves. This can be proved by hand, and is a suitable but hard exercise, see for
example [Stacks, tag 01PJ]. One of these finite type ideal sheaves doesn’t vanish at
p; use this as I instead.)
We now have to start working harder.
We next show that (c”) implies (b).
We wish to show that F ⊗ L ⊗n is globally generated for n ≫ 0. We first show
that (c”) implies that for some N > 0, L ⊗N is globally generated, as follows. Use
quasicompactness of X to select a finite number ofQaffine Xf1 , . . . , Xfn , that cover
⊗( Nj )/Ni
X, where fi ∈ Γ (X, L ⊗Ni ). By replacing fi with fi j , we
Q may assume that
they are all sections of the same power L ⊗N of L (taking N = j Nj ). Then L ⊗N
is generated by these global sections.
We next show that it suffices to show that for all finite type quasicoherent
sheaves F , F ⊗ L ⊗mN is globally generated for m ≫ 0. For if we knew this, we
could apply it to F , F ⊗ L , . . . , F ⊗ L ⊗(N−1) (a finite number of times), and the
result would follow. For this reason, we can replace L by L ⊗N . In other words, to
show that (c”) implies (b), we may also assume the additional hypothesis that L is
globally generated.
For each closed point p, choose an affine open neighborhood of the form Xf ,
using (c”). Then F |Xf is generated by a finite number of global sections (Easy
Exercise 16.1.B — a finite type quasicoherent sheaf on Spec A corresponds to a
finitely generated A-module). By Exercise 15.4.N, each of these generators can be
expressed as a quotient of a section (over X) of F ⊗ L ⊗M(p) by fM(p) . (Note: we
can take a single M(p) for each p.) Then F ⊗ L ⊗M(p) is globally generated at p by
a finite number of global sections. By Exercise 16.1.D(b), F ⊗ L ⊗M(p) is globally
generated at all points in some open neighborhood Up of p. As L is also globally
′
generated, this implies that F ⊗ L ⊗M is globally generated at all points of Up for
M ′ ≥ M(p) (cf. Easy Exercise 16.1.C). From quasicompactness of X, a finite number
of these Up cover X, so we are done (by taking the maximum of these M(p)).
464 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Our penultimate step is to show that (c”) implies (a). We assume (c”). Our
goal is to find sections of some L ⊗N that embed X into projective space. Choose a
cover of (quasicompact) X by n affine open subsets Xa1 , . . . , Xan , where a1 , . . . , an
are all sections of powers of L . By replacing each section with a suitable power,
we may assume that they are all sections of the same power of L , say L ⊗N . Say
Xai = Spec Ai , where (using that π is finite type) Ai = Spec A[ai1 , . . . , aiji ]/Ii . By
m
Exercise 15.4.N, each aij is of the form sij /ai ij , where sij ∈ Γ (X, L ⊗Nmij ) (for
m−mij
some mij ). Let m = maxi,j mij . Then for each i, j, aij = (sij ai )/ami . For
m m−mij
convenience, let bi = ai , and bij = sij ai ; these are all global sections of
L ⊗mN . Now consider the linear series generated by the bi and bij . As the D(bi ) =
Xai cover X, this linear series is base-point-free, and hence (by Exercise 15.2.A)
gives a morphism to PQ (where Q = #bi + #bij − 1). Let x1 , . . . , xn , . . . , xij , . . . be
the projective coordinates on PQ , so f∗ xi = bi , and f∗ xij = bij . Then the morphism
of affine schemes Xai → D(xi ) is a closed embedding, as the associated maps of
rings is a surjection (the generator aij of Ai is the image of xij /xi ).
At this point, we note for future reference that we have shown the following. If
X → Spec A is finite type, and L satisfies (c”), then X is a locally closed embedding
into a projective A-scheme. (We did not use separatedness.) We conclude our proof
that (c”) implies (a) by using Exercise 11.4.C (“image of proper is proper”) to show
that the image of this locally closed embedding into a projective A-scheme is in fact
closed, so X is a projective A-scheme. (Do you see why a locally closed embedding
with closed image is in fact a closed embedding?)
Finally, we note that (a) and (b) together imply (a’): if L ⊗N is very ample (from
(a)), and L ⊗n is base-point-free for n ≥ n0 (from (b)), then L ⊗n is very ample for
n ≥ n0 + N by Exercise 16.2.C. □

16.2.4. ⋆⋆ Semiample line bundles. Just as an invertible sheaf is ample if some tensor
power of it is very ample, an invertible sheaf L is said to be semiample if some
tensor power of it is base-point-free. (Translation: L is ample if some power gives
a closed embedding into projective space, and L is semiample if some power gives
just a morphism to projective space.) We won’t use this notion.

16.2.5. ⋆ Ampleness in the absolute setting. (We will not use this section in
any serious way later.) Note that global generation is already an absolute notion,
i.e., is defined for a quasicoherent sheaf on a scheme, with no reference to any
morphism. An examination of the proof of Theorem 16.2.2 shows that ampleness
may similarly be interpreted in an absolute setting. We make this precise. Suppose
L is an invertible sheaf on a quasicompact scheme X. We say that L is ample if as
f runs over the sections of L ⊗n (n > 0), the open subsets Xf = {p ∈ X : f(p) ̸= 0}
form a base for the topology of X. (We emphasize that quasicompactness in X is
part of the condition of ampleness of L .) For example, (i) if X is an affine scheme,
every invertible sheaf is ample, and (ii) if X is a projective A-scheme, O(1) is ample.

16.2.L. E ASY E XERCISE ( PROPERTIES OF ABSOLUTE AMPLENESS ).

(a) Fix a positive integer n. Show that L is ample if and only if L ⊗n is ample.
(b) Show that if Z ,→ X is a closed embedding, and L is ample on X, then L |Z is
ample on Z.
 © 2024 Ravi Vakil. Published by Princeton University Press. 465

The following result will give you some sense of how ampleness behaves. We
will not use it, and hence omit the proof (which is given in [Stacks, tag 01Q3]).
However, many parts of the proof are identical to (or generalize) the corresponding
arguments in Theorem 16.2.2. The labeling of the statements parallels the labelling
of the statements in Theorem 16.2.2.

16.2.6. Theorem (cf. Theorem 16.2.2). — Suppose L is an invertible sheaf on a

quasicompact scheme X. Then the following are equivalent.
(b) X is quasiseparated, and for every finite type quasicoherent sheaf F , there is an
n0 such that for n ≥ n0 , F ⊗ L ⊗n is globally generated.
(c) As f runs over the global sections of L ⊗n (n > 0), the open subsets Xf = {p ∈
X : f(p) ̸= 0} form a base for the topology of X (i.e., L is ample).
(c’) As f runs over the global sections of L ⊗n (n > 0), those open subsets Xf which
are affine form a base for the topology of X.
(c”) As f runs over the global sections of L ⊗n (n > 0), those open subsets Xf which
are affine cover X.
(d) Let S• be the graded ring ⊕n≥0 Γ (X, L ⊗n ). Then the open sets Xf (with f ∈ S+
homogeneous of positive degree) cover X, and the associated map X → Proj S• is
an open embedding. (Warning: S• need not be finitely generated, and Proj S• is
not necessarily finite type.)

Part (d) implies that X is separated (and thus quasiseparated).

16.2.7. ⋆ Transporting global generation, base-point-freeness, and ampleness to

the relative situation.
These notions can be “relativized”. We could do this right now, but we wait
until §17.3.5, when we will have defined the notion of a projective morphism, and
thus a “relatively very ample” line bundle.

16.3 Applications to curves

We now apply what we have learned to curves. (The only reason this discussion
comes so late is because we need Exercise 16.2.G, which implies that if X → Y is
finite, and Y is projective over k, then X is projective over k too.) Because the word
“curve” can have different meanings, we will try to be pedantic about hypotheses
in statements of theorems. (In common usage, a curve often means a variety over a
field k of pure dimension one.)

16.3.1. Theorem (every integral curve has a birational model that is regular and
projective). — If C is an integral finite type one-dimensional k-scheme, then there exists
a regular projective k-variety C ′ (of dimension 1) birational to C.

Proof. We can assume C is affine. By the Noether Normalization Lemma 12.2.4 , we

can find some x ∈ K(C)\k with K(C)/k(x) a finite extension of fields. By identifying
a standard open subset of P1k with Spec k[x], and taking the normalization of P1k
in the function field K(C) (defined in Exercise 10.7.I), we obtain a finite morphism
C ′ → P1k , where C ′ is a curve (finiteness by Exercise 10.7.M, dim C ′ = dim P1k by
Exercise 12.1.G), and regular (it is reduced hence regular at the generic point, and
466 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

F IGURE 16.3. Constructing a projective regular model of a curve

C over k via a finite cover of P1

regular at the closed points by the main theorem on discrete valuation rings in §13.5).
Also, C ′ is birational to C as they have isomorphic function fields (Exercise 7.5.E).
Finally, C ′ → P1k is finite, and P1k → Spec k is projective, so by Exercise 16.2.G,
C → Spec k is projective.
′
□

16.3.2. Theorem. — If C is an irreducible, regular, one-dimensional k-variety, then there

is an open embedding C ,→ C ′ into some projective regular curve C ′ (over k).

Proof. We first prove the result in the case where C is affine. As in the proof of
Theorem 16.3.1, we have the diagram

C C′
π finite π ′ finite
  
 / P1
A1k k

where π ′ is the normalization of P1k in the function field extension K(C)/K(A1k ),

and C ′ is projective. By the universal property of normalization in a function field
extension, the normalization of A1k in the function field extension K(C)/K(A1k ) is
C itself, as C is already normal. Thus (π ′ )−1 (A1k ) = C, yielding the desired open
embedding C → C ′ .
We next consider the case of general C. Let C1 by any nonempty affine open
subset of C. By the discussion in the previous paragraph, we have a regular pro-
jective compactification C e . The Curve-to-Projective Extension Theorem 15.3.1
1
(applied successively to the finite number of points C \ C1 ) implies that the mor-
phism C ,→ C e extends to a birational morphism C → C e . Because points of a
1 1 1
separated regular curve are determined by their valuation (Exercise 13.7.B), this is
an inclusion of sets. Because the topology on curves is the “cofinite topology”, it
 © 2024 Ravi Vakil. Published by Princeton University Press. 467

expresses C as an open subset of C

e . But we are not done yet!
1

C _ 1  p

 
C1
e

We must show that C → C e is an open embedding of schemes. We show it is

1
an open embedding near a point p ∈ C as follows. Let C2 be an affine open
neighborhood of p in C. We repeat the construction we used on C1 , to obtain the
following diagram, with open embeddings marked.
C _ 1  p Nn C2 _

~
C

   
C1
e C
e
2

By the Curve-to-Projective Extension Theorem 15.3.1, the map C1 → C e extends

2

to π12 : C1 → C2 , and we similarly have a morphism π21 : C2 → C1 , extending

e e e e

C →C
2
e . The composition π ◦ π is the identity morphism (as it is the identity
1 21 12
rational map, see Theorem 11.3.2). The same is true for π12 ◦ π21 , so π12 and π21
are isomorphisms. The enhanced diagram
C _ 1  p Nn C2 _

~
C

   
e o ∼ /C
C 1
e
2

commutes (by Theorem 11.3.2 again, implying that morphisms of reduced separated
schemes are determined by their behavior on dense open sets). But C → C e is
2 1

an open embedding (in particular, at p), so C → C

e is an open embedding at p as
1
well. □

16.3.A. E XERCISE . Show that all regular proper curves over k are projective.

16.3.3. Theorem (various categories of curves are the same). — The following
categories are equivalent.
(i) integral regular projective one-dimensional k-varieties, and surjective k-morphisms.
(ii) integral regular projective one-dimensional k-varieties, and dominant k-morphisms.
468 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

(iii) integral regular projective one-dimensional k-varieties, and dominant rational

maps over k.
(iv) integral one-dimensional k-varieties, and dominant rational maps over k.
(v) the opposite category (defined in §1.1.16) of finitely generated fields of transcen-
dence degree 1 over k, and k-homomorphisms.
All morphisms and maps in the following discussion are assumed to be defined
over k.
(Aside: The interested reader can tweak the proof below to show the following
variation of the theorem: in (i)–(iv), consider only geometrically irreducible curves,
and in (v), consider only fields K such that K ∩ ks = k in K. This variation allows
us to exclude “weird” curves we may not want to consider. For example, if k = R,
then we are allowing `curves such as P1C which are not geometrically irreducible, as
1 ∼ 1
PC ×R Spec C = PC PC .) 1

Proof. Every surjective morphism is a dominant morphism, and every dominant

morphism is a dominant rational map, and each integral regular projective curve is
a quasiprojective curve, so we have shown (informally speaking) how to get from
(i) to (ii) to (iii) to (iv). To get from (iv) to (i), suppose we have a dominant rational
map C1 99K C2 of integral curves. Replace C1 by a dense open set so the rational
map is a morphism C1 → C2 . This induces a map of normalizations C e1 → Ce2 of
regular irreducible curves. Let C ei be a regular projective compactification of C
ei (for
e1 → C
i = 1, 2), as in Theorem 16.3.2. Then the morphism C e2 extends to a morphism
C1 → C2 by the Curve-to-Projective Extension Theorem 15.3.1. This morphism is
e e
surjective (Do you see why?), so we have produced a morphism in category (i).

16.3.B. E XERCISE . Put the above pieces together to describe equivalences of

categories (i) through (iv).
It remains to connect (v). This is essentially the content of Exercise 7.5.E; details
are left to the reader. □
Theorem 16.3.3 has a number of implications. For example, each quasiprojective
reduced curve is birational to precisely one projective regular curve. Here is another
interesting consequence.

16.3.4. Important notion: Degree of a projective morphism from a curve to a

regular curve.
You might already have a reasonable sense that a map of compact Riemann
surfaces has a well-behaved degree, that the number of preimages of a point of
C ′ is constant, so long as the preimages are counted with appropriate multiplicity.
For example, if f locally looks like z 7→ zm = y, then near y = 0 and z = 0 (but
not at z = 0), each point has precisely m preimages, but as y goes to 0, the m
preimages coalesce. Enlightening Example 10.3.4 showed this phenomenon in a
more complicated context.
We now show the algebraic version of this fact. Suppose π : C → C ′ is a
surjective (or equivalently, dominant) map of regular projective curves. We will
show that π has a well-behaved degree, in a sense that we will now make precise.
First we show that π is finite. Theorem 18.1.6 (finite = projective + finite fibers)
implies this, but we haven’t proved it yet. So instead we show the finiteness of
 © 2024 Ravi Vakil. Published by Princeton University Press. 469

π as follows. Let C ′′ be the normalization of C ′ in the function field of C. Then

we have an isomorphism K(C) = ∼ K(C ′′ ) (really, equality) which leads to birational
maps C o / C ′′ which extend to morphisms as both C and C ′′ are regular and
projective (by the Curve-to-Projective Extension Theorem 15.3.1). Thus this yields
an isomorphism of C and C ′′ . But C ′′ → C ′ is a finite morphism by the finiteness
of normalization (Exercise 10.7.M).

16.3.5. Proposition. — Suppose that π : C → C ′ is a finite morphism, where C ′ is a

regular Noetherian scheme of pure dimension one, and C is a Noetherian scheme of pure
dimension one with no embedded points (the most important case: C is reduced). Then
π∗ OC is locally free of finite rank.
The “no embedded points” hypothesis is the same as requiring that every
associated point of C maps to a generic point of (some component of) C ′ . Notice
that if C is reduced, then there are no restrictions on how bad the singularities of C
can be!
We will prove Proposition 16.3.5 in §16.3.11, after showing how useful it is.
The regularity hypothesis on C ′ is necessary: the normalization of a nodal curve
(Figure 8.4) is an example where most points have one preimage, and one point
(the “node”) has two. (We will later see, in Remark 24.3.8 and §24.3.12, that what
matters in the hypotheses of Proposition 16.3.5 is that the morphism is finite and
flat.)

16.3.6. Definition. If C ′ is irreducible, the rank of this locally free sheaf is the degree
of π.

16.3.C. E XERCISE . Recall that the degree of a rational map from one integral curve
to another is defined as the degree of the function field extension (Definition 12.2.2).
Show that (with the notation of Proposition 16.3.5) if C and C ′ are integral, the
degree of π as a rational map is the same as the rank of π∗ OC .

16.3.7. Remark for those with complex-analytic background (algebraic degree = analytic
degree). If C → C ′ is a finite map of regular complex algebraic curves, Proposi-
tion 16.3.5 establishes that the algebraic degree as defined above is the same as the
analytic degree (counting preimages, with multiplicity).

16.3.D. E XERCISE . We continue the notation and hypotheses of Proposition 16.3.5.

Suppose p is a closed point of C ′ . The scheme-theoretic preimage π−1 (p) of p is a
dimension 0 scheme over k.
(a) Suppose C ′ is finite type over a field k, and n is the dimension of the k-
vector space of global sections of the structure sheaf of π−1 (p). Show that n =
(deg π)(deg p). (The degree of a point was defined in §5.3.8.)
(b) Suppose that C is regular, and π−1 (p) = {p1 , . . . , pm }. Suppose t is a uniformizer
of the discrete valuation ring OC ′ ,p . Show that

X
m
deg π = (valpi π∗ t) deg(κ(pi )/κ(p)),
i=1

where deg(κ(pi )/κ(p)) denotes the degree of the field extension of the residue
fields. (Can you extend (a) to remove the hypotheses of working over a field? If
470 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

you are a number theorist, can you recognize (b) in terms of splitting prime ideals
in extensions of rings of integers in number fields?)

16.3.E. E XERCISE . Suppose that C is an irreducible regular projective curve over

k, and s is a nonzero rational function on C. Show that the number of zeros of s
(counted with appropriate multiplicity) equals the number of poles. Hint: recognize
this as the degree of a morphism s : C → P1k . Use Exercise 16.3.D(b) for 0 ∈ P1 and
∞ ∈ P1 . (In the complex category, this is an important consequence of the Residue
Theorem.)

16.3.8. ⋆⋆ Important definition we could make now (although we wait until §18.4.2).
At this point, we can easily make the important definition of the degree of a line
bundle on a smooth projective curve. Feel free to think this through now, although
we wait until §18.4.2 to make a (seemingly different) definition, and show it is
equivalent to this in Important Easy Exercise 18.4.C. Suppose C is an irreducible
regular projective curve over k, and L is a line bundle on C. Then L has a nonzero
rational section s (do you see why?). If s and t are any two nonzero rational sections,
then by Exercise 16.3.E, for the the rational function s/t, the number of zeros equals
the number of poles (counted correctly). Thus the number of zeros minus the
number of poles (counted correctly) of a nonzero rational section s is independent
of the choice of s. This integer could be taken to be the definition of the degree of
the line bundle L on the curve C. (Because we are not taking this as our “official
definition”, we do not write this phrase in bold.)

16.3.9. Remark. In Exercise 18.4.E, we will see that the number of zeros (and poles)
in the previous exercise is the degree of the line bundle giving the map to P1 (via
Important Theorem 15.2.2).

16.3.10. Revisiting Example 10.3.4. Proposition 16.3.5 and Exercise 16.3.D make
precise what general behavior we observed in Example 10.3.4. Suppose C ′ is
irreducible, and that d is the rank of this allegedly locally free sheaf. Then the fiber
over any point of C ′ with residue field K is the Spec of an algebra of dimension d
over K. This means that the number of points in the fiber, counted with appropriate
multiplicity, is always d.
As a motivating example, we revisit Example 10.3.4, the map Q[y] → Q[x]
given by x 7→ y2 , the projection of the parabola x = y2 to the x-axis. We observed
the following.
(i) The fiber over x = 1 is Q[y]/(y2 − 1), so we get 2 points.
(ii) The fiber over x = 0 is Q[y]/(y2 ) — we get one point, with multiplicity 2,
arising because of the nonreducedness.
(iii) The fiber over x = −1 is Q[y]/(y2 + 1) = ∼ Q(i) — we get one point, with
multiplicity 2, arising because of the field extension.
(iv) Finally, the fiber over the generic point Spec Q(x) is Spec Q(y), which is
one point, with multiplicity 2, arising again because of the field extension
(as Q(y)/Q(x) is a degree 2 extension).
We thus see three sorts of behaviors ((iii) and (iv) are really the same). Note that
even if you only work with algebraically closed fields, you will still be forced to
deal with this third type of behavior, because residue fields at generic points are
usually not algebraically closed (witness case (iv) above).
 © 2024 Ravi Vakil. Published by Princeton University Press. 471

16.3.11. Proof of Proposition 16.3.5. The key idea, useful in other circumstances, is to
reduce to a fact about discrete valuation rings.
The question is local on the target, so we may assume that C ′ is affine. Now C ′
is normal, so by Exercise 5.4.B, its connected components are irreducible. Replacing
C ′ by each of its irreducible components, we may also assume C ′ is integral.
By Exercise 14.3.K, if the rank of the finite type quasicoherent sheaf π∗ OC is
constant, then (as C ′ is reduced) π∗ OC is locally free. We will show this by showing
the rank at any closed point p of C ′ is the same as the rank at the generic point.
(This is another example of the usefulness of the generic point.)
Suppose C ′ = Spec A ′ , where A ′ is an integral domain, and p = [m]. As π is an
affine morphism, C is an affine scheme as well; say C = Spec A.
We wish to show that (i) dimA ′ /m (A/m) (the rank of π∗ OC at p) equals (ii)
dimK(A ′ ) (A ′× )−1 A (the rank of π∗ OC at the generic point). In other words, we
take A (considered as an A ′ -module), and (i) quotient by m, and (ii) invert all
nonzero elements of A ′ , and in each case compute the result’s dimension over the
appropriate field.
Both (i) and (ii) factor through localizing at m, so it suffices to show that Am is
′
a finite rank free Am -module, of rank d, say, as the answers to both (i) and (ii) will
then be d.
′
Now Am is a discrete valuation ring; let t be its uniformizer. We can assume that
t ∈ A (as otherwise, we replace A ′ by a suitable localization, at the “denominators”
′
′
of t). Then Am is a finitely generated Am -module, and hence by Remark 6.4.4 is
′ ′
a finite sum of principal modules, of the form Am or Am /(tn ) (for various n). We
wish to show that there are no summands of the latter type. But if there were,
then t (interpreted as an element of Am ) would be a zerodivisor of Am , and thus
(interpreted as an element of A) a zerodivisor of A. But then by Proposition 6.6.13,
there is an associated point of C in π−1 (p), contradicting the hypotheses that C has
no embedded points. □

16.4 ⋆ The Grassmannian as a moduli space

(In this section and others, the symbol k is used both for a field, and for a
dimension. For this we apologize, and hope it does not cause too much confusion.)
We first defined projective space inelegantly in §4.4.9, and in §15.2.3 we gave
a clean (if perhaps surprising) functorial definition. Similarly, in §7.7, we gave a
preliminary description of the Grassmannian. We are now in a position to give
a better definition. Before proceeding, you should first go back and read the
doubled-starred §10.1.7, if you have not done so already.
We describe the “Grassmannian functor” (which we also denote G(k, n)), then
show that it is representable (§7.6.2). The construction works over an arbitrary base
scheme, so we work over the final object Spec Z. (You should think through what
to change if you wish to work with, for example, complex schemes.) The functor is
defined as follows. To a scheme B, we associate the set of locally free rank k quotients
of the rank n free sheaf,

(16.4.0.1) OB⊕n //Q

472 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

up to isomorphism. An isomorphism of two such quotients ϕ : OB⊕n → Q → 0 and

∼
ϕ ′ : OB⊕n → Q ′ → 0 is an isomorphism σ : Q −→ Q ′ such that the diagram

O ⊕n
ϕ
/Q

σ
ϕ′ " 
Q′
commutes. By Exercise 14.2.H(a), ker ϕ is locally free of rank n − k. (Thus if you
prefer, you can extend (16.4.0.1), and instead consider the functor to take B to short
exact sequences
(16.4.0.2) 0 /S / O ⊕n /Q /0
of locally free sheaves over B, of ranks n − k, n, and k respectively.)
It may surprise you that we are considering rank k quotients of a rank n
sheaf, not rank k subobjects, given that the Grassmannian should parametrize
k-dimensional subspace of an n-dimensional space. This is done for several rea-
sons. One is that the kernel of a surjective map of locally free sheaves must be
locally free, while the cokernel of an injective map of locally free sheaves need
not be locally free (Exercise 14.2.H(a) and (b) respectively). (If you prefer, rather
than surjections of finite rank locally free sheaves, you can equivalently consider
injections of finite rank locally free sheaves that are “injections of vector bundles”
in the sense of Definition 14.2.4.) Another reason: we will see in §25.3.3 that the
geometric incarnation of this problem indeed translates to this. We can already
see a key example here: if k = 1, our definition yields one-dimensional quotients
O ⊕n → L → 0. But this is precisely the data of n sections of L , with no common
zeros, which by Theorem 15.2.2 (the functorial description of projective space)
corresponds precisely to maps to Pn−1 , so the k = 1 case parametrizes what we
want.
We now show that the Grassmannian functor is representable for given n and
k.
So far we have described what the Grassmannian functor does on objects B; to
fully describe a functor, we need to say what it does to morphisms B1 → B2 . Can
you see what the definition must be?

16.4.A. E XERCISE . Show that the Grassmannian functor is a Zariski sheaf (§10.1.8).
Hence by Key Exercise 10.1.H, to show that the Grassmannian functor is repre-
sentable, we need only cover it with open subfunctors that are representable.
Throughout the rest of this section, a k-subset is a subset of {1, . . . , n} of size k.

16.4.B. E XERCISE .
(a) Suppose I is a k-subset. Make the following statement precise: there is an
open subfunctor G(k, n)I of G(k, n) where the k sections of Q corresponding to
I (of the n sections of Q coming from the surjection ϕ : O ⊕n → Q) are linearly
independent. Hint: in a trivializing open neighborhood of Q, where we can choose
∼ /
an isomorphism Q O ⊕k , ϕ can be interpreted as a k × n matrix M, and this
locus is where the determinant of the k × k matrix consisting of the I columns of
M is nowhere zero. Show that this locus behaves well under transitions between
trivializations.
 © 2024 Ravi Vakil. Published by Princeton University Press. 473

(b) Show that these open subfunctors G(k, n)I cover the functor G(k, n) (as I runs
through the k-subsets).
Hence by Exercise 10.1.H, to show G(k, n) is representable, we need only show
that G(k, n)I is representable for arbitrary I. After renaming the summands of O ⊕n ,
without loss of generality we may assume I = {1, . . . , k}.

16.4.C. E XERCISE . Show that G(k, n){1...,k} is represented by Ak(n−k) as follows.

(You will have to make this precise.) Given a surjection ϕ : O ⊕n → Q, let ϕi : O →
Q be the map from the ith summand of O ⊕n . (Really, ϕi is just a section of Q.) For
the open subfunctor G(k, n)I , show that

ϕ1 ⊕ · · · ⊕ ϕk : O ⊕k /Q

is an isomorphism. For a scheme B, the bijection G(k, n)I (B) ↔ Hom(B, Ak(n−k) )
is given as follows. Given an element ϕ ∈ G(k, n)I (B), for j ∈ {k + 1, . . . , n},
ϕj = a1j ϕ1 + a2j ϕ2 + · · · + akj ϕk , where aij are functions on B. But k(n − k)
functions on B is the same as a map to Ak(n−k) (Exercise 7.6.E). Conversely, given
k(n − k) functions aij (1 ≤ i ≤ k < j ≤ n), define a surjection ϕ : O ⊕n → O ⊕k as
follows: (ϕ1 . . . , ϕk ) is the identity, and ϕj = a1j ϕ1 + a2j ϕ2 + · · · + akj ϕk for j > k.

You have now shown that G(k, n) is representable, by covering it with n

k
copies of Ak(n−k) . (You might wish to relate this to the description you gave in
§7.7.) In particular, the Grassmannian over a field is smooth, and irreducible of
dimension k(n − k). (The Grassmannian over any base is smooth over that base,
k(n−k)
because AB → B is smooth, see §13.6.2.)

16.4.1. The universal exact sequence over the Grassmannian. Note that we have a
tautological exact sequence

0 /S / O ⊕n /Q / 0.

The bundle Q is called the tautological quotient bundle on the Grassmannian, and
S is called the tautological subbundle. (Exercise 1.2.Z(b) may give some inkling
as to why this might be called the “universal exact sequence”.)

16.4.2. The Plücker embedding.

By applying ∧k to a surjection ϕ : O ⊕n → Q (over an arbitrary base B), we get
a surjection
/ det Q
n
∧k ϕ : O ⊕( k )
(Exercise 14.2.N). But a surjection from a rank N free sheaf to a line bundle is the
same as a map to PN−1 (Theorem 15.2.2).
n
16.4.D. E XERCISE . Use this to describe a map P : G(k, n) → P( k )−1 . (This is just a
tautology: a natural transformation of functors induces a map of the representing
schemes. This is Yoneda’s Lemma, although if you didn’t do Exercise 1.2.Z, you
may wish to do this exercise by hand. But once you do, you may as well go back
to prove Yoneda’s Lemma and do Exercise 1.2.Z, because the argument is just the
same!)
474 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

n
16.4.E. E XERCISE . The projective coordinate xI on P( k )−1 corresponding to the Ith
n
factor of O ⊕( k ) may be interpreted as the determinant of the map ϕI : O ⊕k → Q,
where the O ⊕k consists of the summands of O ⊕n corresponding to I. Make this
precise.
n
16.4.F. E XERCISE . Show that the standard open set UI of P( k )−1 corresponding to
k-subset I (i.e., where the corresponding coordinate xI doesn’t vanish) pulls back
to the open subscheme G(k, n)I ⊂ G(k, n). Denote this map PI : G(k, n)I → UI .

16.4.G. E XERCISE . Show that PI is a closed embedding as follows. We may

deal with the case I = {1, . . . , k}. Note that G(k, n)I is affine — you described it
Spec Z[aij ]1≤i≤k<j≤n in Exercise 16.4.C. Also, UI is affine, with coordinates xI ′ /I ,
as I ′ varies over the other k-subsets. (The case k = 1 is the useful notation xi/j we
used to name coordinates on charts for projective space, §4.4.9.) You want to show
that the map

PI♯ : Z[xI ′ /I ]I ′ ⊂{1,...,n},|I ′ |=k} /(xI/I − 1) / Z[aij ]1≤i≤k<j≤n

is a surjection. By interpreting the map ϕ : O ⊕n → O ⊕k as a k × n matrix M whose

left k columns are the identity matrix and whose remaining entries are aij (1 ≤ i ≤
k < j ≤ n), interpret PI♯ as taking xI ′ /I to the determinant of the k × k submatrix
corresponding to the columns in I ′ . For each (i, j) (with 1 ≤ i ≤ k < j ≤ n), find
some I ′ so that xI ′ /I 7→ ±aij . (Let I ′ = {1, . . . , i − 1, i + 1, . . . , k, j}.)
n
Hence G(k, n) → P( k )−1 is a closed embedding, so G(k, n) is projective over Z.

16.4.H. U NIMPORTANT E XERCISE . As an entertaining geometric consequence: if V

is a vector space over a field, show that the “pure tensors in ∧k V are pure in exactly
one way”: if v1 ∧ · · · ∧ vk = w1 ∧ · · · ∧ wk ̸= 0 in ∧k V, show that there is a k × k
matrix of determinant 1 relating the vi to the wi .
n
16.4.3. The Plücker equations. The equations of G(k, n) → P( k )−1 are particularly
nice. There are quadratic relations among the k × k minors of a k × (n − k) matrix,
called the Plücker equations. By our construction, they are equations satisfied by
G(k, n). It turns out that these equations cut out G(k, n), and in fact generate the
homogeneous ideal of G(k, n), but this takes more work (see [F1, §9.1, Lem. 1]). We
explore this in one example.

16.4.I. E XERCISE . Suppose v1 , v2 , v3 , and v4 are four vectors in a two-dimensional

vector space V over some field. Show that
(v1 ∧ v2 )(v3 ∧ v4 ) − (v1 ∧ v3 )(v2 ∧ v4 ) + (v1 ∧ v4 )(v2 ∧ v3 ) = 0.
The tricky part of this exercise is figuring out how to interpret multiplication (of
Plücker coordinates) here.

16.4.J. E XERCISE . Note that the Plücker embedding embeds the Grassmannian
G(2, 4) into P5 .
(a) Show that G(2, 4) is cut out by the quadratic equation
x12 x34 − x13 x24 + x14 x23 = 0.
 © 2024 Ravi Vakil. Published by Princeton University Press. 475

(Hint: Use Exercise 16.4.I to show that the quadratic vanishes on G(2, 4). But that
isn’t enough.)
(b) Show that every smooth quadric hypersurface in P5 over an algebraically closed
field k of characteristic not 2 is isomorphic to the Grassmannian (over k). (For
comparison, every smooth quadric hypersurface in Pk1 is two points; every smooth
quadric hypersurface in P2k is isomorphic to P1k , §7.5.9; and every smooth quadric
hypersurface in P3k is isomorphic to P1k ×k P1k , Example 10.6.2.)

16.4.K. E XERCISE ( RULINGS ON QUADRICS IN P5 ). Suppose k = k. From Re-

mark 9.3.10, we expect two three-dimensional families of planes in G(2, 4) (inter-
preted as a hypersurface in P5 via the Plücker embedding, see Exercise 16.4.I).
One of them may be described as follows: for each point p ∈ P3k , we have a
two-dimensional family of lines through p; this is a plane in G(2, 4). There is a
three-dimensional family of planes corresponding to the choice of p. This is one of
the two rulings. What is the other one? Prove (as rigorously as you can manage,
given what you know) that these are both rulings.

16.4.4. Further discussion.

16.4.L. E XERCISE . Show that the group scheme GLn acts on the Grassmannian
G(k, n). (The action of a group scheme appeared earlier in Exercise 7.6.S(a).) Hint:
this is much more easily done with the language of functors, §7.6, than with the
description of the Grassmannian in terms of patches, §7.7. (Exercise 15.5.F was the
special case of projective space.)

16.4.M. ⋆⋆ E XERCISE (G RASSMANNIAN BUNDLES ). Suppose F is a rank n locally

free sheaf on a scheme X. Define the Grassmannian bundle G(k, F ) over X.
Intuitively, if F is a varying family of n-dimensional vector spaces over X, G(k, F )
should parametrize k-dimensional quotients of the fibers. You may want to define
the functor first, and then show that it is representable. Your construction will
behave well under base change.

16.4.5. (Partial) flag varieties. The discussion here extends without change to partial
flag varieties (§7.7.1), and the interested reader should think this through.
 CHAPTER 17

Projective morphisms, and relative versions of Spec

and Proj

In this chapter, we will use universal properties to define two useful construc-
tions, Spec of a sheaf of algebras R, and Proj of a sheaf of graded algebras S• on
a scheme X. These will both generalize (globalize) our constructions of Spec of
A-algebras and Proj of graded A-algebras. We will see that affine morphisms are
precisely those of the form Spec R → X, and so we will define projective morphisms
to be those of the form Proj S• → X.
In both cases, our plan is to make a notion we know well over a ring work more
generally over a scheme. The main issue is how to glue the constructions over each
affine open subset together. The slick way we will proceed is to give a universal
property, then show that the affine construction satisfies this universal property,
then that the universal property behaves well with respect to open subsets, then
to use the idea that lets us glue together the fibered product (or normalization)
together to do all the hard gluing work. The most annoying part of this plan is
finding the right universal property, especially in the Proj case.

17.1 Relative Spec of a (quasicoherent) sheaf of algebras

Given an A-algebra, R, we can take its Spec to get an affine scheme over
Spec A: Spec R → Spec A. We will now see a universal property description of a
globalization of that notation. Consider an arbitrary scheme X, and a quasicoherent
sheaf of algebras R on it. We will define how to take Spec of this sheaf of algebras,
and we will get a scheme Spec R → X that is “affine over X”, i.e., the structure
morphism is an affine morphism. You can think of this in two ways.

17.1.1. First, and most concretely, for any affine open set Spec A ⊂ X, Γ (Spec A, R)
is some A-algebra; call it R. Then above Spec A, Spec R will be Spec R.

17.1.2. Second, it will satisfy a universal property. We could define the A-scheme
Spec R by the fact that morphisms to Spec R (from an A-scheme W, over Spec A) cor-
respond to maps of A-algebras R → Γ (W, OW ) (this is our old friend Exercise 7.3.G).
The universal property for β : Spec R → X generalizes this. Given a morphism
µ : W → X, the X-morphisms W → Spec R are in functorial (in W) bijection with

This material is published by Princeton University Press as The Rising Sea: Foundations of Algebraic
Geometry by Ravi Vakil. This version is free to view and download for personal use only. Not for
redistribution, resale, or use in derivative works. © 2024 Ravi Vakil.

477
478 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

morphisms α of sheaves of algebras making

(17.1.2.1) OX

~ #
R
α / µ∗ OW

commute. Here the map OX → µ∗ OW is that coming from the map of ringed spaces,
and the map OX → R comes from the OX -algebra structure on R. (For experts: it
needn’t be true that µ∗ OW is quasicoherent, but that doesn’t matter. Non-experts
should completely ignore this parenthetical comment.)
By universal property nonsense, these data determine β : Spec R → X up to
unique isomorphism, assuming that it exists.
Fancy translation: in the category of X-schemes, β : Spec R → X represents the
functor
 / {(α : R → µ∗ OW )}.
(µ : W → X)

17.1.A. E XERCISE . Show that if X is affine, say Spec A, and R = R,

e where R is
an A-algebra, then Spec R → Spec A satisfies this universal property.(Hint: Exer-
cise 7.3.G.)

17.1.3. Proposition. — Suppose β : Spec R → X satisfies the universal property for

(X, R), and i : U ,→ X is an open subscheme. Then β|U : Spec R ×X U = (Spec R)|U →
U satisfies the universal property for (U, R|U ).

Proof. Suppose W is a U-scheme, with structure morphism ν : W → U. A U-

morphism W → Spec R ×X U is the same as an X-morphism W → Spec R (where
we already know by assumption that µ = i ◦ ν : W → X factors through U).
W

%
Spec R ×X U / Spec R

ν β
 
, U  i /X
By the universal property of Spec R, this is the same information as a map R →
µ∗ OW , which by adjointness of (µ∗ , µ∗ ) (§14.5.5) is the same as µ∗ R → OW , which
(from ν∗ (R|U ) = ν∗ i∗ R = µ∗ R) is ν∗ (R|U ) → OW . By adjointness of (ν∗ , ν∗ ), this
is the same as R|U → ν∗ OW , verifying the universal property. □
Combining Exercise 17.1.A and Proposition 17.1.3, we have shown the existence
of Spec R in the case that Y is an open subscheme of an affine scheme, and R is the
restriction to Y of a quasicoherent sheaf of algebras on the affine scheme.

17.1.B. E XERCISE . Show the existence of Spec R in general, following the philoso-
phy of our construction of the fibered product, normalization, and so forth.

17.1.4. We make some quick observations. First, Spec R can be “computed affine-
∼
locally on X”. We also have an isomorphism ϕ : R −→ β∗ OSpec R .
 © 2024 Ravi Vakil. Published by Princeton University Press. 479

17.1.C. E XERCISE . Given an X-morphism γ:

W
γ
/ Spec R

µ β
 |
X
show that the morphism α (in (17.1.2.1)) is the composition

R
ϕ
/ β∗ OSpec R / β∗ γ∗ OW = µ∗ OW .

The Spec construction gives an important way to understand affine morphisms.

Note that Spec R → X is an affine morphism. The “converse” is also true:

17.1.D. E XERCISE . Show that if µ : Z → X is an affine morphism, then we have a

∼
natural isomorphism Z ←→ Spec µ∗ OZ of X-schemes.
Hence we can recover any affine morphism using the Spec construction. More
precisely, a morphism is affine if and only if it is of the form µ : Spec R → X
constructed above.

17.1.E. E XERCISE . Suppose µ : Spec R → X is such a morphism. Show that

the category of quasicoherent sheaves on Spec R is equivalent to the category of
quasicoherent sheaves on X with the structure of R-modules (quasicoherent R-
modules on X). Show that the category of finite type quasicoherent sheaves on
Spec R is equivalent to the category of quasicoherent sheaves on X with the structure
of “finite type” R-modules (finite type quasicoherent R-modules on X).
This is useful if X is quite simple but Spec R is complicated. We will use this
when X =∼ P1 , and Spec R is a more complicated curve, and more generally when
proving Serre duality in Chapter 29.

17.1.F. E XERCISE ( THE FORMATION OF Spec “ COMMUTES WITH BASE CHANGE ”).
Suppose µ : Z → X is any morphism, and R is a quasicoherent sheaf of algebras
∼
on X. Show that there is a natural isomorphism Z ×X Spec R ←→ Spec µ∗ R. (If you
think about it, you will see that this generalizes the statement and possibly proof of
Proposition 17.1.3.)

17.1.5. Definition. An important example of the Spec construction is the total space
of a finite rank locally free sheaf F , which we define to be Spec (Sym• F ∨ ).

17.1.G. E XERCISE . Suppose F is a locally free sheaf of rank n. Show that the total
space of F is a rank n vector bundle, i.e., that given any point p ∈ X, there is an
open neighborhood p ∈ U ⊂ X such that
Spec Sym• F ∨ |U = ∼ An .


U

Show that F is isomorphic to the sheaf of sections of the total space Spec (Sym• F ∨ ).
(Possible hint: use transition functions.) For this reason, the total space is also called
the vector bundle associated to a locally free sheaf F . (Caution: some authors,
e.g., [Stacks, tag 01M2], call Spec (Sym• F ), the dual of this vector bundle, the vector
bundle associated to F .)
480 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

In particular, if F = OX⊕n , then Spec (Sym• F ∨ ) is An

X.
(Aside: you may notice that the construction Spec Sym• can be applied to any
coherent sheaf F (without dualizing, i.e., Spec (Sym• F )). This is sometimes called
the abelian cone associated to F . This concept can be useful, but we won’t need
it.)

17.1.H. E XERCISE ( THE TAUTOLOGICAL BUNDLE ON Pn IS O(−1)). Suppose k is a

field. Define the subset X ⊂ An+1 k × Pnk corresponding to “points of Ak
n+1
on the
corresponding line of Pk ”, so that the fiber of the map π : X → P corresponding
n n

to a point l = [x0 , · · · , xn ] is the line in An+1

k corresponding to l, i.e., the scalar
multiples of (x0 , . . . , xn ). Show that π : X → Pn k is (the line bundle corresponding
to) the invertible sheaf O(−1). (Possible hint: work first over the usual affine open
sets of Pnk , and figure out transition functions.) For this reason, O(−1) is often
called the tautological bundle of Pn k (even over an arbitrary base, not just a field).
(Side Remark: The projection X → An+1 k is the blow-up of An+1k at the “origin”, see
Exercise 10.3.F.)

17.1.6. To read later: The weird functor π!? , an occasional right-adjoint to push-
forward.
Because the next topic is used at two different places (Tricky Exercise 18.8.C
and §29.3) that you may or may not read, there is no other natural place to put this
discussion. But you should not read this now, and instead wait until you need it, as
otherwise it will be completely unmotivated.
We now define a right adjoint π!? to the pushforward π∗ , when π is a finite
morphism. (Caution: the notation π!? is nonstandard, and is introduced only for
the purposes of the arguments we will give.) This is surprising, as we usually
think of π∗ as a right-adjoint (to the pullback π∗ ), not a left adjoint. This can be
seen as roughly analogous to the surprising occasional left-adjoint to the pullback:
extension by zero (§23.4.7).

17.1.7. Definition. The definition is simple but strange. Suppose π : X → Y is a finite

morphism of locally Noetherian schemes. If G is a quasicoherent sheaf on Y, then
Hom Y (π∗ OX , G ) is also a quasicoherent sheaf on Y (as π∗ OX is finitely presented,
Exercise 14.3.A(b)). It furthermore has the structure of a (π∗ OX )-module, which
means that it corresponds to a quasicoherent sheaf on X (by Exercise 17.1.E, which
shows that this correspondence is functorial in G ). Let π!? G be this quasicoherent
sheaf. Then π!? gives a covariant functor QCohY → QCohX , and also CohY → CohX .

17.1.I. E XERCISE . Describe a natural isomorphism of quasicoherent sheaves on Y

∼ / Hom Y (π∗ OX , G ),
π∗ π!? G
functorial in G . Equivalently, describe an isomorphism of functors QCohY →
QCohY :
∼ /
π∗ π!? (·) Hom Y (π∗ OX , ·).

17.1.8. To show that π!? is right-adjoint to π∗ , we study the affine case, or equiv-
alently, the ring-theoretic version. Suppose B → A is a ring morphism, M is an
A-module, and N is a B-module. Note that HomB (A, N) naturally has the structure
of an A-module. Also, M naturally carries the structure of a B-module; when
 © 2024 Ravi Vakil. Published by Princeton University Press. 481

we wish to emphasize its structure as a B-module, we sometimes call it MB (see

Exercise 1.4.E).
Consider the (very strange) map
(17.1.8.1) HomA (M, HomB (A, N)) / HomB (MB , N)

defined as follows. Given m ∈ M, and an element ϕ of HomA (M, HomB (A, N)),
send m to ϕm (1).

17.1.J. E XERCISE .
(a) Show that (17.1.8.1) is a homomorphism of B-modules.
(b) Show that (17.1.8.1) is a bijection. Thus (M 7→ MB , N 7→ HomB (A, N)) is an
adjoint pair ModA ↔ ModB .

17.1.K. E XERCISE . Suppose π : X → Y is a finite morphism of locally Noetherian

schemes.
(a) Explain how the functor ModB → ModA given by N 7→ HomB (A, N) in the
previous exercise is the affine-local description of π!? in Definition 17.1.7.
(b) Show that (π∗ , π!? ) is an adjoint pair between QCohX and QCohY .
(c) Show that (π∗ , π!? ) is an adjoint pair between CohX and CohY .
(d) If F ∈ QCohX and G ∈ QCohY , describe a natural isomorphism
∼ / Hom Y (π∗ F , G )
(17.1.8.2) π∗ Hom X (F , π!? G )
(functorial in F and G ), which affine-locally is the isomorphism (17.1.8.1) described
in Exercise 17.1.J.

17.2 Relative Proj of a (quasicoherent) sheaf of graded algebras

In parallel with the relative version Spec of Spec, we define a relative version of
Proj, denoted Proj (called “relative Proj” or “sheaf Proj”), of a quasicoherent graded
sheaf of algebras (satisfying some hypotheses) on a scheme X. We have already
done the case where the base X is affine, in §4.5.7, using the regular Proj construction
over a ring A. The elegant way to proceed would be to state the right universal
property, and then use this cleverly to glue together the constructions over each
affine, just as we did in the constructions of fibered product, normalization, and
Spec . But because graded rings and graded modules make everything confusing,
we do not do this. Instead we guiltily take a more pedestrian approach. (The
universal property can be made to work, see [Stacks, tag 01O0]. But I recommend
against it — a universal property should make your life easier, not harder.)

17.2.A. E XERCISE (Proj COMMUTES WITH AFFINE BASE CHANGE ). Suppose A → B

is map of rings, and S• is a Z≥0 -graded ring over A.
(a) Give a canonical isomorphism
∼ / (Proj S• ) ×Spec A Spec B
(17.2.0.1) α : ProjB (S• ⊗A B) A

(The subscripts on the Proj are new notation. They are redundant, and just remind
us that we are thinking of ProjA S• as an A-scheme, for example.)
(b) (easy) Suppose X is a projective A-scheme (§4.5.10). Show that X ×Spec A Spec B
482 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

is a projective B-scheme.
(c) Suppose S• is generated in degree 1, so OProjA S• (1) is an invertible sheaf (Exer-
cise 15.1.G). Clearly S• ⊗A B is generated in degree 1 as a B-algebra. Describe an
isomorphism
∼ / ∗ ∗
OProj (S ⊗ B) (1) o
B • A
α γ OProj S• (1), A

where γ is the top morphism in the Cartesian diagram

(ProjA S• ) ×Spec A Spec B

γ
/ Proj S•
A

 
Spec B / Spec A

Possible hint: transition functions.

We now give a general means of constructing schemes over X (from [Stacks,
tag 01LH]), if we know what they should be over any affine open set, and how
these behave under open embeddings of one affine open set into another.

17.2.B. E XERCISE . Suppose we are given a scheme X, and the following data:
(i) For each affine open subset U ⊂ X, we are given some morphism πU : ZU →
U (a “scheme over U”).
(ii) For each (open) inclusion of affine open subsets V ⊂ U ⊂ X, we are given
an open embedding ρU V : ZV ,→ ZU .
Assume this data satisfies:
∼ −1
(a) for each V ⊂ U ⊂ X, ρU V induces an isomorphism ZV −→ πU (V) of
schemes over V, and
(b) whenever W ⊂ V ⊂ U ⊂ X are three nested affine open subsets, ρU W =
ρU
V ◦ ρ V
W .
∼
Show that there exists an X-scheme π : Z → X, and isomorphisms iU : π−1 (U) −→
ZU over each affine open set U, such that for nested affine open sets V ⊂ U, ρU
V
agrees with the composition

/ π−1 (V)  
i−1
ZV
V
/ π−1 (U) iU
/ ZU

Hint (cf. Exercise 4.4.A): construct Z first as a set, then as a topological space, then
as a scheme. (Your construction will be independent of choices. Your solution
will work in more general situations, for example when the category of schemes is
replaced by ringed spaces, and when the affine open subsets are replaced by any
base of the topology.)

17.2.C. I MPORTANT E XERCISE AND D EFINITION (relative Proj ). Suppose S• =

⊕n≥0 Sn is a quasicoherent sheaf of Z≥0 -graded algebras on a scheme X. Over each
∼ U ⊂ X, we have an U-scheme Proj S• (U) → U. Show
affine open subset Spec A = A
that these can be glued together to form an X-scheme, which we call Proj X S• ; we
have a “structure morphism” β : Proj X S• → X. (The structure morphism β is part
of the definition.) Hint/Warning: This problem would be easier if Exercise 17.2.B
required considering only distinguished affine inclusions. But that would make
Exercise 17.2.B notably harder. Instead, use Exercise 17.2.A.
 © 2024 Ravi Vakil. Published by Princeton University Press. 483

By the construction of Exercise 17.2.B, the preimage over any affine open set can
be computed using the original Proj construction. (You may enjoy going back and
giving constructions of Xred , the normalization of X, and Spec of a quasicoherent
sheaf of O-algebras using this idea. But there is a moral price to be paid by giving
up the universal property.)

17.2.1. Ongoing (reasonable) hypotheses on S• : “finite generation in degree 1”.

The Proj construction is most useful when applied to an A-algebra S• satisfying
some reasonable hypotheses (§4.5.6), notably when S• is a finitely generated Z≥0 -
graded A-algebra, and ideally if it is generated in degree 1. For this reason, in the
rest of the book, we will enforce these assumptions on S• , once we make sense
of them for quasicoherent sheaves of algebras. (If you later need to relax these
hypotheses — for example, to keep the finite generation hypothesis but remove the
“generation in degree 1” hypothesis — it will not be too difficult.) Precisely, we now
always require that (i) S0 = OX , (ii) S• is “generated in degree 1”, and (iii) S1 is
finite type. The cleanest way to make condition (ii) precise is to require the natural
map

Sym•OX S1 / S•

to be surjective. Because the Sym• construction may be computed affine-locally

(§14.2.6), we can check generation in degree 1 on any affine cover. The motivation
for these hypotheses is the following construction.

17.2.D. I MPORTANT E XERCISE : O(1) ON Proj S• . If S• is finitely generated in

degree 1 (Hypotheses 17.2.1), construct an invertible sheaf OProj X S• (1) on Proj X S•
that “restricts to OProjA S• (Spec A) (1) over each affine open subset Spec A ⊂ X”.

17.2.E. E XERCISE (“Proj COMMUTES WITH BASE CHANGE ”). Suppose S• is a

quasicoherent sheaf of Z≥0 -graded algebras on X. Let ρ : Z → X be any morphism.
Give a natural isomorphism

∼
(Proj ρ∗ S• , OProj ρ∗ S• (1)) o / (Z ×X Proj S• , ψ∗ OProj S• (1))

where ψ is the “top” morphism in the base change diagram

Z ×X Proj S•
ψ
/ Proj S•

β
 
Z
ρ
/ X.

17.2.2. Definition: π-very ample. Suppose π : X → Y is proper. If L is an invertible

sheaf on X, then we say that L is very ample (with respect to π), relatively very
ample, or (awkwardly) π-very ample if we can write X = Proj Y S• with L = ∼ O(1),
where S• is a quasicoherent sheaf of algebras on Y satisfying Hypotheses 17.2.1
(“finite generation in degree 1”). (The notion of very ampleness can be extended to
more general situations, see for example [Stacks, tag 01VM]. But this is of interest
only to people with esoteric tastes.)
484 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

17.2.F. E XERCISE . Suppose S• is finitely generated in degree 1 (Hypotheses 17.2.1).

Describe a map of graded quasicoherent sheaves ϕ : S• → ⊕∞ n=0 β∗ O(n) (β is the
structure morphism, see Exercise 17.2.C). Hint: Exercise 15.7.C.

17.2.G. E XERCISE . Suppose L is an invertible sheaf on X, and S• is a quasicoherent

sheaf of graded algebras on X generated in degree 1 (Hypotheses 17.2.1). Define
S•′ = ⊕∞ n=0 (Sn ⊗ L
⊗n
). Then S•′ has a natural algebra structure inherited from
S• ; describe it. Give a natural isomorphism of “X-schemes with line bundles”

∼
(Proj S•′ , OProj S•′ (1)) o / (Proj S• , OProj S• (1) ⊗ β∗ L ),

where β : Proj S• → X is the structure morphism. In other words, informally

speaking, the Proj is the same, but the O(1) is twisted by L .

17.2.3. Projectivization of vector bundles.

17.2.4. Definition. If F is a finite rank locally free sheaf on X, then Proj (Sym• F ∨ ) is
called its projectivization, and is denoted PF . (The reason for the dual is the same
as for Spec (Sym• F ∨ ) in Definition 17.1.5.) You can check that this construction
⊕(n+1)
behaves well with respect to base change. Define Pn X := P(OX ). (Then PnSpec A
agrees with our earlier definition of PA , cf. Exercise 4.5.Q, and Pn
n
X agrees with
our earlier usage, see for example the proof of Theorem 11.4.5.) If F is locally
free of rank n + 1, then PF is a projective bundle or Pn -bundle over X. By
Exercise 17.2.G, if F is a finite rank locally free sheaf on X, there is a canonical
∼
isomorphism PF ←→ P(L ⊗ F ).
More generally, if F is a finite type quasicoherent sheaf on X, then one might
define similarly its projectivization Proj (Sym• F ∨ ). Be careful, though. For ex-
ample, if G is a torsion sheaf on an integral scheme, then F ∨ = 0, so with this
definition, PF = ∅. So this isn’t a great notion.

17.2.5. Huge notation caution. There is violent disagreement on whether PF should

be defined as Sym• F or Sym• F ∨ , parallel to the disagreement as to whether
the projectivization of a vector space parametrizes one-dimensional subspaces or
quotients (cf. Exercise 17.2.I). Hence it is safest to avoid the notation PF , or at least
to state at the outset which convention you are following. In this book, we will try to
avoid the notation PF , and will use it only when F is finite rank locally free (“a vector
bundle”), and we will mean it as defined here, Proj Sym• F ∨ .

17.2.6. Example: Ruled surfaces. If C is a regular curve and F is locally free of

rank 2, then PF is called a ruled surface over C. If C is further isomorphic to
P1 , PF is called a Hirzebruch surface. All vector bundles on P1 split as a direct
sum of line bundles (see §18.5.5 for a proof), so each Hirzebruch surface is of the
form P(O(n1 ) ⊕ O(n2 )). By Exercise 17.2.G, this depends only on n2 − n1 . The
Hirzebruch surface P(O ⊕ O(n)) (n ≥ 0) is often denoted Fn . We will discuss the
Hirzebruch surfaces in greater length in §20.2.10. We will see that the Fn are all
distinct in Exercise 20.2.Q.
 © 2024 Ravi Vakil. Published by Princeton University Press. 485

17.2.H. E XERCISE . If S• is finitely generated in degree 1 (Hypotheses 17.2.1),

describe a canonical closed embedding

Proj S•  i / Proj Sym• S1
X

β
" y
X
∼
and an isomorphism OProj S• (1) ←→ i∗ OProj X Sym• S1 (1) arising from the surjection

Sym• S1 / / S• .

In particular, if S1 is locally free, then Proj Sym• S1 = PS1∨ , so we have embedded

Proj S• in a projective bundle on X.
17.2.I. E XERCISE . Suppose F is a locally free sheaf of rank n + 1 on X. Exhibit a
bijection between the set of sections s : X → PF of PF → X and the set of surjective
homomorphisms F → L → 0 of F onto invertible sheaves on X. This functorial
description of PF in some sense generalizes the functorial description of projective
space in §15.2.3.

17.2.7. Remark (the relative version of the projective and affine cone). There is a natural
morphism from Spec S• minus the zero-section to Proj S• (cf. Exercise 9.3.N). Just
as Proj S• [T ] contains a closed subscheme identified with Proj S• whose complement
can be identified with Spec S• (Exercise 9.3.O), Proj S• [T ] contains a closed sub-
scheme identified with Proj S• whose complement can be identified with Spec S• .
You are welcome to think this through.

17.3 Projective morphisms

In §17.1, we reinterpreted affine morphisms: X → Y is an affine morphism if
∼
there is an isomorphism X ←→ Spec B of Y-schemes for some quasicoherent sheaf
of algebras B on Y. We will define the notion of a projective morphism similarly.
You might think that because projectivity is such a classical notion, there should
be some obvious definition, that is reasonably behaved. But this is not the case, and
there are many possible variant definitions of projective (see [Stacks, tag 01W8]).
All are imperfect, including the accepted definition we give here (see the warnings
in §17.3.4).

17.3.1. Definition. A morphism π : X → Y is projective if there is an isomorphism

∼ / Proj S•
X

π
 |
Y
for a quasicoherent sheaf of algebras S• on Y satisfying “finite generation in degree
1” (Hypotheses 17.2.1). We say X is a projective Y-scheme, or X is projective over
Y. Using Exercise 7.4.D, this generalizes the notion of a projective A-scheme.
486 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

17.3.A. E XERCISE .
(a) (useful characterization of projective morphisms) Suppose π : X → Y is a morphism.
Show that π is projective if and only if there exist a finite type quasicoherent sheaf
S1 on Y, and a closed embedding i : X ,→ Proj Y Sym• S1 (over Y, i.e., commuting
with the maps to Y). Hint: Exercise 17.2.H.
(b) (useful characterization of projective morphisms, with line bundle) Suppose L is an
invertible sheaf on X, and π : X → Y is a morphism. Show that π is projective, with
O(1) = ∼ L , if and only if there exist a finite type quasicoherent sheaf S1 on Y, a
closed embedding i : X ,→ Proj Y Sym• S1 (over Y, i.e., commuting with the maps
∼
to Y), and an isomorphism i∗ OProj Y Sym• S1 (1) ←→ L .
(c) Suppose, furthermore, that Y admits an ample line bundle in the sense of §16.2.5,
as is the case whenever Y is projective, affine or, more generally, quasi-projective.
Show that π is projective if and only if there exists a closed embedding X → Pn Y
(over Y) for some n. This is the definition of projective morphism given in [Ha1,
p. 103]. (If you want to avoid the starred section §16.2.5, you can assume that Y is
projective over Spec A and use the definition of ample from §16.2.1. You will then
have dealt with the important case where Y is projective, but missed out on other
potentially interesting cases, such as when Y is affine or otherwise quasiprojective
(but not proper).) Hint: the harder direction is the forward implication. Use the
finite type quasicoherent sheaf S1 from (a). Tensor S1 with a high enough power
of M so that it is finitely globally generated (Theorem 16.2.6, or Theorem 16.2.2 in
the proper setting), to obtain a surjection

OY
⊕(n+1) / / S1 ⊗ M ⊗N .

Then use Exercise 17.2.G.

17.3.2. Properties of projective morphisms.

The property of a morphism being projective is clearly preserved by base
change, as the Proj construction behaves well with respect to base change (Ex-
ercise 17.2.E). Also, projective morphisms are proper: properness is local on the
target (Theorem 11.4.4(b)), and we saw earlier that projective A-schemes are proper
over A (Theorem 11.4.5). In particular (by definition of properness), projective
morphisms are separated, finite type, and universally closed.

17.3.B. I MPORTANT E XERCISE : F INITE MORPHISMS ARE PROJECTIVE ( CF. E XERCISE

8.3.I). Show that finite morphisms are projective as follows. Suppose Z → X is
∼ Spec B where B is a finite type quasicoherent sheaf of algebras
finite, so that Z =
on X. Describe a sheaf of graded algebras S• where S0 = ∼ OX and Sn = ∼ B for
∼
n > 0. Describe an X-isomorphism Z ←→ Proj S• .
In particular, closed embeddings are projective. We have the sequence of
implications for morphisms
closed embedding =⇒ finite =⇒ projective =⇒ proper.

17.3.3. We have shown that finite morphisms are projective (Exercise 17.3.B), and
have finite fibers (Exercise 8.3.J). We will show the converse in Theorem 18.1.6, and
state the extension to proper morphisms immediately after (although we won’t
prove it until Theorem 28.5.2).
 © 2024 Ravi Vakil. Published by Princeton University Press. 487

17.3.4. Warnings about projective morphisms.

First, notice that O(1), an important part of the concept of Proj , is not mentioned
in the definition. (I would prefer that it be part of the definition, but this isn’t
accepted practice.)
Second, [Ha1, p. 103] defines projective morphism differently; we follow the
more general definition of Grothendieck. Although these definitions are the same
in most circumstances (Exercise 17.3.A(c)), you should be cautious when reading
the literature. For example, finite morphisms are not always projective in the sense
of [Ha1], while they are projective with this definition (Exercise 17.3.B).
Third, projective morphisms are not reasonable in the sense of §8.1! Although
projective morphisms are preserved by base change (§17.3.2), the property of being
projective is not local on the target (see §24.7.7 for an example). (With the additional
data of the invertible sheaf O(1), and in Noetherian circumstances, it is, see §17.3.8.)
Furthermore, we will show that projective morphisms are preserved by composition
only when the final target is affine or Noetherian (Exercise 17.3.I).
But many of the consequences of reasonableness still hold, as the next few
exercises show. We first prove that the class of projective morphisms, despite its
unreasonable nature, still satisfies the Cancellation Theorem 11.1.1.

17.3.C. E XERCISE ( THE C ANCELLATION T HEOREM FOR P ROJECTIVE M ORPHISMS ).

(a) Suppose π : X → Y is a closed embedding, and ρ : Y → Z is projective. Show that

ρ ◦ π is projective.
(b) Show that a morphism ρ : Y → Z has projective diagonal if and only if ρ is
separated.
(c) Prove the Cancellation Theorem for Projective Morphisms: Suppose π : X → Y
is a morphism, ρ : Y → Z is a separated morphism (i.e., by (b), has projective
diagonal), and ρ ◦ π is projective. Show that π is projective. (Possible hint: use the
diagram in the hint to Exercise 11.1.A.)

17.3.D. E XERCISE . Show that a morphism (over Spec k) from a projective k-scheme
to a separated k-scheme is always projective. (Hint: the previous exercise.)

17.3.E. E XERCISE . Suppose π : X → Y and π ′ : X ′ → Y ′ are projective morphisms of

S-schemes. Show that π × π ′ : X ×S X ′ → Y ×S Y ′ is also projective. Hint: Because of
the unreasonable nature of projective morphisms, you cannot invoke Exercise 8.1.A.
Use Exercise 10.6.D instead.

17.3.5. ⋆ Global generation and (very) ampleness in the relative setting.

We extend the discussion of §16.1 to the relative setting, in order to give our-
selves the language of relatively base-point-freeness. We won’t use this discussion,
so on a first reading you should skip it. But these ideas come up repeatedly in the
research literature.
Suppose π : X → Y is a quasicompact quasiseparated morphism. If F is a
quasicoherent sheaf on X, we say that F is relatively globally generated or globally
generated with respect to π if the natural map of quasicoherent sheaves π∗ π∗ F →
F is surjective. Quasicompactness and quasiseparatedness are needed ensure
that π∗ F is a quasicoherent sheaf, Exercise 14.4.C. But these hypotheses are not
very restrictive. Global generation is most useful only in the quasicompact setting,
488 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

and most people won’t be bothered by quasiseparated hypotheses. (Unimportant

aside: these hypotheses can be relaxed considerably. If π : X → Y is a morphism
of locally ringed spaces — not necessarily schemes — with no other hypotheses,
and F is a quasicoherent sheaf on X, then we say that F is relatively globally
generated or globally generated with respect to π if the natural map π∗ π∗ F → F
of OX -modules is surjective.)
Thanks to our hypotheses, as the natural map π∗ π∗ F → F is a morphism
of quasicoherent sheaves, the condition of being relatively globally generated is
affine-local on Y.
Suppose now that L is a locally free sheaf on X, and π : X → Y is a morphism.
We say that L is relatively base-point-free or base-point-free with respect to π if
it is relatively globally generated.

17.3.F. E XERCISE . Suppose L is a relatively base-point-free line bundle on X,

π : X → Y is a quasicompact separated morphism, and π∗ L is finite type on Y.
(We will later show in Grothendieck’s Coherence Theorem 18.9.1 that this latter
statement is true if π is proper and Y is Noetherian. This is much easier if π is
projective, see Theorem 18.7.1. We could work hard and prove it now, but it isn’t
worth the trouble.) Describe a canonical morphism ψ : X → Proj Y Sym• (π∗ L ).
(Possible hint: this generalizes the fact that base-point-free line bundles give maps
to projective space, so generalize that argument, see §15.2.)
We say that L is relatively ample or π-ample or relatively ample with re-
spect to π if for every affine open subset Spec B of Y, L |π−1 (Spec B) is ample on
π−1 (Spec B) over B, or equivalently (by §16.2.5), L |π−1 (Spec B) is (absolutely) ample
on π−1 (Spec B). By the discussion in §16.2.5, if L is ample then π is necessarily
quasicompact, and (by Theorem 16.2.6) separated; if π is affine, then all invertible
sheaves are ample; and if π is projective, then the corresponding O(1) is ample.
By Exercise 16.2.L, L is π-ample if and only if L ⊗n is π-ample, and if Z ,→ X is a
closed embedding, then L |Z is ample over Y.
From Theorem 16.2.6(d), we have a natural open embedding X ,→ Proj Y ⊕π∗ L ⊗d .
(Do you see what this map is? Also, be careful: ⊕π∗ L ⊗d need not be a finitely
generated graded sheaf of algebras, so we are using the Proj construction where
one of the usual hypotheses doesn’t hold.)
The notions of relative global generation and relative ampleness are most useful
in the proper setting, because of Theorem 16.2.2.

17.3.6. Many statements of §16.1 carry over without change. For example, we
have the following. Suppose π : X → Y is proper, F and G are quasicoherent
sheaves on X, and L and M are invertible sheaves on X. If π is affine, then
F is relatively globally generated (from Easy Exercise 16.1.B). If F and G are
relatively globally generated, so is F ⊗ G (Easy Exercise 16.1.C). If L is π-very
ample (Definition 17.2.2), then it is π-base-point-free (Easy Exercise 16.2.B). If L is
π-very ample, and M is π-base-point-free (if for example it is π-very ample), then
L ⊗ M is π-very ample (Exercise 16.2.C). Exercise 16.2.G extends immediately to
 © 2024 Ravi Vakil. Published by Princeton University Press. 489

show that if

X
π /Y

τ ρ
 
S

is a finite morphism of S-schemes, and if L is a ρ-ample invertible sheaf on Y, then

π∗ L is τ-ample.
By the nature of the statements, some of the statements of §16.1 require quasi-
compactness hypotheses on Y, or other patches. For example:

17.3.7. Theorem. — Suppose π : X → Y is proper, L is an invertible sheaf on X, and Y is

quasicompact. The following are equivalent.

(a) For some N > 0, L ⊗N is π-very ample.

(a’) For all n ≫ 0, L ⊗n is π-very ample.
(b) For all finite type quasicoherent sheaves F , there is an n0 such that for n ≥ n0 ,
F ⊗ L ⊗n is relatively globally generated.
(c) The invertible sheaf L is π-ample.

17.3.G. E XERCISE . Prove Theorem 17.3.7 using Theorem 16.2.2. (Unimportant

remark: The proof given of Theorem 16.2.2 used Noetherian hypotheses, but as
stated there, they can be removed.)
After doing the above Exercise, it will be clear how to adjust the statement of
Theorem 17.3.7 if you need to remove the quasicompactness assumption on Y.

17.3.H. E XERCISE ( USEFUL EQUIVALENT DEFINITION OF VERY AMPLENESS ). Sup-

pose π : X → Y is a proper morphism, Y is locally Noetherian (hence X is too, as π is
finite type), and L is an invertible sheaf on X. Suppose that you know that in this
situation π∗ L is finite type. (We will later show this, as described in Exercise 17.3.F.
This is where Noetherian hypotheses are used.) Show that L is π-very ample
if and only if (i) L is relatively base-point-free, and (ii) the canonical morphism
ψ : X → Proj Y Sym• (π∗ L ) of Exercise 17.3.F is a closed embedding. Conclude that
the notion of relative very ampleness is affine-local on Y (it may be checked on any
affine cover Y), if π is proper.

17.3.8. As a consequence, Theorem 17.3.7 implies the notion of relative ampleness

is affine-local on Y (if π is proper and Y is locally Noetherian).

17.3.I. I MPORTANT C HALLENGING E XERCISE ( THE COMPOSITION OF PROJECTIVE

MORPHISMS IS USUALLY PROJECTIVE ). Suppose π : X → Y and ρ : Y → Z are
projective morphisms, and Z is affine or Noetherian. Show that ρ ◦ π is projective.
Hint: the criterion for projectivity given in Exercise 17.3.A(b) will be useful. (i) Deal
first with the case where Z is affine. Build the following commutative diagram,
thereby finding a closed embedding X ,→ PF ⊕n over Z. (See §17.2.3 for the
definition of PV , including a caution about duals.) In this diagram, all inclusions are
490 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

closed embeddings, and all script fonts refer to finite type quasicoherent sheaves.
 / PE   / Pn−1 ×Z Y   / Pn−1 ×Z PF  
X
(†) Segre
/ P (F ⊕n )
Z Z cf. Ex. 10.6.C, 10.6.D

π
%)    
Y / PF
ρ
 v
Zs
Construct the closed embedding (†) as follows. Suppose M is the very ample line
bundle on Y over Z. Then M is ample, and so by Theorem 16.2.2, for m ≫ 0,
E ⊗ M ⊗m is generated by a finite number of global sections. Suppose OY⊕n ↠
E ⊗ M ⊗m is the corresponding surjection. This induces a closed embedding
P(E ⊗ M ⊗m ) ,→ Pn−1 . But P(E ⊗ M ⊗m ) =∼ PE (Exercise 17.2.G), and Pn−1 =
Y Y
Pn−1
Z ×Z Y. (ii) Unwind this diagram to show that (for Z affine) if L is π-very
ample and M is ρ-very ample, then for m ≫ 0, L ⊗ π∗ M ⊗m is (ρ ◦ π)-very ample.
Then deal with the general case by covering Z with a finite number of affine open
subsets, using Exercise 17.3.H.

17.3.9. ⋆⋆ Ample vector bundles. The notion of an ample vector bundle is useful
in some parts of the literature, so we define it, although we won’t use the notion.
A locally free sheaf E on a proper A-scheme X is ample if OProj X Sym• E /X (1) is an
ample invertible sheaf. In particular, using Exercise 17.2.G, you can verify that an
invertible sheaf is ample as a locally free sheaf (this definition) if and only if it is
ample as an invertible sheaf (Definition 16.2.1), preventing a notational crisis. (The
proper hypotheses can be relaxed; it is included only because Definition 16.2.1 of
ampleness is only for proper schemes.)

17.3.10. ⋆⋆ Quasiprojective morphisms. In analogy with projective and quasipro-

jective A-schemes (§4.5.10), one may define quasiprojective morphisms. If Y is
quasicompact, we say that π : X → Y is a quasiprojective morphism if π can be ex-
pressed as a quasicompact open embedding into a scheme projective over Y. This is
not a great notion, and we will not use it. (The general definition of quasiprojective
morphism is slightly delicate — see [Gr-EGA, II.5.3] — and we won’t need it.)

17.3.11. ⋆⋆ Quasiaffine morphisms.

Because we have introduced quasiprojective morphisms, we briefly introduce
quasiaffine morphisms (and quasiaffine schemes), as some readers may have cause
to use them. Many of these ideas could have been introduced long before, but
because we will never use them, we deal with them all at once.
A scheme X is quasiaffine if it admits a quasicompact open embedding into an
affine scheme. This implies that X is quasicompact and separated. If X is Noetherian
(the most relevant case for most people), then any open embedding is of course
automatically quasicompact.

17.3.J. E XERCISE . Show that X is quasiaffine if and only if the canonical map
X → Spec Γ (X, OX ) (defined in Exercise 7.3.G and the paragraph following it) is a
quasicompact open embedding. Thus a quasiaffine scheme comes with a canonical
quasicompact open embedding into an affine scheme. Hint: Let A = Γ (X, OX ) for
 © 2024 Ravi Vakil. Published by Princeton University Press. 491

convenience. Suppose X ,→ Spec R is a quasicompact open embedding. We wish

to show that X ,→ Spec A is a quasicompact open embedding. Factor X ,→ Spec R
through X → Spec A → Spec R. Show that X → Spec A is an open embedding in an
open neighborhood of any chosen point p ∈ X, as follows. Choose r ∈ R such that
p ∈ D(r) ⊂ X. Notice that if Xr = {q ∈ X : r(q) ̸= 0}, then Γ (Xr , OX ) = Γ (X, OX )r by
Exercise 15.4.N, using the fact that X is quasicompact and quasiseparated. Use this
to show that the map Xr → Spec Ar is an isomorphism.
It is not hard to show that X is quasiaffine if and only if OX is ample, but we
won’t use this fact.
A morphism π : X → Y is quasiaffine if the inverse image of every affine
open subset of Y is a quasiaffine scheme. By Exercise 17.3.J, this is equivalent to
π being quasicompact and separated, and the natural map X → Spec π∗ OX being
a quasicompact open embedding. This implies that the notion of quasiaffineness
is local on the target (may be checked on an open cover), and also affine-local on
the target (one may choose an affine cover, and check that the preimages of these
open sets are quasiaffine). Quasiaffine morphisms are preserved by base change:
if a morphism X ,→ Z over Y is a quasicompact open embedding into a scheme
affine over Y, then for any W → Y, X ×Y W ,→ Z ×Y W is a quasicompact open
embedding into a scheme affine over W. (Interestingly, Exercise 17.3.J is not the
right tool to use to show this base change property.)
One may readily check that quasiaffine morphisms are preserved by compo-
sition, [Stacks, tag 01SN] (so quasiaffine morphisms form a “reasonable class”).
Thus quasicompact locally closed embeddings are quasiaffine. If X is affine, then
X → Y is quasiaffine if and only if it is quasicompact (as the preimage of any affine
open subset of Y is an open subset of an affine scheme, namely X). In particular,
from the Cancellation Theorem 11.1.1 for quasicompact morphisms, any morphism
from an affine scheme to a quasiseparated scheme is quasiaffine.
 CHAPTER 18

Čech cohomology of quasicoherent sheaves

This topic is surprisingly simple and elegant. You may think cohomology must
be complicated, and that this is why it appears so late in the book. But you will
see that we need very little background. After defining schemes, we could have
immediately defined quasicoherent sheaves, and then defined cohomology, and
verified that it had many useful properties.

18.1 (Desired) properties of cohomology

Rather than immediately defining cohomology of quasicoherent sheaves, we
first discuss why we care, and what properties it should have.
As Γ (X, ·) is a left-exact functor, if 0 → F → G → H → 0 is a short exact
sequence of quasicoherent sheaves on a scheme X, then
0 / F (X) / G (X) / H (X)

is exact. We dream that this sequence continues to the right, giving a long exact
sequence. More explicitly, there should be some covariant functors Hi (i ≥ 0) from
quasicoherent sheaves on X to groups such that H0 is the global section functor Γ ,
and so that there is a “long exact sequence in cohomology”.

(18.1.0.1) 0 / H0 (X, F ) / H0 (X, G ) / H0 (X, H )

/ H1 (X, F ) / H1 (X, G ) / H1 (X, H ) / ···

(In general, whenever we see a left-exact or right-exact functor, we should hope

for this, and in good cases our dreams will come true. The machinery behind this
usually involves derived functors, which we will discuss in Chapter 23.)
Before defining cohomology groups of quasicoherent sheaves explicitly, we
first describe their important properties, which are in some ways more important
than the formal definition. The boxed properties will be the important ones.
Suppose X is a separated and quasicompact A-scheme. For each quasicoherent
sheaf F on X, we will define A-modules Hi (X, F ). In particular, if A = k, they are
k-vector spaces. In this case, we define hi (X, F ) = dimk Hi (X, F ) (where k is left
implicit on the left side).

This material is published by Princeton University Press as The Rising Sea: Foundations of Algebraic
Geometry by Ravi Vakil. This version is free to view and download for personal use only. Not for
redistribution, resale, or use in derivative works. © 2024 Ravi Vakil.

493
494 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

(i) Each Hi is a covariant functor QCohX → ModA .

(ii) The functor H0 is identified with the functor Γ : H0 (X, F ) = Γ (X, F ), and
the covariance of (i) for i = 0 is just the usual covariance for Γ (F → G induces
Γ (X, F ) → Γ (X, G )).
(iii) If 0 → F → G → H → 0 is a short exact sequence of quasicoher-
ent sheaves on X, then we have a long exact sequence (18.1.0.1). The maps
Hi (X, F ) → Hi (X, G ) come from covariance, and similarly for Hi (X, G ) → Hi (X, H ).
The connecting homomorphisms Hi (X, H ) → Hi+1 (X, F ) will have to be defined.
(iv) If π : X → Y is any morphism of quasicompact separated A-schemes, and F
is a quasicoherent sheaf on X, then there is a natural morphism Hi (Y, π∗ F ) → Hi (X, F )
extending Γ (Y, π∗ F ) → Γ (X, F ). (Note that π is quasicompact and separated by the
Cancellation Theorem 11.1.1 for quasicompact and separated morphisms, taking
Z = Spec A in the statement of the Cancellation Theorem, so π∗ F is indeed a
quasicoherent sheaf by Exercise 14.4.C.) We will later see this as part of a larger
story, the Leray spectral sequence (Theorem 23.4.5). If G is a quasicoherent sheaf
on Y, then setting F := π∗ G and using the adjunction map G → π∗ π∗ G and co-
variance (property (i)) gives a natural pullback map Hi (Y, G ) → Hi (X, π∗ G ) (via
Hi (Y, G ) → Hi (Y, π∗ π∗ G ) → Hi (X, π∗ G )) extending Γ (Y, G ) → Γ (X, π∗ G ). In this
way, Hi is a “contravariant functor in the space”.
(v) If π : X → Y is an affine morphism, and F is a quasicoherent sheaf on X,
∼
the natural map of (iv) is an isomorphism: Hi (Y, π∗ F ) −→ Hi (X, F ). When π
is a closed embedding and Y = PN A , this isomorphism translates calculations on
arbitrary projective A-schemes to calculations on PN A.
(vi) (affine cover cohomology vanishing) If X can be covered by n affine open
sets, then Hi (X, F ) = 0 for i ≥ n for all F . In particular, on affine schemes, all
higher (i > 0) quasicoherent cohomology groups vanish. The vanishing of H1
in this case, along with the long exact sequence (iii) implies that Γ is an exact
functor for quasicoherent sheaves on affine schemes, something we already knew
(Exercise 6.3.A).

18.1.1. Remarks (not part of property (vi)). It is also true that if dim X = n, then
Hi (X, F ) = 0 for all i > n and for all F (dimensional cohomology vanishing). We
will prove this for projective k-schemes (Theorem 18.2.6) and even quasiprojective
k-schemes (Exercise 22.4.T). See §18.2.7 for discussion of the general case.
The converse to (vi) in the case when n = 1 is Serre’s cohomological criterion for
affineness, see §18.8.4.
Let’s get back to our list.
(vii) The functor Hi behaves well under direct sums, and more generally under
filtered colimits: Hi (X, colim Fj ) = colim Hi (X, Fj ).
(viii) We will also identify the cohomology of all O(m) on Pn
A:

18.1.2. Theorem. —

A , OPA (m)) is a free A-module of rank

n+m


(a) H0 (Pn n
m if m ≥ 0.
(b) H (PA , OPnA (m)) is a free A-module of rank −m−1
n n


−n−m−1 if m ≤ −n − 1.
 © 2024 Ravi Vakil. Published by Princeton University Press. 495

A , OPA (m)) = 0 otherwise.

(c) Hi (Pn n

We have already shown the first statement in Essential Exercise 15.1.C (see
also §15.1.2). We saw there (and in Exercise 15.1.D) that it is advantageous to
interpret to interpret H0 (Pn A , OPA (m)) as the degree m homogeneous polynomials
n

in A[x0 , . . . , xn ]. Similarly, we will find that Hn (PnA , OPA (m)) can be interpreted
n

as the homogeneous degree −m terms of (x0 x1 · · · xn )−1 A[x−1 −1

0 , . . . , xn ] (see Re-
mark 18.3.5).
Theorem 18.1.2 has a number of features that will be the first appearances of
facts that we will prove later.
• The cohomology of these bundles vanish in degree above n ((vi) above,
“affine cover cohomology vanishing”)
• These cohomology groups are always finitely generated A-modules. This
will be true for all coherent sheaves on projective A-schemes (Theo-
rem 18.1.3(i)), and indeed (with more work) on proper A-schemes (Grothendieck’s
Coherence Theorem 18.9.1).
• The top cohomology group vanishes for m > −n − 1. We will later see
this as an example of Kodaira vanishing, see §21.5.7.
If A is a field k:
• The top cohomology group is one-dimensional for m = −n − 1. This is
the first appearance of the dualizing sheaf.
• There is a natural perfect pairing

Hi (Pn , O(m)) × Hn−i (Pn , O(−n − 1 − m)) / Hn (Pn , O(−n − 1))

This is the first appearance of Serre duality. (For the case n = 1, see
Example 18.5.4.) P
• The alternating sum (−1)i hi (Pn , O(m)) is a polynomial in m. This is a
first example of a Hilbert polynomial.
Before proving these facts, let’s first use them to prove interesting things, as
motivation.
By Theorem 16.1.1, for any coherent sheaf F on Pn A we can find a surjection
O(m)⊕j → F , which yields the exact sequence

(18.1.2.1) 0 /G / O(m)⊕j /F /0

for some coherent sheaf G . We can use this to prove the following.

18.1.3. Theorem. — (i) For any coherent sheaf F on a projective A-scheme X where A is
Noetherian, Hi (X, F ) is a coherent (finitely generated) A-module.
(ii) (Serre vanishing) Furthermore, for m ≫ 0, Hi (X, F (m)) = 0 for all i > 0 (even
without Noetherian hypotheses).
(A slightly fancier version of Serre vanishing will be given in Theorem 18.7.F.)

Proof. Because cohomology of a closed subscheme can be computed on the ambient

space ((v) above), we may immediately reduce to the case X = Pn
A.
496 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

(i) Consider the long exact sequence:

0 / H0 (Pn , G ) / H0 (Pn , O(m)⊕j ) / H0 (Pn , F ) /

A A A

/ H1 (Pn , O(m)⊕j ) / H1 (Pn , F ) / ···

A, G )
H1 (Pn A A

··· / Hn−1 (Pn , G ) / Hn−1 (Pn , O(m)⊕j ) / Hn−1 (Pn , F ) /

A A A

/ Hn (Pn , O(m)⊕j ) / Hn (Pn , F ) /0

A, G )
Hn (Pn A A

The exact sequence ends here because Pn A is covered by n + 1 affine open sets ((vi)
A , O(m) ) is finitely
⊕j
above, “affine cover cohomology vanishing”). Then Hn (Pn
generated by Theorem 18.1.2(b) and (vii), hence H (PA , F ) is finitely generated
n n

for all coherent sheaves F . Hence in particular Hn (Pn A , G ) is finitely generated.

As Hn−1 (Pn A , O(m) ⊕j
) is finitely generated, and H n n
(PA , G ) is too, we have that
Hn−1 (Pn A , F ) is finitely generated for all coherent sheaves F . We continue induc-
tively downwards.
(ii) Twist (18.1.2.1) by O(N) for N ≫ 0. Then

A , O(m + N) ) = H (PA , O(m + N))

⊕j ⊕j
Hn (Pn n n
=0

A , F (N)) = 0. Translation: for any coherent sheaf, its top

(by (vii) above), so Hn (Pn
cohomology vanishes once you twist by O(N) for N sufficiently large. Hence this
is true for G as well. Hence from the long exact sequence, Hn−1 (Pn A , F (N)) = 0 for
N ≫ 0. As in (i), we induct downwards, until we get that H1 (Pn A F (N)) = 0. (The
,
induction stops here, as it is not true that H0 (Pn
A , O(m + N) ⊕j
) = 0 for large N —
quite the opposite.) □
In particular, we have proved the following, that we would have cared about
even before we knew about cohomology.

18.1.4. Corollary. — Any projective k-scheme has a finite-dimensional space of functions.

More generally, if A is Noetherian and F is a coherent sheaf on a projective A-scheme,
then H0 (X, F ) is a coherent A-module.
(We will generalize this in Theorem 18.7.1.) I want to emphasize how remark-
able this proof is. It is a question about global sections, i.e., H0 , which we think
of as the most down-to-earth cohomology group, yet the proof is by downward
induction for Hn , starting with n large.
Corollary 18.1.4 is true more generally for proper k-schemes, not just projective
k-schemes (Grothendieck’s Coherence Theorem 18.9.1).
Here are some important consequences. They can also be shown directly,
without the use of cohomology, but with much more elbow grease.

18.1.A. C RUCIAL E XERCISE ( PUSHFORWARDS OF COHERENT SHEAVES BY PROJEC -

TIVE MORPHISMS ARE COHERENT ). Suppose π : X → Y is a projective morphism of
locally Noetherian schemes. Show that the pushforward of a coherent sheaf on X is
 © 2024 Ravi Vakil. Published by Princeton University Press. 497

a coherent sheaf on Y. (See Grothendieck’s Coherence Theorems 18.7.1 and 18.9.1

for generalizations.)
Finite morphisms are affine (from the definition) and projective (Exercise 17.3.B).
We can now show that this is a characterization of finiteness.

18.1.5. Corollary (projective + affine = finite). — Suppose Y is locally Noetherian.

Then a morphism π : X → Y is projective and affine if and only if π is finite.
We will see in Exercise 18.9.A that the projective hypotheses can be relaxed to
proper.

Proof. We already know that finite morphisms are affine (by definition) and projec-
tive (Exercise 17.3.B), so we show the converse. Suppose π is projective and affine.
By Exercise 18.1.A, π∗ OX is coherent and hence finite type. □
The following result was promised in §17.3.3, and has a number of useful
consequences.

18.1.6. Theorem (projective + finite fibers = finite). — Suppose π : X → Y is a

morphism, with Y Noetherian. Then π is projective and has finite fibers if and only if it is
finite. Equivalently, π is projective and quasifinite if and only if it is finite.
(Recall that quasifinite = finite fibers + finite type. But projective includes finite
type.) It is true more generally that (with Noetherian hypotheses) proper + finite
fibers = finite, see Theorem 28.5.2.

Proof. We have already shown that finite morphisms are projective (Exercise 17.3.B)
and have finite fibers (Exercise 8.3.J). So we now assume that π is projective and has
finite fibers, and show that π is finite. We show π is finite near a point q ∈ Y. Fix an
affine open neighborhood Spec A of q in Y. For notational convenience, we rename
Y = Spec A and X = π−1 (Spec A). Projectivity of π means we can factor π as

 cl.emb. / Pn
X A

ρ
π
" 
Spec A Y.

Choose a hypersurface Hq in Pn κ(q) missing π

−1
(q) ⊂ X (Exercise 12.3.C). Choose
any lift of its equation to A, to obtain a hypersurface H in Pn A also missing π
−1
(q).
n
(See Figure 18.1.) Now H ∩ X is closed in PA , so by the Fundamental Theorem of
Elimination Theory 8.4.10, H ′ := π(H ∩ X) is closed in Spec A. Also, H ′ doesn’t
contain q, so U := Spec A \ H ′ is an open subset of Y = Spec A containing q.
Finally, π−1 (U) → U is a projective morphism. It is also affine, because it is the
composition of two affine morphisms
 cl.emb.
π−1 (U)  / Pn \ H /U
ρ
U

(the second of which is affine by Exercise 4.5.I). We are then done by Corollary 18.1.5.
□
A similar trick was used in the proof of the Rigidity Lemma 11.4.12.
498 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

F IGURE 18.1. The key construction in the proof of Theorem 18.1.6

18.1.B. E XERCISE ( UPPER SEMICONTINUITY OF FIBER DIMENSION ON THE TARGET,

FOR PROJECTIVE MORPHISMS ). Use a similar argument as in Theorem 18.1.6
to prove upper semicontinuity of fiber dimension of projective morphisms: suppose
π : X → Y is a projective morphism where Y is locally Noetherian (or more generally
OY is coherent over itself). Show that {q ∈ Y : dim π−1 (q) > k} is a Zariski-closed
subset of Y. In other words, the dimension of the fiber “jumps over Zariski-closed
subsets” of the target. (You can interpret the case k = −1 as the fact that projective
morphisms are closed, which is basically the Fundamental Theorem of Elimination
Theory 8.4.10, cf. §17.3.2.) This exercise is rather important for having a sense
of how projective morphisms behave. (The case of varieties was done earlier, in
Theorem 12.4.3(b). This approach is much simpler.)
The final exercise of the section is on a different theme.

18.1.C. E XERCISE . Suppose 0 → F → G → H → 0 is an exact sequence of

quasicoherent sheaves on X, projective over A, with F coherent. Show that for
n ≫ 0,

0 / H0 (X, F (n)) / H0 (X, G (n)) / H0 (X, H (n)) /0

is also exact. (Hint: for n ≫ 0, H1 (X, F (n)) = 0.)

18.2 Definitions and proofs of key properties

This section could be read much later; the facts we will use are all stated in the
previous section. However, the arguments are not complicated, so you want to read
 © 2024 Ravi Vakil. Published by Princeton University Press. 499

this right away. As you read this, you should go back and check off all the facts in
the previous section, to assure yourself that you understand everything promised.

18.2.1. Čech cohomology. Čech cohomology in general settings is defined using a

limit over finer and finer covers of a space. In our algebro-geometric setting, the
situation is much cleaner, and we can use a single cover.
Suppose X is quasicompact and separated, which is true for example if X is
quasiprojective over A. In particular, X may be covered by a finite number of affine
open sets, and the intersection of any two affine open sets is also an affine open
set (by separatedness, Proposition 11.2.13). We will use quasicompactness and
separatedness only in order to ensure these two nice properties.
Suppose F is a quasicoherent sheaf, and U = {Ui }n i=1 is a finite collection of
affine open sets covering X. For I ⊂ {1, . . . , n} define UI = ∩i∈I Ui , which is affine
by the separated hypothesis. (Here is a strong analogy for those who have seen
cohomology in other contexts: cover a topological space X with a finite number of
open sets Ui , such that all intersections ∩i∈I Ui are contractible.) Consider the Čech
complex
Q
(18.2.1.1) 0 / F (UI ) / ··· /
|I| = 1
I ⊂ {1, . . . , n}

Q Q
F (UI ) / F (UI ) / ··· .
|I| = i |I| = i + 1
I ⊂ {1, . . . , n} I ⊂ {1, . . . , n}

The maps are defined as follows. The map from F (UI ) → F (UJ ) is 0 unless I ⊂ J,
i.e., J = I ∪ {j}. If j is the kth element of J, then the map is (−1)k−1 times the
restriction map resUI ,UJ .

18.2.A. E ASY E XERCISE ( FOR THOSE WHO HAVEN ’ T SEEN ANYTHING LIKE THE
Č ECH COMPLEX BEFORE ). Show that the Čech complex is indeed a complex, i.e.,
that the composition of two consecutive arrows is 0.
Define HiU (X, F ) to be the ith cohomology group of the complex (18.2.1.1).
(The indexing starts with i = 0.) Note that if X is an A-scheme, then HiU (X, F ) is an
A-module. We have almost succeeded in defining the Čech cohomology group Hi ,
except our definition a priori depends on a choice of a cover U . Note that HiU (X, ·)
is clearly a covariant functor QCohX → ModA .

18.2.B. E ASY E XERCISE . Identify H0U (X, F ) with Γ (X, F ). (Hint: use the sheaf
axioms for F .)

18.2.C. E XERCISE . Suppose

(18.2.1.2) 0 /F /G /H /0

is a short exact sequence of sheaves of abelian groups on a topological space, and

U is a finite open cover such that on any intersection UI of open subsets in U , the
map Γ (UI , G ) → Γ (UI , H ) is surjective. Show that we get a “long exact sequence of
cohomology for HiU ” (where we take the same definition of HiU ). In our situation,
where X is a quasicompact separated A-scheme, and (18.2.1.2) is a short exact
500 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

sequence of quasicoherent sheaves on X, show that we get a long exact sequence

for the A-modules HiU .
In the proof of Theorems 18.2.4 and 18.7.1, we will make use of the fact that your
construction of the connecting homomorphism will “commute with localization of
A”. More precisely, we will need the following.

18.2.D. E XERCISE . Suppose we are given a short exact sequence (18.2.1.2) of

quasicoherent sheaves on a quasicompact separated A-scheme π : X → Spec A,
a cover U of X by affine open sets, and some f ∈ A. The restriction of the sets
of U to Xf yields an affine open cover U ′ of Xf = π−1 (D(f)). Identify the long
exact sequence associated to (18.2.1.2) using HiU , localized at f, with the long exact
sequence associated to the restriction of (18.2.1.2) to Xf , using the affine open cover
U ′ . (First check that the maps such as HiU (F ) → HiU (G ) given by covariance
“commute with localization”, and then check that the connecting homomorphisms
do as well.)

18.2.2. Theorem/Definition. — Our standing assumption is that X is quasicompact and

separated. HiU (X, F ) is independent of the choice of (finite) cover {Ui }. More precisely,
for any two covers {Ui } ⊂ {Vi }, the maps Hi{Vi } (X, F ) → Hi{Ui } (X, F ) induced by the
natural map of Čech complexes (18.2.1.1) are isomorphisms. Define the Čech cohomology
group Hi (X, F ) to be this group.
If you are unsure of what the “natural map of Čech complexes” is, by (18.2.3.1)
it should become clear.

18.2.3. For experts: maps of complexes inducing isomorphisms on cohomology

groups are called quasiisomorphisms. We are actually getting a finer invariant
than cohomology out of this construction; we are getting an element of the derived
category of A-modules.

Proof. We need only prove the result when |{Vi }| = |{Ui }| + 1. We will show that
if {Ui }1≤i≤n is a cover of X, and U0 is any other affine open set, then the map
Hi{Ui }0≤i≤n (X, F ) → Hi{Ui }1≤i≤n (X, F ) is an isomorphism. Consider the exact
 © 2024 Ravi Vakil. Published by Princeton University Press. 501

sequence of complexes
(18.2.3.1)
0 0 0

Q  Q  Q 
··· / F (UI ) / F (UI ) / F (UI ) / ···
|I| = i − 1 |I| = i |I| = i + 1
0∈I 0∈I 0∈I

Q  Q  Q 
··· / F (UI ) / F (UI ) / F (UI ) / ···
|I| = i − 1 |I| = i |I| = i + 1

Q  Q  Q 
··· / F (UI ) / F (UI ) / F (UI ) / ···
|I| = i − 1 |I| = i |I| = i + 1
0∈/ I 0∈ / I 0∈/ I

  
0 0 0

Throughout, I ⊂ {0, . . . , n}. The bottom two rows are Čech complexes with respect
to two covers, and the map between them induces the desired map on cohomology.
We get a long exact sequence of cohomology from this short exact sequence of
complexes (Theorem 1.5.8). Thus we wish to show that the top row is exact and
thus has vanishing cohomology. (Note that U0 ∩ Uj is affine by our separatedness
hypothesis, Proposition 11.2.13.) But the ith cohomology of the top row is precisely
{Uj ∩U0 }j>0 (U0 , F ) except at the start, where we get 0 (because the complex starts
Hi−1
Qn
off 0 → F (U0 ) → j=1 F (U0 ∩ Uj )). So it suffices to show that higher Čech
groups of affine schemes are 0. Hence we are done by the following result. □

18.2.4. Theorem. — The higher Čech cohomology HiU (X, F ) of an affine A-scheme X
vanishes (for any affine cover U , i > 0, and quasicoherent F ).
This is another “partition of unity” argument, in the spirit of the proof of
Theorem 4.1.2.

Proof. We want to show that the “extended” complex

(18.2.4.1) 0 / F (X) / Q F (UI ) / Q F (UI ) / ···

|I|=1 |I|=2

(where the global sections F (X) have been appended to the start) has no cohomol-
ogy, i.e., is exact. (Incidentally, this extended complex is called the Čech resolution
of F (X), with respect to this choice of cover U .) We do this with a trick.
Suppose first that some Ui , say U0 , is X. We deal with this case by sleight
of hand. The complex (18.2.4.1) is the middle row of the following short exact
502 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

sequence of complexes.
Q Q
(18.2.4.2) 0 /0 / F (UI ) / F (UI ) / ···
|I|=1,0∈I |I|=2,0∈I

 Q  Q 
0 / F (X) / F (UI ) / F (UI ) / ···
|I|=1 |I|=2

 Q  Q 
0 / F (X) / F (UI ) / F (UI ) / ···
|I|=1,0∈
/I |I|=2,0∈
/I

The top row is the same as the bottom row, slid over by 1. The corresponding long
exact sequence of cohomology shows that the central row has vanishing cohomol-
ogy. (You should show that the “connecting homomorphism” on cohomology is
indeed an isomorphism.) This might remind you of the mapping cone construction
(Exercise 1.6.E). (Here is a trick to accomplish the goal of this paragraph from
a different point of view. Let id be the identity map between complex (18.2.4.1)
and itself, and let 0 be the zero map between the complex (18.2.4.1) and itself. De-
scribe an explicit homotopy between id and 0, cf. §1.5.7. The homotopy will use
the “same idea” as (18.2.3.1). Then the identity map on homology equals the zero
map on homology, Exercise 1.5.E, so the homology must be zero. This idea can be
enlightening because it is used in other circumstances.)

F IGURE 18.2. The final step in the proof of Theorem 18.2.2

Finally, we prove the general case by another sleight of hand. Say X = Spec R.
We wish to show that the complex of A-modules (18.2.4.1) is exact. It is also a
complex of R-modules, so we wish to show that the complex of R-modules (18.2.4.1)
is exact. To show that it is exact, it suffices to show that for a cover of Spec R by
distinguished open sets D(fi ) (1 ≤ i ≤ r) (i.e., (f1 , . . . , fr ) = 1 in R) the complex is
exact. (Translation: exactness of a sequence of sheaves may be checked locally.) We
choose a cover so that each D(fi ) is contained in one of our affine open subsets Uj
(see Figure 18.2). The complex (18.2.4.1) localized at fi is the Čech complex for the
affine cover of D(fi ) by {Uk ∩ D(fi )}. We want to show that this complex (for our
fixed i and j) is exact. But this is the case dealt with in the previous paragraph —
one of the affine open sets, Uj ∩ D(fi ), equals D(fi ). □
We have now proved properties (i)–(iii) of the previous section. The “affine
cover cohomology vanishing” property (vi) is also straightforward: if X is covered
 © 2024 Ravi Vakil. Published by Princeton University Press. 503

by n affine open sets, use these as the cover U , and notice that the Čech complex
ends by the nth step.

18.2.E. E XERCISE ( PROPERTY (v)). Suppose π : X → Y is an affine morphism, and Y

is a quasicompact and separated A-scheme (and hence X is too, as affine morphisms
are both quasicompact and separated). If F is a quasicoherent sheaf on X, describe
∼
a natural isomorphism Hi (Y, π∗ F ) ←→ Hi (X, F ). (Hint: if U is an affine cover of
Y, “π−1 (U )” is an affine cover of X. Use these covers to compute the cohomology
of F .)

18.2.F. E XERCISE ( PROPERTY (iv)). Suppose π : X → Y is any quasicompact

separated morphism, F is a quasicoherent sheaf on X, and Y is a quasicom-
pact separated A-scheme. The hypotheses on π ensure that π∗ F is a quasico-
herent sheaf on Y. Describe a natural morphism Hi (Y, π∗ F ) → Hi (X, F ) extending
Γ (Y, π∗ F ) → Γ (X, F ). (Aside: this morphism is an isomorphism for i = 0, but
need not be an isomorphism for higher i: consider i = 1, X = P1k , F = O(−2), and
let Y be a point Spec k.)

18.2.G. E XERCISE . Prove Property (vii) of the previous section, that “cohomology
commutes with filtered colimits”. (This can be done by hand. Hint: in the category
of modules over a ring, taking the colimit over a filtered sets is an exact functor,
§1.5.14.)
We have now proved all of the properties of the previous section, except for
(viii), which we will get to in §18.3.

18.2.5. Useful facts about cohomology for k-schemes.

18.2.H. E XERCISE ( COHOMOLOGY AND CHANGE OF BASE FIELD ). Suppose X is a

quasicompact separated k-scheme, and F is a quasicoherent sheaf on X. Give an
isomorphism

∼
Hi (X, F ) ⊗k K ←→ Hi (X ×Spec k Spec K, F ⊗k K)

for all i, where K/k is any field extension. Here F ⊗k K means the pullback of F
to X ×Spec k Spec K. Hence hi (X, F ) = hi (X ×Spec k Spec K, F ⊗k K). (This is useful
for relating facts about k-schemes to facts about schemes over algebraically closed
fields. Your proof might use vector spaces — i.e., linear algebra — in a fundamental
way. If it doesn’t, you may prove something more general, if k → K is replaced
by a flat ring map B → A. Recall that B → A is flat if ⊗B A is an exact functor
ModB → ModA . A hint for this harder exercise: the FHHF Theorem, Exercise 1.5.I.
See Exercise 18.7.B(b) for the next generalization of this.)

18.2.I. E XERCISE ( BASE - POINT- FREENESS IS INDEPENDENT OF EXTENSION OF BASE

FIELD ). Suppose X is a quasicompact separated k-scheme, L is an invertible sheaf
on X, and K/k is a field extension. Show that L is base-point-free if and only if its
pullback to X ×Spec k Spec K is base-point-free. (Hint: Exercise 18.2.H with i = 0
implies that a basis of sections of L over k becomes, after tensoring with K, a basis
of sections of L ⊗k K.)
504 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

18.2.6. Theorem (dimensional cohomology vanishing for quasicoherent sheaves

on projective k-schemes). — Suppose X is a projective k-scheme, and F is a quasico-
herent sheaf on X. Then Hi (X, F ) = 0 for i > dim X.
In other words, cohomology vanishes above the dimension of X.

Proof. Suppose X ,→ PN , and let n = dim X. We show that X may be covered by

n + 1 affine open sets. Exercise 12.3.D shows that there are n + 1 effective Cartier
divisors on PN such that their complements U0 , . . . , Un cover X. Then Ui is affine,
so Ui ∩ X is affine, and thus we have covered X with n + 1 affine open sets. □
(It turns out that n + 1 affine open sets are always necessary. One way of
proving this is by showing that the complement of a dense affine open subset is
always pure codimension 1, see for example [RotV, Lem. 2.3].)

18.2.7. ⋆ Dimensional cohomology vanishing more generally. Using the the-

ory of blowing up, Theorem 18.2.6 can be extended to quasiprojective k-schemes,
see §22.4.15. Dimensional cohomology vanishing is even true in much greater
generality. To state it, we need to define cohomology with the more general ma-
chinery of derived functors (Chapter 23). If X is a Noetherian topological space
(§3.6.14) and F is any sheaf of abelian groups on X, we have Hi (X, F ) = 0 for all
i > dim X. (Grothendieck sketches his elegant proof in [GrS, p. 29-30]; see [Ha1,
Theorem III.2.7] for a more detailed explanation.) In particular, if X is a k-variety of
dimension n, we always have dimensional cohomology vanishing, even for crazy
varieties that can’t be covered with n + 1 affine open subsets (see §22.4.15).

18.3 Cohomology of line bundles on projective space

We finally prove the last promised basic fact about cohomology, property (viii)
of §18.1, Theorem 18.1.2, on the cohomology of line bundles on projective space. But
first we discuss a seemingly different problem, of the cohomology of the structure
sheaf of affine space minus the origin.

18.3.1. The Problem. For any ring A, let X = An+1 A \ V(x0 , . . . , xn ) for n ≥ 1. Here
n+1
the coordinates on A are x0 , . . . , xn , and the motivation for the indexing will
soon become clear. Find Hi (X, OX ).

18.3.2. The Answer. We will show that Hi (X, OX ) = 0 unless i = 0 (in which case
it will be A[x0 , . . . , xn ]) or i = n. We will see that Hn (X, OX ) can be identified
with those “Laurent polynomials” of the form (x0 x1 · · · xn )−1 A[x−1 −1 −1
0 , x1 , . . . , xn ].
(This gives another proof that Ak \ {(0, 0)} is not affine, cf. §4.4.3.) (Cf. Serre’s
2

cohomological criterion for affineness, §18.8.4.)

There is a natural open cover of X by {Ui } where Ui = D(xi ), and we use the
corresponding Čech complex.
We discuss the case n = 2, because all of the complications present themselves
here. In this case the Čech complex is
(18.3.2.1)
0 / A[x0 , x1 , x2 , x−1 ] × A[x0 , x1 , x2 , x−1 ] × A[x0 , x1 , x2 , x−1 ] /
0 1 2
 © 2024 Ravi Vakil. Published by Princeton University Press. 505

A[x0 , x1 , x2 , x−1 −1 −1 −1 −1 −1
0 , x1 ] × A[x0 , x1 , x2 , x1 , x2 ] × A[x0 , x1 , x2 , x0 , x2 ]

/ A[x0 , x1 , x2 , x−1 , x−1 , x−1 ] / 0.

0 1 2

We extend (18.3.2.1) by replacing the “0 →” on the left by “0 → A[x0 , x1 , x2 ] →”:

(18.3.2.2)
H0 U0 U1 U2 U012

0 / A[x0 , x1 , x2 ] / ··· / ··· / A[x0 , x1 , x2 , x−1

0
, x−1
1
x−1
2
] / 0.

18.3.A. E XERCISE . Show that if (18.3.2.2) is exact, except that at U012 the coho-
mology/cokernel is x−1 −1 −1 −1 −1 −1
0 x1 x2 A[x0 , x1 , x2 ], then Answer 18.3.2 is correct for
n = 2.
Because the maps in (18.3.2.2) preserve multidegree (degrees of each xi inde-
pendently), we can study exactness of (18.3.2.2) monomial by monomial.
The “3 negative exponents” case. Consider first the monomial xa 0 a1 a2
0 x1 x2 , where
the exponents ai are all negative. Then (18.3.2.2) in this multidegree is:

0 / 0H0 / 00 × 01 × 02 / 001 × 012 × 002 / A012 / 0.

Here the subscripts serve only to remind us which “Čech” terms the factors cor-
respond to. For example, A012 corresponds to the coefficient of xa 0 a1 a2
0 x1 x2 in
−1 −1 −1
A[x0 , x1 , x2 , x0 , x1 , x2 ]. Clearly this complex only has (co)homology at the U012
spot, as desired.
The “2 negative exponents” case. Consider next the case where two of the expo-
nents, say a0 and a1 , are negative. Then the complex in this multidegree is

0 / 0H0 / 00 × 01 × 02 / A01 × 012 × 002 / A012 / 0,

which is clearly exact.

The “1 negative exponent” case. We next consider the case where one of the
exponents, say a0 , is negative. Then the complex in this multidegree is

0 / 0H0 / A0 × 01 × 02 / A01 × 012 × A02 / A012 /0

With a little thought (paying attention to the signs on the arrows A → A), you will
see that it is exact. (The subscripts, by reminding us of the subscripts in the original
Čech complex, remind us what signs to take in the maps.)
The “0 negative exponent” case. Finally, consider the case where none of the
exponents are negative. Then the complex in this multidegree is
(18.3.2.3)
0 / AH0 / A0 × A1 × A2 / A01 × A12 × A02 / A012 /0
506 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

We wish to show that this is exact. We use the same strategy as we used at (18.2.4.2).
We write (18.3.2.3) as the middle of a short exact sequence of complexes:
(18.3.2.4)
0 /0 / A2 / A02 × A12 / A012 /0

     
0 / AH0 / A0 × A1 × A2 / A01 × A12 × A02 / A012 /0

     
0 / AH0 / A0 × A1 / A01 /0 /0

Thus we get a long exact sequence in cohomology (Theorem 1.5.8). But the top and
bottom rows are exact (basically from the “1-negative” case), i.e., cohomology-free,
so the middle row must be exact too. (Alternatively, as with (18.2.4.2), let id be the
identity map between complex (18.3.2.3) and itself. Describe an explicit homotopy
between id and the zero map. Then the identity map on homology equals the zero
map on homology, Exercise 1.5.E, which means the homology is zero.)

18.3.B. E XERCISE . Prove that Answer 18.3.2 is true for all n ≥ 1. (I could of course
just have given you the proof for general n, but seeing the argument in action may
be enlightening. In particular, your argument may be much shorter. For example,
the “2-negative” case could be done in the same way as the “1-negative” case, so
you will not need n + 1 separate cases if you set things up carefully.)

18.3.3. Remark. This argument is basically the proof that the reduced homology
of the boundary of a simplex S (known in some circles as a “sphere”) is 0, unless
S is the empty set, in which case it is one-dimensional. The “empty set” case
corresponds to the “3-negative” case.
We are now ready to prove Theorem 18.1.2.

18.3.4. Proof of Theorem 18.1.2. We use the standard cover U0 = D(x0 ), . . . ,

Un = D(xn ) of Pn A.
We come to the central trick (Exercise 18.3.C): the Čech complex for O(m) is the
degree m part of the Čech complex for Hi (An+1A , OX )! (You can stare at the n = 2
case (18.3.2.1) if it helps.)

18.3.C. E XERCISE ( CF. R EMARK 18.3.5). If I ⊂ {0, . . . , n}, then give an isomorphism
(of A-modules) of Γ (UI , O(m)) with the homogeneous degree m Laurent mono-
mials (in x0 , . . . , xn , with coefficients in A) where each xi for i ∈/ I appears with
non-negative degree. Your construction should be such that the restriction map
Γ (UI , O(m)) → Γ (UJ , O(m)) (I ⊂ J) corresponds to the natural inclusion: a Laurent
polynomial in Γ (UI , O(m)) maps to the same Laurent polynomial in Γ (UJ , O(m)).

18.3.D. E XERCISE . Prove Theorem 18.1.2. □

18.3.5. Remark. Exercises 15.1.C and 15.1.D interpreted H0 (PnA , OPA (m)) as the
n

homogeneous degree m polynomials in x0 , . . . , xn (with A-coefficients). We can

 © 2024 Ravi Vakil. Published by Princeton University Press. 507

now state this in a better way. Let

π : An+1 \ V(x0 , . . . , xn ) / Pn
A A

be the morphism of Exercise 7.3.F and Example 7.3.12. You can show that the
induced map
/ Hn (An+1 \ V(x0 , . . . , xn ), O)
A , O(m))
π∗ : Hn (Pn A

is an inclusion, and “identify the image”.

18.4 Riemann-Roch, and arithmetic genus

We have seen some powerful uses of Čech cohomology, to prove things about
spaces of global sections, and to prove Serre vanishing. We will now see some
classical constructions come out very quickly and cheaply.
In this section, we will work over a field k. Suppose F is a coherent sheaf on
a projective k-scheme X. Recall the notation (§18.1) hi (X, F ) := dimk Hi (X, F ).
By Theorem 18.1.3, hi (X, F ) is finite. (The arguments in this section will extend
without change to proper X once we have this finiteness for proper morphisms, by
Grothendieck’s Coherence Theorem 18.9.1.) Define the Euler characteristic of F by
XX
dim
χ(X, F ) := (−1)i hi (X, F ).
i=0

We will see repeatedly here and later that Euler characteristics behave better than
individual cohomology groups. As one sign, notice that for fixed n, and m ≥ 0,
 
n+m (m + 1)(m + 2) · · · (m + n)
h (Pk , O(m)) =
0 n
= .
m n!
Notice that the expression on the right is a polynomial in m of degree n. (For later
reference, notice also that the leading term is mn /n!.) But it is not true that
(m + 1)(m + 2) · · · (m + n)
k , O(m)) =
h0 (Pn
n!
for all m — it breaks down for m ≤ −n − 1. Still, you can check (using Theo-
rem 18.1.2) that
(m + 1)(m + 2) · · · (m + n)
k , O(m)) =
χ(Pn .
n!
So one lesson is this: if one cohomology group (usually the top or bottom) behaves
well in a certain range, and then messes up, likely it is because (i) it is actually
the Euler characteristic which behaves well always, and (ii) the other cohomology
groups vanish in that certain range.
In fact, we will see that it is often hard to calculate cohomology groups (even
h0 ), but it can be easier calculating Euler characteristics. So one important way of
getting a hold of cohomology groups is by computing the Euler characteristic, and
then showing that all the other cohomology groups vanish. Hence the ubiquity and
importance of vanishing theorems. (A vanishing theorem usually states that a certain
508 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

cohomology group vanishes under certain conditions.) We will see this in action
when discussing curves. (One of the first applications will be (19.2.5.1).)
The following exercise shows another way in which Euler characteristic behaves
well: it is additive in exact sequences.

18.4.A. E XERCISE . Show that if 0 → F → G → H → 0 is an exact sequence of

coherent sheaves on a projective k-scheme X, then χ(X, G ) = χ(X, F ) + χ(X, H ).
(Hint: consider the long exact sequence in cohomology.) More generally, if

0 / F1 / ··· / Fn /0

is an exact sequence of coherent sheaves, show that

X
n
(−1)i χ(X, Fi ) = 0.
i=1

(This exercise both generalizes the “exact” case of Exercise 1.5.B — consider the
case where X = Spec k — and uses it in the proof.)

18.4.1. The Riemann-Roch Theorem for line bundles on projective curves.

18.4.B. E SSENTIAL E XERCISE : T HE R IEMANN -R OCH T HEOREM

P FOR LINE BUNDLES
ON A PROJECTIVE CURVE ( FIRST VERSION ). Suppose D = p∈Creg ap [p] is a divisor
on a projective (pure dimension 1) curve C over a field k (where ap ∈ Z, and all but
finitely many ap are 0). Here Creg are the dimension 0 regular points of C. Define
the degree of D by
X
deg D = ap deg p.
(The degree of a point p was defined in §5.3.8, as the degree of the field extension
of the residue field over k.) Show that
χ(C, OC (D)) = deg D + χ(C, OC )
P P
by induction on |ap | (where D = ap [p] as above). Hint: to show that
χ(C, OC (D)) = deg p + χ(C, OC (D − p)), tensor the closed subscheme exact se-
quence (9.1.2.1)

0 / OC (−p) / OC / O|p /0

by OC (D), and use additivity of Euler characteristics in exact sequences (Exer-

cise 18.4.A).
As every invertible sheaf L on a regular curve is of the form OC (D) for some
D (see §15.4), this exercise is very powerful.

18.4.2. Important definition: degree of a line bundle on a projective curve (generalized in

§18.4.7). Suppose L is an invertible sheaf on a projective curve C over k. (We make
no assumption on the regularity, or even reducedness, of C!) Define the degree of
L (denoted deg L or degC L ) by
deg L := χ(C, L ) − χ(C, OC ).
Thus we have a degree map deg : Pic C → Z. The set of degree d line bundles on C
is denoted Picd C := deg−1 (d).
 © 2024 Ravi Vakil. Published by Princeton University Press. 509

18.4.C. I MPORTANT E ASY E XERCISE : DEGREE OF A DIVISOR = DEGREE OF THE

CORRESPONDING LINE BUNDLE ( JUSTIFYING R EMARK 16.3.8). Suppose L is an
invertible sheaf on a projective (pure dimension 1) curve C over k, and let s be a
nonzero rational section of L on C, which is regular at all singular points of C. Let
D be the divisor of zeros and poles of s:
X
D := vp (s)[p].
p∈C

Show that deg L = deg D. In particular, the degree of a line bundle on a regular
curve can be computed by counting zeros and poles of any rational section not
vanishing on a component of C.
The next few exercises give useful applications.

18.4.D. E XERCISE . Give a new solution to Exercise 16.3.E (a nonzero rational

function on a projective curve has the same number of zeros and poles, counted
appropriately) using the ideas above.

18.4.E. E XERCISE . Suppose s1 and s2 are two sections of a degree d line bundle
L on an irreducible regular projective curve C, with no common zeros. Then s1
and s2 determine a morphism π : C → P1k . Show that π is a finite morphism, and
that the degree of π is d. (Translation: a two-dimensional base-point-free degree d
linear series on C defines a degree d finite cover of P1 .)

18.4.F. E XERCISE ( TO BE EXTENDED IN E XERCISE 18.4.M). If L and M are two line

bundles on a regular projective curve C, show that deg L ⊗ M = deg L + deg M .
(Hint: choose nonzero rational sections of L and M .)

18.4.3. Remark: The “degree” map Pic → Z is a homomorphism. Thus for a regular
projective curve C (over a field k), the degree map deg : Pic → Z is a homomor-
phism of abelian groups. (This result will be extended to all proper curves over k
in Exercise 18.4.M.)

18.4.G. E XERCISE . Suppose π : C → C ′ is a degree d morphism of integral projec-

tive regular curves, and L is an invertible sheaf on C ′ . Show that degC π∗ L =
d degC ′ L . Hint: compute degC ′ L using any nonzero rational section s of L ,
and compute degC π∗ L using the rational section π∗ s of π∗ L . Note that zeros
pull back to zeros, and poles pull back to poles. Depending on your approach,
Exercise 16.3.D might help.

18.4.H. ⋆⋆ E XERCISE ( COMPLEX - ANALYTIC INTERPRETATION OF DEGREE ; ONLY FOR

THOSE WITH SUFFICIENT ANALYTIC BACKGROUND ). Suppose X is a connected
regular projective complex curve. Show that the degree map is the composition of
group homomorphisms
∩[Xan ]
Pic X / Pic Xan c1
/ H2 (Xan , Z) ∼ Z.
/ H0 (Xan , Z) =

Hint: show it for a generator O(p) of the group Pic X, using explicit transition
functions. (The first map was discussed in Exercise 14.1.J. The second map takes a
line bundle to its first Chern class, and can be interpreted as follows. The transition
functions for a line bundle yield a Čech 1-cycle for OX∗ an ; this yields a map Pic Xan →
510 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

H1 (Xan , OX∗ an ). Combining this with the map H1 (Xan , OX∗ an ) → H2 (Xan , Z) from
the long exact sequence in cohomology corresponding to the exponential exact
sequence (2.4.9.1) yields the first Chern class map.)

18.4.4. Arithmetic genus.

Motivated by geometry (Miracle 18.4.5 below), we define the arithmetic genus
of a dimension n scheme X as pa (X) := (−1)n (χ(X, OX ) − 1). For integral projective
curves over an algebraically closed field, h0 (X, OX ) = 1 (§11.4.7), so pa (X) =
h1 (X, OX ). (In higher dimension, this is a less natural notion.)
We can restate the Riemann-Roch formula for curves (Exercise 18.4.B) as:
(18.4.4.1) h0 (C, L ) − h1 (C, L ) = deg L − pa (C) + 1.
This is the most common formulation of the Riemann-Roch formula.

18.4.5. Miracle. If C is a regular irreducible projective complex curve, then the cor-
responding complex-analytic object, a compact Riemann surface, has a notion called
the genus g, which is (informally speaking) the number of holes (see Figure 18.3).
Miraculously, g = pa in this case (see Exercise 21.4.I), and for this reason, we will
often write g for pa when discussing regular (projective irreducible) curves, over
any field. We will discuss genus further in §18.6.7, when we will be able to compute
it in many interesting cases. (Warning: the arithmetic genus of P1C as an R-variety is
−1!)

F IGURE 18.3. A genus 3 Riemann surface

18.4.6. Degree of a coherent sheaf (e.g., degree of a vector bundle).

Suppose C is an irreducible reduced curve (possibly singular), over a field k. If
F is a coherent sheaf on C, recall (from §14.3.3) that the rank of F , denoted rank F ,
is its rank at the generic point of C.

18.4.7. Definition. If C is projective, define the degree of F by

(18.4.7.1) deg F = χ(C, F ) − (rank F ) · χ(C, OC ).
The statement (18.4.7.1) is often called Riemann-Roch for coherent sheaves (or
vector bundles) on a projective curve.
If F is an invertible sheaf (or if more generally the rank is the same on each
irreducible component), we can drop the irreducibility hypothesis.

18.4.I. E ASY E XERCISE . Show that degree (as a function of coherent sheaves on a
fixed curve C) is additive in exact sequences.

18.4.J. E XERCISE . If F is a torsion sheaf on C, show that deg F ≥ 0, with equality

if and only F = 0.
 © 2024 Ravi Vakil. Published by Princeton University Press. 511

18.4.K. E XERCISE . If k = k, and F is a torsion sheaf on C, show that deg F = ℓ(F ).

(The length ℓ(F ) of F was defined in §6.5.8.) What goes wrong if k ̸= k?

18.4.L. ⋆ E XERCISE . Suppose C is a projective curve over a field k, L is a line

bundle on C, and F is a coherent sheaf on C. Using the following steps, show that
χ(C, L ⊗ F ) − χ(C, F ) is the sum over the irreducible components Ci of C of the
degree of L on Credi (the reduction was defined in §9.4.10) times the length of F
at the generic point ηi of Ci (the length of Fηi as an Oηi -module — length was
defined in Definition 6.5.2). This could reasonably be called Riemann-Roch for
nonreduced curves.
(a) First reduce to the case where F is scheme-theoretically supported on Cred
(you defined the scheme-theoretic support of a finite type quasicoherent sheaf in
Exercise 9.1.N), by showing that both sides of the alleged equality are additive in
short exact sequences, and using the filtration
0 = I r F ⊂ I r−1 F ⊂ · · · ⊂ I F ⊂ F
of F , where I is the ideal sheaf cutting out Cred in C. Thus we need only consider
the case where C is reduced.
(b) In this “reduced” case, by Exercise 16.2.K, we can write L = ∼ O(P nj pj ) where
the pj are regular points distinct from the associated points of F . By an appropriate
induction, prove what remains to be proved.

18.4.M. ⋆ E XERCISE . If L and M are line bundles on a projective curve C over a

field k, show that degC (L ⊗ M ) = degC (L ) + degC (M ). Thus the degree map
deg : Pic C → Z is a homomorphism of abelian groups, extending Remark 18.4.3.

18.4.8. In fact, all proper curves over k are projective (Remark 19.2.I), so “projective”
can be replaced by “proper” in Exercises 18.4.L and 18.4.M. In this guise, we will
use Exercise 18.4.L when discussing intersection theory in Chapter 20.

18.4.9. Two properties of ample bundles on curves. For Exercises 18.4.N and 18.4.O,
either assume C is regular (so you can invoke Remark 18.4.3), or use starred
Exercise 18.4.M.

18.4.N. E XERCISE . If C is a projective curve, and L is an ample line bundle on C,

show that deg L > 0. (Hint: show it if L is very ample.)

18.4.O. E XERCISE . Suppose L is a base-point-free invertible sheaf on a proper

variety X, and hence induces some morphism ϕ : X → Pn . Then L is ample if and
only if ϕ is finite. (Hint: if ϕ is finite, use Exercise 16.2.G. If ϕ is not finite, show
that there is a curve C contracted by ϕ, using Theorem 18.1.6. Show that L has
degree 0 on C.)

18.4.10. ⋆ Extended example: the universal plane conic has no rational sections.
We use the theory of the degree to get an interesting consequence. We work
over a fixed field k. We consider the following diagram.
cl. emb. 2
C / P × P5 / P2

# 
P5
512 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

If the P2 has projective coordinates x0 , x1 , x2 , then P5 has coordinates a00 , a01 , a11 ,
a02 , a12 , a22 , and C is cut out by the single equation
a00 x20 + a01 x0 x1 + · · · + a22 x22 = 0.
We interpret P5 as the parameter space of conics (in P2 ), and C as the universal
conic over P5 (parametrizing a conic C along with a point p ∈ C), which comes
with a canonical projection C → P2 .

18.4.P. E XERCISE . By interpreting C as a P4 -bundle over P2 (with two irreducible

divisors whose complement is an A4 -bundle over A2 ), show that C is a smooth
∼ Z × Z.
sixfold, and that Pic C =

18.4.Q. E XERCISE . Fix a line ℓ ⊂ P2 and a k-valued point q ∈ P2 . Let Dℓ be the

divisor on C corresponding to (C, p) with p lying on ℓ. Let Dq be the divisor on C
corresponding to (C, p) with q ∈ C. Using your description of C as a P4 -bundle
over P2 , show that Dℓ and Dq generate Pic C .

18.4.R. E XERCISE . Suppose K is a fiber of C → P5 over a point r ∈ P5 — i.e., a conic

in P2 over the field κ(r). Suppose further that K is integral, and that neither q nor ℓ
are contained in K. (This hypothesis is unnecessary, but simplifies the problem.)
Show that degK O(Dq ) = 0 and degK O(Dℓ ) = 2. Hence show that if L is any
invertible sheaf on C , then degK (L ) is even.

18.4.S. E XERCISE . Show that there is no rational section to the projection π : C → P5 .

Hint: if there were a regular section over an open subset U of P5 , it would be a
divisor on π−1 (U); let D be its closure in C . Show that D meets any fiber of π over
U in multiplicity 1. Use Exercise 18.4.R to obtain a contradiction.
We can restate Exercise 18.4.S in the following dramatic way: there is no way
to write down three rational functions X0 , X1 , X2 in a00 , . . . , a22 such that
a00 X20 + a01 X0 X1 + · · · + a22 X22 = 0
without X0 = X1 = X2 = 0.
The question of a rational point on a conic is one of arithmetic. (Think: x2 +y2 =
2
z .) Our solution was topological. The unification of topology and arithmetic in
this example is the beginning of a long and fruitful story in algebraic geometry.

18.4.11. ⋆ Numerical equivalence, the Néron-Severi group, nef line bundles, and
the nef and ample cones.
Suppose X is a proper k-variety, and L is an invertible sheaf on X. If i : C ,→ X
is a one-dimensional closed reduced subscheme of X, define the degree of L on
C by degC L := degC i∗ L . If degC L = 0 for all C, we say that L is numerically
trivial.

18.4.T. E XERCISE .
(a) Show that L is numerically trivial if and only if degC L = 0 for all integral
curves C in X.
(b) Show that if π : X → Y is a proper morphism, and L is a numerically trivial
invertible sheaf on Y, then π∗ L is numerically trivial on X.
(c) Show that L is numerically trivial if and only if L is numerically trivial on each
of the irreducible components of X.
 © 2024 Ravi Vakil. Published by Princeton University Press. 513

(d) Show that if L and L ′ are numerically trivial, then L ⊗ L ′ and L ∨ are both
numerically trivial.

18.4.12. Numerical equivalence. By part (d), the numerically trivial invertible sheaves
form a subgroup of Pic X, denoted Picτ X. Two lines bundles equivalent modulo
the subgroup of numerically trivial line bundles are called numerically equivalent.
A property of invertible sheaves stable under numerical equivalence is said to
be a numerical property. We will see that “nefness” and ampleness are numerical
properties (Definition 18.4.13 and Remark 20.4.2 respectively).
We will later define the Néron-Severi group NS(X) of X as Pic X modulo algebraic
equivalence (Exercise 24.7.5). (We will define algebraic equivalence once we have
discussed flatness.) The highly nontrivial Néron-Severi Theorem (or Theorem
of the Base) states that NS(X) is a finitely generated group. (See [ChJLO] for a
thorough modern discussion, and [GH1, p. 462] for a simpler proof over C.) The
group Pic X/ Picτ X is denoted N1 (X). We will see (in §24.7.5) that it is a quotient of
NS(X), so it is also finitely generated. As the group N1 (X) is clearly abelian and
torsion-free, it is finite free Z-module (by the classification of finitely generated
modules over a principal ideal domain, see §0.3). The rank of N1 (X) is called
the Picard number, and is denoted ρ(X) (although we won’t have need of this
notion, except in our discussion of the Hodge Index Theorem in §20.2.11). For
example, ρ(Pn ) = 1 and ρ((P1 )n ) = n. We define N1Q (X) := N1 (X) ⊗Z Q (so
ρ(X) = dimQ N1Q (X)), and call the elements of this group Q-line bundles, for lack
of any common term in the literature.

18.4.U. ⋆⋆ E XERCISE ( FINITENESS OF P ICARD NUMBER IN THE COMPLEX CASE , ONLY

FOR THOSE WITH SUFFICIENT BACKGROUND ). Show (without the Néron-Severi
Theorem) that if X is a complex proper variety, then ρ(X) is finite, by interpreting it
as a subquotient of H2 (X, Z). Hint: show that the image of (L , C) under the map
H2 (X, Z) × H2 (X, Z) → H0 (X, Z) → Z is degC L . Hint: figure out how to reduce to
the case where C is a smooth projective curve, then use Exercise 18.4.H.

18.4.13. Definition. We say that an invertible sheaf L is numerically effective, or

nef if for all such C, degC L ≥ 0. Clearly nefness is a numerical property.

18.4.V. E ASY E XERCISE ( CF. E XERCISE 18.4.T).

(a) Show that L is nef if and only if degC L ≥ 0 for all integral curves C in X.
(b) Show that if π : X → Y is a proper morphism, and L is a nef invertible sheaf on
Y, then π∗ L is nef on X. (Hint: Exercise 18.4.G will be needed.)
(c) Show that L is nef if and only if L is nef on each of the irreducible components
of X.
(d) Show that if L and L ′ are nef, then L ⊗ L ′ is nef. Thus the nef elements of
Pic X form an abelian semigroup.
(e) Show that ample invertible sheaves are nef.
(f) Suppose n ∈ Z>0 . Show that L is nef if and only if L ⊗n is nef.

18.4.W. E XERCISE . Define what it means for a Q-line bundle to be nef. Show that
the nef Q-line bundles form a closed cone in N1Q (X). This is called the nef cone.

18.4.X. E XERCISE . Describe the nef cones of P2k and P1k ×k P1k . (Notice in the latter
case that the two boundaries of the cone correspond to linear series contracting one
514 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

of the P1 ’s. This is true in general: informally speaking, linear series corresponding
to the boundaries of the cone give interesting contractions. Another example will
be given in Exercise 20.2.F.)
It is a surprising fact that whether an invertible sheaf L on X is ample depends
only on its class in N1Q (X), i.e., on how it intersects the curves in X. Because of this
(as for any n ∈ Z≥0 , L is ample if and only if L ⊗n is ample, see Theorem 16.2.2),
it makes sense to define when a Q-line bundle is ample. Then by Exercise 16.2.H,
the ample divisors form a cone in N1Q (X), necessarily contained in the nef cone by
Exercise 18.4.V(e). It turns out that if X is projective, the ample divisors are precisely
the interior of the nef cone. The new facts in this paragraph are a consequence of
Kleiman’s criterion for ampleness, Theorem 20.4.6.

18.5 A first glimpse of Serre duality

A common version of Riemann-Roch involves Serre duality, which unlike
Riemann-Roch, is hard to prove.

18.5.1. Theorem (Serre duality for smooth projective varieties). — Suppose X is

a geometrically irreducible smooth projective k-variety, of dimension n. Then there is an
invertible sheaf ωX (or simply ω) on X such that
hi (X, F ) = hn−i (X, ωX ⊗ F ∨ )
for all i ∈ Z and all finite rank locally free sheaves F .
The invertible sheaf ωX is a called the dualizing sheaf, which will be formally
defined in §29.1.5. We will see in Chapter 29 that Theorem 18.5.1 is a consequence
of a perfect pairing

(18.5.1.1) Hi (X, F ) × Hn−i (X, ωX ⊗ F ∨ ) ∼ k,

/ Hn (X, ωX ) =

and that smoothness can be relaxed somewhat.

18.5.2. Further miracle: the sheaf of algebraic volume forms is Serre-dualizing. The
invertible sheaf ωX turns out to be the canonical sheaf (or canonical bundle) KX ,
which is defined as the determinant of the cotangent bundle ΩX/k of X (see §21.5.3);
we will define the cotangent bundle in Chapter 21. We connect the dualizing sheaf
to the canonical bundle in §29.4; see Desideratum 29.1.1.

18.5.3. Back to Riemann-Roch. For the purposes of restating Riemann-Roch for a

curve C, it suffices to note that h1 (C, L ) = h0 (C, ωC ⊗ L ∨ ). Then the Riemann-
Roch formula can be rewritten as
h0 (C, L ) − h0 (C, ωC ⊗ L ∨ ) = deg L − pa (C) + 1.

18.5.A. E XERCISE ( ASSUMING S ERRE DUALITY ). Suppose C is a geometrically

integral smooth projective curve over k.
(a) Show that h0 (C, ωC ) is the genus g of C. (The genus g, in the guise of the
arithmetic genus pa , was defined in §18.4.4–18.4.5.)
(b) Show that deg ωC = 2g − 2. (Hint: Riemann-Roch for L = ωC .)
 © 2024 Ravi Vakil. Published by Princeton University Press. 515

∼ O(−2). Moreover,
18.5.4. Example. If C = P1k , Exercise 18.5.A implies that ωC =
we also have a natural perfect pairing (cf. (18.5.1.1))

H0 (P1 , O(n)) × H1 (P1 , O(−2 − n)) ∼ H1 (P1 , O(−2)).

/k=

We can interpret this pairing as follows. If n < 0, both factors on the left are 0,
so we assume n > 0. Then H0 (P1 , O(n)) corresponds to homogeneous degree
n polynomials in x and y, and H1 (P1 , O(−2 − n)) corresponds to homogeneous
degree −2−n Laurent polynomials in x and y so that the degrees of x and y are both
at most −1 (see Remark 18.3.5). You can quickly check that the dimension of both
vector spaces are n + 1. The pairing is given as follows: multiply the polynomial by
the Laurent polynomial, to obtain a Laurent polynomial of degree −2. Read off the
coefficient of x−1 y−1 . (This works more generally for PnA ; see the discussion after
the statement of Theorem 18.1.2.)

18.5.B. E XERCISE ( ASSUMING S ERRE DUALITY ): AMPLE DIVISORS ON A CONNECTED

SMOOTH PROJECTIVE VARIETY ARE CONNECTED . Suppose X is a connected smooth
projective k-variety of dimension at least 2, and D is an effective ample divisor.
Show that D is connected. Hint: Suppose D = V(s), where s is a section of an
ample invertible sheaf L . Then V(sn ) = V(s) (as vanishing sets, not vanishing
schemes) for all n > 0, so we may replace L with a high power of our choosing.
Use the long exact sequence for 0 → OX (−nD) → OX → OV(sn ) → 0 to show that
for n ≫ 0, h0 (X, OV(sn ) ) = 1. (This may remind you of §11.4.7.)
Once we know that Serre duality holds for Cohen-Macaulay projective schemes
(§29.3), this result will automatically extend to these schemes. (A related result is
Exercise 18.6.U, which doesn’t use Serre duality.) On the other hand, the result is
false if X is the union of two 2-planes in P4 meeting at a point (why?), so this will
imply that this X is not Cohen-Macaulay. (We will show this in another way in
Counterexample 26.2.2.)

18.5.5. ⋆ Classification of vector bundles on P1k .

As promised in Example 17.2.6, we classify the vector bundles on P1k . We
discuss this in §18.5 because you might use Serre duality at one step (although you
might not).
∼ O(a1 ) ⊕ · · · ⊕ O(ar ),
18.5.6. Theorem. — If E is a rank r vector bundle on P1k , then E =
for a unique nondecreasing sequence of integers a1 , . . . , ar .
This result was proved independently many times (see for example Dedekind
and Weber’s article [DW, §22]), and is a special case of a theorem of Grothendieck,
[Gr2]. It is sometimes called Grothendieck’s Theorem, because Grothendieck
doesn’t have enough theorems named after him.

18.5.7. For Pn k more generally, the case r = 1 was shown in §15.4.14, but the
statement is false for r > 1 and n > 1. A counterexample is given in Exercise 21.3.H,
which shows that it is not even possible to filter the rank two bundle ΩP2 on P2
by line bundles. One true generalization is a theorem of Horrocks, which states
that a finite rank locally free sheaf E on Pn
k splits precisely when “all the middle
cohomology of all of its twists is zero” — when Hi (Pn k , E (m)) = 0 for 0 < i < n
and all m ∈ Z. This has the surprising consequence that if n ≥ 2, then a finite
516 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

rank locally free sheaf on Pn k splits if and only if its restriction to some previously
chosen 2-plane (P2k ) splits — all of the additional complication turns up already in
dimension 2. See [OSS, §2.3] for more.

Proof. Note that the classification makes no reference to cohomology, but the proof
uses cohomology in an essential way. It is possible to prove Theorem 18.5.6 with no
cohomological machinery (see for example [HM] or [GW, p. 314-5]), or with more
cohomological machinery (see for example [Ha1, Ex. V.2.6]).

18.5.C. E XERCISE . Suppose that E = ∼ O(a1 ) ⊕ · · · ⊕ O(ar ), with a1 ≤ · · · ≤ ar .

Show that the ai can be determined from the numbers dimk Hom(O(m), E ), as m
ranges over the integers. Use this to show the uniqueness part of Theorem 18.5.6.
We now begin the proof, by induction on r. Fix a rank r locally free sheaf E .
The case r = 1 was established in §15.4.14, so we assume r > 1.

18.5.D. E XERCISE . Show that for m ≪ 0, Hom(O(m), E ) ̸= 0, and that for

m ≫ 0, Hom(O(m), E ) = 0. Hint: show that Hom (O(m), E ) = E (−m) using
Exercise 14.1.G. Use Serre’s Theorem A (Theorem 16.1.5) for the first part. Feel free
to use Serre duality and Serre vanishing for the second part. But you may prefer to
come up with an argument without Serre duality, to avoid invoking something we
have not yet proved.
Thus there is some ar for which dim Hom(O(ar ), E ) > 0, but for which
Hom(O(m), E ) = 0 for all m > ar . Choose a nonzero map ϕ : O(ar ) → E .

18.5.E. E XERCISE . Show that ϕ is an injection. (Hint: O(ar ) is torsion-free, and

thus the kernel is torsion-free.)

18.5.F. E XERCISE . Let F be the cokernel of ϕ. Show that F is locally free. Hint:
Exercise 14.3.F(c) gives an exact sequence 0 → Ftors → F → Flf → 0, where Flf is
locally free. Let L be the kernel of the surjection E → Flf . Show that L is locally
free, and thus is isomorphic to O(N) for some N. Show that there is a nonzero
map ϕ ′ : O(ar ) → O(N). Show (using the same idea as the previous exercise) that
this nonzero map ϕ ′ must be an injection, so ar ≤ N. Show that N ≤ ar because
there is a nonzero map O(N) → E (recall how ar was chosen). Show that ϕ ′ is an
isomorphism, and thus that F = Flf .
By our inductive hypothesis, we have F = O(a1 ) ⊕ · · · ⊕ O(ar−1 ), where
a1 ≤ · · · ≤ ar−1 , so we have a short exact sequence

(18.5.7.1) 0 / O(ar ) /E / O(a1 ) ⊕ · · · ⊕ O(ar−1 ) / 0.

We next show that ar ≥ ai for i < r. We are motivated by the fact that for a
quasicoherent sheaf G on P1 ,
HomP1 (O(ar + 1), G ) = H0 (P1 , G (−ar − 1))
(Exercise 14.1.G). Tensor (18.5.7.1) with O(−ar − 1) (preserving exactness, by Exer-
cise 14.1.F), and take the long exact sequence in cohomology. Part of the long exact
sequence is

H0 (P1 , E (−ar − 1)) / H0 (P1 , F (−ar − 1)) / H1 (P1 , O(−1)).

 © 2024 Ravi Vakil. Published by Princeton University Press. 517

Notice that H0 (P1 , E (−ar − 1)) = 0 (as Hom(O(ar + 1), E ) = 0, by the definition
of ar ), and H1 (P1 , O(−1)) = 0 (Theorem 18.1.2(b)). Thus
0 = H0 (P1 , F (−ar − 1))
= H0 (P1 , ⊕i O(ai − ar − 1))
= ⊕i H0 (P1 , O(ai − ar − 1)).
Hence ai − ar − 1 < 0 (by Exercise 15.1.A), so ar ≥ ai as desired.
Finally, we wish to show that exact sequence (18.5.7.1) expresses E as a direct
sum (of the subsheaf and quotient sheaf). For simplicity, we focus on the case r = 2,
and return to the general case in Exercise 18.5.I.

18.5.G. E XERCISE . Show that the transition functions for the vector bundle, in
appropriate coordinates, are given by
 −a 
t 2 α(t)
,
0 t−a1
where α(t) is a Laurent polynomial in t. Hint/reminder: Recall that all vector
bundles on A1k are trivial, Exercise 14.2.C. Transition matrices for extensions of
one vector bundle (with known transition matrices) by another were discussed in
Exercise 14.2.G.

18.5.H. E XERCISE . Implicit in the above 2 × 2 matrix is a choice of a basis of a rank

2 free module M over the ring k[t] corresponding to one of the standard affine open
subsets Spec k[t], and a rank 2 free module M ′ over the ring k[1/t] over the other
standard open subset. Show that by an appropriate “upper-triangular” change
of basis of M, you can arrange for α(t) to have no monomials of degree ≥ −a2 .
Show that by an appropriate “upper-triangular” change of basis of M ′ , you can
arrange for α(t) to have no monomials of degree ≤ −a1 . Thus by choosing bases
of M and M ′ appropriately, we can take α(t) = 0. Show that this implies that
E = O(a1 ) ⊕ O(a2 ).

18.5.I. E XERCISE . Finish the proof of Theorem 18.5.6 for arbitrary r. Hint: there are
no new ideas beyond the case r = 2.
□

18.6 Hilbert functions, Hilbert polynomials, and genus

If F is a coherent sheaf on a projective k-scheme X ⊂ Pn , define the Hilbert
function of F by
hF (m) := h0 (X, F (m)).
Note that the Hilbert function depends on the projective embedding of X (by way
of its line bundle), although this is dependency is suppressed in the notation. The
Hilbert function of X is the Hilbert function of the structure sheaf. The ancients
were aware that the Hilbert function is “eventually polynomial”, i.e., for large
enough m, it agrees with some polynomial. This polynomial contains lots of
interesting geometric information, as we will soon see. In modern language, we
expect that this “eventual polynomiality” arises because the Euler characteristic
518 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

should be a polynomial, and that for m ≫ 0, the higher cohomology vanishes. This
is indeed the case, as we now verify.

18.6.1. Theorem. — If F is a coherent sheaf on a projective k-scheme X ,→ Pn k,

χ(X, F (m)) is a polynomial of degree equal to dim Supp F . Hence by Serre vanish-
ing (Theorem 18.1.3 (ii)), for m ≫ 0, h0 (X, F (m)) is a polynomial pF (m) of degree
dim Supp F . In particular, for m ≫ 0, h0 (X, OX (m)) is polynomial with degree equal to
dim X.

18.6.2. Definition. The polynomial pF (m) defined in Theorem 18.6.1 is called the
Hilbert polynomial. If X ⊂ Pn is a projective k-scheme, define pX (m) := pOX (m).

18.6.3. In Theorem 18.6.1, OX (m) is the restriction or pullback of OPnk (m). Both the
degree of the 0 polynomial and the dimension of the empty set are defined to be −1.
In particular, the only coherent sheaf with Hilbert polynomial 0 is the zero-sheaf.
This argument uses the notion of associated points of a coherent sheaf on a
locally Noetherian scheme, §6.6.2. (The resolution given by the Hilbert Syzygy
Theorem 16.1.3 can give a shorter proof; but we won’t prove it until § 24.4.)

Proof. Define pF (m) = χ(X, F (m)). We will show that pF (m) is a polynomial of
the desired degree.
We first use Exercise 18.2.H to reduce to the case where k is algebraically closed,
and in particular infinite. (This is one of those cases where even if you are concerned
with potentially arithmetic questions over some non-algebraically closed field like
Fp , you are forced to consider the “geometric” situation where the base field is
algebraically closed.)
The coherent sheaf F has a finite number of associated points (§6.6.17). We
show a useful fact that we will use again.

18.6.A. E XERCISE . Suppose X is a projective k-scheme with k infinite, and F is

a coherent sheaf on X. Show that if L is a very ample invertible sheaf on X, then
there is an effective Cartier divisor D on X with L = ∼ O(D), and where D does not
contain the associated points of F . (Hint: show that given any finite set of points
of Pn
k , there is a hyperplane not containing any of them. This is a variant of the key
step in Exercise 12.3.D(c).)
Thus there is a hyperplane x = 0 (x ∈ Γ (X, O(1))) missing this finite number of
points. (This is where we use the infinitude of k.)
Then the map F (−1)
×x
/ F is injective. (On any affine open subset F
corresponds to a module, and x is not a zerodivisor on that module, as it doesn’t
vanish at any associated point of that module, see Proposition 6.6.13.) Thus we
have a short exact sequence

(18.6.3.1) 0 / F (−1) /F /G /0

where G is a coherent sheaf.

Now Supp G = (Supp F )∩V(x). (Reason: Exercise 6.6.D gives the affine/module
version of this statement.) Hence V(x) meets all positive-dimensional components
 © 2024 Ravi Vakil. Published by Princeton University Press. 519

of Supp F (Exercise 12.3.D(a)), so dim Supp G = dim Supp F − 1 by Krull’s Princi-

pal Ideal Theorem 12.3.3 unless F = 0 (in which case we already know the result,
so assume this is not the case).
Twisting (18.6.3.1) by O(m) yields

0 / F (m − 1) / F (m) / G (m) /0

Euler characteristics are additive in exact sequences (Exercise 18.4.A), from which
pF (m)−pF (m−1) = pG (m). Now pG (m) is a polynomial of degree dim Supp F −
1.
The result is then a consequence from the following elementary fact about
polynomials in one variable.

18.6.B. E XERCISE . Suppose f and g are functions on the integers, f(m + 1) − f(m) =
g(m) for all m, and g(m) is a polynomial of degree d ≥ 0. Show that f is a
polynomial of degree d + 1.
□
Example 1. The Hilbert polynomial of projective space is pPn (m) = m+n


m ,
where we interpret this as the polynomial (m + 1) · · · (m + n)/n!.
Example 2. Suppose H is a degree d hypersurface in Pn , with i : H ,→ Pn the
closed embedding. Then from the closed subscheme exact sequence

0 / OPn (−d) / OPn / i∗ OH / 0,

we have
   
m+n m+n−d
(18.6.3.2) pH (m) = p Pn (m) − p Pn (m − d) = − .
n n
∼ i∗ (OH ⊗ i∗ OPn (m)).
(Implicit in this argument is the fact that (i∗ OH ) ⊗ OPn (m) =
This follows from the projection formula, Exercise 14.5.L(b). You can also show this
directly with transition functions.)

18.6.C. E XERCISE . Show that the twisted cubic (P1 embedded in P3 by the complete
linear series |OP1 (3)|) has Hilbert polynomial 3m+1. (The twisted cubic was defined
in Exercise 9.3.A.)

18.6.D. E XERCISE . More generally, find the Hilbert polynomial for the dth Veronese
embedding of Pn (i.e., the closed embedding of Pn in a bigger projective space by
way of the line bundle O(d), §9.3.6).

18.6.E. E XERCISE ( TO BE USED SEVERAL TIMES IN C HAPTER 19). Suppose

 
X / Y / Pn
i
k

is a sequence of closed embeddings.

(a) Show that pX (m) ≤ pY (m) for m ≫ 0. Hint: let IX/Y be the ideal sheaf of X in
Y. Consider the exact sequence

0 / IX/Y (m) / OY (m) / i∗ OX (m) / 0.

520 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

(b) If pX (m) = pY (m) for m ≫ 0, show that X = Y. Hint: Show that if the Hilbert
polynomial of a coherent sheaf F is 0, then F = 0. (Handy trick: For m ≫ 0,
F (m) is generated by global sections.) Apply this to F = IX/Y .
From the Hilbert polynomial, we can extract many invariants, of which two
are particularly important. The first is the degree, and the second is the arithmetic
genus (§18.6.7). The degree of a projective k-scheme of dimension n is defined to
be leading coefficient of the Hilbert polynomial (the coefficient of mn ) times n!.
Using the examples above, we see that the degree of Pn in itself is 1. The degree
of the twisted cubic is 3.

18.6.F. E XERCISE . Show that the degree is always an integer. Hint: by induction,
show that any polynomial in m of degree k taking on only integer values must
have coefficient of mk an integral multiple of 1/k!. Hint for this: if f(x) takes on
only integral values and is of degree k, then f(x + 1) − f(x) takes on only integral
values and is of degree k − 1.

18.6.G. E XERCISE . Show that the degree of a degree d hypersurface (Defini-

tion 9.3.2) is d (preventing a notational crisis).

18.6.H. E XERCISE ( DEGREE = DEGREE ). Suppose a regular curve C is embedded

in projective space via an invertible sheaf of degree d (as defined in §18.4.6). In
other words, this line bundle determines a closed embedding. Show that the
degree of C under this embedding is d, preventing another notational crisis. (A
similar notational crisis was averted in Exercise 18.4.E.) Hint: Riemann-Roch,
Exercise 18.4.B. (The hypothesis of regularity is not necessary. It is only included
to make this exercise easier.)

18.6.I. E XERCISE . Show that the degree of the dth Veronese embedding of Pn is dn .

18.6.J. E XERCISE (B ÉZOUT ’ S T HEOREM , GENERALIZING E XERCISES 9.3.D AND 15.2.I).

Suppose X ⊂ Pn k is a projective scheme of dimension at least 1, and H is a hypersur-
face not containing any associated points of X. (For example, if X is reduced and
thus has no embedded points, we are just requiring H not to contain any irreducible
components of X.) Show that deg(H ∩ X) = (deg H)(deg X).

18.6.4. Bézout’s Theorem for plane curves. As a consequence, we have Bézout’s Theorem
for plane curves: if C and D are plane curves of degrees m and n respectively, with no
common components, then C and D meet at mn points, counted with appropriate
multiplicity. (To apply Exercise 18.6.J, you need to know that plane curves have no
embedded points. You can either do this using Exercise 6.6.F, or by assuming that
one of the curves is reduced.) The notion of multiplicity of intersection is just the
degree of the intersection as a k-scheme.

18.6.K. E XERCISE . Suppose C is a degree 1 curve in P3k (or more precisely, a degree
1 pure-one-dimensional closed subscheme of P3k , with no embedded points). Show
that C is a line. Hint: reduce to the case k = k. Suppose p and q are distinct closed
points on C. Use Bézout’s Theorem (Exercise 18.6.J) to show that any hyperplane
containing p and q must contain C, and thus that C ⊂ pq.
 © 2024 Ravi Vakil. Published by Princeton University Press. 521

18.6.L. F UN E XERCISE . Let k be a field, which we assume to be algebraically

closed for convenience (although you are free to remove this hypothesis if you
wish). Suppose C is a degree d integral curve in PN k with N ≥ d. Show that C is
contained in a linear Pdk ⊂ P N
k . In particular, if N > d, C is degenerate (defined in
Exercise 15.2.C). (Exercise 18.6.K is not quite a special case of this problem, but the
hint may still be helpful.)

18.6.M. E XERCISE ( ANOTHER FORM OF B ÉZOUT ’ S T HEOREM ). Classically, the

degree of a complex projective variety of dimension n was defined as follows. We
slice the variety with n generally chosen hyperplanes. Then the intersection will be
a finite number of reduced points, by Exercise 13.4.B (a consequence of Bertini’s
Theorem 13.4.2). The degree is this number of points. Use Bézout’s Theorem to
make sense of this in a way that agrees with our definition of degree.
Thus the classical definition of the degree, which involved making a choice
and then showing that the result is independent of choice, has been replaced by
making a cohomological definition involving Euler characteristics. This should
remind you of how we got around to “correctly” understanding the degree of a
line bundle. It was traditionally defined as the degree of a divisor of any nonzero
rational section (Important Exercise 18.4.C), and we found a better definition in
terms of Euler characteristics (§18.4.6).

18.6.5. ⋆⋆ Aside: Connection to the topological definition of degree (for those with sufficient
background). Another definition of degree of a dimension d complex projective
variety X ⊂ Pn C is as the number d such that [X] is d times the “positive” generator
of H2d (PnC , Z). You can show this by induction on d as follows. Suppose X is a
complex projective variety of dimension d. For a generally chosen hyperplane,
H ∩ X is a complex projective variety of (complex) dimension d − 1. (Do you see
why?) Show that [H ∩ X] = c1 (O(1)) ∪ [X] in H2(d−1) (X, Z), by suitably generalizing
the solution to Exercise 18.4.H. (A further generalization is given in §20.1.9.) For
this reason, c1 (O(1)) is often called the “hyperplane class”.

18.6.6. Revisiting an earlier example. We revisit the enlightening example of Exam-

ple 10.3.4 and §16.3.10: let k = Q, and consider the parabola x = y2 . We intersect it
with the four lines, x = 1, x = 0, x = −1, and x = 2, and see that we get 2 each time
(counted with the same convention as with the last time we saw this example).
If we intersect it with y = 2, we only get one point — but that’s because this
isn’t a projective curve, and we really should be doing this intersection on P2k , and
in this case, the conic meets the line in two points, one of which is “at ∞”.

18.6.N. E XERCISE . Show that the degree of the d-fold Veronese embedding of
Pn is dn in a different way from Exercise 18.6.I as follows. Let vd : Pn → PN be
the Veronese embedding. To find the degree of the image, we intersect it with
n hyperplanes in PN (scheme-theoretically), and find the number of intersection
points (counted with multiplicity). But the pullback of a hyperplane in PN to Pn is
a degree d hypersurface. Perform this intersection in Pn , and use Bézout’s theorem
(Exercise 18.6.J).
522 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

18.6.O. E XERCISE ( DEGREE IS ADDITIVE FOR UNIONS ). Suppose X and Y are

two d-dimensional closed subschemes of Pn
k , with no d-dimensional irreducible
components in common. Show that deg X ∪ Y = deg X + deg Y.

18.6.7. Arithmetic genus, again.

There is another central piece of information residing in the Hilbert polynomial.
Notice that pX (0) = χ(X, OX ) is an intrinsic invariant of the scheme X, independent
of the projective embedding. It has the same information as the arithmetic genus
(−1)dim X (χ(X, OX ) − 1) (§18.4.4).
Imagine how amazing this must have seemed to the ancients: they defined
the Hilbert function by counting how many “functions of various degrees” there
are; then they noticed that when the degree gets large, it agrees with a polynomial;
and then when they plugged 0 into the polynomial — extrapolating backwards, to
where the Hilbert function and Hilbert polynomials didn’t agree — they found a
magic invariant! Furthermore, in the case when X is a complex curve, this invariant
was basically the topological genus!
We can now see a large family of curves over an algebraically closed field that
is provably not P1 ! Note that the Hilbert polynomial of P1 is (m + 1)/1 = m + 1, so
χ(OP1 ) = 1. Suppose C is a degree d curve in P2 . Then the Hilbert polynomial of C
is
pP2 (m) − pP2 (m − d) = (m + 1)(m + 2)/2 − (m − d + 1)(m − d + 2)/2.
Plugging in m = 0 gives us −(d2 − 3d)/2. Thus when d > 2, we have a curve that
cannot be isomorphic to P1 ! (And it is not hard to show that there exists a regular
degree d curve, Exercise 13.3.E.)
Now from 0 → OP2 (−d) → OP2 → OC → 0, using h1 (P2 , OP2 (−d)) = 0, we
have that h0 (C, OC ) = 1. As h0 − h1 = χ, we have
(18.6.7.1) h1 (C, OC ) = (d − 1)(d − 2)/2.
We now revisit an interesting question we first saw in §7.5.10. If k is an alge-
braically closed field, is every finitely generated transcendence degree 1 extension
of k isomorphic to k(x)? In that section, we found ad hoc (but admittedly beautiful)
examples showing that the answer is “no”. But we now have a better answer. The
question initially looks like an algebraic question, but we now recognize it as a
fundamentally geometric one. There is an integer-valued cohomological invariant
of such field extensions that has good geometric meaning: the genus.
Equation (18.6.7.1) yields examples of curves of genus 0, 1, 3, 6, . . . (corresponding
to degree 1 or 2, 3, 4, 5, . . . ). This begs some questions, such as: are there curves of
other genera? (We will see soon, in §19.5.5, that the answer is yes.) Are there other
genus 0 curves? (Not if k is algebraically closed, but sometimes yes otherwise —
consider x2 + y2 + z2 = 0 in P2R , which has no R-points and hence is not isomorphic
to P1R — we will discuss this more in §19.3.) Do we have all the curves of genus 3?
(Almost all, but not quite. We will see more in §19.7.) Do we have all the curves of
genus 6? (We are missing “most of them”, as will be suggested by §19.8.4.)

18.6.8. Caution. The Euler characteristic of the structure sheaf is an incomplete

invariant. We will now see that it doesn’t always distinguish between isomorphism
classes of irreducible smooth projective varieties.

k has degree ≤ n, show that χ(X, OX ) = 1.

18.6.P. E XERCISE . If a hypersurface X ⊂ Pn
 © 2024 Ravi Vakil. Published by Princeton University Press. 523

Thus for example if X = P1k × P1k , then χ(X, OX ) = 1, as X an be described

as a quadric surface in P3k . Then P2k and P1k × P1k both have structure sheaf Euler
characteristic 1 (see Theorem 18.1.2 for the case of P2k ), but are not isomorphic —
∼ Z (§15.4.14) while Pic(P1 × P1 ) =
Pic P2k = ∼ Z ⊕ Z (Exercise 15.5.K).
k k

18.6.9. Complete intersections in Pn k . Recall that a codimension r complete inter-

section in Pn
k is the complete intersection of r hypersurfaces in Pn
k (§9.5.8), which
by Krull’s Height Theorem 12.3.10 has pure codimension r.

18.6.Q. E XERCISE . Suppose X is a dimension m complete intersection in PN

k . Show
that h (X, OX (n)) = 0 for 0 < i < m.
i

18.6.R. E XERCISE . Find the genus of the complete intersection of two quadric
surfaces in P3k .

18.6.S. E XERCISE . More generally, find the genus of the complete intersection of a
degree m surface with a degree n surface in P3k . (If m = 2 and n = 3, you should
get genus 4. We will see in §19.8 that in some sense most genus 4 curves arise in
this way. Note that Bertini’s Theorem 13.4.2 ensures that there are regular curves of
this form.)

18.6.T. E XERCISE . Show that the rational normal curve of degree d in Pd is not a
complete intersection if d > 2. (Hint: If it were the complete intersection of d − 1
hypersurfaces, what would the degree of the hypersurfaces be? Why could none of
the degrees be 1?)

18.6.U. E XERCISE ( POSITIVE - DIMENSIONAL COMPLETE INTERSECTIONS IN Pn k ARE

CONNECTED , CF. E XERCISE 18.5.B). Show that complete intersections in Pn k of
positive dimension are connected. Hint: show that h0 (X, OX ) = 1. (This argument
will even show that they are geometrically connected, defined in §10.5.1, as h0 is
preserved by field extension, see Exercise 18.2.H.)
Notice that this statement does not require X to be reduced, so for example this
implies that that if X is the scheme cut out by x2 = 0 in P2k , then h0 (X, OX ) = 1 —
there are no nonzero nilpotent functions, even though there are local functions that
are nonzero and nilpotent! This example came up earlier in §5.2.3 and §11.4.7.

18.6.V. E XERCISE . Show that the union of two planes in P4 meeting at a point is
not a complete intersection. Hint: it is connected, but you can slice with another
hyperplane and get something not connected (see Exercise 18.6.U).

18.7 Higher pushforward (or direct image) sheaves

Cohomology groups were defined for X → Spec A where the structure mor-
phism is quasicompact and separated; for any quasicoherent F on X, we defined
Hi (X, F ). We will now define a “relative” version of this notion, for quasicompact
and separated morphisms π : X → Y: for any quasicoherent F on X, we will define
Ri π∗ F , a quasicoherent sheaf on Y. (Now would be a good time to do Exercise 1.5.I,
the FHHF Theorem, if you haven’t done it before.)
524 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

We have many motivations for doing this. In no particular order:

(1) It “globalizes” what we did before with cohomology.
(2) If 0 → F → G → H → 0 is a short exact sequence of quasicoherent
sheaves on X, then we know that 0 → π∗ F → π∗ G → π∗ H is exact, and
higher pushforwards will extend this to a long exact sequence.
(3) We will later see that this will show how cohomology groups vary in
families, especially in “nice” situations. Intuitively, if we have a nice
family of varieties, and a family of sheaves on them, we could hope that
the cohomology varies nicely in families, and in fact in “nice” situations,
this is true. (As always, “nice” usually means “flat”, whatever that means.
We will see that Euler characteristics are locally constant in proper flat
families in §24.7, and the Cohomology and Base Change Theorem 25.1.6
will show that in particularly good situations, dimensions of cohomology
groups are constant.)
All of the important properties of cohomology described in §18.1 will carry over to
this more general situation. Best of all, there will be no extra work required.
In the notation Ri π∗ F for higher pushforward sheaves, the “R” stands for
“right derived functor”, and corresponds to the fact that we get a long exact sequence
in cohomology extending to the right (from the 0th terms). In Chapter 23, we will
see that in good circumstances, if we have a left-exact functor, there is a long exact
sequence going off to the right, in terms of right derived functors. Similarly, if
we have a right-exact functor (e.g., if M is an A-module, then ⊗A M is a right-
exact functor from the category of A-modules to itself), there may be a long exact
sequence going off to the left, in terms of left derived functors.
Suppose π : X → Y, and F is a quasicoherent sheaf on X. For each Spec A ⊂ Y,
we have A-modules Hi (π−1 (Spec A), F ). We now show that these patch together
to form a quasicoherent sheaf, in the sense of §6.2.3. We need check only one fact:
that this behaves well with respect to taking distinguished open sets. In other
words, we must check that for each f ∈ A, the natural map
Hi (π−1 (Spec A), F ) / Hi (π−1 (Spec Af ), F )

(induced by the map of spaces in the opposite direction — Hi is contravariant in

the space) is precisely the localization ⊗A Af . But this can be verified easily: let
{Ui } be an affine open cover of π−1 (Spec A). We can compute Hi (π−1 (Spec A), F )
using the Čech complex (18.2.1.1). But this induces a cover of π−1 (Spec Af ) in a
natural way: if Ui = Spec Ai is an affine open subset of π−1 (Spec A), we define
Ui′ := Spec(Ai )π♯ f . The resulting Čech complex for Spec Af is the localization of
the Čech complex for Spec A. As taking cohomology of a complex commutes with
localization (as discussed in the FHHF Theorem, Exercise 1.5.I), we have defined a
quasicoherent sheaf on Y by the characterization of quasicoherent sheaves in §6.2.3.
Define the ith higher pushforward sheaf or the ith higher direct image sheaf
Ri π∗ F to be this quasicoherent sheaf.
Caution: Although when U ⊂ Y is an affine open subset, H0 (U, R• π∗ F ) =
H (π (U), F ), this is not true for open subsets U ⊂ Y in general. You might later
• −1

try to construct an example of this behavior.

18.7.1. Theorem. — Suppose π : X → Y is a quasicompact separated morphism of

schemes. Then:
 © 2024 Ravi Vakil. Published by Princeton University Press. 525

(a) Ri π∗ is a covariant functor QCohX → QCohY .

(b) We can identify R0 π∗ with π∗ . (More precisely: we have a natural isomorphism
of functors.)
(c) (the long exact sequence of higher pushforward sheaves) A short exact
sequence 0 → F → G → H → 0 of sheaves on X induces a long exact sequence

(18.7.1.1) 0 / R0 π∗ F / R0 π∗ G / R0 π∗ H /

R1 π∗ F / R1 π∗ G / R1 π∗ H / ···

of sheaves on Y.
(d) (projective pushforwards of coherent are coherent: Grothendieck’s Coherence
Theorem for projective morphisms) If π is a projective morphism, Y is locally
Noetherian, and F is a coherent sheaf on X, then for all i, Ri π∗ F is a coherent
sheaf on Y.

18.7.2. Unimportant Remark. If X and Y are Noetherian, the hypotheses “quasicom-

pact and separated” on π can be removed; see Unimportant Remark 23.5.8 for the
“separated” hypothesis.

18.7.3. Proof of Theorem 18.7.1. We first show covariance: if F → G is a morphism

of quasicoherent sheaves on X, we define a map Ri π∗ F → Ri π∗ G . (It will be clear
we will have shown that Ri π∗ is a functor.) It suffices to define this map on the
“distinguished affine base” of Y (Definition 6.2.1). Thus it suffices to show the
following: if X ′ is a quasicompact separated A-scheme, and F → G is a morphism
of quasicoherent sheaves on X, then the map Hi (X ′ , F ) → Hi (X ′ , G ) constructed
in §18.2 (property (i) of §18.1) “commutes with localization at f ∈ A”. But this was
shown in Exercise 18.2.D.
In a similar way, we construct the connecting homomorphism Ri π∗ H →
Ri+1 π∗ F in the long exact sequence (18.7.1.1), by showing that the construction in
the case where Y = Spec A “commutes with localization at f ∈ A”. Again, this was
shown in Exercise 18.2.D.
It suffices to check all other parts of this statement on affine open subsets of Y,
so they all follow from the analogous statements in Čech cohomology (§18.1). □
The following result is handy, and essentially immediate from our definition.

18.7.A. E ASY E XERCISE . Show that if π is affine, then for i > 0, Ri π∗ F = 0.

18.7.4. How higher pushforwards behave with respect to base change.

18.7.B. E XERCISE ( HIGHER PUSHFORWARDS AND BASE CHANGE ).

(a) (the push-pull map for cohomology) Suppose ψ : Y ′ → Y is any morphism, and
π : X → Y is quasicompact and separated. Suppose F is a quasicoherent sheaf on X.
Let
ψ′
(18.7.4.1) X′ /X

π′ π
 
Y′
ψ
/Y
526 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

be a Cartesian diagram. Describe a natural morphism

ψ∗ (Ri π∗ F ) / Ri π∗′ (ψ ′ )∗ F

of quasicoherent sheaves on Y ′ . (Hint: the FHHF Theorem, Exercise 1.5.I.) Show

that the i = 0 case is the push-pull map of Exercise 14.5.K. So this might reasonablly
be called the push-pull map for cohomology.
(b) (cohomology commutes with affine flat base change) If ψ : Y ′ → Y is an affine mor-
phism, and for a cover Spec Ai of Y, where ψ−1 (Spec Ai ) = Spec Bi , Bi is a flat
Ai -algebra (§1.5.13: ⊗Ai Bi is exact), and the diagram in (a) is a Cartesian diagram,
show that the push-pull map of (a) is an isomorphism. (Exercise 18.2.H was a
special case of this exercise. You can likely generalize this to non-affine morphisms
— and thus show that cohomology commutes with flat base change (Theorem 24.1.9)
— but we wait until Chapter 24 to discuss flatness at length.)

18.7.C. E XERCISE ( CF. E XERCISE 14.5.K). Prove Exercise 18.7.B(a) without the
hypothesis that (18.7.4.1) is a Cartesian diagram, but adding the requirement that
π ′ is quasicompact and separated (just so our definition of Ri π∗′ applies). In the
course of the proof, you will see a map arising in the Leray spectral sequence
(Theorem 23.4.5). (Hint: use Exercise 18.7.B(a).)
A useful special case of Exercise 18.7.B(a) is the following.

18.7.D. E XERCISE . If q ∈ Y, describe a natural morphism (Ri π∗ F ) ⊗ κ(q) →

Hi (π−1 (q), F |π−1 (q) ). (See §14.5.8 for the case i = 0.) Be sure you understand the
meaning of the left side! Hint: the FHHF Theorem, Exercise 1.5.I.
Thus the fiber of the pushforward may not be the cohomology of the fiber, but
at least it always maps to it. We will later see that in good situations this map is an
isomorphism, and thus the higher pushforward sheaf indeed “patches together”
the cohomology on fibers (the Cohomology and Base Change Theorem 25.1.6).

18.7.E. E XERCISE ( THE PROJECTION FORMULA , GENERALIZING E XERCISE 14.5.L).

Suppose π : X → Y is quasicompact and separated, and F and G are quasicoherent
sheaves on X and Y respectively.

F π∗ G

π∗ F X G


Y
(a) Describe a natural morphism

(Ri π∗ F ) ⊗ G / Ri π∗ (F ⊗ π∗ G ).

(Hint: the FHHF Theorem, Exercise 1.5.I.)

(b) If G is locally free, show that this natural morphism is an isomorphism.
The following fact uses the same trick as Theorem 18.1.6 and Exercise 18.1.B.
 © 2024 Ravi Vakil. Published by Princeton University Press. 527

18.7.5. Theorem (relative dimensional cohomology vanishing). — If π : X → Y is

a projective morphism and Y is locally Noetherian (or more generally OY is coherent over
itself), then the higher pushforwards vanish in degree higher than the maximum dimension
of the fibers.
This is false without the projective hypothesis (see Example 18.7.6 below).
In particular, you might hope that just as dimensional cohomology vanishing
generalized from projective varieties to quasiprojective varieties or more general
settings (§18.2.7) that relative dimensional cohomology vanishing would generalize
from projective morphisms to quasiprojective morphisms, but this is not the case.

k − {0} → Ak for n > 1. From

18.7.6. Example. Consider the open embedding π : An n

§18.3.1–18.3.2, Rn−1
π∗ OAk −{0} ̸= 0.
n

Proof of Theorem 18.7.5. Let m be the maximum dimension of all the fibers.
The question is local on Y, so we will show that the result holds near a point p
of Y. We may assume that Y is affine, and hence that X ,→ Pn Y.
Let k be the residue field at p. Then π−1 (p) is a projective k-scheme of dimen-
sion at most m. By Exercise 12.3.D we can find affine open sets D(f1 ), . . . , D(fm+1 )
that cover π−1 (p). In other words, the intersection of the V(fi ) does not intersect
π−1 (p).
If Y = Spec A and p = [p] (so k = Ap /pAp ), then arbitrarily lift each fi from
an element of k[x0 , . . . , xn ] to an element fi′ of Ap [x0 , . . . , xn ]. Let F be the prod-
uct of the denominators of the fi′ ; note that F ∈ / p, i.e., p = [p] ∈ D(F). Then
fi′ ∈ AF [x0 , . . . , xn ]. The intersection of their zero loci ∩i V(fi′ ) ⊂ Pn AF is a closed
subscheme of Pn AF . Intersect it with X to get another closed subscheme of Pn
AF . Take
its image under π; as projective morphisms are closed, we get a closed subset of
D(F) = Spec AF . But this closed subset does not include p; hence we can find an
affine open neighborhood Spec B of p in Y missing the image. But if fi′′ are the
restrictions of fi′ to B[x0 , . . . , xn ], then D(fi′′ ) cover π−1 (Spec B); in other words,
π−1 (Spec B) is covered by m + 1 affine open sets, so by affine cover cohomology
vanishing (Property (vi) of §18.1), its cohomology vanishes in degree at least m + 1.
But the higher direct image sheaf is computed using these cohomology groups,
hence the higher pushforward sheaf Ri π∗ F vanishes on Spec B too. □

18.7.F. E XERCISE ( RELATIVE S ERRE VANISHING , CF. T HEOREM 18.1.3( II )). Suppose
π : X → Y is a proper morphism of Noetherian schemes, and L is a π-ample
invertible sheaf on X. Show that for any coherent sheaf F on X, for m ≫ 0,
Ri π∗ (F ⊗ L ⊗m ) = 0 for all i > 0.

18.8 ⋆ Serre’s characterizations of ampleness and affineness

Theorem 16.2.2 gave a number of characterizations of ampleness, in terms of

projective geometry, global generation, and the Zariski topology. Here is another
characterization, this time cohomological, under Noetherian hypotheses. Because
(somewhat surprisingly) we won’t use this result, this section is starred.
528 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

18.8.1. Theorem (Serre’s cohomological criterion for ampleness). — Suppose A is

a Noetherian ring, X is a proper A-scheme, and L is an invertible sheaf on X. Then the
following are equivalent.
(a-c) The invertible sheaf L is ample on X (over A).
(e) For all coherent sheaves F on X, there is an n0 such that for n ≥ n0 , Hi (X, F ⊗
L ⊗n ) = 0 for all i > 0.
The label (a-c) is intended to reflect the statement of Theorem 16.2.2. We avoid
the label (d) because it appeared in Theorem 16.2.6. (Aside: the “properness”
assumption cannot be removed, see Example 18.7.6.) Before getting to the proof,
we motivate this result by giving some applications. (As a warm-up, you can give
a second solution to Exercise 16.2.G in the Noetherian case, using the affineness of
π to show that Hi (X, F ⊗ (π∗ L )⊗m ) = Hi (Y, (π∗ F ) ⊗ L ⊗m ).)

18.8.A. E XERCISE . Suppose X is a proper A-scheme (A Noetherian), and L is an

invertible sheaf on X. Show that L is ample on X if and only if L |Xred is ample on
Xred . Hint: for the “only if” direction, use Exercise 16.2.G. For the “if” direction,
let I be the ideal sheaf cutting out the closed subscheme Xred in X. Filter F by
powers of I :

0 = I r F ⊂ I r−1 F ⊂ · · · ⊂ I F ⊂ F .

(Essentially the same filtration appeared in Exercise 18.4.L, for similar reasons.)
Show that each quotient I n F /I n−1 F , twisted by a high enough power of L ,
has no higher cohomology. Use descending induction on n to show each part I n F
of the filtration (and hence in particular F ) has this property as well.

18.8.B. E XERCISE . Suppose X is a proper A-scheme (A Noetherian), and L is an

invertible sheaf on X. Show that L is ample on X if and only if L is ample on each
component. Hint: follow the outline of the solution to the previous exercise, taking
instead I as the ideal sheaf of one component. Perhaps first reduce to the case
where X = Xred .

18.8.C. T RICKY E XERCISE . Suppose C is a proper reduced curve (over a field

k), and L is a line bundle on C, such that ν∗ L is an ample line bundle on the
normalization C̃ (where the normalization as usual is ν : C̃ → C). Show that
L is ample on C, as follows. First read §17.1.6 to understand the morphism
ν!? : QCohC → QCohC̃ . Then for every quasicoherent sheaf F on C, describe an
exact sequence of quasicoherent sheaves on C:

0 /T / ν∗ ν!? F /F /Q /0

where T and Q are both supported on a set of dimension 0 (a finite set of points).
Then show that for n ≫ 0, Hi (C, F ⊗ L ⊗n ) = 0 for i > 0. You will use the
projection formula (Exercise 18.7.E(b)) to show that Hi (C, (ν∗ ν!? F ) ⊗ L ⊗n ) = 0
for i > 0 and n ≫ 0.

18.8.2. Very ample versus ample. The previous exercises don’t work with “ample”
replaced by “very ample”, which shows again how the notion of ampleness is
better-behaved than very ampleness.
 © 2024 Ravi Vakil. Published by Princeton University Press. 529

18.8.3. Proof of Theorem 18.8.1. For the fact that (a-c) implies (e), use the fact that
L ⊗N is very ample for some N (Theorem 16.2.2(a)), and apply Serre vanishing
(Theorem 18.1.3(ii)) to F , F ⊗ L , . . . , and F ⊗ L ⊗(N−1) .
So we now assume (e), and show that L is ample by criterion (b) of Theo-
rem 16.2.2: we will show that for any coherent sheaf F on X, F ⊗ L ⊗n is globally
generated for n ≫ 0.
We begin with a special case: we will show that L ⊗n is globally generated (i.e.,
base-point-free) for n ≫ 0. To do this, it suffices to show that every closed point p
has an open neighborhood U so that there exists some Np so that n ≥ Np , L ⊗n
is globally generated for all points of Up . (Reason: by quasicompactness, every
closed subset of X contains a closed point, by Exercise 5.1.E. So as p varies over the
closed points of X, these Up cover X. By quasicompactness again, we can cover X
by a finite number of these Up . Let N be the maximum of the corresponding Np .
Then for n ≥ N, L ⊗n is globally generated in each of these Up , and hence on all of
X.)
Let p be a closed point of X. For all n, Ip ⊗L ⊗n is coherent (by our Noetherian
hypotheses). (Here Ip is the ideal sheaf of p.) By (e), there exists some n0 so that
for n ≥ n0 , H1 (X, Ip ⊗ L ⊗n ) = 0. By the long exact sequence arising from the
closed subscheme exact sequence

0 / Ip ⊗ L ⊗n / L ⊗n / L ⊗n |p / 0,

we have that L ⊗n is globally generated at p for n ≥ n0 . By Exercise 16.1.D(b),

there is an open neighborhood V0 of p such that L ⊗n0 is globally generated at all
points of V0 . Thus L ⊗kn0 is globally generated at all points of V0 for all positive
integers k (using Easy Exercise 16.1.C). For each i ∈ {1, . . . , n0 − 1}, there is an open
neighborhood Vi of p such that L ⊗(n0 +i) is globally generated at all points of Vi
(again by Exercise 16.1.D(b)). We may take each Vi to be contained in V0 . By Easy
Exercise 16.1.C, L ⊗(kn0 +n0 +i) is globally generated at every point of Vi (as this is
the case for L ⊗kn0 and L ⊗(n0 +i) ). Thus in the open neighborhood Up := ∩n−1 i=0 Vi ,
L ⊗n is globally generated for n ≥ Np := 2n0 .
We have now shown that there exists some N such that for n ≥ N, L ⊗n is
globally generated. Now suppose F is a coherent sheaf. To conclude the proof,
we will show that F ⊗ L ⊗n is globally generated for n ≫ 0. This argument has a
similar flavor to what we have done so far, so we give it as an exercise.

18.8.D. E XERCISE . Suppose p is a closed point of X.

(a) Show that for n ≫ 0, F ⊗ L ⊗n is globally generated at p.
(b) Show that there exists an open neighborhood Up of p such that for n ≫ 0,
F ⊗ L ⊗n is globally generated at every point of Up . Caution: while it is true that
by Exercise 16.1.D(b), for each n ≫ 0, there is some open neighborhood Vn of p
such that F ⊗ L ⊗n is globally generated there, it need not be true that
(18.8.3.1) ∩n≫0 Vn
is an open set. You may need to use the fact that L ⊗n is globally generated for
n ≥ N to replace (18.8.3.1) by a finite intersection.

18.8.E. E XERCISE . Conclude the proof of Theorem 18.8.1 by showing that F ⊗ L ⊗n

is globally generated for n ≫ 0. □
530 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

18.8.4. Serre’s cohomological characterization of affineness. Serre’s characteriza-

tion of affineness is similar to Theorem 18.8.1.

18.8.5. Theorem (Serre’s cohomological characterization of affineness). —

Suppose X is a quasicompact separated scheme. Then the following are equivalent.
(a) The scheme X is affine.
(b) For every quasicoherent sheaf F on X, Hi (X, F ) = 0 for all i > 0.
(c) For every quasicoherent sheaf F on X, H1 (X, F ) = 0.

Proof. Clearly (a) implies (b) implies (c) (the former from “affine cover cohomology
vanishing”, §18.1 (vi)), so we show that (c) implies (a).
Let A = Γ (X, OX ), so we have a canonical map π : X → Spec A (defined in
Exercise 7.3.G and the paragraph following it), and it suffices to prove that π is an
affine morphism. In other words, we wish to find fi ∈ A such that (i) the fi generate
A (so ∪D(fi ) = Spec A), and also (ii) Xfi is affine (where Xfi = {p ∈ X : fi (p) ̸= 0},
Definition 6.2.8).
We first produce affine Xfi covering X. For any point p ∈ X, let U be an affine
neighborhood of p ∈ X, and let I be the ideal sheaf of X \ U (with induced reduced
subscheme structure). Let q be any closed point of U (it needn’t be a closed point
of X!) in the closure of p.

18.8.F. E XERCISE . Show that I → I |q is a surjection of quasicoherent sheaves,

and use the hypotheses (c) to show that it is surjective on global sections. Show
that there exists f ∈ A, such that Xf ⊂ U is affine, and contains p.
By quasicompactness of X, we may cover X with a finite number of such Xf ,
say f ∈ {f1 , . . . , fn }. In order to show that X → Spec A is an affine morphism, we
need that D(f1 ) ∪ · · · ∪ D(fn ) = Spec A, i.e., (f1 , . . . , fn ) = A.

18.8.G. E XERCISE . Prove this, by showing the map of quasicoherent sheaves on X

·(f1 , . . . , fn ) : OX⊕n / OX

is a surjection, and then taking global sections and invoking (c). □

18.8.H. E XERCISE . With the same hypotheses as in Theorem 18.8.5 above, show
that (a)–(c) are equivalent to:
(d) For every quasicoherent sheaf of ideals I on X, H1 (X, I ) = 0.

18.8.6. Serre proved an analogous result in complex analytic geometry: Stein spaces
are also characterized by the vanishing of cohomology of coherent sheaves.

18.9 ⋆ From projective to proper hypotheses: Chow’s Lemma and

Grothendieck’s Coherence Theorem
The main proofs in this section are double-starred because the results are
not absolutely necessary in the rest of our discussions, and may not be worth
reading right now. But just knowing the statement of Grothendieck’s Coherence
Theorem 18.9.1, (generalizing Theorem 18.7.1(d)) will allow you to immediately
 © 2024 Ravi Vakil. Published by Princeton University Press. 531

translate many of our arguments about projective schemes and morphisms to

proper schemes and morphisms, and Chow’s Lemma is a multi-purpose tool to
extend results from the projective situation to the proper situation in general.
So you have four choices: (i) read the proof of Grothendieck’s Coherence
Theorem; (ii) skim the proof and convince yourself that you could understand it
if you wanted to; (iii) skip the proof and assume it is true; or (iv) just retreat to
projective hypotheses whenever Grothendieck’s Coherence Theorem is invoked.
Life is full of compromises, and I won’t judge you.

18.9.1. Grothendieck’s Coherence Theorem. — Suppose π : X → Y is a proper

morphism of locally Noetherian schemes. Then for any coherent sheaf F on X, Ri π∗ F is
coherent on Y.
The case i = 0 has already been mentioned a number of times.

18.9.A. E XERCISE . Recall that finite morphisms are affine (by definition) and
proper. Use Theorem 18.9.1 to show that if π : X → Y is proper and affine and Y
is Noetherian, then π is finite. (Hint: mimic the proof of the weaker result where
proper is replaced by projective, Corollary 18.1.5.)
The proof of Theorem 18.9.1 requires two sophisticated facts. The first is the
Leray spectral sequence (Theorem 23.4.5, which applies in this situation because
of Exercise 23.5.F). Suppose π : X → Y and ρ : Y → Z are quasicompact separated
morphisms. Then for any quasicoherent sheaf F on X, there is a spectral sequence
with E2 term given by Rp ρ∗ (Rq π∗ F ) converging to Rp+q (ρ ◦ π)∗ F . Because this
would be a reasonable (but hard) exercise in the case we need it (where Z is affine),
we will feel comfortable using it. But because we will later prove it, we won’t prove
it now.
We will also need Chow’s Lemma.

18.9.2. Chow’s Lemma (proper = projective - birational projective). — Suppose

π : X → Spec A is a proper morphism, and A is Noetherian. Then there exists µ : X ′ → X
which is surjective and projective, such that π ◦ µ is also projective, and such that µ is an
isomorphism on a dense open subset of X:

X′
µ
/X

π◦µ π
# {
Spec A.

In particular, if X is a proper k-variety, it admits a projective birational mor-

phism from a projective k-variety.
Many generalizations of results from projective to proper situations go through
Chow’s Lemma. We will prove this version, and state other versions of Chow’s
Lemma, in §18.9.4. Assuming these two facts, we now prove Grothendieck’s
Coherence Theorem 18.9.1 in a series of exercises.

18.9.3. ⋆⋆ Proof of Grothendieck’s Coherence Theorem 18.9.1, assuming Chow’s Lemma 18.9.2.
The question is local on Y, so we may assume Y is affine, say Y = Spec A. We work
532 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

by Noetherian induction on Supp F , with the base case when Supp F = ∅, i.e.,
F = 0, which is obvious. So fix F , and assume the result is known for all coherent
sheaves with strictly smaller support.

18.9.B. E XERCISE . Show that we may assume that Supp F = X. (Hint: Replace X
by the scheme-theoretic support of F , defined in Exercise 9.1.N.)
We now invoke Chow’s Lemma to construct a projective morphism µ : X ′ → X
that is an isomorphism on a dense open subset U of X (so X \ U is a closed subset of
X, strictly smaller than X), and such that π ◦ µ : X ′ → Spec A is projective.
Then G = µ∗ F is a coherent sheaf on X ′ , µ∗ G is a coherent sheaf on X (by the
projective case, Theorem 18.7.1(d)) and the adjunction map F → µ∗ G = µ∗ µ∗ F
is an isomorphism on U. The kernel E and cokernel H are coherent sheaves on X
that have strictly smaller support:

0 /E /F / µ∗ G /H / 0.

18.9.C. E XERCISE . By the inductive hypothesis, the higher pushforwards of E and

H are coherent. Show that if all the higher pushforwards of µ∗ G are coherent, then
the higher pushforwards of F are coherent.
So we are reduced to showing that the higher pushforwards of µ∗ G are coherent
for any coherent G on X ′ .
The Leray spectral sequence (Theorem 23.4.5) for X ′
µ
/ X π / Spec A
has E2 page given by Rp π∗ (Rq µ∗ G ) converging to Rp+q (π ◦ µ)∗ G . Now Rq µ∗ G
is coherent by Theorem 18.7.1(d). Furthermore, as µ is an isomorphism on a
dense open subset U of X, Rq µ∗ G is zero on U, and is thus supported on the
complement of U, whose support is strictly smaller than X. Hence by our inductive
hypothesis, Rp π∗ (Rq µ∗ G ) is coherent for all p, and all q ≥ 1. The only possibly
noncoherent sheaves on the E2 page are in the row q = 0 — precisely the sheaves
we are interested in. Also, by Theorem 18.7.1(d) applied to π ◦ µ, Rp+q (π ◦ µ)∗ G is
coherent.

18.9.D. E XERCISE . Show that Ep,q

n is always coherent for any n ≥ 2, q > 0. Show
that En is coherent for a given n ≥ 2 if and only if Ep,0
p,0
2 is coherent. Show that
p,0
Ep,q
∞ is coherent, and hence that E2 is coherent, thereby completing the proof of
Grothendieck’s Coherence Theorem 18.9.1.
□

18.9.4. ⋆⋆ Proof of Chow’s Lemma 18.9.2.

We use the properness hypothesis on π through each of its three constituent
parts: finite type, separated, universally closed. The parts using separatedness are
particularly tricky.
As X is Noetherian, it has finitely many irreducible components. Cover X with
affine open sets U1 , . . . , Un . We may assume that each Ui meets each irreducible
component. (If some Ui does not meet an irreducible component Z, then take any
affine open subset Z ′ of Z \ X − Z, and replace Ui by Ui ∪ Z ′ .) Then U := ∩i Ui is a
dense open subset of X. As each Ui is finite type over A, we can choose a closed
embedding Ui ⊂ An ni
A . Let Ui be the (scheme-theoretic) closure of Ui in PA .
i
 © 2024 Ravi Vakil. Published by Princeton University Press. 533

Q
Now we have the diagonal morphism U → X ×A Ui (where the product is
over Spec A), which is a locally closed embedding (the composition of the closed
Q
embedding U ,→ Un+1 with the open embedding Un+1 ,→ X ×A Ui ). Let X ′ be
Q
the scheme-theoretic closure of U in X ×A Ui . Let µ be the composed morphism
Q
X ′ → X ×A Ui → X, so we have a diagram

X ′ _
µ
cl. emb.
Q
&/
X ×A Ui X
proj.
proper proper

Q 
Ui / Spec A
proj.

(where the square is Cartesian). The morphism µ is projective (as it is the composi-
tion of two projective morphisms and X is quasicompact, Exercise 17.3.I). We will
conclude the argument by showing that µ−1 (U) = U (or more precisely, µ is an
Q
isomorphism above U), and that X ′ → Ui is a closed embedding (from which
the composition
Q
X / Ui / Spec A

is projective).

18.9.E. E XERCISE . Suppose T0 , . . . , Tn are separated schemes over A with isomor-

phic open sets, which we sloppily call V in each case. Then V is a locally closed
subscheme of T0 ×A · · · ×A Tn . Let V be the closure of this locally closed subscheme.
Show that
∼ V ∩ (V ×A T1 ×A · · · ×A Tn )
V=
= V ∩ (T0 ×A V ×A T2 ×A · · · ×A Tn )
= ···
= V ∩ (T0 ×A · · · ×A Tn−1 ×A V).

(Hint for the first isomorphism: the graph of the morphism V → T1 ×A · · · ×A Tn is

a closed embedding, as T1 ×A · · · ×A Tn is separated over A, by Proposition 11.2.6.
Thus the scheme-theoretic closure of V in V ×A T1 ×A · · · ×A Tn is V itself. Finally,
the scheme-theoretic closure can be computed locally, essentially by Theorem 9.4.4.)

18.9.F. E XERCISE . Using (the idea behind) the previous exercise, show that
µ−1 (U) = U.
Q Q
It remains to show that X ′ → Ui is a closed embedding. Now X ′ → Ui
is closed (it is the composition of two closed maps), so it suffices to show that
Q
X′ → Ui is a locally closed embedding.

18.9.G. E XERCISE . Let Ai be the closure of U in

Bi := X ×A U1 ×A · · · ×A Ui ×A · · · ×A Un
534 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

(only the ith term is missing the bar), and let Ci be the closure of U in
Di := U1 ×A · · · ×A Ui ×A · · · ×A Un .
∼
Show that there is an isomorphism Ai −→ Ci induced by the projection Bi → Di .
Hint: note that the section Di → Bi of the projection Bi → Di , given informally by
(t1 , . . . , tn ) 7→ (ti , t1 , . . . , tn ), is a closed embedding, as it can be interpreted as the
graph of a morphism to a separated A-scheme (Proposition 11.2.6). So U can be
interpreted as a locally closed subscheme of Di , which in turn can be interpreted as
a closed subscheme of Bi . Thus the closure of U in Di may be identified with its
closure in Bi .
As the Ui cover X, the µ−1 (Ui ) cover X. But µ−1 (Ui ) = Ai (closure can be be
computed locally — the closure of U in Bi is the intersection of Bi with the closure
X of U in X ×A U1 ×A · · · ×A Un ).
Hence over each Ui , we get a closed embedding of Ai ,→ Di , and thus X ′ →
Q
Ui is a locally closed embedding as desired. □

18.9.5. Other versions of Chow’s Lemma. We won’t use these versions, but their
proofs are similar to what we have already shown.

18.9.6. Remark. Notice first that if X is reduced (resp., irreducible, integral), then X ′
can be taken to be reduced (resp., irreducible, integral) as well.

18.9.H. E XERCISE . By suitably crossing out lines in the proof above, weaken
the hypothesis “π is proper” to “π finite type and separated”, at the expense of
weakening the conclusion “π ◦ µ is projective” to “π ◦ µ is quasiprojective”.

18.9.7. Remark. The target Spec A can be generalized to a scheme S that is (i)
Noetherian, or (ii) separated and quasicompact with finitely many irreducible
components. This can be combined with Remark 18.9.6 and Exercise 18.9.H. See
[Gr-EGA, II.5.6.1] for a proof. See also [GW, Thm. 13.100] for a version that is
slightly more general.
 CHAPTER 19

Application: Curves

We now use what we have developed to study something explicit — curves.

Throughout this chapter, we will assume that all curves are projective, geometrically
integral, regular curves over a field k. We will sometimes add the hypothesis that k
is algebraically closed. Most people are happy with working over algebraically
closed fields, and those people should ignore the adverb “geometrically”.
We certainly do not need the massive machinery we have developed in order
to understand curves, but with the perspective we have gained, the development is
quite clean. The key ingredients we will need are as follows. We use a criterion for
a morphism to be a closed embedding, that we prove in §19.1. We use the “black
box” of Serre duality (to be proved in Chapter 29). In §19.2, we use this background
to observe a very few useful facts, which we will use repeatedly. Finally, in the
course of applying them to understand curves of various genera, we develop the
theory of hyperelliptic curves in a hands-on way (§19.5), in particular proving a
special case of the Riemann-Hurwitz formula.
If you are jumping into this chapter without reading much beforehand, you should skip
§19.1, taking Theorem 19.1.1 as a black box. Depending on your background, you may
want to skip §19.2 as well, taking the “crucial tools” as a black box.

19.1 A criterion for a morphism to be a closed embedding

We will repeatedly use a criterion for when a morphism is a closed embedding,

which is not special to curves. Before stating it, we recall some facts about closed
embeddings. Suppose π : X → Y is a closed embedding. Then π is projective, and it
is injective on points. This is not enough to ensure that it is a closed embedding, as
the example of the normalization of the cusp shows (Figure 10.5). Another example
is the following.

19.1.A. E XERCISE (F ROBENIUS , CF. §8.3.Q). Show that π : A1Fp → A1Fp given by
x 7→ xp is a bijection on points, and induces an isomorphism of residue fields on
closed points, yet is not a closed embedding.
The additional information you need to ensure that a morphism π is a closed
embedding is that the tangent map is injective at all closed points.

This material is published by Princeton University Press as The Rising Sea: Foundations of Algebraic
Geometry by Ravi Vakil. This version is free to view and download for personal use only. Not for
redistribution, resale, or use in derivative works. © 2024 Ravi Vakil.

535
536 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

19.1.B. E XERCISE . Show (directly, not invoking Theorem 19.1.1) that in the two ex-
amples described above (the normalization of a cusp and the Frobenius morphism),
the tangent map is not injective at all closed points.

19.1.1. Theorem. — Suppose k = k, and π : X → Y is a projective morphism of finite

type k-schemes that is injective on closed points and injective on tangent vectors at closed
points. Then π is a closed embedding.
The converse is straightforward, so you should think of this result as “if and
only if”.
(Remark: In this case of finite type schemes over an algebraically closed field,
“injective on tangent vectors at closed points” is equivalent to “unramified”, see
Exercise 21.7.E. We will define unramified in §21.7.)
The example Spec C → Spec R shows that we need the hypothesis that k is
algebraically closed in Theorem 19.1.1. Those allergic to algebraically closed fields
should still pay attention, as we will use this to prove things about curves over k
where k is not necessarily algebraically closed (see also Exercises 10.2.K and 19.1.F).
We need the hypothesis that the morphism be projective, as shown by the
example of Figure 19.1. It is the normalization of the node, except we erase one of
the preimages of the node. We map A1 to the plane, so that its image is a curve
with one node. We then consider the morphism we get by discarding one of the
preimages of the node. Then this morphism is an injection on points, and is also
injective on tangent vectors, but it is not a closed embedding. (In the world of
differential geometry, this fails to be an embedding because the map doesn’t give a
homeomorphism onto its image.)

F IGURE 19.1. (variant of Figure 8.4) The projectivity hypothesis in

Theorem 19.1.1 cannot be dropped

Theorem 19.1.1 appears to be fundamentally a statement about varieties, but it

isn’t. We will reduce it to the following result.

19.1.2. Theorem. — Suppose π : X → Y is a finite morphism of Noetherian schemes

whose degree at every point of Y (§14.3.4) is 0 or 1. Then π is a closed embedding.

19.1.C. E XERCISE . Suppose π : X → Y is a finite morphism whose degree at every

point of Y is 0 or 1. Show that π is injective on points (easy). If p ∈ X is any point,
∼
show that π induces an isomorphism of residue fields κ(π(p)) −→ κ(p). Show that
π induces an injection of tangent spaces. Thus key hypotheses of Theorem 19.1.1
are implicitly in the hypotheses of Theorem 19.1.2.
 © 2024 Ravi Vakil. Published by Princeton University Press. 537

19.1.3. Reduction of Theorem 19.1.1 to Theorem 19.1.2. The property of being a closed
embedding is local on the target, so we may assume that Y is affine, say Spec B.
I next claim that π has finite fibers, not just finite fibers above closed points: the
fiber dimension for projective morphisms is upper semicontinuous (Exercise 18.1.B,
or Theorem 12.4.3(b)), so the locus where the fiber dimension is at least 1 is a closed
subset, so if it is nonempty, it must contain a closed point of Y. Thus the fiber over
any point is a dimension 0 finite type scheme over that point, hence a finite set.
Hence π is a projective morphism with finite fibers, thus finite by Theo-
rem 18.1.6.

19.1.D. E XERCISE . Show that the degree of π is at most 1 at all closed points of Y.
(Hint 1: You will use the “injectivity on tangent vectors” hypothesis here. Hint 2:
what are the finite morphisms to Spec k?)
But the degree of a finite morphism is upper semicontinuous, (§14.3.4), hence is
at most 1 at all points (do you see why?). We have thus shown that Theorem 19.1.1
would follow from Theorem 19.1.2.

19.1.4. Proof of Theorem 19.1.2. The problem is local on Y, so we may assume Y

is affine, say Y = Spec B. Thus X is affine too, say Spec A, and π corresponds to
a ring morphism B → A. We wish to show that this is a surjection of rings, or
(equivalently) of B-modules. Let K be the cokernel of this morphism of B-modules:

(19.1.4.1) B /A /K / 0.

We wish to show that K = 0. It suffices to show that for any maximal ideal n of B,
Kn = 0. (Do you remember why?) Localizing (19.1.4.1) at n, we obtain the exact
sequence
Bn / An / Kn / 0.

Applying the right-exact functor ⊗Bn (Bn /nBn ), we obtain

(19.1.4.2) Bn /nBn
α / An /nAn / Kn /nKn / 0.

Now Spec An /nAn is the scheme theoretic preimage of [n] ∈ Spec B, so by hypoth-
esis, it is either empty, or the map α is an isomorphism (in which case we will
see that the map α will turn out to be the map of residue fields). In the first case,
An /nAn = 0, from which Kn /nKn = 0. In the second case, Kn /nKn = 0 as well.
Applying Nakayama’s Lemma 8.2.9 (noting that A is a finitely generated B-module,
hence K is too, hence Kn is a finitely generated Bn -module), Kn = 0 as desired. □
Exercise 10.2.K can be used to extend Theorem 19.1.1’s reach to general fields k,
not necessarily algebraically closed. The next exercise is an example of how.

19.1.E. E XERCISE . Use Theorem 19.1.1 to show that the dth Veronese embedding
from Pn k , corresponding to the complete linear series |OPk (d)|, is a closed embed-
n

ding. Do the same for the Segre embedding from Pm k × Spec

n
k Pk . (This is just for
practice for using this criterion. It is a weaker result than we had before; we have
earlier checked both of these statements over an arbitrary base ring in Remark 9.3.8
and §10.6 respectively, and we are now checking it only over fields. However, see
Exercise 19.1.F below.)
538 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

19.1.F. L ESS IMPORTANT EXERCISE . Using the ideas from this section, prove that
the dth Veronese embedding from Pn Z (over the integers!), is a closed embedding.
(Again, we have done this before. This exercise is simply to show that these
methods can easily extend to work more generally.)

19.2 A series of crucial tools

We are now ready to start understanding curves in a hands-on way. We will

repeatedly make use of the following series of crucial remarks, and it will be
important to have them at the tip of your tongue.

19.2.1. In what follows, C will be a projective, geometrically regular, geometrically

integral curve over a field k, and L is an invertible sheaf on C.

19.2.2. Reminder: Serre duality. Serre duality (Theorem 18.5.1) on a geometrically

irreducible regular genus g curve C over k involves an invertible sheaf ωC (of
degree 2g − 2, with g linearly independent sections, Exercise 18.5.A), such that
for any finite rank locally free sheaf F on C, hi (C, F ) = h1−i (C, ωC ⊗ F ∨ ) for
i = 0, 1. (Better: there is a duality between the two cohomology groups.)

19.2.3. Negative degree line bundles have no nonzero section. If deg L < 0, then
h0 (C, L ) = 0. Reason: deg L is the number of zeros minus the number of poles
(suitably counted) of any rational section (Important Exercise 18.4.C). If there is a
regular section (i.e., with no poles), then this is necessarily non-negative. Refining
this argument yields the following.

19.2.4. Degree 0 line bundles, and recognizing when they are trivial. If deg L = 0,
then h0 (C, L ) = 0 or 1; and if h0 (C, L ) = 1 then L = ∼ OC . Reason: if there is
a nonzero section s, it has no poles, and hence no zeros, because deg L = 0.
Then div s = 0, so L = ∼ OC (div s) = OC . (Recall how this works, cf. Important
Exercise 15.4.G: s gives a trivialization for the invertible sheaf. We have a natural
∼
bijection for any open set Γ (U, L ) ←→ Γ (U, OU ), where the map from left to right is
s 7→ s /s, and the map from right to left is f 7→ sf.) Conversely, for a geometrically
′ ′

integral projective variety, h0 (O) = 1. (§11.4.7 shows this for k algebraically closed
— this is where geometric integrality is used — and Exercise 18.2.H shows that
cohomology commutes with base field extension.)
Serre duality turns these statements about line bundles of degree at most 0 into
statements about line bundles of degree at least 2g − 2, as follows.

19.2.5. We know h0 (C, L ) if the degree is sufficiently high. If deg L > 2g − 2,

then

(19.2.5.1) h0 (C, L ) = deg L − g + 1.

So we know h0 (C, L ) if deg L ≫ 0. (This is important — remember this!) Reason:

h1 (C, L ) = h0 (C, ωC ⊗ L ∨ ); but ωC ⊗ L ∨ has negative degree (as ωC has degree
2g − 2), and thus this invertible sheaf has no sections. The result then follows from
the Riemann-Roch Theorem in the guise of (18.4.4.1).
 © 2024 Ravi Vakil. Published by Princeton University Press. 539

19.2.A. U SEFUL EXERCISE ( RECOGNIZING ωC AMONG DEGREE 2g − 2 LINE BUN -

DLES ). Suppose L is a degree 2g − 2 invertible sheaf. Show that it has g − 1 or g
∼ ωC .
sections, and it has g sections if and only if L =

19.2.B. E XERCISE . Suppose C is a curve of genus g > 1, over a field k that is not
necessarily algebraically closed. Show that C has a closed point of degree at most
2g − 2 over the base field. (For comparison: if g = 1, for any n, there is a genus 1
curve over Q with no point of degree less than n, [MO286390]!)

19.2.6. Twisting L by a (degree 1) point changes h0 by at most 1. Suppose p

is any closed point of degree 1 (i.e., the residue field of p is k). Then h0 (C, L ) −
h0 (C, L (−p)) = 0 or 1. (The twist of L by a divisor, such as L (−p), was defined
in §15.4.16.) Reason: consider 0 → OC (−p) → OC → O|p → 0, tensor with L (this
is exact as L is locally free) to get

0 / L (−p) /L / L |p / 0.

Then h0 (C, L |p ) = 1, so as the long exact sequence of cohomology starts off

0 / H0 (C, L (−p)) / H0 (C, L ) / H0 (C, L |p ),

we are done.

19.2.7. Criterion for L to be base-point-free. Suppose for this remark that k is

algebraically closed, so all closed points have degree 1 over k. Then if h0 (C, L ) −
h0 (C, L (−p)) = 1 for all closed points p, then L is base-point-free, and hence
induces a morphism from C to projective space (Theorem 15.2.2). Reason: given
any p, our equality shows that there exists a section of L that does not vanish at p
— so by definition, p is not a base point of L .

19.2.8. Next, suppose p and q are distinct (closed) points of degree 1. Then
h0 (C, L ) − h0 (C, L (−p − q)) = 0, 1, or 2 (by repeating the argument of Re-
mark 19.2.6 twice). If h0 (C, L ) − h0 (C, L (−p − q)) = 2, then necessarily
(19.2.8.1)
h0 (C, L ) = h0 (C, L (−p)) + 1 = h0 (C, L (−q)) + 1 = h0 (C, L (−p − q)) + 2.
Then the linear series L separates points p and q, i.e., the corresponding map f to
projective space satisfies f(p) ̸= f(q). Reason: there is a hyperplane of projective
space passing through f(q) but not passing through f(p), or equivalently, there is
a section of L vanishing at q but not vanishing at p. This is because of the last
equality in (19.2.8.1).

19.2.9. By the same argument as above, if p is a (closed) point of degree 1, then

h0 (C, L ) − h0 (C, L (−2p)) = 0, 1, or 2. I claim that if this is 2, then the map
corresponding to L (which is already seen to be base-point-free at p from the above)
separates the tangent vectors at p. To show this, we need to show that the cotangent
map is surjective. To show surjectivity onto a one-dimensional vector space, we just
need to show that the map is nonzero. So it suffices to find a function on the target
vanishing at the image of p that pulls back to a function that vanishes at p to order
1 but not 2. In other words, we want a section of L vanishing at p to order 1 but
not 2. But that is the content of the statement h0 (C, L (−p)) − h0 (C, L (−2p)) = 1.
540 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

19.2.10. Criterion for L to be very ample. Combining some of our previous

comments: suppose C is a curve over an algebraically closed field k, and L is
an invertible sheaf such that for all closed points p and q, not necessarily distinct,
h0 (C, L ) − h0 (C, L (−p − q)) = 2, then L gives a closed embedding into projective
space, as it separates points and tangent vectors, by Theorem 19.1.1.

19.2.C. E XERCISE . Suppose that k is algebraically closed, so the previous remark

applies. Show that C \ {p} is affine. Hint: Show that if j ≫ 0, then O(jp) is base-
point-free and has at least two linearly independent sections, one of which has
divisor jp. Use these two sections to map to P1 so that the set-theoretic preimage
of ∞ is p. Argue that the map is finite, and that C \ {p} is the preimage of A1 . (A
trivial variation of this argument shows that C \ {p1 , . . . , pn } is affine if n > 0.)

19.2.11. Powerful conclusion: Numerical criteria for L to be base-point-free

and very ample. We can combine much of the above discussion to give the
following useful fact. If k is algebraically closed, then deg L ≥ 2g implies that L
is base-point-free (and hence determines a morphism to projective space). Also,
deg L ≥ 2g + 1 implies that for any basis of sections s0 , . . . , sdeg L −g of L , the
morphism given by [s0 , . . . , sdeg L −g ] is in fact a closed embedding into Pdeg L −g
(so L is very ample). Remember this!

19.2.D. E XERCISE ( ON A PROJECTIVE , REGULAR , INTEGRAL CURVE OVER k, AMPLE

= POSITIVE DEGREE ). Show that an invertible sheaf L on a projective, regular,
integral curve over k is ample if and only if deg L > 0.
(This can be extended to curves over general fields using Exercise 19.2.E below.)
Thus there is a blunt purely numerical criterion for ampleness of line bundles on
curves. This generalizes to projective varieties of higher dimension; this is called
the Nakai-Moishezon criterion for ampleness, Theorem 20.4.1.

19.2.E. E XERCISE ( EXTENSION TO NON - ALGEBRAICALLY CLOSED FIELDS ). Show

that the statements in §19.2.11 hold even without the hypothesis that k is alge-
braically closed. (Hint: to show one of the facts about some curve C and line bundle
L , consider instead C ×Spec k Spec k. Then show that if the pullback of L here
has sections giving you one of the two desired properties, then there are sections
downstairs with the same properties. You may want to use facts that we have used,
such as the fact that base-point-freeness is independent of extension of base field,
Exercise 18.2.I, or that the property of an affine morphism over k being a closed
embedding holds if and only if it does after an extension of k, Exercise 10.2.K.)

19.2.F. E XERCISE ( EXTENDING E XERCISE 19.2.D). Suppose L is an invertible sheaf

on a projective, geometrically regular, geometrically integral curve C (over a field
k). Show that L is ample if and only if deg L > 0. Hint: reduce to the case where
k is algebraically closed, with the help of Exercise 10.2.K.

19.2.G. E XERCISE . Suppose C is a curve (in the sense of §19.2.1 as usual in this
section), and p ∈ C is a closed point. Show that C\{p} is affine. Hint: Exercise 19.2.F.
We are now ready to take these facts and go to the races, in §19.3.

19.2.12. ⋆ Remark: Proper curves are projective.

 © 2024 Ravi Vakil. Published by Princeton University Press. 541

(In this final part of §19.2, we no longer assume our curves are smooth, or
even regular or reduced.) Our discussion here, along with Serre’s cohomological
criterion for ampleness (§18.8), is the key ingredient for showing that every proper
curve over a field is projective.

19.2.H. E XERCISE . Show that every reduced proper curve C is projective. Hint:
Choose a regular point on each irreducible component of C, and let L be the
corresponding invertible sheaf. We hope that L will be ample. Recall that a line
bundle on C is ample if its pullback to the normalization of C is ample (Tricky
Exercise 18.8.C). So show that the pullback to the normalization is ample.

19.2.I. E XERCISE . Show that every proper (not necessarily reduced) curve C is
projective, using Exercise 18.8.A, after first finding an invertible sheaf on C that you
will show to be ample.

19.2.J. E XERCISE . Show that a line bundle on a projective curve is ample if and
only if it has positive degree on each component.

19.3 Curves of genus 0

We are now ready to (in some form) answer the question: what are the curves
of genus 0?
In §7.5.9, we saw a genus 0 curve (over a field k) that was not isomorphic to P1 :
x + y2 + z2 = 0 in P2R . (It has genus 0 by (18.6.7.1).) We have already observed that
2

this curve is not isomorphic to P1R , because it doesn’t have an R-valued point. On
the other hand, we haven’t seen a genus 0 curve over an algebraically closed field
with this property. This is no coincidence: the lack of an existence of a k-valued
point is the only obstruction to a genus 0 curve being P1 .

19.3.1. Proposition. — Suppose C is genus 0, and C has a k-valued (degree 1) point.

Then C =∼ P1 .
k

Thus we see that all genus 0 (integral, regular, projective) curves over an
algebraically closed field are isomorphic to P1 .

Proof. Let p be the point, and consider L = OC (p). Then deg L = 1, so we can
apply what we know above: first, h0 (C, L ) = 2 (Remark 19.2.5), and second, these
two sections give a closed embedding into P1k (Remark 19.2.11, as extended in
Exercise 19.2.E). But the only closed embedding of a curve into the integral curve
P1k is an isomorphism! □
As a bonus, Proposition 19.3.1 implies that x2 + y2 + z2 = 0 in P2R has no line
bundles of degree 1 over R; otherwise, we could just apply the above argument to
the corresponding line bundle. This example shows us that over a non-algebraically
closed field, there can be genus 0 curves that are not isomorphic to P1k . The next
result lets us get our hands on them as well.

19.3.2. Claim. — All genus 0 curves can be described as conics in P2k .

542 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Proof. Any genus 0 curve has a degree −2 line bundle — the dualizing sheaf
ωC . Thus any genus 0 curve has a degree 2 line bundle: L = ω∨ C . We apply
Exercise 19.2.E: deg L = 2 ≥ 2g + 1, so this line bundle gives a closed embedding
i : C ,→ P2 . Exercise 18.6.H ensures that the image is degree 2 as a closed subscheme,
but we haven’t quite shown that the curve can be cut out by one homogeneous
polynomial of degree 2. (This is more subtle than you might think — you can
try to make your own argument before reading on.) We have the isomorphism
∼
Sym2 H0 (P2 , O(1)) −→ H0 (P2 , O(2)) (both sides are the homogeneous quadratic
polynomials in three variables). Consider the map Sym2 H0 (C, L ) → H0 (C, L ⊗2 ),
which (by (19.2.5.1)) is a map from a vector space of dimension 6 to a vector space
of dimension 5. This map has nontrivial kernel, so i(C) is scheme-theoretically
contained in V(f) for some homogeneous conic f.

19.3.A. E XERCISE . Use Exercise 18.6.E to show that i(C) = V(f) scheme-theoretically,
completing the proof.
□

19.3.B. E XERCISE . Suppose C is a genus 0 curve (projective, geometrically integral

and regular). Show that C has a point of degree at most 2. (The degree of a point
was defined in §5.3.8.)
The geometric means of finding Pythagorean triples presented in §7.5.8 looked
quite different, but was really the same. There was a genus 0 curve C (a plane conic)
with a k-valued point p, and we proved that it was isomorphic to P1k . The line
bundle used to show the isomorphism wasn’t the degree 1 line bundle OC (p); it
was the degree 1 line bundle OP2 (1)|C ⊗ OC (−p).

19.4 Classical geometry arising from curves of positive genus

We will use the following Proposition and Corollary later, and we take this as
an excuse to revisit some very classical geometry from a modern standpoint.

19.4.1. Proposition. — Recall our standing assumptions for this chapter (§19.2.1), that
C is a projective, geometrically regular, geometrically integral curve over a field k. Suppose
C is not isomorphic to P1k (with no assumptions on the genus of C), and L is an invertible
sheaf of degree 1. Then h0 (C, L ) < 2.

Proof. We could prove this similarly to Proposition 19.3.1 (and the reader may wish
to do so). We instead prove it “by hand” because Proposition 19.3.1 requires Serre
Duality, which we will not prove until Chapter 29 (and which is hard), and we wish
to emphasize that we do not need Serre Duality for this result. Also, some readers
may read this argument without ever reading about Serre Duality.
Assume the statement is false, and let s1 and s2 be two (independent) sections.
As the degree of the divisor of zeros of si is the degree of L , each vanishes at a
single point pi (to order 1).
Now p1 ̸= p2 . (Otherwise, if p1 = p2 , then s1 /s2 has no poles or zeros, and
hence is an invertible function by Algebraic Hartogs’s Lemma, see §13.5.20, and a
 © 2024 Ravi Vakil. Published by Princeton University Press. 543

constant function using §11.4.7 after base change to k. This contradicts the linear
independence of s1 and s2 .)
Thus s1 and s2 generate a base-point-free linear series, so we have an induced
morphism C → P1 . This is a finite degree 1 morphism from one regular curve to
another (Exercise 18.4.E), which hence induces a degree 1 extension of function
fields, i.e., an isomorphism of function fields, which means that the curves are
isomorphic (cf. Theorem 16.3.3). But we assumed that C is not isomorphic to P1k , so
we have a contradiction. □

19.4.2. Corollary. — If C is a projective, geometrically regular, geometrically integral

∼ OC (q)
curve over k, not isomorphic to P1k and p and q are degree 1 points, then OC (p) =
if and only if p = q.

19.4.A. E XERCISE . Prove this. (Possibly useful: show/use that OC (p) has a global
section s with div(s) = p.)

19.4.B. E XERCISE . Show that if k is algebraically closed, then C has genus 0 if and
only if all degree 0 line bundles are trivial.

19.4.C. E XERCISE . Suppose C is a regular plane curve of degree e > 2, and D1

and D2 are two plane curves of the same degree d not containing C. By Bézout’s
Theorem for plane curves (§18.6.4), Di meets C at de points, counted “correctly”.
Suppose D1 and D2 meet C at de − 1 of the “same points”, plus one more. Show
that the remaining points are the same as well. More precisely, suppose there is
a divisor E on C of degree de − 1, and degree 1 (k-valued) points p1 and p2 such
that Di ∩ C = E + pi (as divisors on C). Show that p1 = p2 . (The case d = e = 3
is Chasles’ Theorem, and is the first case of the Cayley-Bacharach Theorem, see
[EGH].)
As an entertaining application of Exercise 19.4.C, we can prove two classical
results. For convenience, in this discussion we will assume k is algebraically closed,
although this assumption can be easily removed.

19.4.3. Pappus’s Theorem (Pappus of Alexandria). — (See Figure 19.2.) Suppose

ℓ and m are distinct lines in the plane, and α, β, γ are distinct points on ℓ, and α ′ , β ′ ,
and γ ′ are distinct points on m, and all six points are distinct from ℓ ∩ m. Then points
x = αβ ′ ∩ α ′ β, y = aγ ∩ αγ ′ , and z = βγ ′ ∩ β ′ γ are collinear.
Pascal’s “Mystical Hexagon” Theorem was discovered by Pascal at age 16.
(What were you doing at age 16?)

19.4.4. Pascal’s “Mystical Hexagon” Theorem. — (See Figure 19.3.) If a hexagon

αγ ′ βα ′ γβ ′ is inscribed in a smooth conic K, and opposite pairs of sides are extended until
they meet, the three intersection points x = αβ ′ ∩α ′ β, y = α ′ γ∩αγ ′ , and z = βγ ′ ∩β ′ γ
are collinear.
Pappus’s Theorem can be seen as a degeneration of Pascal’s Theorem: the conic
degenerates into the union of two lines, and the six points degenerate so that α,
β, and γ are on one, and α ′ , β ′ , and γ ′ are on the other. We thus prove Pascal’s
Theorem, and you should check that the proof readily applies to prove Pappus’s
Theorem.
544 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

F IGURE 19.2. Pappus’ Theorem

F IGURE 19.3. Pascal’s “Mystical Hexagon” Theorem

19.4.D. E XERCISE . Suppose α, β, γ, α ′ , β ′ are five points in P2R , no three on a line.

Show that there is a unique conic K passing through the five points. Show that K is
regular. Explain how to construct the tangent to K at α using only a straightedge.
(Hint for the last part: apply Pascal’s Theorem, taking γ ′ = α. You may need to
first figure out why you can apply Pascal’s Theorem in this degenerate case.)
In particular, given an ellipse K ⊂ R2 and a point α ∈ K, you will now be able
to construct the tangent to K at α using only a straightedge.

19.4.5. Proof of Pascal’s Theorem 19.4.4. We wish to show that the line xy meets βγ ′
and β ′ γ at the same point. Let C be the curve that is the union of K and the line
xy. (Warning, cf. §19.2.1: for one time in this chapter, we are not assuming C to be
regular!)
Let D1 be the union of the three lines αβ ′ , βγ ′ , and γα ′ , and let D2 be the
union of the three lines α ′ β, β ′ γ, and γ ′ α. Note that D1 meets C at the nine points
α, β, γ, α ′ , β ′ , γ ′ , x, y, and xy ∩ βγ ′ , and D2 meets C at the same nine points,
 © 2024 Ravi Vakil. Published by Princeton University Press. 545

except xy ∩ βγ ′ is replaced by xy ∩ β ′ γ. Thus if we knew Corollary 19.4.2 so that it

applied to our C (which is singular), then we would be done (cf. Exercise 19.4.C).
So we extend Corollary 19.4.2 to our situation. So to do this, we extend
Proposition 19.4.1 to our situation. We have a plane cubic C which is the union
of a line L and a conic K, and two points p = xy ∩ βγ ′ and q = xy ∩ β ′ γ,
with O(p) = ∼ O(q). (Reason: both are the restriction of OP2 (3) to C, twisted by
−(α + β + γ + α ′ + β ′ + γ ′ + x + y).) Call this invertible sheaf L . Suppose p ̸= q.
Then the two sections of L with zeros at p and q give a base-point-free linear series,
and thus a morphism π : C → P1 , with π−1 (0) = p and π−1 (∞) = q.
By the argument in the proof of Proposition 19.4.1, π gives an isomorphism of
L with P1 . As π is proper, π(K) is closed, and as K is irreducible, π(K) is irreducible.
As π(K) does not contain 0 (or ∞), it can’t be all of P1 . Hence π(K) is a point.
The conic K meets the line L in two points with multiplicity (by Bézout’s
Theorem for plane curves, §18.6.4, or by simple algebra). If K ∩ L consists of two
points a and b, we have a contradiction: K can’t be contracted to both π(a) and
π(b). But what happens if K meets L at one point, i.e., if the conic is tangent to the
line?

19.4.E. E XERCISE . Finish the proof of Pascal’s Theorem by dealing with this case.
Hint: nilpotents will come to the rescue. The intuition is as follows: K ∩ L is a
subscheme of L of length 2 (the length of a scheme was defined in §6.5.8), but the
scheme theoretic image π(K) can be shown to be a reduced closed point.
□

19.4.6. Remark. The key motivating fact that makes our argument work, Proposi-
tion 19.4.1, is centrally about curves not of genus 0, yet all the curves involved in
Pappus’s Theorem and Pascal’s Theorem have genus 0. The insight to keep in mind
is that union of curves of genus 0 need not have genus 0. In our case, it mattered
that cubic curves have genus 1, even if they are union of P1 ’s.

19.5 Hyperelliptic curves

We next discuss an important class of curves, the hyperelliptic curves. In this

section, we assume k is algebraically closed of characteristic not 2. (These hypotheses
can be relaxed, at some cost.)
A (projective regular integral) genus g curve C is hyperelliptic if it admits a
double cover of (i.e., degree 2, necessarily finite, morphism to) P1k . (See §16.3.4
for discussion of degrees of maps from curves to curves.) For convenience, when
we say C is hyperelliptic, we will implicitly have in mind a choice of double cover
π : C → P1 . (We will later see that if g ≥ 2, then there is at most one such double
cover, Proposition 19.5.7, so this is not a huge assumption.) The map π is called the
hyperelliptic map.
By Exercise 16.3.D, the preimage of any closed point p of P1 consists of either
one or two points. If π−1 (p) is a single point, we say p is a branch point, and
π−1 (p) is a ramification point of π. (The notion of ramification will be defined
more generally in §21.7.)
546 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

19.5.1. Theorem (hyperelliptic Riemann-Hurwitz formula). — Suppose k = k and

char k ̸= 2, π : C → P1k is a double cover by a projective regular irreducible genus g curve
over k. Then π has 2g + 2 branch points.
This is a special case of the Riemann-Hurwitz formula, which we will state and
prove in §21.4. You may have already heard about genus 1 complex curves double
covering P1 , branched over 4 points.
To prove Theorem 19.5.1, we first prove the following.

19.5.2. Proposition. — Assume char k ̸= 2 and k = k. Given r distinct points

p1 , . . . , pr ∈ P1 , there is precisely one double cover branched at precisely these points if r
is even, and none if r is odd.

Proof. Pick points 0 and ∞ of P1 distinct from the r branch points. All r branch
points are in P1 − ∞ = A1 = Spec k[x]. Suppose we have a double cover of A1 ,
C ′ → A1 , where x is the coordinate on A1 . This induces a quadratic field extension
K over k(x). As char k ̸= 2, this extension is Galois. Let σ : K → K be the Galois
involution. Let y be a nonzero element of K such that σ(y) = −y, so 1 and y
form a basis for K over the field k(x), and are eigenvectors of σ. Now σ(y2 ) = y2 ,
so y2 ∈ k(x). We can replace y by an appropriate k(x)-multiple so that y2 is
a polynomial, with no repeated factors, and monic. (This is where we use the
hypothesis that k is algebraically closed, to get leading coefficient 1.)
Thus y2 = xN + aN−1 xN−1 + · · · + a0 , where the polynomial on the right
(call it f(x)) has no repeated roots. The Jacobian criterion for regularity (in the
guise of Exercise 13.2.F) implies that this curve C0′ in A2 = Spec k[x, y] is regular.
Then C0′ is normal and has the same function field as C ′ . Thus C0′ and C ′ are both
normalizations of A1 in the finite extension of fields generated by y, and hence are
isomorphic. Thus we have identified C ′ in terms of an explicit equation.
The branch points correspond to those values of x for which there is exactly
one value of y, i.e., the roots of f(x). In particular, N = r, and
f(x) = (x − p1 ) · · · (x − pr ),
where the pi are interpreted as elements of k.
Having mastered the situation over A1 , we return to the situation over P1 . We
will examine the branched cover over the affine open set P1 \ {0} = Spec k[u], where
u = 1/x. The previous argument applied to Spec k[u] rather than Spec k[x] shows
that any such double cover must be of the form
C ′′ = Spec k[Z, u]/ Z2 − (u − 1/p1 ) · · · (u − 1/pr )


 Y  
= Spec k[Z, u]/ (−1)r pi Z2 − ur f (1/u)
−→ Spec k[u] = A1 .
So if there is a double cover over all of P1 , it must be obtained by gluing C ′′ to
′
C , “over” the gluing of Spec k[x] to Spec k[u] to obtain P1 .
Thus in K(C), we must have
z2 = ur f(1/u) = f(x)/xr = y2 /xr
Q
(where z is obtained from Z by multiplying by a square root of (−1)r pi ) from
which z2 = y2 /xr .
 © 2024 Ravi Vakil. Published by Princeton University Press. 547

If r is even, considering K(C) as generated by y and x, there are two possible

values of z: z = ±y/xr/2 . After renaming z by −z if necessary, there is a single way
of gluing these two patches together (we choose the positive square root).
If r is odd, the result follows from Exercise 19.5.A below.

19.5.A. E XERCISE . Suppose char k ̸= 2. Show that x does not have a square root in
the field k(x)[y]/(y2 −√f(x)), where
√ f is a polynomial with nonzero roots p1 , . . . , pr .
(Possible hint: why is 3 ∈ / Q( 2)?) Explain how this proves Proposition 19.5.2 in
the case where r is odd.
□
For future reference, we collect here our explicit (two-affine) description of the
hyperelliptic cover C → P1 .
z=y/xr/2
(19.5.2.1) Spec k[x, y]/(y2 − f(x)) Spec k[u, z]/(z2 − ur f(1/u))
y=z/ur/2

 u=1/x 
Spec k[x] Spec k[u]
x=1/u

19.5.3. If k is not algebraically closed. If k is not algebraically closed (but of character-

istic not 2), the above argument shows that if we have a double cover of A1 , then it
is of the form y2 = af(x), where f is monic (and a ̸= 0). Furthermore, if a and a ′
differ (multiplicatively) by an element of (k× )2 , then y2 = af(x) is isomorphic to
y2 = a ′ f(x). You may be able to use this to show that (assuming that k× ̸= (k× )2 )
a double cover is not determined by its branch points. Moreover, this failure is
classified by k× /(k× )2 . Thus we have lots of curves that are not isomorphic over k,
but become isomorphic over k. These are often called twists of each other.
(In particular, once we define elliptic curves, you will be able to show that there
exist two elliptic curves over Q with the same j-invariant, that are not isomorphic,
see Exercise 19.10.H.)

19.5.4. Back to proving the hyperelliptic Riemann-Hurwitz formula, Theorem 19.5.1.

Our explicit description of the unique double cover of P1 branched over r differ-
ent points will allow us to compute the genus, thereby completing the proof of
Theorem 19.5.1.
We continue the notation (19.5.2.1) of the proof of Proposition 19.5.2. Suppose
P1 has affine cover by Spec k[x] and Spec k[u], with u = 1/x, as usual. Suppose
C → P1 is a double cover, given by y2 = f(x) over Spec k[x], where f has degree r,
and z2 = ur f(1/u). Then C has an affine open cover by Spec k[x, y]/(y2 − f(x)) and
Spec k[u, z]/(z2 − ur f(1/u)). The corresponding Čech complex for OC is

0 / k[x, y]/(y2 − f(x)) × k[u, z]/(z2 − ur f(1/u)) d /

/ 0.


k[x, y]/(y2 − f(x)) x

The second (and last) nonzero part of the complex has basis consisting of monomials
xn yϵ , where n ∈ Z and ϵ = 0 or 1. To compute the genus g = h1 (C, OC ), we
548 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

must compute coker d. We can use the first factor k[x, y]/(y2 − f(x)) to hit the
monomials xn yϵ where n ∈ Z≥0 , and ϵ = 0 or 1. The image of the second
factor is generated by elements of the form um zϵ , where m ≥ 0 and ϵ = 0 or 1.
But um zϵ = x−m (y/xr/2 )ϵ . By inspection, the cokernel has basis generated by
monomials x−1 y, x−2 y, . . . , x−r/2+1 y, and thus has dimension r/2 − 1. Hence
g = r/2 − 1, from which Theorem 19.5.1 follows. □

19.5.5. Curves of every genus. As a consequence of the hyperelliptic Riemann-

Hurwitz formula (Theorem 19.5.1), we see that there are curves of every genus
g ≥ 0 over an algebraically closed field of characteristic not 2: to get a curve of
genus g, consider the branched cover branched over 2g + 2 distinct points. The
unique genus 0 curve is of this form, and we will see in §19.6.2 that every genus 2
curve is of this form. We will soon see that every genus 1 curve (reminder: over an
algebraically closed field!) is too (§19.9.5). But it is too much to hope that all curves
are of this form, and we will soon see (§19.7.4) that there are genus 3 curves that
are not hyperelliptic, and we will even get heuristic evidence that “most” genus 3
curves are not hyperelliptic. We will later give vague evidence (that can be made
precise) that “most” genus g curves are not hyperelliptic if g > 2 (§19.8.4).

19.5.B. E XERCISE . Verify that a curve C of genus at least 1 admits a degree 2 cover
of P1 if and only if it admits a degree 2 invertible sheaf L with h0 (C, L ) = 2.
Possibly in the course of doing this, verify that if C is a curve of genus at least 1, and
C has a degree 2 invertible sheaf L with at least 2 (linearly independent) sections,
then L has precisely two sections, and that this L is base-point-free and gives a
hyperelliptic map.

19.5.6. Proposition. — Assume g ≥ 2. If L corresponds to a hyperelliptic cover C → P1 ,

then L ⊗(g−1) =∼ ωC .

Proof. Compose the hyperelliptic map with the (g − 1)th Veronese embedding
|L | νg−1 =|OP1 (g−1)|
(19.5.6.1) C / P1 / Pg−1 .

The composition corresponds to L ⊗(g−1) . This invertible sheaf has degree 2g − 2

(by Exercise 18.4.G).
We next show that the pullback H0 (Pg−1 , O(1)) → H0 (C, L ⊗(g−1) ) is injec-
tive, i.e., any nonzero s ∈ H0 (Pg−1 , O(1)) pulls back to a nonzero element of
H0 (C, L ⊗(g−1) ), i.e., the image of C in Pg−1 is nondegenerate. This is because the
image of C in Pg−1 is the image of P1 by νg−1 (a rational normal curve, see §9.3.6);
but the image of P1 under the complete linear system |OP1 (g − 1)| is nondegenerate
(Exercise 15.2.C).
Thus L ⊗(g−1) has at least g sections. But by Exercise 19.2.A, the only invertible
sheaf of degree 2g − 2 with (at least) g sections is the dualizing sheaf. □
As an added bonus, we see that the composition of (19.5.6.1) is the complete
linear series |L ⊗(g−1) | — all sections of L ⊗(g−1) come up in this way.

19.5.7. Proposition (a genus ≥ 2 curve can be hyperelliptic in only one way). —

Any curve C of genus at least 2 admits at most one double cover of P1 . More precisely, if
L and M are two degree two line bundles yielding maps C → P1 , then L = ∼ M.
 © 2024 Ravi Vakil. Published by Princeton University Press. 549

Proof. If C is hyperelliptic, then we can recover the hyperelliptic map by considering

the canonical linear series given by ωC (the canonical map, which we will use again
repeatedly in the next few sections): it is a double cover of a degree g − 1 rational
normal curve (by the previous proposition), which is isomorphic to P1 . This double
cover is the hyperelliptic cover (also by the proof of the previous proposition). Thus
the hyperelliptic cover C → P1 is determined by the canonical map (so there can be
at most one). This map C → P1 must be induced by both L and M , from which
we have L = ∼ M (using Theorem 15.2.2, relating maps to projective space and line
bundles). □

19.5.C. E XERCISE . Suppose C is a hyperelliptic curve of genus at least 2. Show that

the hyperelliptic involution commutes with any other automorphism g of C. (Hint:
what happens when you conjugate the involution by g?)

19.5.8. The “space of hyperelliptic curves”. Thanks to Proposition 19.5.7, we can now
classify hyperelliptic curves of genus at least 2. Hyperelliptic curves of genus g ≥ 2
correspond to precisely 2g + 2 distinct points on P1 modulo S2g+2 , and modulo
automorphisms of P1 . Thus “the space of hyperelliptic curves” has dimension

2g + 2 − dim Aut P1 = 2g − 1.

This is not a well-defined statement, because we haven’t rigorously defined “the

space of hyperelliptic curves” — it is an example of a moduli space. For now, take
this as a plausibility statement. It is also plausible that this space is irreducible and
reduced — it is the image of something irreducible and reduced.

19.6 Curves of genus 2

19.6.1. The reason for leaving genus 1 for later. It might make most sense to jump
to genus 1 at this point, but the theory of elliptic curves is especially rich and subtle,
so we will leave it for §19.9.
In general, curves have quite different behaviors (topologically, arithmetically,
geometrically) depending on whether g = 0, g = 1, or g ≥ 2. This trichotomy
extends to varieties of higher dimension. We already have some inkling of it in
the case of curves. Arithmetically, genus 0 curves can have lots and lots of rational
points, genus 1 curves can have lots of rational points, and by Faltings’s Theorem
(Mordell’s Conjecture) any curve of genus at least 2 has at most finitely many
rational points. (Thus even before the proof of important cases of the Taniyama-
Shimura conjecture [Wi, TWi], and hence Fermat’s Last Theorem, we knew that
xn + yn = zn in P2 has at most finitely many rational solutions for n ≥ 4, as
such curves have genus n−1


2 > 1, see (18.6.7.1).) In the language of differential
geometry, Riemann surfaces of genus 0 are positively curved, Riemann surfaces
of genus 1 are flat, and Riemann surfaces of genus at least 2 are negatively curved.
It is a fact that curves of genus at least 2 have finite automorphism groups (see
§21.4.9), while curves of genus 1 have one-dimensional automorphism groups, see
Question 19.10.6), and the unique curve of genus 0 over an algebraically closed field
550 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

has a three-dimensional automorphism group (see Exercises 15.5.A and 15.5.B).

(See Exercise 21.5.F for more on this issue.)

19.6.2. Back to curves of genus 2.

Over an algebraically closed field, we saw in §19.3 that there is only one genus
0 curve. In §19.5 we established that there are hyperelliptic curves of genus 2. How
can we get a hold of curves of genus 2? For example, are they all hyperelliptic?
“How many” are there? We now tackle these questions.
Fix a curve C of genus g = 2. Then ωC is degree 2g − 2 = 2, and has 2 sections
(Exercise 19.2.A). By Exercise 19.5.B, ωC is base-point-free, and thus induces a
double cover C → P1 (unique up to automorphisms of P1 , which we studied in
§15.5.2). Conversely, any double cover C → P1 arises from a degree 2 invertible
sheaf with at least 2 sections. Hence if g(C) = 2, this invertible sheaf must be the
dualizing sheaf, just because it is a degree 2g − 2 = 2 line bundle with g = 2 sections
(or alternatively, by the easiest case of Proposition 19.5.6).
Hence we have a natural bijection between genus 2 curves and genus 2 dou-
ble covers of P1 (up to automorphisms of P1 ). If the characteristic is not 2, the
hyperelliptic Riemann-Hurwitz formula (Theorem 19.5.1) shows that the double
cover is branched over 2g + 2 = 6 geometric points. In particular, we have a “three-
dimensional space of genus 2 curves”. This is not rigorous, but we can certainly
show that there are an infinite number of genus 2 curves. Precisely:

19.6.A. E XERCISE . Fix an algebraically closed field k of characteristic not 2. Show

that there are an infinite number of (pairwise) non-isomorphic genus 2 curves over
k.

19.6.B. E XERCISE . Show that every genus 2 curve (over any field of characteristic
not 2) has finite automorphism group.

19.7 Curves of genus 3

19.7.1. The canonical embedding.

We begin with some general discussion about genus g > 1, then specialize to
genus 3.

19.7.A. I MPORTANT E XERCISE . Assume k = k (purely to avoid distraction — feel

free to remove this hypothesis). Suppose C is a genus g curve, with g > 1. Show
that if C is not hyperelliptic, then the dualizing invertible sheaf gives a closed
embedding C ,→ Pg−1 . (In the hyperelliptic case, we have already seen that the
dualizing sheaf gives us a double cover of a rational normal curve.) Hint: Recall
that h0 (C, ωC ) = g (Exercise 19.2.A). Use our “closed embedding test” of §19.1,
in the form of §19.2.10. Recall that if h0 (C, O(p + q)) = 2, then C is hyperelliptic
(Exercise 19.5.B).
This exercise gives a second proof that all genus 2 curves are hyperelliptic.
 © 2024 Ravi Vakil. Published by Princeton University Press. 551

19.7.2. Definition. Such a curve embedded in Pg−1 by the dualizing sheaf (also
known as the canonical line bundle) is called a canonical curve, and this closed
embedding is called the canonical embedding of C.

19.7.3. Back to genus 3.

Suppose now that C is a curve of genus 3. Then ωC has degree 2g − 2 = 4, and
has g = 3 sections. By Exercise 19.7.A, if (and only if) C is not hyperelliptic, the
canonical map describes C as a degree 4 curve in P2 .
Conversely, any quartic plane curve is canonically embedded. Reason: the
curve has genus 3 (see (18.6.7.1)), and is mapped by an invertible sheaf of degree 4
with 3 sections. But by Exercise 19.2.A, the only invertible sheaf of degree 2g − 2
with g sections is ωC .
In particular, each nonhyperelliptic genus 3 curve can be described as a quartic
plane curve in precisely one way (up to automorphisms of P2 , which were described
in Exercise 15.5.A). We have shown a bijection between nonhyperelliptic genus 3
curves, and plane quartic curves up to projective linear transformations (§15.5.2).

19.7.4. Remark. In particular, as there exist regular plane quartics (Exercise 13.3.E),
there exist nonhyperelliptic genus 3 curves.

19.7.B. E XERCISE ( CF. §19.5.8). Give a heuristic (non-rigorous) argument that the
nonhyperelliptic curves of genus 3 form a family of dimension 6. (Hint: Count the
dimension of the family of regular quartic plane curves, and quotient by Aut P2 =
PGL(3).)
The genus 3 curves thus seem to come in two families: the hyperelliptic curves
(a family of dimension 5), and the nonhyperelliptic curves (a family of dimension 6).
This is misleading — they actually come in a single family of dimension 6. In fact,
hyperelliptic curves are naturally limits of nonhyperelliptic curves. We can write
down an explicit family. (This explanation necessarily requires some hand-waving,
as it involves topics we haven’t seen yet.) Suppose we have a hyperelliptic curve
branched over 2g + 2 = 8 points of P1 . Choose an isomorphism of P1 with a conic
in P2 . There is a regular quartic plane curve meeting the conic at precisely those 8
points. (This requires a short argument using Bertini’s Theorem 13.4.2, which we
omit.) Then if f is the equation of the conic, and g is the equation of the quartic,
then f2 + t2 g is a family of quartics that are smooth for most t (smoothness is an
open condition in t, as we will see in Theorem 21.6.6 on generic smoothness). The
t = 0 case is a double conic. Then it is a fact that if you normalize the total space of
the family, the central fiber (above t = 0) turns into our hyperelliptic curve. Thus
we have expressed our hyperelliptic curve as a limit of nonhyperelliptic curves.

19.7.C. U NIMPORTANT E XERCISE . A (projective) curve (over a field k) admitting

a degree 3 cover of P1 is called trigonal. Show that every nonhyperelliptic genus
3 complex curve is trigonal, by taking the quartic model in P2 , and projecting to
P1 from any point on the curve. Do this by choosing coordinates on P2 so that p is
at [0, 0, 1]. (The same idea, applied to cubics rather than quartics, will be used in
§19.9.9.)

19.7.5. ⋆ A genus 3 curve with no nontrivial automorphisms.

552 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

We have seen that a (smooth projective integral) curve of genus at most 2

always has nontrivial automorphisms. It turns out that there are genus 3 curves
with no nontrivial automorphisms.

19.7.D. E XERCISE . Suppose C ′ ⊂ P2 is a smooth plane quartic curve (over any field
k). Show that there is bijection between automorphisms of C ′ and automorphisms
of P2 preserving C ′ .
Thus to find a genus 3 curve with no nontrivial automorphisms, we need only
find a smooth quartic plane curve C ′ such that the only automorphism of P2 fixing
C ′ as a set must be the identity. Your intuition may (correctly) tell you that most
quartics have this property. But exhibiting a specific C ′ (with proof) requires rolling
up our sleeves and getting to work. Poonen gives automorphism free curves over
any field in [P1]; an example in characteristic 0 is
y3 z − 3yz3 = 3x4 − 4x3 z + z4 .

19.7.E. E XERCISE . Suppose C is a smooth projective curve with no nontrivial

automorphisms. Show that no two distinct open subsets of C are isomorphic.

19.7.6. Genus 3 curves with nontrivial automorphisms.

Certainly genus 3 curves can have nontrivial automorphisms: witness hyperel-
liptic curves. More impressive is the Klein quartic
x3 y + y3 z + z3 x = 0
in P2C , which has 168 automorphisms. (Can you find them all?) In fact, the auto-
morphism group of the Klein quartic is the unique finite simple group of order 168
(the second-smallest nonabelian finite simple group). In §21.4.9, we will see that no
genus 3 curve can have more automorphisms (assuming the hard fact that curves
of genus greater than 1 have finite automorphism groups).

19.8 Curves of genus 4 and 5

We first consider nonhyperelliptic curves C of genus 4. Note that deg ωC = 6
and h0 (C, ωC ) = 4, so the canonical map expresses C as a sextic curve in P3 . Let
i : C ,→ P3 be the canonical (closed) embedding. We shall see that all such C are
complete intersections of quadric surfaces and cubic surfaces, and conversely all
regular complete intersections of quadric surfaces and cubic surfaces are genus 4
nonhyperelliptic curves, canonically embedded.
By (19.2.5.1) (Riemann-Roch and Serre duality),
h0 (C, i∗ OP3 (2)) = h0 (C, ω⊗2 ⊗2
C ) = deg ωC − g + 1 = 12 − 4 + 1 = 9.

On the other hand, h0 (P3 , OP3 (2)) = 3+2

2 = 10 (Exercise 15.1.C). Thus the restric-
tion map
H0 (P3 , OP3 (2)) / H0 (C, i∗ OP3 (2))

must have a nontrivial kernel, so there is at least one quadric surface Q in P3 that
contains our curve C. Now quadric surfaces are either double planes, or the union
of two planes, or cones, or regular quadrics. (They correspond to quadric forms
 © 2024 Ravi Vakil. Published by Princeton University Press. 553

of rank 1, 2, 3, and 4 respectively. The rank of a quadratic form was defined in

Exercise 5.4.J.) But C can’t lie in a plane, so Q must be a cone or regular. In particular,
Q is irreducible.
Now C can’t lie on two (distinct) such quadrics, say Q and Q ′ . Otherwise, as
Q and Q ′ have no common components (they are irreducible and not the same!),
Q ∩ Q ′ is a curve (not necessarily reduced or irreducible). By Bézout’s Theorem
(Exercise 18.6.J), Q ∩ Q ′ is a curve of degree 4. Thus our curve C, being of degree 6,
cannot be contained in Q ∩ Q ′ . (If you don’t see why directly, Exercise 18.6.E might
help.)

19.8.1. We next consider cubic surfaces. By (19.2.5.1) again,

h0 (C, ω⊗3 ⊗3
C ) = deg ωC − g + 1 = 18 − 4 + 1 = 15.

Now Sym3 Γ (C, ωC ) has dimension 3+3

3 = 20. Thus C lies on at least a 5-
dimensional vector space of cubics. Now a 4-dimensional subspace comes from
multiplying the quadratic form Q by a linear form (?w + ?x + ?y + ?z). Hence there is
still one cubic K whose underlying form is not divisible by the quadric form Q (i.e.,
K doesn’t contain Q.) Then K and Q share no component, so K ∩ Q is a complete
intersection containing C as a closed subscheme. Now K ∩ Q and C are both degree
6 (the former by Bézout’s Theorem, Exercise 18.6.J, and the latter because C is
embedded by a degree 6 line bundle, Exercise 18.6.H). Also, K ∩ Q and C both have
arithmetic genus 4 (the former by Exercise 18.6.S). These two invariants determine
the (linear) Hilbert polynomial, so K ∩ Q and C have the same Hilbert polynomial.
Hence C = K ∩ Q by Exercise 18.6.E.
We now show the converse, and that any regular complete intersection C of a
quadric surface with a cubic surface is a canonically embedded genus 4 curve. By
Exercise 18.6.S, such a complete intersection has genus 4.

19.8.A. E XERCISE . Show that OC (1) has at least 4 sections. (Translation: C doesn’t
lie in a hyperplane.)
The only degree 2g − 2 invertible sheaf with (at least) g sections is the dualizing
∼ ωC , and C is indeed canonically embedded.
sheaf (Exercise 19.2.A), so OC (1) =

19.8.B. E XERCISE . Give a heuristic argument suggesting that the nonhyperelliptic

curves of genus 4 “form a family of dimension 9”.
On to genus 5!

19.8.C. E XERCISE . Suppose C is a nonhyperelliptic genus 5 curve. Show that the

canonical curve is degree 8 in P4 . Show that it lies on a three-dimensional vector
space of quadric surfaces (i.e., it lies on 3 linearly independent quadrics). Show that
a regular complete intersection of 3 quadric surfaces is a canonical(ly embedded)
genus 5 curve.
Unfortunately, not all canonical genus 5 curves are the complete intersection of
3 quadric surfaces in P4 . But in the same sense that most genus 3 curves can be de-
scribed as plane quartics, most canonical genus 5 curves are complete intersections
of 3 quadric surfaces, and most genus 5 curves are nonhyperelliptic. The correct
way to say this is that there is a dense Zariski-open locus in the moduli space of
554 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

genus 5 curves consisting of nonhyperelliptic curves whose canonical embedding

is cut out by 3 quadric surfaces.
(Those nonhyperelliptic genus 5 canonical curves not cut out by a 3-dimensional
vector space of quadric surfaces are precisely the trigonal curves, see Exercise 19.7.C.
The triplets of points mapping to the same point of P1 under the trigonal map turn
out to lie on a line in the canonical embedding. Any quadric vanishing along those
3 points must vanish along the line — basically, any quadratic polynomial with
three zeros must be the zero polynomial.)

19.8.D. E XERCISE . Assuming the discussion above, count complete intersections of

three quadric surfaces to give a heuristic argument suggesting that the curves of
genus 5 “form a family of dimension 12”.

19.8.2. We have now understood curves of genus 3 through 5 by thinking of

canonical curves as complete intersections. Sadly our luck has run out.

19.8.E. E XERCISE . Show that if C ⊂ Pg−1 is a canonical curve of genus g ≥ 6, then

C is not a complete intersection. (Hint: Bézout’s Theorem, Exercise 18.6.J.)

19.8.3. There is still something beautiful going on. For example, for most genus 6
curves over k = k, the canonical embedding (degree 10 in P5 ) arises in the following
way. Start with the complete intersection of G(2, 5) (dimension 6; do you see why?)
under its Plücker embedding in P9 . Intersect it with a linear space of codimension
4, yielding a smooth surface in P5 . This surface is isomorphic to P2 blown up at
4 points. (It is a del Pezzo surface, to be defined in Exercise 21.5.H. This particular
surface appears in the table in Remark 27.3.7.) Slice (intersect) the surface with a
quadric hypersurface to obtain a genus 6 curve.

19.8.4. Some discussion on curves of general genus. However, we still have some data.
If Mg is this ill-defined “moduli space of genus g curves”, we have heuristics to
find its dimension for low g. In genus 0, over an algebraically closed field, there is
only one genus 0 curve (Proposition 19.3.1), so it appears that dim M0 = 0. In genus
1, over an algebraically closed field, we will soon see that the elliptic curves are
classified by the j-invariant (Exercise 19.10.G), so it appears that dim M1 = 1. We
have also informally computed dim M2 = 3, dim M3 = 6, dim M4 = 9, dim M5 =
12. What is the pattern? In fact in some strong sense it was known by Riemann
that dim Mg = 3g − 3 for g > 1. What goes wrong in genus 0 and genus 1? As a
clue, recall our insight when discussing Hilbert functions (§18.6) that whenever
some function is “eventually polynomial”, we should assume that it “wants to be
polynomial”, and there is some better function (usually an Euler characteristic) that
is polynomial, and that cohomology-vanishing ensures that the original function
and the better function “eventually agree”. Making sense of this in the case of Mg
is far beyond the scope of our current discussion, so we will content ourselves
by observing the following facts. Every regular curve of genus greater than 1 has
a finite number of automorphisms — a zero-dimensional automorphism group.
Every regular curve of genus 1 has a one-dimensional automorphism group (see
Question 19.10.6). And the only regular curve of genus 0 has a three-dimensional
automorphism group (Exercise 15.5.B). (See Aside 21.5.12 for more discussion.) So
 © 2024 Ravi Vakil. Published by Princeton University Press. 555

notice that for all g ≥ 0,

dim Mg − dim Aut Cg = 3g − 3

where Aut Cg means the automorphism group of any curve of genus g.

In fact, in the language of stacks (or orbifolds), it makes sense to say that the
dimension of the moduli space of (smooth projective geometrically irreducible)
genus 0 curves is −3, and the dimension of the moduli space of genus 1 curves is 0.

19.9 Curves of genus 1

Finally, we come to the very rich case of curves of genus 1. We will present the
theory by discussing interesting things about line bundles of steadily increasing
degree. Although it is not particularly essential to us now, recall that the degree d
line bundles on C are denoted Picd C (Definition 18.4.2).

19.9.1. Line bundles of degree 0 (Pic0 C).

Suppose C is a genus 1 curve. Then deg ωC = 2g − 2 = 0 and h0 (C, ωC ) = g =
1 (by Exercise 19.2.A). But the only degree 0 invertible sheaf with a section is the
structure sheaf (§19.2.4), so we conclude that ωC =∼ OC . (If you know that complex
genus 1 curves are of the form C modulo a lattice, and Miracle 18.5.2 that the sheaf
of differentials is dualizing, then you might not be surprised.)
We move on to line bundles of higher degree. Next, note that if deg L > 0,
then Riemann-Roch and Serre duality (19.2.5.1) give

h0 (C, L ) = deg L − g + 1 = deg L .

19.9.2. Line bundles of degree 1 (Pic1 (C)).

Each degree 1 (k-valued) point q determines a line bundle OC (q), and two
distinct points determine two distinct line bundles (as a degree 1 line bundle has
only one section, up to scalar multiples). Conversely, any degree 1 line bundle L is
of the form OC (q) (as L has a section — then just take its divisor of zeros).
Thus we have a canonical bijection between degree 1 line bundles and degree 1
(closed) points. (If k is algebraically closed, as all closed points have residue field k,
this means that we have a canonical bijection between degree 1 line bundles and
closed points.)
Define an elliptic curve to be a genus 1 curve E with a choice of k-valued point
p. The choice of this point should always be considered part of the definition of
an elliptic curve — “elliptic curve” is not a synonym for “genus 1 curve”. (Note: a
genus 1 curve need not have any k-valued points at all! For example, the hyperelip-
tic genus 1 curve obtained (as in §19.5.2) from y2 = −x4 − 1 in A2Q has no R-points,
and hence no Q-points. Of course, if k = k, then any closed point is k-valued, by
the Nullstellensatz 3.2.5.) We will often denote elliptic curves by E rather than C.
If (E, p) is an elliptic curve, then there is a canonical bijection between the set
of degree 0 invertible sheaves (up to isomorphism) and the set of degree 1 points of
556 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

E: twist the degree 1 line bundles by O(−p). Explicitly, the bijection is given by

L  / div(L (p))


O(q − p) o q
But the degree 0 invertible sheaves form a group (under tensor product), so we
have proved:

19.9.3. Proposition (the group law on the degree 1 points of an elliptic curve). —
The above bijection defines an abelian group structure on the degree 1 points of an elliptic
curve, where p is the identity.
From now on, we will identify closed points of E with degree 0 invertible
sheaves on E without comment.
For those familiar with the complex analytic picture, this is not surprising: E is
isomorphic to the complex numbers modulo a lattice: E = ∼ C/Λ. (We haven’t shown
this, of course.)
Proposition 19.9.3 is currently just a bijection of sets. Given that E has a much
richer structure (it has a generic point, and the structure of a variety), this is a sign
that there should be a way of defining some scheme Pic0 (E), and that this should be
an isomorphism of schemes. We will soon show (Theorem 19.10.4) that this group
structure on the degree 1 points of E comes from a group variety structure on E.

19.9.4. Aside: The Mordell-Weil Theorem, group, and rank. This is a good excuse to
mention the Mordell-Weil Theorem: for any elliptic curve E over Q, the Q-points of E
form a finitely generated abelian group, often called the Mordell-Weil group. By the
classification of finitely generated abelian groups (a special case of the classification
of finitely generated modules over a principal ideal domain, Remark 6.4.4), the
Q-points are a direct sum of a torsion part, and of a free Z-module. The rank of the
Z-module is called the Mordell-Weil rank.

19.9.5. Line bundles of degree 2 (Pic2 (E)), see Figure 19.4.

Note that OE (2p) has 2 sections, so E admits a double cover of P1 (Exer-
cise 19.5.B). One of the ramification points is p: one of the sections of OE (2p)
vanishes at p to order 2, so there is a point of P1 whose preimage consists of p
(with multiplicity 2). Assume now that k = k and char k ̸= 2, so we can use the
hyperelliptic Riemann-Hurwitz formula (Theorem 19.5.1), which implies that E has
4 ramification points (p and three others). Conversely, given 4 points in P1 , there
exists a unique double cover branched at those 4 points (Proposition 19.5.2). Thus
elliptic curves correspond to 4 distinct points in P1 , where one is marked P, up to
automorphisms of P1 . Equivalently, by placing P at ∞, elliptic curves correspond
to 3 points in A1 , up to affine maps x 7→ ax + b.

19.9.A. E XERCISE . Show that the other three ramification points are precisely the
(non-identity) 2-torsion points in the group law. (Hint: if one of the points is q,
show that O(2q) = ∼ O(2p), but O(q) is not congruent to O(p).)

Thus (if char k ̸= 2 and k = k) every elliptic curve has precisely four 2-torsion
points. If you are familiar with the complex picture E =∼ C/Λ, this isn’t surprising.
 © 2024 Ravi Vakil. Published by Princeton University Press. 557

F IGURE 19.4. Elliptic curves yield double covers of the projective

line, branched at four points (§19.9.5)

19.9.6. Follow-up remark. An elliptic curve with full level n structure is an elliptic
curve with an isomorphism of its n-torsion points with (Z/n)2 . (This notion has
problems if n is divisible by char k.) Thus an elliptic curve with full level 2 structure is
the same thing as an elliptic curve with an ordering of the three other branch points
in its degree 2 cover description. Thus (if k = k) these objects are parametrized by
the λ-line, which we discuss below.
Follow-up to the follow-up. There is a notion of moduli spaces of elliptic curves
with full level n structure. Such moduli spaces are smooth curves (where this is
interpreted appropriately — they are stacks), and have smooth compactifications.
A weight k level n modular form is a section of ω⊗k
M where ωM is the dualizing (or
canonical) sheaf of this moduli space (“modular curve”).

19.9.7. The cross-ratio and the j-invariant. (We maintain the assumption that k is
algebraically closed and characteristic not 2.) If the three other branch points are
temporarily labeled Q1 , Q2 , Q3 ∈ P1 , there is a unique automorphism of P1 taking
P, Q1 , Q2 to (∞, 0, 1) respectively (as Aut P1 is three-transitive, Exercise 15.5.B).
Suppose that Q3 is taken to some number λ under this map, where necessarily
λ ̸= 0, 1, ∞.
The value λ is called the cross-ratio of the four points (P, Q1 , Q2 , Q3 ) of P1
(invented by Pappus of Alexandria, [MO152008]).

19.9.B. E XERCISE . Show that isomorphism classes of four ordered distinct (closed)
points on P1 (over an algebraically closed field k), up to projective equivalence
(automorphisms of P1 ), are classified by the cross-ratio.

19.9.C. E XERCISE . Show that the cross-ratio of (a, b, c, d) is the same as the cross
ratio of (b, a, d, c). The group S4 acts on 4-tuples of points on P1 , and hence on their
cross ratios. Show that the action of S4 on the cross-ratio is trivial when restricted
to the “Klein 4-group” V4 ⊂ S4 .
We have not defined the notion of moduli space, but the previous exercise
illustrates the fact that P1 \ {0, 1, ∞} (the image of the cross-ratio map) is the moduli
space for four ordered distinct points of P1 up to projective equivalence.
Notice:
• If we had instead sent P, Q2 , Q1 to (∞, 0, 1), then Q3 would have been
sent to 1 − λ.
558 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

• If we had instead sent P, Q1 , Q3 to (∞, 0, 1), then Q2 would have been

sent to 1/λ.
• If we had instead sent P, Q3 , Q1 to (∞, 0, 1), then Q2 would have been
sent to 1 − 1/λ = (λ − 1)/λ.
• If we had instead sent P, Q2 , Q3 to (∞, 0, 1), then Q1 would have been
sent to 1/(1 − λ).
• If we had instead sent P, Q3 , Q2 to (∞, 0, 1), then Q1 would have been
sent to 1 − 1/(1 − λ) = λ/(λ − 1).
Thus these six values (which correspond to S3 , the symmetric group permuting
Q1 , Q2 , and Q3 ) yield the same elliptic curve, and this elliptic curve will (upon
choosing an ordering of the other 3 branch points) yield one of these six values.
This is fairly satisfactory already. To check if two elliptic curves (E, p), (E ′ , p ′ )
over k = k are isomorphic, we write both as double covers of P1 ramified at p
and p ′ respectively, then order the remaining branch points, then compute their
respective λ’s (say λ and λ ′ respectively), and see if they are related by one of the
six expressions above:
(19.9.7.1) λ ′ = λ, 1/λ, 1 − λ, 1/(1 − λ), (λ − 1)/λ, or λ/(λ − 1).
It would be far more convenient if, instead of a “six-valued invariant” λ, there
were a single invariant (let’s call it j), such that j(λ) = j(λ ′ ) if and only if one of
the equalities of (19.9.7.1) holds. This j-function should presumably be algebraic,
so it would give a map j from the λ-line A1 \ {0, 1} to the A1 . By the Curve-to-
Projective Extension Theorem 15.3.1, this would extend to a morphism j : P1 → P1 .
By Exercise 16.3.D, because this is (for most λ) a 6-to-1 map, the degree of this cover
is 6 (or more correctly, at least 6).
We can make this dream more precise as follows. The elliptic curves over k
correspond to k-valued points of P1 \ {0, 1, ∞}, modulo the action of S3 on λ given
above. Consider the subfield K of k(λ) fixed by S3 . Then the extension k(λ)/K
is necessarily Galois (see for example [DF, §14.2, Thm. 9]), and degree 6. We are
hoping that this subfield is of the form k(j), and if so, we would obtain the j-map
P1 → P1 as described above. One could show that K is finitely generated over k,
and then invoke Lüroth’s Theorem, which we will soon prove in Example 21.4.7;
but we won’t need this.
Instead, we will just hunt for such a j. Note that λ should satisfy a sextic poly-
nomial over k(j) (or more precisely given what we know right now, a polynomial
of degree at least six), as for each j-invariant, there are six values of λ in general.
As you are undoubtedly aware, there is such a j-invariant. Here is the formula
for the j-invariant that everyone uses:
(λ2 − λ + 1)3
(19.9.7.2) j = 28 .
λ2 (λ − 1)2
You can readily check that j(λ) = j(1/λ) = j(1 − λ) = · · · , and that as j (in
lowest terms) has a degree 6 numerator and degree < 6 denominator, j indeed
determines a degree 6 map from P1 (with coordinate λ) to P1 (with coordinate j).
But this complicated-looking formula begs the question: where did this formula
come from? How did someone think of it? We will largely answer this, but we will
ignore the 28 . This, as you might imagine, arises from characteristic 2 issues, and
in order to invoke the results of §19.5 we have been assuming char k ̸= 2. (To see
 © 2024 Ravi Vakil. Published by Princeton University Press. 559

why the 28 is forced upon us by characteristic 2, see [De, p. 64]. From a different,
complex-analytic, point of view, the 28 comes from the fact that the q-expansion of
j begins j = q−1 + 744 + · · · : the q−1 term has coefficient 1; see [Se5, p. 90, Rem. 2].
These seemingly different explanations are related by the theory of the Tate curve,
see [Si2, Thm. V.3.1].)
Rather than using the formula handed to us, let’s try to guess what j is. We
won’t expect to get the same formula as (19.9.7.2), but our answer should differ by
an automorphism of the j-line (P1 ) — we will get j ′ = (aj + b)/(cj + d) for some a,
b, c, d (Exercise 15.5.A).
We are looking for some j ′ (λ) such that j ′ (λ) = j ′ (1/λ) = · · · . Hence we want
some expression in λ that is invariant under this S3 -action. A first possibility would
be to take the product of the six numbers
1 λ−1 1 λ
λ · (1 − λ) · · · ·
λ λ 1−λ λ−1
This is silly, as the product is obviously 1.
A better idea is to add them all together:
1 λ−1 1 λ
λ + (1 − λ) + + + +
λ λ 1−λ λ−1
This also doesn’t work, as they add to 3 — the six terms come in pairs adding to 1.
(Another reason you might realize this can’t work: if you look at the sum, you
will realize that you will get something of the form “degree at most 3” divided by
“degree at most 2”. Then if j ′ = p(λ)/q(λ), then λ is a root of a cubic over j. But we
said that λ should satisfy a sextic over j ′ . The only way we avoid a contradiction is
if j ′ ∈ k.)
But you will undoubtedly have another idea immediately. One good idea is
to take the second symmetric function in the six roots. An equivalent one that is
easier to do by hand is to add up the squares of the six terms. Even before doing
the calculation, we can see that this will work: it will clearly produce a fraction
whose numerator and denominator have degree at most 6, and it is not constant, as
when λ is some fixed small number (say 1/2), the sum of squares is some small real
number, while when λ is a large real number, the sum of squares will have to be
some large real number (different from the value when λ = 1/2).
When you add up the squares by hand (which is not hard), you will get
2λ6 − 6λ5 + 9λ4 − 8λ3 + 9λ2 − 6λ + 2
j′ = .
λ2 (λ − 1)2
∼ k(j ′ ): you can check (again by hand) that
Indeed k(j) =
2λ6 − 6λ5 + 12λ4 − 14λ3 + 12λ2 − 6λ + 2
2j/28 = .
λ2 (λ − 1)2
Thus 2j/28 − j ′ = 3.

19.9.8. Line bundles of degree 3 (Pic3 (E)).

In the discussion of degree 2 line bundles 19.9.5, we assumed char k ̸= 2 and
k = k, in order to invoke the Riemann-Hurwitz formula. In this section, we will
start with no assumptions, and add them as we need them. In this way, you will
see what partial results hold with weaker assumptions.
560 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Consider the degree 3 invertible sheaf OE (3p). By Riemann-Roch (19.2.5.1),

h0 (E, OE (3p)) = deg(3p) − g + 1 = 3. As deg OE (3p) > 2g, this gives a closed
embedding (Remark 19.2.11 and Exercise 19.2.E). Thus we have a closed embedding
E ,→ P2k as a cubic curve. Moreover, there is a line in P2k meeting E at point p with
multiplicity 3, corresponding to the section of O(3p) vanishing precisely at p with
multiplicity 3. (A line in the plane meeting a smooth curve with multiplicity at least
2 is a tangent line, see Definition 13.3.3. A line in the plane meeting a smooth curve
with multiplicity at least 3 is said to be a flex line, and that point is a flex point or
flex of the curve.) (See Figure 19.5.)

F IGURE 19.5. Finding the Weierstrass cubic incarnation of an el-

liptic curve

Choose projective coordinates on P2k so that p maps to [0, 1, 0], and the flex line
is the line at infinity z = 0. Then the cubic is of the following form:

? x3 + 0 x2 y + 0 xy2 + 0 y3

+ ? x2 z + ? xyz + ? y2 z =0

+ ? xz2 + ? yz2

+ ? z3

The coefficient of x3 is not 0 (or else this cubic is divisible by z). Dividing the
entire equation by this coefficient, we can assume that the coefficient of x3 is 1. The
coefficient of y2 z is not 0 either (or else this cubic is singular at x = z = 0). We
can scale z (i.e., replace z by a suitable multiple) so that the coefficient of y2 z is
−1. If the characteristic of k is not 2, then we can then replace y by y + ?x + ?z
so that the coefficients of xyz and yz2 are 0, and if the characteristic of k is not 3,
we can replace x by x + ?z so that the coefficient of x2 z is also 0. In conclusion, if
 © 2024 Ravi Vakil. Published by Princeton University Press. 561

char k ̸= 2, 3, the elliptic curve may be written as

(19.9.8.1) y2 z = x3 + axz2 + bz3 .
This is called the Weierstrass normal form of the curve.

19.9.9. From the Weierstrass cubic to the double cover. We see the hyperelliptic (double
cover) description of the curve by setting z = 1, or more precisely, by working in
the distinguished open set z ̸= 0 and using inhomogeneous coordinates. Here is
the geometric explanation of why the double cover description is visible in the
cubic description. Project the cubic from p = [0, 1, 0] (see Figure 19.6). This is a map
E \ p → P1 (given by [x, y, z] 99K [x, z]), and is basically the vertical projection of
the cubic to the x-axis. (Figure 19.6 may help you visualize this.) By the Curve-to-
Projective Extension Theorem 15.3.1, the morphism E − p → P1 extends over p. If
L := O(3p) is the line bundle giving the morphism E ,→ P2 (via Theorem 15.2.2
describing maps to projective space in terms of line bundles), then the two sections
x and z giving the map to P1 vanish at p to order 1 and 3 respectively. To “resolve”
the rational map into an honest morphism, we interpret x and z as sections of
L (−p) = O(2p), and now they generate a base-point-free linear series, and thus a
morphism to P1 . (You may be able to interpret this as implementing the proof of
Theorem 15.2.2 in a different language.) A similar idea, applied to quartics rather
than cubics, was used in Exercise 19.7.C.

F IGURE 19.6. From the Weierstrass cubic to the double cover in-
terpretation of an elliptic curve (cf. Figure 19.5)

As a consequence, with a little sweat, we can compute the j-invariant as a

function of a and b:
28 33 a3
j(a, b) = .
4a3 + 27b2

19.9.D. E XERCISE . Show that the flexes of the cubic are the 3-torsion points in the
group E. (“Flex” was defined in §19.9.8: it is a point where the tangent line meets
562 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

the curve with multiplicity at least 3 at that point. In fact, if k is algebraically closed
and char k ̸= 3, there are nine of them. This won’t be surprising if you are familiar
with the complex story, E = ∼ C/Λ.)

19.9.10. The group law, geometrically.

The group law has a beautiful classical description in terms of the Weierstrass
form. Consider Figure 19.7. In the Weierstrass coordinates, the origin p is the only
point of E meeting the line at infinity (z = 0); in fact the line at infinity corresponds
to the tautological section of O(3p). If a line meets E at three points p1 , p2 , p3 , then
∼ O(3p)
O(p1 + p2 + p3 ) =
from which (in the group law) p1 + p2 + p3 = 0.
Hence to find the inverse of a point s, we consider the intersection of E with
the line sp; −s is the third point of intersection. To find the sum of two points q and
r, we consider the intersection of E with the line qr, and call the third point s. We
then compute −s by connecting s to p, obtaining q + r.

F IGURE 19.7. The group law on the elliptic curve, geometrically

We could give this description of a group law on a cubic curve in Weierstrass

normal form to anyone familiar with the notion of projective space, and the notion
of a group, but we would then have to prove that the construction we are giving
indeed defines a group. In particular, we would have to prove associativity, which
is not a priori clear. But in this case, we have already established that the degree 1
points form a group, by giving a bijection to Pic0 E, and we are merely interpreting
the group law on Pic0 E.
Note that this description works even in characteristic 2 and 3; we do not need
the cubic to be in Weierstrass normal form, and we need only that O(3p) gives a
closed embedding into P2 .

19.9.11. Degree 4 line bundles. You have probably forgotten that we began by
studying line bundles degree by degree. The story doesn’t stop in degree 3. In the
same way that we showed that a canonically embedded nonhyperelliptic curve of
genus 4 is the complete intersection in P3k of a quadric surface and a cubic surface
(§19.8), we can show the following.

19.9.E. E XERCISE . Show that the complete linear series for O(4p) embeds E in P3
as the complete intersection of two quadric surfaces. Hint: Show the image of E is
contained in at least 2 linearly independent quadric surfaces. Show that neither can
 © 2024 Ravi Vakil. Published by Princeton University Press. 563

be reducible, so they share no components. Use Bézout’s Theorem (Exercise 18.6.J)

and Exercise 18.6.E, as in §19.8.1.
The beautiful structure doesn’t stop with degree 4, but it gets more compli-
cated. For example, the degree 5 embedding is not a complete intersection (of
hypersurfaces), but is the complete intersection of G(2, 5) under its Plücker em-
bedding with five hyperplanes (or perhaps better, a codimension 5 linear space).
In seemingly different terminology, its equations are 4 × 4 Pfaffians of a general
5 × 5 skew-symmetric matrix of linear forms, although I won’t say what this means.
(This construction may remind you of a similar construction for genus 6 curves in
§19.8.3. This is no coincidence.)

19.10 Elliptic curves are group varieties

Ever since 1949, I considered the construction of an algebraic theory of the Picard
variety as the task of greatest urgency in abstract algebraic geometry.
— A. Weil [Weil, p. 537]
We initially described the group law on the degree 1 points of an algebraic
curve in a rather abstract way. From that definition, it was not clear that over C the
group operations (addition, inverse) are continuous. But the explicit description in
terms of the Weierstrass cubic makes this clear. In fact we can observe even more:
addition and inverse are algebraic in general. Better yet, elliptic curves are group
varieties. (Thus they are abelian varieties, see Definition 11.4.11.)

19.10.1. (This is a clue that Pic0 (E) really wants to be a scheme, and not just a group.
Once the notion of “moduli space of line bundles on a variety” is made precise, this
can be shown, see for example [FGIKNV, Part 5].)
We begin with the “group inverse” morphism, as a warm-up.

19.10.2. Proposition. — If char k ̸= 2, 3, there is a morphism of k-varieties E → E

sending a (degree 1) point to its inverse, and this construction behaves well under field
extension of k.
In other words, the “inverse map” in the group law actually arises from a
morphism of schemes — it isn’t just a set map. (You are welcome to think through
the two remaining characteristics, and to see that essentially the same proof applies.
But the proof of Theorem 19.10.4 will give you a better sense of how to proceed.)

Proof. Consider the map (the hyperelliptic involution) y 7→ −y of the Weierstrass

normal form. □
The algebraic description of addition would be a big mess if we were to write
it down. We will be able to show algebraicity by a trick — not by writing it down
explicitly, but by thinking through how we could write it down explicitly. The main
part of the trick is the following proposition. We give it in some generality just
because it can be useful, but you may prefer to assume that k = k and C is a regular
cubic. If C is a curve, let Creg be the regular points of C, or equivalently, the normal
points (Theorem 13.5.8), which form an open subset of C (Exercise 10.7.O).
564 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

19.10.3. Proposition. — Suppose C ⊂ P2k is a geometrically integral cubic curve (so in

particular C contains no lines). There is a unique morphism t : Creg × Creg → Creg such
that

(a) if p and q are distinct regular k-valued points of C, then t(p, q) is obtained by
intersecting the line pq with C, and taking the third “residual” point of inter-
section with C. More precisely, pq will meet C at three points with multiplicity
(Exercise 9.3.D), including p and q; t(p, q) is the third point; and
(b) this property remains true after extension to k.

Furthermore, if p is a k-valued point of Creg , then t(p, p) is where the tangent line ℓ to C
at p meets C again. More precisely, ℓ will meet C at three points with multiplicity, which
includes p with multiplicity 2; t(p, p) is the third point.
We will need property (b) because C may have few enough k-valued points
(perhaps none!) that the morphism t cannot be determined by its behavior on them.
In the course of the proof, we will see that (b) can be extended to “this property
remains true after any field extension of k”.

Proof. We first show (in this paragraph) that if p and q are distinct regular points,
then the third point r of intersection of pq with C is also regular. If r = p or r = q,
we are done. Otherwise, the cubic obtained by restricting C to pq has three distinct
(hence reduced, i.e., multiplicity 1) roots, p, q, and r. Thus C ∩ pq is regular at r, so
r is a regular point of C by the slicing criterion for regularity, Exercise 13.2.C.
We now assume that k = k, and leave the general case to the end. Fix p, q,
and r, where p ̸= q, and r is the “third” point of intersection of pq with C. We
will describe a morphism tp,q in an open neighborhood of (p, q) ∈ Creg × Creg . By
Exercise 11.3.B, showing that morphisms of varieties over k are determined by their
behavior on closed (k-valued) points, these morphisms glue together (uniquely) to
give a morphism t, completing the proof in the case k = k.
Choose projective coordinates on P2 in such a way that U0 = ∼ Spec k[x1 , x2 ] con-
tains p, q, and r, and the line pq is not “vertical”. More precisely, in Spec k[x1 , x2 ],
say p = (p1 , p2 ) (in terms of “classical coordinates” — more pedantically, p =
[(x1 − p1 , x2 − p2 )]), q = (q1 , q2 ), r = (r1 , r2 ), and p1 ̸= q1 . In these coordi-
nates, the curve C is cut out by some cubic, which we also sloppily denote C:
C(x1 , x2 ) = 0.
Now suppose P = (P1 , P2 ) and Q = (Q1 , Q2 ) are two points of C ∩ U0 (not
necessarily our p and q). We attempt to compute the third point of intersection of
PQ with C, in a way that works on an open subset of C × C that includes (p, q). To
do this explicitly requires ugly high school algebra, but because we know how it
looks, we will be able to avoid dealing with any details!
The line PQ is given by x2 = mx1 + b, where m = PP21 −Q −Q1 and b = P2 − mP1
2

are both rational functions of P and Q. Then m and b are defined for all P and Q
such that P1 ̸= Q1 (and hence for an open neighborhood of (p, q), as p1 ̸= q1 , and
as P1 ̸= Q1 is an open condition).
Now we solve for C ∩ PQ, by substituting x2 = mx1 + b into C, to get
C(x1 , mx1 + b). This is a cubic in x1 , say

γ(x1 ) = Ax31 + Bx21 + Cx1 + D = 0.

 © 2024 Ravi Vakil. Published by Princeton University Press. 565

The coefficients of γ are rational functions of P1 , P2 , Q1 , and Q2 . The cubic γ has 3

roots (with multiplicity) so long as A ̸= 0 , which is a Zariski-open condition on m
and b, and hence a Zariski-open condition on P1 , P2 , Q1 , Q2 . As P, Q ∈ C∩PQ∩U0 ,
P1 and Q1 are two of the roots of γ(x1 ) = 0. Since the line PQ contains p q, and
r, the third root of γ(x1 ) = 0 is R1 . The sum of the roots of γ(x1 ) = 0 is −B/A (by
Viète’s formula), so the third root of γ is R1 := −B/A − P1 − Q1 . Thus if we take
R2 = mR1 + b, we have found the third point of intersection of PQ with C (which
happily lies in U0 ) We have thus described a morphism from the open subset of
(Creg ∩ U0 ) × (Creg ∩ U0 ), containing (p, q), that does what we want. (Precisely, the
open subset is defined by A ̸= 0, which can be explicitly unwound.) We have thus
completed the proof of Proposition 19.10.3 (except for the last paragraph) for k = k.
(Those who believe they are interested only in algebraically closed fields can skip
ahead.)
We extend this to Proposition 19.10.3 for every field k except F2 . Suppose
[x ,x ]
U0 1 2 = Spec k[x1 , x2 ] is any affine open subset of P2k , along with choice of
coordinates. (The awkward superscript “[x1 , x2 ]” is there to emphasize that the
particular coordinates are used in the construction.) Then the construction above
[x ,x ]
gives a morphism defined over k from an open subset of (Creg ∩ U0 1 2 ) × (Creg ∩
[x ,x ]
U0 1 2 ) (note that all of the hypothetical algebra was done over k), that sends P
and Q to the third point of intersection of PQ with C. Note that this construction
commutes with any field extension, as the construction is insensitive to the field
we are working over. Thus after base change to the algebraic closure, the map
also has the property that it takes as input two points, and spits out the third
point of intersection of the line with the cubic. Furthermore, all of these maps (as
[x ,x ]
U0 1 2 varies over all complements U0 of lines “with k-coefficients”, and choices
of coordinates on U0 ) can be glued together: they agree on their pairwise overlaps
(as after base change to k they are the same, by our previous discussion, and two
maps that are the same after base change to k were the same to begin with by
Exercise 10.2.J), and this is what is required to glue them together (Exercise 7.2.A).
[x ,x ]
We can geometrically interpret the open subset (Creg ∩ U0 1 2 ) × (Creg ∩
[x ,x ]
U0 1 2 ) by examining the construction: it is defined in the locus {P = (P1 , P2 ), Q =
(Q1 , Q2 )} where (i) P1 ̸= Q1 , and (ii) the third point of intersection R of PQ with C
also lies in U0 .
So which points (P, Q) of Creg × Creg are missed? Most important: we have
omitted the case where P = Q.

19.10.A. E XERCISE . Use the algebra of Extended example 15.3.4 to deal with the
case where P = Q. (This will take some time and care. Ideally you should avoid
doing any messy algebra.)
We now deal with the remaining cases where P ̸= Q. Condition (i) isn’t
important; if (P, Q) satisfies (ii) but not (i), we can swap the roles of x1 and x2 , and
(P, Q) will then satisfy (i). The only way (P, Q) can not be covered by one of these
open sets is if there is no U0 (a complement of a line defined over k) that includes P,
Q, and R.
566 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

19.10.B. E XERCISE . Use |k| > 2 to show that there is a linear form on P2 with
coefficients in k that misses P, Q, and R. (This is sadly not true if k = F2 — do you
see why?)

19.10.C. E XERCISE . Prove the last statement of Proposition 19.10.3.

19.10.D. ⋆⋆ U NIMPORTANT E XERCISE . Complete the proof by dealing with the case
k = F2 . Hint: first produce the morphism t over F4 . The goal is then to show that
this t is really “defined over” F2 (“descends to” F2 ). The morphism t is initially
described locally by considering the complement of a line defined over F4 (and
then letting the line vary). Instead, look at the map by looking at the complement of
a line and its “conjugate”. The complement of the line and its conjugate is an affine
F2 -variety. The partially-defined map t on this affine variety is a priori defined
over F4 , and is preserved by conjugation. Show that this partially defined map is
“really” defined over F2 . (If you figure out what all of this means, you will have an
important initial insight into the theory of “descent”.)
□
We can now use this to define the group variety structure on E.

19.10.4. Theorem. — Suppose (E, p) is an elliptic curve (a regular genus 1 curve over
k, with a k-valued point p). Take the Weierstrass embedding of E in P2k , via the complete
linear series |OE (3p)|. Define the k-morphism e : Spec k → E by sending Spec k to p.
Define the k-morphism i : E → E via q 7→ t(p, q), or more precisely, as the composition

(id,e)
E / E×E t / E.

Define the k-morphism m : E × E → E via (q, r) 7→ t(p, t(q, r)). Then (E, e, i, m) is a
group variety over k.
By the construction of t, all of these morphisms “commute with arbitrary base
extension”.

Proof. We need to check that various pairs of morphisms described in axioms

(i)–(iii) of §7.6.4 are equal. For example, in axiom (iii), we need to show that
m ◦ (i, id) = m ◦ (id, i); all of the axioms are clearly of this sort.
Assume first that k = k. Then each of these pairs of morphisms agree as
maps of k-points: Pic E is a group, and under the bijection between Pic E and E of
Proposition 19.9.3, the group operations translate into the maps described in the
statement of Theorem 19.10.4 by the discussion of §19.9.10.
But morphisms of k-varieties are determined by their maps on the level of
k-points (Exercise 11.3.B), so each of these pairs of morphisms are the same.
For general k, we note that from the k case, these morphisms agree after base
change to the algebraic closure. Then by Exercise 10.2.J, they must agree to begin
with. □

19.10.5. Features of this construction. The most common derivation of the proper-
ties of an elliptic curve is to describe it as a cubic, and then to describe addition
using the explicit construction with lines. Then one has to work to prove that the
multiplication described is associative.
 © 2024 Ravi Vakil. Published by Princeton University Press. 567

Instead, we started with something that was patently a group (the degree 0 line
bundles). We interpreted the maps used in the definition of the group (addition
and inverse) geometrically using our cubic interpretation of elliptic curves. This
allowed us to see that these maps were algebraic.
As a bonus, we see that in some vague sense, the Picard group of an elliptic
curve wants to be an algebraic variety.

19.10.E. E XERCISE . Suppose p and q are k-points of a genus 1 curve E. Show that
there is an automorphism of E sending p to q.

19.10.F. E XERCISE . Suppose (E, p) is an elliptic curve over an algebraically closed

field k of characteristic not 2 or 3. Show that the automorphism group of (E, p)
is isomorphic to Z/2, Z/4, or Z/6. (An automorphism of an elliptic curve (E, p)
over k = k is an automorphism of E fixing p scheme-theoretically, or equivalently,
fixing the k-valued point p by Exercise 11.3.B.) Hint: reduce to the question of
automorphisms of P1 fixing a point ∞ and a set of distinct three points {p1 , p2 , p3 } ∈
P1 \ {∞}. (The algebraic closure of k is not essential, so feel free to remove this
hypothesis, using Exercise 10.2.J. What happens if the characteristic is 3?)

19.10.G. E XERCISE . Explain why genus 1 curves over an algebraically closed field
of characteristic not 2 are classified (up to isomorphism) by j-invariant. (Caution:
note that the problem says “genus 1”, not “elliptic”.)

19.10.H. E XERCISE . Give (with proof) two elliptic curves over Q with the same
j-invariant that are not isomorphic. (Hint: §19.5.3.)

19.10.6. Vague question. What are the possible automorphism groups of a genus 1
curve over an algebraically closed k of characteristic not 2? You should be able to
convince yourself that the group has “dimension 1”.

19.10.I. I MPORTANT E XERCISE : A DEGENERATE ELLIPTIC CURVE . Suppose char k ̸=

2. Consider the arithmetic genus 1 curve C ⊂ P2k given by y2 z = x3 + x2 z, with
the point p = [0, 1, 0]. Emulate the above argument to show that C \ {[0, 0, 1]} is a
group variety (where the operation is that given by tensoring line bundles). Show
that it is isomorphic to Gm (the multiplicative group scheme Spec k[t, t−1 ], see
Exercise 7.6.E) with coordinate t = (y − x)/(y + x), by showing an isomorphism
of schemes, and showing that multiplication and inverse in both group varieties
agree under this isomorphism.

19.10.J. E XERCISE : AN EVEN MORE DEGENERATE ELLIPTIC CURVE . Consider the

arithmetic genus 1 curve C ⊂ P2k given by y2 z = x3 , with the point p = [0, 1, 0].
Emulate the above argument to show that C \ {[0, 0, 1]} is a group variety (where
the operation is that given by tensoring line bundles). Show that it is isomorphic
to A1 (with additive group structure) with coordinate t = x/y, by showing an
isomorphism of schemes, and showing that multiplication/addition and inverse in
both group varieties agree under this isomorphism.

19.10.7. ⋆ Towards Poncelet’s Porism.

These ideas lead to a beautiful classical fact, Poncelet’s Porism. (A porism is an
archaic name for a type of mathematical result.)
568 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

F IGURE 19.8. Poncelet’s Porism 19.10.8

19.10.8. Poncelet’s Porism. — Suppose C and D are two ellipses in R2 , with C

containing D. Choose any point p0 on C. Choose one of the two tangents ℓ1 from p0 to
D. Then ℓ1 meets C at two points in total: p0 and another point p1 . From p1 , there are
two tangents to D, ℓ1 and another line ℓ2 . The line ℓ2 meets C at some other point p2 .
Continue this to get a sequence of points p0 , p1 , p2 , . . . . Suppose this sequence starting
with p0 is periodic, i.e., p0 = pn for some n. Then it is periodic with any starting point
p ∈ C (see Figure 19.8).
It is possible to prove Poncelet’s Porism in an elementary manner, but a proof
involving elliptic curves is quite beautiful, and gives connections to more sophis-
ticated ideas. Rather than proving Poncelet’s Porism, we discuss some related
facts.

19.10.K. E XERCISE . Suppose E is a smooth degree 3 projective plane curve over an

algebraically closed field, q, r ∈ E, and n is a positive integer. Let F : E → E be the
morphism x 7→ t(r, t(q, x)). Suppose that Fn (p) = p for some closed point p ∈ E.
Show that Fn (x) = x for all points x ∈ E.
You should feel the urge to improve this result. (Can you show that if Fn (p) = p
for some p ∈ E, then Fn is the identity morphism on E? Can you extend this to the
case where the base field is not algebraically closed?)

19.10.L. E XERCISE . Suppose E is an ellipse in R2 . For each point (x, y) ∈ E, there is

another point G(x, y) := (x, y ′ ) (possibly the same), with the same first coordinate,
corresponding to the “other” intersection of the vertical line through (x, y) with E.
Similarly, there is another point H(x, y) := (x ′ , y), with the same second coordinate.
Let F = G ◦ H. Show that if F17 (p) = p for some point p ∈ E, then F17 (x) = x for all
points x ∈ E. (See Figure 19.9, where 17 is replaced by 3.) Hint: this might be best
done not using any fancy methods from algebraic geometry.
Again, you should wish to improve this. To what extent is this dependent on
the real numbers? What if the ellipse was instead a different conic section? (And
what does this have to do with elliptic curves? Why is the conic called E?!)
 © 2024 Ravi Vakil. Published by Princeton University Press. 569

F IGURE 19.9. Exercise 19.10.L (with 17 replaced by 3)

19.10.M. ⋆⋆ E XERCISE . Let k be a field, algebraically closed purely for convenience.

Suppose Q is a smooth quadric surface in P3k , and K is a cone (a rank 3 quadric) in
P3 , such that E = Q ∩ K is a smooth curve.
(a) Show that E has genus 1.
(b) If ℓ is a line on Q, show that E meets ℓ at two points (with multiplicity).
(c) Fix one of the rulings (family of lines) of Q. Show that there exists a morphism
G : E → E that takes a closed point p of E, and sends it to the other point of E on the
line ℓ in the ruling containing p.
(d) Define H : E → E similarly, using the other ruling of Q. Fix a positive integer n.
Let F = G ◦ H. Show that if Fn (p) = p for one closed point p ∈ E, then Fn (x) = x
for all x ∈ E.
We now connect Exercise 19.10.M to Poncelet’s Porism 19.10.8. Let v be the
cone point of K. Then v ∈ / Q, as otherwise v would be a singular point of Q ∩ K
(Do you see why?). It is a fact that projection from v gives a morphism Q → P2k
that is branched over a smooth conic D. (You may be able to make this precise,
and prove it, after reading Chapter 21.) The lines on Q project to tangent lines
to D in P2k . The projection from v contracts K \ {v} to a conic C. You may be able
to interpret the statement of Poncelet’s Porism 19.10.8 (for this C and D) in terms
of Exercise 19.10.M. (Caution: the field is not algebraically closed in Poncelet’s
Porism 19.10.8, and is algebraically closed in Exercise 19.10.M.) You may even
be able to take an arbitrary C and D as in Poncelet’s Porism 19.10.8, and reverse-
engineer K and Q as in Exercise 19.10.M, thereby proving Poncelet’s Porism (and
discovering why there is an elliptic curve hidden in its statement).

19.11 Counterexamples and pathologies using elliptic curves

We now give some fun counterexamples using our understanding of elliptic
curves. The extra juice elliptic curves give us comes from the fact that elliptic curves
are the simplest varieties with “continuous Picard groups”.

19.11.1. A scheme that is factorial, yet no affine open neighborhood of any point
is Spec of a unique factorization domain.
Suppose E is an elliptic curve over C (or some other uncountable algebraically
closed field). Consider p ∈ E. The local ring OE,p is a discrete valuation ring
570 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

and hence a unique factorization domain (Theorem 13.5.8). Then any affine open
neighborhood of E is of the form E \ q1 \ · · · \ qn for some points q1 , . . . , qn . I claim
that its Picard group is nontrivial. Recall the exact sequence:
(a1 ,...,an )7→a1 q1 +···+an qn
Z⊕n / Pic E / Pic(E \ q1 \ · · · \ qn ) /0.

But the group on the left is countable, and the group in the middle is uncountable,
so the group on the right, Pic(E\q1 \· · ·\qn ), is nonzero. We have shown that every
nonempty open subset of E has nonzero line bundles, as promised in Remark 14.2.
As E \ q1 \ · · · \ qn is affine, by Exercise 15.5.P the corresponding ring is not a
unique factorization domain. To summarize: complex elliptic curves are factorial,
but no affine open subset has a ring that has unique factorization, as promised in
§5.4.4.

19.11.A. E XERCISE . The above argument shows that over an uncountable field,
Pic E is not a finitely generated group. Show that even over the countable field Q,
Pic E is not a finitely generated group, as follows. If the elliptic curve E is generated
by q1 , . . . , qn , then there is a finite extension of fields K of Q over which all qi are
defined — the compositum of the residue fields of the qi . (Careful — do you see
what it means for the qi to be “defined” over a smaller field than Q?) Show that any
point in the subgroup of E generated by the qi must also be defined over K. Show
that E has a point not defined over K. Use this to show that Pic E is not finitely
generated. (The same argument works with Q replaced by Fp .)

19.11.2. Remark. In contrast to the above discussion, over Q, the Mordell-Weil

Theorem states that Pic E is finitely generated (see Aside 19.9.4).

19.11.3. ⋆⋆ A complex surface with infinitely many algebraic structures (for those
with sufficient complex geometric background).
As remarked in §7.3.16, a complex manifold may have many algebraic struc-
tures. The following example of M. Kim gives an example with infinitely many
algebraic structures. Suppose E is an elliptic curve over C with origin p, and let L
be a nontrivial line bundle on E − p. Then L is analytically trivial because E − p is
a Stein space, so the analytification of the total space is independent of the choice
of L . However, you know enough to show (with work) that there are infinitely
many pairwise (algebraically) nonisomorphic L , and also that their total spaces are
likewise pairwise (algebraically) nonisomorphic. This gives a complex surface with
infinitely many algebraic structures (indeed a continuum of them). See [MO68421]
for more details.

19.11.4. Counterexamples using a non-torsion point.

We next give a number of counterexamples using the existence of a non-torsion
point of a complex elliptic curve. We first show the existence of such a point.

19.11.5. We have a “multiplication by n” map [n] : E → E, which sends p to np.

If n = 0, the map [n] has degree 0 (even though [n] isn’t dominant, degree is still
defined, see Definition 12.2.2). If n = 1, the map [n] has degree 1. Given the
complex picture of a torus, you might not be surprised that the degree of [n] is n2 .
If n = 2, we have almost shown that it has degree 4, as we have checked that there
are precisely 4 points q such that 2p = 2q (Exercise 19.9.A). All that really shows
 © 2024 Ravi Vakil. Published by Princeton University Press. 571

(using Exercise 16.3.D(b)) is that the degree is at least 4. (We could check by hand
that the degree is 4 is we really wanted to.)

19.11.6. Proposition. — Suppose E is an elliptic curve over a field k of characteristic not

2. For each n > 0, the “multiplication by n” morphism [n] has positive degree, so there are
only a finite number of n-torsion points.

Proof. We may assume k = k, as the degree of a map of curves is independent of

field extension.
We prove the result by induction; it is true for n = 1 and n = 2.
If n is odd (2k + 1, say), then assume otherwise that nq = 0 for all closed
points q. Let r be a non-trivial 2-torsion point, so 2r = 0. But nr = 0 as well, so
r = (n − 2k)r = 0, contradicting r ̸= 0.
If n is even, then [n] = [2] ◦ [n/2] (degree is multiplicative under composition
of rational maps, §12.2.2), and by our inductive hypothesis both [2] and [n/2] have
positive degree. □
In particular, the total number of torsion points on E is countable. If k is an
uncountable field, then E has an uncountable number of closed points (consider an
open subset of the curve as y2 = x3 + ax + b; there are uncountably many choices
for x, and each of them has 1 or 2 choices for y). Thus we have the following.

19.11.7. Corollary. — If E is a curve over an uncountable algebraically closed field of

characteristic not 2 (e.g., C), then E has a non-torsion point.

Proof. For each n, there are only finitely many n-torsion points. Thus there are (at
most) countably many torsion points. The curve E has uncountably many closed
points. (One argument for this: take a double cover π : E → P1 . Then P1 has
uncountably many closed points, and π is surjective on closed points.) □

19.11.8. Remark. This argument clearly breaks down over countable fields. In fact,
over Fp , all points of an elliptic curve E are torsion. (Any given point is defined
over some finite field Fpr . The points defined over Fpr form a subgroup of E, using
the explicit geometric construction of the group law, and there are finite number
of points over Fpr — certainly no more than the number of Fpr -points of P2 .) But
over Q, there are elliptic curves with non-torsion points. Even better, there are
examples over Q: [2, 1, 8] is a Q-point of the elliptic curve y2 z = x3 + 4xz2 − z3 that
is not torsion. The proof would carry us too far afield, but one method is to use the
Nagell-Lutz Theorem (see for example [Si1, Cor. 7.2]).
We now use the existence of a non-torsion point to create some interesting
pathologies.

19.11.9. A map of projective varieties not arising from a map of graded rings,
even after regrading.

19.11.B. E XERCISE . Suppose E ,→ P2C is a smooth complex plane cubic (hence

genus 1), yielding a graded ring S• . Let t : E → E be translation by a non-torsion
point. Show that t does not correspond to a map of graded rings S• → S• , even
after regrading (cf. Remark 15.2.9). (If you wish, you can show the following. If
572 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

u : E → E is a translation, show that u corresponds to some map of regraded rings

Sn• → Sn• if and only if u is translation by a torsion point.)

19.11.10. An affine open subset of an affine scheme that is not a distinguished

open set.
We can use this to construct an example of an affine scheme X and an affine
open subset Y that is not distinguished in X. Let X = E − p, which is affine (see
Exercise 19.2.C, or better, note that the linear series O(3p) sends E to P2 in such a
way that the “line at infinity” meets E only at p; then E − p has a closed embedding
into the affine scheme A2 ).
Let q be another point on E so that q − p is non-torsion. Then E − p − q is
affine (Exercise 19.2.C). Assume that it is distinguished. Then there is a function f
on E − p that vanishes on q (to some positive order d). Thus f is a rational function
on E that vanishes at q to order d, and (as the total number of zeros minus poles of
f is 0) has a pole at p of order d. But then d(p − q) = 0 in Pic0 E, contradicting our
assumption that p − q is non-torsion.
In particular, E − p is an affine scheme, and q is locally cut out by one equation,
but it is not globally cut out even set-theoretically by one equation. This was promised
in Exercise 8.3.E and §9.5.1.

19.11.11. A proper (nonprojective) surface with no nontrivial line bundles.

We next use a non-torsion point p on an elliptic curve E to construct a proper
nonprojective surface with no nontrivial line bundles. Let X1 be P2 , and Z1 a
smooth cubic in X1 , identified with E. Let Z2 ⊂ X2 be exactly the same. Glue X1 to
∼
X2 along the isomorphism Z1 ←→ Z2 given, not by the identity, but by translation
by p. (This gluing construction was described in §15.2.12.) Call the result X. By
Exercise 15.2.M (and the sentence thereafter), X is proper.

19.11.C. E XERCISE . Show that every line bundle on X is trivial. Hint: Suppose
L is an invertible sheaf on X. Then L |Xi is O(di ) for some di (for i = 1, 2). The
restriction of L |X1 to Z1 must agree with the restriction of L |X2 to Z2 . Use this
to show that d1 = d2 = 0. Explain why gluing two trivial bundles (on X1 and X2
respectively) together in this way yields a trivial bundle on X.

19.11.D. E XERCISE . Show that X is not projective.

See Exercise 20.2.G for a somewhat simpler example of a proper nonprojective
surface.

19.11.12. A Picard group that has no chance of being a scheme.

We informally observed that the Picard group of an elliptic curve “wants to
be” a scheme (see §19.10.5). This is true of projective (and even proper) varieties in
general (see [FGIKNV, Ch. 9]). On the other hand, if we work over C, the affine
scheme E − p − q (in the language of §19.11.10 above) has a Picard group that can
be interpreted as C modulo a lattice modulo a non-torsion point (e.g., C/⟨1, i, π⟩).
This has no reasonable interpretation as a manifold, let alone a variety. So the fact
that the Picard group of proper varieties turns out to be a scheme should be seen as
quite remarkable.

19.11.13. A variety whose ring of functions is not finitely generated.

 © 2024 Ravi Vakil. Published by Princeton University Press. 573

We next show an example of a complex variety whose ring of global sections

is not finitely generated. (An example over Q can be constructed in the same way
using the curve of Remark 19.11.8. If you wish, you can show that these examples
are even quasiaffine.) This is related to Hilbert’s fourteenth problem (see [Mu5,
§3]).

19.11.E. P RELIMINARY E XERCISE . Suppose X is a scheme, and L is the total space of

a line bundle corresponding to invertible sheaf L , so L = Spec ⊕n≥0 (L ∨ )⊗n . (This
construction first appeared in Definition 17.1.5.) Show that if X is quasicompact,
then H0 (L, OL ) = ⊕H0 (X, (L ∨ )⊗n ). (Possible hint: choose a trivializing cover for
L . Rhetorical question: can you figure out the more general statement if L is a
rank r locally free sheaf?)
Let E be an elliptic curve over some ground field k, N a degree 0 non-
torsion invertible sheaf on E, and P a positive-degree invertible sheaf on E. Then
H0 (E, N m ⊗ P n ) is nonzero if and only if either (i) n > 0, or (ii) m = n = 0 (in
which case the sections are elements of k).

19.11.F. E ASY E XERCISE . Show that the ring R = ⊕m,n≥0 H0 (E, N m

⊗ P n ) is not
finitely generated.

19.11.G. E XERCISE . Let X be the total space of the vector bundle associated to
(N ⊕ P)∨ over E. Show that the ring of global sections of X is R, and hence is not
finitely generated. (Hint: interpret X as a line bundle over a line bundle over E.)

19.11.H. E XERCISE . Show that X (as in the above exercise) is a variety (and in
particular, Noetherian) whose ring of global sections is not Noetherian.
 CHAPTER 20

⋆ Application: A glimpse of intersection theory

This section is intended to show some of the interesting things we can do with
what we have built so far. But it can be skipped on a first reading, by young people
in a hurry.
The only reason this chapter appears after Chapter 19 is because we will use
Exercise 19.2.F in the double-starred proof of the Nakai-Moishezon criterion for
ampleness (§20.4.5).
Throughout this chapter, X will be a k-variety. In most applications, X will be
projective.

20.1 Intersecting n line bundles with an n-dimensional variety

The central tool in this chapter is the following.

20.1.1. Definition: intersection product, or intersection number (L1 · · · Ln ),

(L1 · · · Ln · F ), (L1 · · · Ln · Y).
Suppose F is a coherent sheaf on X with proper support (automatic if X is
proper) of dimension at most n, and L1 , . . . , Ln are invertible sheaves on X. Let
(L1 · L2 · · · Ln · F ) be the signed sum over the 2n subsets of {1, . . . , n}
X
(20.1.1.1) (−1)m χ(X, Li∨1
⊗ · · · ⊗ Li∨
m
⊗ F ).
{i1 ,...,im }⊂{1,...,n}

We call this the intersection of L1 , . . . , Ln with F . (Never forget that whenever

we write (L1 · · · Ln · F ), we are implicitly assuming that dim Supp F ≤ n.)
The case we will find most useful is if F is the structure sheaf of a closed
subscheme Y (of dimension at most n). In this case, we may write it (L1 ·L2 · · · Ln ·
Y), and call it the intersection of L1 , . . . , Ln with Y.
If Y = X, we may write it (L1 · L2 · · · Ln ), and call it the intersection of L1 ,
. . . , Ln (with the “X” left implicit).
If the Li are all the same, say L , one often writes (L n · F ) or (L n · Y) or
(L ). (Be careful with this confusing notation: L n does not mean L ⊗n .)
n

In some circumstances the convention is to omit the parentheses.

We will prove many things about the intersection product in this chapter. One
fact will be left until we study flatness (Exercise 24.7.D): that it is “deformation-
invariant” — that it is constant in “nice” families.

This material is published by Princeton University Press as The Rising Sea: Foundations of Algebraic
Geometry by Ravi Vakil. This version is free to view and download for personal use only. Not for
redistribution, resale, or use in derivative works. © 2024 Ravi Vakil.

575
576 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

∼ OX then (L1 · L2 · · · Ln ·
20.1.A. E XERCISE ( REALITY CHECK ). Show that if L1 =
F ) = 0.

20.1.B. E XERCISE . Show that the intersection product (20.1.1.1) is preserved by

field extension of k.
The following exercise suggests that the intersection product might be interest-
ing, as it “interpolates” between two useful notions: the degree of a line bundle on
a curve, and Bézout’s Theorem.

20.1.C. E XERCISE .
(a) If X is a projective curve, and L is an invertible sheaf on X, show that (L · X) =
degX L .
(b) Suppose k is an infinite field, X = PN , and Y is a dimension n subvariety of X. If
H1 , . . . , Hn are generally chosen hypersurfaces of degrees d1 , . . . , dn respectively
(so dim(H1 ∩ · · · ∩ Hn ∩ Y) = 0 by Exercise 12.3.D(d)), then by an appropriate
version of Bézout’s Theorem (see perhaps Exercise 13.4.C or Exercise 18.6.J),
deg(H1 ∩ · · · ∩ Hn ∩ Y) = d1 · · · dn deg(Y).
Show that
(OX (H1 ) · · · OX (Hn ) · Y) = d1 · · · dn deg(Y).

We now describe some of the properties of the intersection product. In the

course of proving Exercise 20.1.C(b) you may in effect solve the following exercise.

20.1.D. E XERCISE . Suppose D is an effective Cartier divisor on X that restricts to

an effective Cartier divisor D|Y = D ∩ Y on Y (i.e., remains locally not a zerodivisor
on Y). Show that
(L1 · · · Ln−1 · O(D) · Y) = (L1 · · · Ln−1 · D|Y ).
More generally, if D is an effective Cartier divisor on X that does not contain any
associated point of F , show that
(L1 · · · Ln−1 · O(D) · F ) = (L1 · · · Ln−1 · F |D ).
(A similar idea came up in the proof that the Hilbert polynomial is actually polyno-
mial; see the discussion around (18.6.3.1).)

20.1.2. Definition. For this reason, if D is an effective Cartier divisor, in the symbol
for the intersection product, we often write D instead of O(D). We interchangeably
think of intersecting effective Cartier divisors rather than line bundles. For example,
we will discuss the special case of intersection theory on a surface in §20.2, and
when we intersect two curves C and D, we will write the intersection as (C · D) or
even C · D.

20.1.3. Proposition. — Assume X is projective. For fixed F , the intersection product

(L1 · · · Ln · F ) is a symmetric multilinear function of the L1 , . . . , Ln .
Proposition 20.1.3 is actually true with “projective” replaced by “proper”, see
[Kl1, Prop. 2] or [Ko1, Prop. VI.2.7]. Unlike most extensions to the proper case,
this is not just an application of Chow’s Lemma. It involves a different approach,
involving a beautiful trick called dévissage.
 © 2024 Ravi Vakil. Published by Princeton University Press. 577

20.1.4. Important remark: additive notation for line bundles. There is a standard,
useful, but confusing convention suggested by the multilinearity of the intersection
product: we sometimes write tensor product of invertible sheaves additively. This is
convenient for working with multilinearity. (Some people try to avoid confusion
by using divisors rather than line bundles, as we add divisors when we “multiply”
the corresponding line bundles. This is psychologically helpful, but may add
more confusion, as one then has to worry about the whether and why and how
and when line bundles correspond to divisors.) We will use this, for example, in
Exercises 20.2.B–20.2.G, §20.2.11, and §20.4.

20.1.5. Proof of Proposition 20.1.3. Symmetry is clear. By Exercise 20.1.B, we may

assume that k is infinite (e.g., algebraically closed). We now prove the result by
induction on n.

20.1.E. E XERCISE ( BASE CASE ). Prove the result when n = 1. Hint: Exercise 18.4.M.

We now assume the result for smaller values of n. We use a trick. We wish to
show that (for arbitrary L1 , L1′ , L2 , . . . , Ln ),

(20.1.5.1) (L1 · L2 · · · Ln · F ) + (L1′ · L2 · · · Ln · F ) − ((L1 ⊗ L1′ ) · L2 · · · Ln · F )

is 0.

20.1.F. E XERCISE . Rewrite (20.1.5.1) as

(20.1.5.2) (L1 · L1′ · L2 · · · Ln · F ).

(There are now n + 1 line bundles appearing in the product, but this does not
contradict the definition of the intersection product, as dim Supp F ≤ n < n + 1.)

20.1.G. E XERCISE . Use the inductive hypothesis to show that (20.1.5.1) is 0 if

Ln =∼ O(D) for D an effective Cartier divisor missing the associated points of F .

In particular, if Ln is very ample, then (20.1.5.1) is 0, as Exercise 18.6.A shows

that there exists a section of Ln nonvanishing at the associated points of F . (Cau-
tion: Exercise 18.6.A requres the field to be infinite. How would you edit this to
work if k were finite? Hint: show a variant of Exercise 18.6.A, that there exists a
section of Ln⊗N nonvanishing at the associated points of F for some N ≫ 0. Hint
for this: Compare Exercise 12.3.D(c) and Exercise 12.3.D(d).)
By the symmetry of its incarnation as (20.1.5.2), expression (20.1.5.1) vanishes
if L1 is very ample. We will use this observation repeatedly in the final equation
display of the proof.
By Exercise 16.2.F, any invertible sheaf on X may be written “as the difference
of two very amples”, so write L1 = A ⊗ B ∨ and L1′ = A ′ ⊗ (B ′ )∨ , where A , B,
578 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

A ′ , and B ′ are very ample. Then

((L1 ⊗ L1′ ) · L2 · · · Ln · F ) = ((A ⊗ A ′ ⊗ (B ⊗ B ′ )∨ ) · L2 · · · Ln · F )

= ((A ⊗ A ′ ) · L2 · · · Ln · F ) − ((B ⊗ B ′ ) · L2 · · · Ln · F )
= (A · L2 · · · Ln · F ) + (A ′ · L2 · · · Ln · F )
− (B · L2 · · · Ln · F ) − (B ′ · L2 · · · Ln · F )
= ((A ⊗ B ∨ ) · L2 · · · Ln · F ) + ((A ′ ⊗ (B ′ )∨ ) · L2 · · · Ln · F )
= (L1 · L2 · · · Ln · F ) + (L1′ · L2 · · · Ln · F )

as desired.
□
We have an added bonus arising from the proof.

20.1.H. E XERCISE . Suppose X is projective. Show that if dim Supp F < n + 1, and
L1 , L1′ , L2 , . . . , Ln are invertible sheaves on X, then (20.1.5.2) vanishes. In other
words, the intersection product of n + 1 invertible sheaves with a coherent sheaf
F vanishes if the dim Supp F < n + 1. (In fact, the result holds with “projective”
replaced by “proper”, as the results it relies on hold in this greater generality.)

20.1.6. Proposition. — Suppose X is projective. The intersection product depends only

on the numerical equivalence classes of the Li .
(Numerical equivalence was defined in §18.4.12 — two line bundles are linearly
equivalent if their degrees are the same on any curve.) Again, the result remains true
with “projective” replaced by “proper”. But we use the fact that every line bundle is
the difference of two very ample line bundles in both the proof of Proposition 20.1.3
and in the proof of Proposition 20.1.6 itself. For the proof of the Proposition in the
proper case, see [FGIKNV, Prop. B.20].

Proof. Suppose L1 is numerically equivalent to L1′ , and L2 , . . . , Ln , and F are

arbitrary. We wish to show that (L1 · L2 · · · Ln · F ) = (L1′ · L2 · · · Ln · F ). We
proceed by induction on n. The case n = 1 follows from Exercise 18.4.L. We
assume that n > 1, and assume the result for “smaller n”. By multilinearity of the
intersection product, and the fact that each Ln may be written as the “difference”
of two very ample invertible sheaves (Exercise 16.2.F), it suffices to prove the result
in the case when Ln is very ample. We may then write Ln = O(D), where D is an
effective Cartier divisor missing the associated points of F (Exercise 18.6.A). Then

(L1 · L2 · · · Ln · F ) = (L1 · L2 · · · Ln−1 · F |D ) (Exercise 20.1.D)

= (L1′ · L2 · · · Ln−1 · F |D ) (inductive hypothesis)
= (L1′ · L2 · · · Ln · F ) (Exercise 20.1.D).

20.1.7. Asymptotic Riemann-Roch.

If Y is a projective (=proper, see Remark 19.2.I) curve, χ(Y, L ⊗m ) = m degY L +
χ(Y, OY ) is a linear polynomial in m (see Exercises 18.4.L and 18.4.M), whose leading
term is an intersection product. This generalizes as follows.
 © 2024 Ravi Vakil. Published by Princeton University Press. 579

20.1.I. E XERCISE (A SYMPTOTIC R IEMANN -R OCH ). Suppose X is projective. (As

usual, the result will remain true with “projective” replaced by “proper”.) Suppose
F is a coherent sheaf on X with dim Supp F ≤ n, and L is a line bundle on X.
Show that q(m) := χ(X, L ⊗m ⊗ F ) is a polynomial in m of degree at most n.
Show that the coefficient of mn in q(m) (the “leading term”) is (L n · F )/n!. Hint:
Exercise 20.1.H implies that

| · L{z· · · L} ·(L ⊗ F )) = 0,
⊗i
(L
n + 1 times

which in turn (using (20.1.1.1)) implies that

X
n+1 
n+1

j
(−1) q(m + j) = 0
j
j=0

for all m. Your argument may resemble the proof of polynomiality of the Hilbert
polynomial, Theorem 18.6.1, so you may find further hints there. Exercise 18.6.B in
particular might help.
We know all the coefficients of this polynomial if X is a curve, by Riemann-
Roch (see (18.4.7.1)), or basically by definition. We will know/interpret all the
coefficients if X is a regular projective surface and F is an invertible sheaf when we
prove Riemann-Roch for surfaces (Exercise 20.2.B(b)). To understand the general
case, we need the theory of Chern classes. The result is the Hirzebruch-Riemann-
Roch Theorem, which can be further generalized to the celebrated Grothendieck-
Riemann-Roch Theorem (see [F2, §15.2]).

20.1.J. E XERCISE ( THE PROJECTION FORMULA ). Suppose π : X1 → X2 is a (projec-

tive) morphism of integral projective varieties of the same dimension n, and L1 ,
. . . , Ln are invertible sheaves on X2 . Show that
(π∗ L1 · · · π∗ Ln ) = deg(X1 /X2 )(L1 · · · Ln ).
(The first intersection is on X1 , and the second is on X2 .) Hint: Let d = deg π, and
assume d > 0. (Deal with the case where π is not dominant separately, so d = 0
by convention, using Chevalley’s Theorem 8.4.2.) Argue that by the multilinearity
of the intersection product, it suffices to deal with the case where the Li are very
ample. Then choose sections of each Li , all of whose intersections lie in an open
subset U where π has “genuine degree d”. To find U: first use Exercise 10.3.G to find
a dense open subset U ′ ⊂ X2 over which π is finite. Then use Useful Exercise 14.3.E
to show that there exists a dense open subset U ⊂ U ′ on which π∗ O is a locally free
sheaf of rank d. In the “flatness” language of Chapter 24, you are showing that
there is a dense open subset U of X2 over which π is finite and flat (and hence has
“constant degree”, see Remark 24.3.8). (As usual, the result holds with “projective”
replaced with “proper”; see [Ko1, Prop. VI.2.11].)

20.1.8. Remark: A more general projection formula. Suppose π : X1 → X2 is a proper

morphism of proper varieties, and F is a coherent sheaf on X1 with dim Supp F ≤
n (so dim Supp π∗ F ≤ n, using Exercise 12.2.C). Suppose also that L1 , . . . , Ln are
invertible sheaves on X2 . Then
(π∗ L1 · · · π∗ Ln · F ) = (L1 · · · Ln · π∗ F ).
580 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

This is also called the projection formula (and generalizes, in a nonobvious way,
Exercise 20.1.J). Proofs are given in [FGIKNV, B.15] and [Ko1, Prop. VI.2.11].
Because we won’t use this version of the projection formula, we do not give a proof
here.

20.1.K. E XERCISE ( INTERSECTING WITH AMPLE LINE BUNDLES ). Suppose X is

a projective k-variety, and L is an ample line bundle on X. Show that for any
subvariety Y of X of dimension n, (L n · Y) > 0. (Hint: use Proposition 20.1.3
and Theorem 16.2.2 to reduce to the case where L is very ample. Then show that
(L n · Y) = deg Y in the embedding into projective space induced by the linear
series |L |.)
The Nakai-Moishezon criterion (Theorem 20.4.1) states that this characterizes
ampleness.

20.1.9. ⋆⋆ Cohomological interpretation in the complex projective case, generalizing Ex-

ercise 18.4.H and §18.6.5. If k = C, we can interpret (L1 · · · Ln · Y) as the degree
of

(20.1.9.1) c1 ((L1 )an ) ∪ · · · ∪ c1 ((Ln )an ) ∩ [Yan ]

in H0 (Yan , Z). (Recall c1 ((Li )an ) ∈ H2 (Xan , Z), as discussed in Exercise 18.4.H.) One
way of proving this is to use multilinearity of both the intersection product and
(20.1.9.1) to reduce to the case where the Ln is very ample, so Ln = ∼ O(D), where
D restricts to an effective Cartier divisor E on Y. Then show that if L is an analytic
line bundle on Yan with nonzero section Ean , then c1 (L ) ∩ [Yan ] = [Ean ]. Finally, use
induction on n and Exercise 20.1.D.

20.2 Intersection theory on a surface

We now apply the general machinery of §20.1 to the case of a regular projective
surface X. (What matters is that X is Noetherian and factorial, so Pic X → Cl X is an
isomorphism, Proposition 15.4.15. Recall that regular schemes are factorial by the
Auslander-Buchsbaum Theorem 13.8.5.)

20.2.A. E XERCISE /D EFINITION . Suppose C and D are effective divisors (i.e.,

curves) on X. (Does this mean “Weil divisor” or “Cartier divisor”?)
(a) Show that

(20.2.0.1) degC OX (D)|C

(20.2.0.2) = (O(C) · O(D) · X)
(20.2.0.3) = degD OX (C)|D .

We call this the intersection number of C and D, and denote it C · D.

(b) If C does not contain any associated point of D (so in particular, C and D have
no components in common), show that

(20.2.0.4) C · D = h0 (C ∩ D, OC∩D )
 © 2024 Ravi Vakil. Published by Princeton University Press. 581

where C ∩ D is the scheme-theoretic intersection of C and D on X. (When speaking

of associated points of D, we by definition must be considering D as an effective
Cartier divisor.)

20.2.1. Important aside. The hypothesis in Exercise 20.2.A(b), that C not contain any
associated point of D, is a red herring. In fact, D can never have any embedded
points, as we will see in §26.2.5 when discussing Cohen-Macaulay rings. (This
should help motivate you to learn about Cohen-Macaulayness.) The case of A2
(and hence P2 ) can be done by hand (Exercise 6.6.F).
Thus the hypothesis in Exercise 20.2.A(b) can be replaced by the more simple
“C and D have no common components”.

20.2.2. Remark. In particular, if C and D have no irreducible component in common,

then C·D ≥ 0. This means that we cannot hope to understand intersection numbers
using moving lemmas as in topology — if we have some irreducible curve E with
E2 < 0, we cannot “move” E to something (call it F) “equivalent” to E, to compute
E · E as E · F.

20.2.3. Remark. The degree of the normal bundle to C in X (defined in §15.6.2) is

the self-intersection C · C of C, as both are degC OX (C)|C .

20.2.4. Advantages and disadvantages. We thus have four descriptions of the

intersection number (20.2.0.1)–(20.2.0.4), each with advantages and disadvantages.
The Euler characteristic description (20.2.0.2) is remarkably useful (for example, in
the exercises below), but the geometry is obscured. The definition degC OX (D)|C ,
(20.2.0.1), is not obviously symmetric in C and D. The definition h0 (C ∩ D, OC∩D ),
(20.2.0.4), is clearly local — to each point of C ∩ D, we have a vector space. For
example, we know that in A2k , y − x2 = 0 meets the x-axis with multiplicity 2,
because h0 of the scheme-theoretic intersection (k[x, y]/(y − x2 , y)) has dimension
2. (This h0 is also the length of the dimension 0 scheme, by Exercise 6.6.Y.)
By Proposition 20.1.3, the intersection number induces a bilinear “intersection
form”
(20.2.4.1) Pic X × Pic X / Z.

By Asymptotic Riemann-Roch (Exercise 20.1.I), χ(X, O(nD)) is a quadratic polyno-

mial in n.

20.2.5. You can verify that Exercise 20.2.A recovers Bézout’s Theorem for plane
curves (see Exercise 18.6.J), using χ(P2 , O(n)) = (n + 2)(n + 1)/2 (from Theo-
rem 18.1.2), and the fact that effective Cartier divisors on P2k have no embedded
points (Exercise 6.6.F).
Before getting to a number of interesting explicit examples, we derive a couple
of fundamental theoretical facts.

20.2.B. E XERCISE . Assuming Serre duality (Theorem 18.5.1) for a smooth projective
surface X, prove the following. (We are mixing divisor and invertible sheaf notation,
as described in Remark 20.1.4, so be careful. Here KX is a divisor corresponding to
ωX .)
(a) (sometimes called the adjunction formula) C · (KX + C) = 2pa (C) − 2 for any curve
C ⊂ X. Hint: compute C · (−KX − C) instead. (See Exercise 21.5.B and §29.4 for
582 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

other versions of the adjunction formula.)

(b) (Riemann-Roch for surfaces)

χ(X, OX (D)) = D · (D − KX )/2 + χ(X, OX )

for any Weil divisor D (cf. Riemann-Roch for curves, Exercise 18.4.B).

20.2.6. Two explicit examples: P1 × P1 and Blp P2 .

20.2.C. E XERCISE : X = P1 × P1 . Recall from Exercise 15.5.K that Pic(P1 × P1 ) =

Zℓ × Zm, where ℓ is the curve P1 × {0} and m is the curve {0} × P1 . Show that
the intersection form (20.2.4.1) is given by ℓ · ℓ = m · m = 0, ℓ · m = 1. (Hint:
You can compute the cohomology groups of line bundles on P1 × P1 using Ex-
ercise 20.2.A(b).) What is the class of the diagonal in P1 × P1 in terms of these
generators?

20.2.D. E XERCISE : THE BLOWN UP PROJECTIVE PLANE . (You needn’t have read
Chapter 22 to do this exercise.) Let X = Blp P2 be the blow-up of P2k at a k-valued
point (the origin, say) p — see Exercise 10.3.F, which describes the blow-up of A2k ,
and “compactify”. Interpret Pic X as generated (as an abelian group) by ℓ and e,
where ℓ is a line not passing through the origin, and e is the exceptional divisor.
Show that the intersection form (20.2.4.1) is given by ℓ · ℓ = 1, e · e = −1, and ℓ · e = 0.
Hence show that Pic X = ∼ Zℓ × Ze (as promised in the aside in Exercise 15.5.L). In
particular, the exceptional divisor has negative self-intersection. (This exercise will
be generalized in §22.4.13.)

20.2.7. Hint. Here is a possible hint to get the intersection form in Exercise 20.2.D.
The scheme-theoretic preimage in Blp P2 of a line through the origin is the scheme-
theoretic union of the exceptional divisor e and the “proper transform” m of the
line through the origin. Show that ℓ = e + m in Pic(Blp P2 ) (writing the Picard
group operation additively, cf. Remark 20.1.4). Show that ℓ · m = e · m = 1 and
m · m = 0.

20.2.8. Definition: (−1)-curve. Notice that the exceptional divisor e has self-
intersection −1. We will see more generally in Exercise 22.4.O that this is the case
for all exceptional divisors of blow-ups of smooth surfaces (over k) blown up at a
k-valued point. We give such curves a name. If X is a surface over k, and C ⊂ X is
∼ P1 , and C · C = −1, we say that
a curve in X consisting of smooth points, with C = k
C is a (−1)-curve. (We can restate “C · C = −1” as “C has normal bundle O(−1)”.)

20.2.E. E XERCISE . Show that the blown up projective plane Blp P2 in Exercise 20.2.D
is not isomorphic to P1 × P1 , perhaps considering their (isomorphic) Picard groups,
and identifying which classes are effective (represented by effective divisors). (This
is an example of a pair of smooth projective birational surfaces that have isomorphic
Picard groups, but which are not isomorphic. This exercise shows that F0 is not
isomorphic to F1 . The method and result are generalized in Exercise 20.2.Q.)
A simpler approach but without the same generalizations: show that the self-
intersections of all the elements of the Picard group give Z in one case and 2Z in
the other.
 © 2024 Ravi Vakil. Published by Princeton University Press. 583

20.2.F. E XERCISE ( CF. E XERCISE 18.4.X). Show that the nef cone (Exercise 18.4.W)
of Blp P2 is generated by ℓ and m. Hint: show that ℓ and m are nef. By intersecting
line bundles with the curves e and m, show that nothing outside the cone spanned
by ℓ and m is nef. (Side Remark: note that as in Exercise 18.4.X, linear series
corresponding to the boundaries of the cone give “interesting contractions”.)

20.2.G. E XERCISE : A NOTHER PROPER NONPROJECTIVE SURFACE . Show the

existence of a proper nonprojective surface over a field as follows, paralleling the
construction of a proper nonprojective threefold in §15.2.11. Take two copies of the
blown up projective plane Blp P2 , gluing ℓ on the first to e on the second, and e on
the first to ℓ on the second. Hint: show that if L is a line bundle having positive
degree on each effective curve, then L · ℓ > L · e, using ℓ = e + m from Hint 20.2.7.
(See §19.11.11 for another example of a proper nonprojective surface.)

20.2.9. Fibrations.
Suppose π : X → B is a morphism from a regular projective surface to a regular
curve, not contracting any irreducible components of X, and b ∈ B is a closed point.
Let F = π−1 (b). Then OX (F) = π∗ OB (b), which is isomorphic to O on F. Thus
F · F = degF OX (F) = 0: “the self-intersection of a fiber is 0”. The same argument
works without X being regular, so long as you phrase it properly: (π∗ OX (b))2 = 0.

20.2.H. E XERCISE . Suppose E is an elliptic curve, with origin p. On E × E, let

∆ be the diagonal. By considering the “difference” map E × E → E, for which
π−1 (p) = ∆, show that ∆2 = 0. Show that N1Q (E × E) (defined in §18.4.12) has rank
at least 3. Show that in general for schemes X and Y, Pic X × Pic Y → Pic(X × Y)
(defined by ⊠, i.e., pulling back and tensoring) need not be isomorphism; the case
of X = Y = P1 is misleading.
Remark: dimQ N1Q (E × E) is always 3 or 4. It is 4 if there is a nontrivial endomor-
phism from E to itself (i.e., not just multiplication [n] by some n, §19.11.5, followed
by a translation); the additional class comes from the graph of this endomorphism.
(See [Mu3, §21, App. III] for an introduction to the tools needed to show this.)
Our next goal is to describe the self-intersection of a curve on a ruled surface
(Exercise 20.2.J). To set this up, we have a useful preliminary result.

20.2.I. E XERCISE ( THE NORMAL BUNDLE TO A SECTION OF THE PROJECTIVIZATION

OF A RANK 2 VECTOR BUNDLE ). Suppose X is a scheme, and V is a rank 2 locally
free sheaf on X. By Exercises 17.2.I and 14.2.H(a), the short exact sequences
(20.2.9.1) 0 /S /V /Q /0

on X, where S and Q have rank 1, correspond to the sections σ : X → PV to the

projection PV → X. (See §17.2.3 for the definition of PV , including a caution
about duals.) Show that the normal bundle (§15.6.2) to σ(X) in PV is Q ⊗ S ∨ .
(A generalization is stated in §21.3.10.) Hint: (i) For simplicity, it is convenient
to assume S = OX , by replacing V by V ⊗ S ∨ , as the statement of the problem
respects tensoring by an invertible sheaf (see Exercise 17.2.G). (ii) Assume now
∼ OX , and the exact sequence splits. Then describe
(with loss of generality) that Q =
the section as σ : X → P × X, with X mapping to the 0 section. Describe an
1

isomorphism of OX with the normal bundle to σ(X) → P1 × X. (Do not just say
that the normal bundle “is trivial”.) (iii) Now consider the case where Q is general.
584 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Choose trivializing open neighborhoods Ui of Q, and let gij be the transition

function for Q. On the overlap between two trivializing open neighborhoods
Ui ∩ Uj , determine how your two isomorphisms of OX with Nσ(X)/P1X from (ii) (one
for Ui , one for Uj ) are related. In particular, show that they differ by gij .

20.2.J. E XERCISE ( SELF - INTERSECTIONS OF SECTIONS OF RULED SURFACES ). Sup-

pose C is a regular curve, and V is a rank 2 locally free sheaf on C. Then PV is a
ruled surface (Example 17.2.6). Fix a section σ of PV corresponding to a filtration
(20.2.9.1). Show that σ(C) · σ(C) = degC (Q ⊗ S ∨ ).

20.2.10. The Hirzebruch surfaces Fn = P(OP1 ⊕ OP1 (n)).

Recall the definition of the Hirzebruch surface Fn = P(OP1 ⊕ OP1 (n)) in Ex-
ample 17.2.6. It is a P1 -bundle over P1 ; let π : Fn → P1 be the structure morphism.
Using Exercise 20.2.J, corresponding to

0 / O(n) / O ⊕ O(n) /O / 0,

we have a section of π of self-intersection −n; call it E ⊂ Fn . Similarly, correspond-

ing to
0 /O / O ⊕ O(n) / O(n) / 0,

we have a section C ⊂ Fn of self-intersection n. Let p be any k-valued point of P1 ,

and let F = π−1 (p).

20.2.K. E XERCISE . Show that the line bundle O(F) is independent of the choice of
p.

20.2.L. E XERCISE . Show that Pic Fn is generated by E and F. In the course of doing
this, you might develop “local charts” for Fn , which could help you solve later
exercises.

20.2.M. E XERCISE . Compute the intersection matrix on Pic Fn . Show that E and F
∼ ZE ⊕ ZF. Calculate C in terms of E and F.
are independent, and thus Pic Fn =

20.2.N. E XERCISE . Show how to identify Fn \ E, along with the structure map π,
with the total space of the line bundle O(n) on P1 , with C as the 0-section. Similarly
show how to identify Fn \ C with the total space of the line bundle O(−n) on P1 ;
with E as the 0-section.

20.2.O. E XERCISE . Show that h0 (Fn , OFn (C)) > 1. Hint: As OFn (C) has a section —
namely the “canonical one” whose divisor is C. Use Exercise 20.2.M to find another
section.

20.2.P. E XERCISE . Show that every effective curve on Fn is a non-negative linear

combination of E and F. (Conversely, it is clear that for every nonnegative a and
b, O(aE + bF) has a section, corresponding to the effective curve “aE + bF”. The
extension of this to N1Q (which was defined in §18.4.12) is called the effective cone,
and this notion, extended to proper varieties more generally, can be very useful.
This exercise shows that E and F generate the effective cone of Fn .) Hint: show
that because “F moves”, any effective curve must intersect F nonnegatively, and
similarly because “C moves” (Exercise 20.2.O), any effective curve must intersect C
 © 2024 Ravi Vakil. Published by Princeton University Press. 585

nonnegatively. If O(aE + bF) has a section corresponding to an effective curve D,

what does this say about a and b?

20.2.Q. E XERCISE . By comparing effective cones, and the intersection pairing,

show that the Fn are pairwise nonisomorphic. (This result was promised in Exam-
ple 17.2.6. Exercise 20.2.E is a special case.)
Exercise 20.2.Q is difficult to do otherwise, and foreshadows the fact that nef
and effective cones are useful tools in classifying and understanding varieties in
general. In particular, they are central to the minimal model program.

20.2.R. E XERCISE . If n = 0, show that there are no curves on Fn of negative self-

intersection. If n ≥ 0, show that E is the unique curve on Fn with self-intersection
−n, and there are no reduced curves on Fn of smaller self-intersection. This again
gives another (related) means of showing that the Fn are pairwise nonisomorphic.

20.2.S. E XERCISE . Show that the nef cone of Fn is generated by C and F. (We will
soon see that by Kleiman’s criterion for ampleness, Theorem 20.4.7, that the ample
cone is the interior of this cone, so we have now identified the ample line bundles
on Fn .)

20.2.T. E XERCISE . We have seen earlier (Exercises 20.2.F and 18.4.X) that the
boundary of the nef cone give “interesting contractions”. Show that the map given
by the linear series corresponding to O(F) is the projection of Fn onto P1 , and the
map given by the linear series corresponding to O(C) sends Fn to the projective
cone over the degree n rational normal curve.
After this series of exercises, you may wish to revisit Exercises 20.2.C-20.2.F,
∼ P1 × P1 and F1 =
and interpret them as special cases: F0 = ∼ Blp P2 .

20.2.11. The Hodge Index Theorem.

We use what we have learned to prove a celebrated result, the Hodge Index
Theorem. The “index” in the name comes from the following notion.

20.2.12. Index of a real quadratic form. By arguments similar to the classification of

quadratic forms over algebraically closed fields (Exercise 5.4.J), one can show that
given a symmetric bilinear form on a real vector space V of (finite) dimension n,
there are integers a+ , a0 , a− such that then given any orthogonal basis v1 , . . . , vn ,
the number of vi · vi that are positive (resp., zero, negative) is a+ (resp., a0 , a− ).
The ordered pair (a+ , a− ) is often called the index, and is sometimes called the
signature. (This result is sometimes called Sylvester’s law of inertia. You can prove it
yourself as follows. Given any basis (v1 , . . . , vn ) for V, we can construct the n × n
intersection matrix (vi · vj )1≤i,j≤n . It is symmetric, so by the spectral theorem, it
has real eigenvalues. If you change basis, how do the eigenvalues change? Why
are the number of positive eigenvalues preserved?)

20.2.13. The Hodge Index Theorem. — Suppose X is an irreducible smooth projective

surface (over a field k) with L , H ∈ Pic X, with H · H > 0 and L · H = 0. Then (a)
L · L ≤ 0, and (b) equality holds if and only if L is numerically trivial.

20.2.U. M OTIVATING E XERCISE . If ρ(X) is finite, show that the Hodge Index Theo-
rem is equivalent to the statement that N1R (X) := N1Q (X) ⊗Q R, with its “intersection
586 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

form”, has index (1, ρ(X) − 1). (We defined N1 (X) and ρ(X) in §18.4.12.) Don’t
forget that X is projective, and thus there is an ample line bundle on X! You should
think of the positive “term” as coming from the ample class; the “rest” is negative.
We won’t use the fact that ρ(X) is finite, because we haven’t proved it, but it will
motivate us.

20.2.14. Proof of The Hodge Index Theorem 20.2.13. Motivated by Exercise 20.2.U, we
first prove the result under the additional assumption that H is ample.

20.2.15. We begin with the following fact. If H is an ample invertible sheaf, M

is any invertible sheaf, and M · H > ωX · H , then h2 (X, M ) = 0. (Translation:
for all invertible sheaves “more positive than ωX ”, there is no top cohomology;
compare this to §19.2.5, the corresponding statement in dimension 1.) Reason: by
Serre duality (Theorem 18.5.1), we wish to show that h0 (X, ωX ⊗ M ∨ ) = 0. But
if otherwise ωX ⊗ M ∨ had a nonzero section, then its (effective nonzero) divisor
would intersect H nonnegatively (Exercise 20.1.K), so (using additive notation, see
Remark 20.1.4) (ωX − M ) · H ≥ 0, contradicting M · H > ωX · H .

20.2.16. We next show that if H is ample, and L is any invertible sheaf with
L · L > 0 and L · H > 0, then for n ≫ 0, L ⊗n has a section. Here is why. For
n ≫ 0,
L ⊗n · H = nL · H > ωX · H ,
so h2 (X, L ⊗n ) = 0 by §20.2.15 (taking M = L ⊗n ). Then for n ≫ 0, by Riemann-
Roch for surfaces (Exercise 20.2.B(b)),
h0 (X, L ⊗n ) − h1 (X, L ⊗n ) = χ(X, L ⊗n )
= (n2 /2)L · L − (n/2)L · ωX + χ(X, OX ).
As L · L > 0, the right side is positive for n ≫ 0, so
h0 (X, L ⊗n ) ≥ h0 (X, L ⊗n ) − h1 (X, L ⊗n ) > 0
as desired.

20.2.17. We are now ready to prove the Hodge Index Theorem for ample H .
Assume first that L · L > 0 (and L · H = 0). Then for n ≫ 0, H ′ := L ⊗ H ⊗n
is ample (indeed very ample) by Exercise 16.2.E. Then L · H ′ = L · L > 0, so by
§20.2.16, L ⊗n is effective, which implies that L · H = (L ⊗n · H )/n > 0 (using
Exercise 20.1.K again), contradicting our hypothesis.

20.2.18. Assume finally that L · H = 0, and (i) L · L = 0, but that (ii) L is not
numerically trivial. By (ii), we can find an invertible sheaf Q such that Q · L ̸= 0.
Then we can find an invertible sheaf R such that R · L ̸= 0 and R · H = 0: take
R = (H · H )Q − (Q · H )H .
Then take L ′ = L ⊗n ⊗ R, so L ′ · H = 0, but because R · L ̸= 0, we can find
some n ∈ Z so that
L ′ · L ′ = 2nL · R + R · R > 0.
Then the argument of §20.2.17 applies, with L ′ in the place of L . We have thus
proved “the Hodge Index Theorem for ample line bundles”.
 © 2024 Ravi Vakil. Published by Princeton University Press. 587

20.2.19. It remains to prove the Hodge Index Theorem without the assumption that
H is ample. So now assume we have H and L as in the statement of the theorem.
Let A ∈ Pic X be an ample line bundle on X. (If you would like to complete the
proof from here on your own, the following hint should suffice: try to apply the
Hodge Index Theorem for ample line bundles three times, where the ample line
bundle is A of course, and the “other” line bundle is sequentially L , H , and
(20.2.19.1).)
If A · L = 0, then we are done without even using H , by the Hodge Index
Theorem for ample line bundles (using A and L ). So now assume that A · L ̸= 0.
Similarly, A · H ̸= 0 (as otherwise appplying the Hodge Index Theorem for ample
line bundles using A and H would give H · H ≤ 0, contradicting the hypothesis
H · H > 0). Take
(20.2.19.1) M = (A · L )H − (A · H )L .
Note that A · M = 0. By the Hodge Index Theorem for ample line bundles (applied
this time to A and M ) we have that M · M ≤ 0, i.e.,
(A · L )2 (H · H ) + (A · H )2 (L · L ) ≤ 0.
But (A · L )2 , H · H , and (A · H )2 are all positive, so L · L < 0 as desired. □

20.2.20. Generalizations and variations. The hypotheses can be weakened consider-

ably. We used smoothness only because we need Serre duality, with an invertible
sheaf ωX . We will see in §29.4 that we need less than smoothness for this. But we
can do much better, and a proof where X is merely required to be geometrically
irreducible and proper requires only a minor modification of our argument, see
[FGIKNV, Thm. B.27].

20.3 The Grothendieck group of coherent sheaves, and an algebraic

version of homology
The construction of the intersection product (20.1.1.1) may leave you hungry for
something more, especially in light of the cohomological interpretation of §20.1.9.
You may want some sort of homology-like theory which is a repository for cycles
of different dimensions, on which (Chern classes of) line bundles can act. We can
actually do this easily, given what we know.

20.3.1. Definition. If X is a k-variety, we define the Grothendieck group of coherent

sheaves, which we denote K(X) (and which is often denoted K0 (X)), as the abelian
group generated symbols of the form [F ] where F is a coherent sheaf on X, subject
to the relations that [F ] = [F ′ ] if F = ∼ F ′ , and [F ′ ] + [F ′′ ] = [F ] if there is a
short exact sequence
0 / F′ /F / F ′′ / 0.
(Caution: the notation K(X) is now used for two concepts, the Grothendieck group
and the function field. The meaning of the symbol should be clear from the context.)
By construction, the Grothendieck group is the universal construction of an operator
on the category of coherent sheaves that “behaves well in exact sequences”. For
example, if X is proper, then:
588 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

(i) We have a map χ : CohX → Z, as the Euler characteristic of a coherent sheaf

is finite by Theorem 18.1.3(i) in the projective case, and more generally
by Grothendieck’s Coherence Theorem 18.9.1 in the proper case. This
map descends to χ : K(X) → Z by Exercise 18.4.A (which extends without
change to the proper case).
(ii) If X is integral, then the rank function

rank : CohX /Z

descends to a function rank : K(X) → Z. (The argument of Exercise 14.3.I

applies.)

20.3.2. Definition and nonstandard notation. The Grothendieck group is filtered by

dimension: let K(X)≤d be the subgroup of K(X) generated by coherent sheaves
supported in dimension at most d. (This is called the dimensional filtration or
coniveau filtration on K(X), although we won’t need these names.) Let Ad′ (X) be
the dth graded piece of K(X), i.e., Ad′ (X) := K(X)≤d /K(X)≤d−1 .
If L is an invertible sheaf on X, define L · : K(X) → K(X) by L · [F ] = [F ] −
[L ∨ ⊗ F ]. (Do you see why this operator is well-defined?)

20.3.A. E XERCISE . If X is projective and k is infinite, show that L · sends K(X)≤d to

K(X)≤d−1 . (Hint: for each fixed F supported on a subset of dimension at most d,
write L is a difference of two very ample invertible sheaves, and choose sections
of those two, missing the associated points of F .)

20.3.3. Remark. The previous exercise holds true without k being infinite or X being
proper; see [Ko1, Prop. VI.2.5].

20.3.4. For the rest of this section, we assume X is projective and k is infinite. (But
in light of Remark 20.3.3, these hypotheses can be removed.) By Exercise 20.3.A,
L · descends to a map Ad′ (X) → Ad−1 ′
(X); we denote this operator c1 (L )∩. (It is
the action of the first Chern class.)

20.3.B. E XERCISE (“c1 IS ADDITIVE ”). Show that c1 (L ⊗ L ′ )∩ = (c1 (L )∩) +

(c1 (L ′ )∩). Hint: show that ((L ⊗ L ′ )·) − (L ·) − (L ′ ·) = −(L ·) ◦ (L ′ ·), and
thus sends K(X)≤d to K(X)≤d−2 . (This will remind you of the trick in the proof of
Proposition 20.1.3, and indeed is the motivation for that trick. Caution: the action
of Pic(X) is not additive on K(X); it only becomes additive once we pass to the
associated graded ring A•′ (X).)
If F has support of dimension at most n, and L1 , . . . , Ln are n invertible
sheaves, we can now reinterpret the intersection product (L1 · L2 · · · Ln · F ) as

c1 (L1 ) ∩ · · · ∩ c1 (Ln ) ∩ [F ]

where [F ] is interpreted as lying in either K(X) or An′ (X). You can now go back and
read §20.1, and reprove all the results with this starting point, sometimes obtaining
interesting generalizations. We have now promoted the intersection product from
an integer to a class in A0′ (X), which may have much more information. (This is
true even if X is a curve of positive genus over C, using the theory developed in
[F2].)
 © 2024 Ravi Vakil. Published by Princeton University Press. 589

These Ad′ (X) behave like homology groups in a number of ways. If Y ⊂ X is a

closed subvariety of pure dimension d, we have a class [Y] := [OY ] ∈ Ad′ (X). The
groups Ad′ (X) have appropriate functorial properties. For example, if π : X1 → X2
is a proper morphism (even if the Xi are not themselves proper), we have a map
π∗ : K(X1 ) → K(X2 ) given by
(20.3.4.1) π∗ [F ] = [π∗ F ] − [R1 π∗ F ] + · · ·
(using the long exact sequence for higher pushforwards, Theorem 18.7.1 (c)) which
descends to a map Ad′ (X1 ) → Ad′ (X2 ), where the “later terms” on the right side of
(20.3.4.1) disappear, so π∗ [F ] = [π∗ F ]. This pushforward interacts well with the
first Chern class of line bundles, yielding the projection formula of Remark 20.1.8.
If π is instead a flat morphism, soon to be discussed at length in Chapter 24, then
π∗ is exact (Exercise 24.1.H), so we have a map π∗ : K(X2 ) → K(X1 ). If the “relative
dimension” of this map is r (to be properly defined in Definition 24.5.7), this yields
a map π∗ : Ad′ (X2 ) → Ad+r ′
(X1 ). This interacts well with first Chern classes and
proper pushforwards.
If k = C and X is proper, there is a map Ad′ (X) → H2d (X, Q), which behaves as
you might hope (for example, in its interaction with Chern classes of line bundles).
If X is not proper (but k = C), then the map is to Borel-Moore homology rather than
usual homology.
Our groups Ad′ (X) are a good approximation of the theory of Chow groups
Ad (X), as developed in [F2]. In fact, there is a surjective map
(20.3.4.2) Ad (X) / A ′ (X)
d

[F2, Examp. 15.1.5], and this map is an isomorphism once tensored with Q [F2,
Thms. 18.2 and 18.3]. This is the beginning of a long and rich story in algebraic
geometry.

20.3.5. Side Remark. The surjection map (20.3.4.2) need not be an isomorphism, see
[SGA6, Exp. XIV, §4.5-4.7] (although its kernel must be torsion, as described above).
As a tantalizing example, if X is the group E8 , the kernel has 2-torsion, 3-torsion,
and 5-torsion, see [KN, DZ].

20.4 ⋆⋆ The Nakai-Moishezon and Kleiman criteria for ampleness

Exercise 20.1.K stated that if X is projective k-variety, and L is an ample line
bundle on X, then for any subvariety Y of X of dimension n, (L n · Y) > 0. The
Nakai-Moishezon criterion states that this is a characterization:

20.4.1. Theorem (Nakai-Moishezon criterion for ampleness). — If L is an

invertible sheaf on a projective k-scheme X, and for every irreducible subvariety Y of X,
(L dim Y · Y) > 0, then L is ample.

20.4.2. Remarks. We note that X need only be proper for this result to hold ([Kl1,
Thm. III.1.1]).
Before proving the Nakai-Moishezon criterion, we point out some conse-
quences related to our discussion of numerical equivalence in §18.4.11. By Propo-
sition 20.1.6, (L dim Y · Y) depends only on the numerical equivalence class of L ,
590 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

so ampleness is a numerical property. As a result, the notion of ampleness makes

sense on N1Q (X). As the tensor product of two ample invertible sheaves is ample
(Exercise 16.2.H), the ample Q-line bundles in N1Q (X) form a cone, called the ample
cone of X.

20.4.3. Proposition. — If X is a projective k-scheme, the ample cone is open.

20.4.4. In the rest of this section, we often use additive notation for the tensor
product of invertible sheaves, as described in Remark 20.1.4. This is because we
want to deal with intersections on the Q-vector space N1Q (X). For example by
((aL1 + bL1′ ) · L2 · · · Ln · F ) (where a, b ∈ Q), we mean a(L1 · L2 · · · Ln · F ) +
b(L1′ · L2 · · · Ln · F ).

Proof. We will need the fact that N1Q (X) is a finite-dimensional vector space (§18.4.12),
which unfortunately we have not explained except over C (Exercise 18.4.U).
Suppose A is an ample invertible sheaf on X. We will describe a small open
neighborhood of [A ] in N1Q (X) consisting of ample Q-line bundles. Choose in-
vertible sheaves L1 , . . . , Ln on X whose classes form a basis of N1Q (X). By Ex-
ercise 16.2.E, there is some m such that A ⊗m ⊗ Li and A ⊗m ⊗ Li∨ are both
very ample for all n. Thus (in the additive notation of §20.4.4), A + m 1
Li and
A − m Li are both ample. As the ample
1
P Q-line bundles form a cone, it follows that
A + ϵ1 L1 + · · · + ϵn Ln is ample for i |ϵi | ≤ 1/m. □

20.4.5. Proof of the Nakai-Moishezon criterion, Theorem 20.4.1. We prove the Nakai-
Moishezon criterion in several steps.

20.4.A. U NIMPORTANT E XERCISE . Prove the case where dim X = 0.

Step 1: initial reductions. Suppose L satisfies the hypotheses of the Theorem;
we wish to show that L is ample. By Exercises 18.8.A and 18.8.B, we may assume
that X is integral. Moreover, we can work by induction on dimension, so we can
assume that L is ample on any closed subvariety. The base case is dimension 1,
which was done in Exercise 19.2.H.
Step 2: sufficiently high powers of L have sections. We show that H0 (X, L ⊗m ) ̸= 0
for m ≫ 0.
Our plan is as follows. By Asymptotic Riemann-Roch (Exercise 20.1.I), χ(X, L ⊗m ) =
m (L n )/n! + · · · grows (as a function of m) without bound. A plausible means of
n

attack is to show that hi (X, L ⊗m ) = 0 for i > 0 and m ≫ 0. We won’t do that, but
will do something similar.
By Exercise 16.2.C, L is the difference of two very ample line bundles, say
L = ∼ A ⊗ B −1 with A = O(A) and B = O(B). From 0 → O(−A) → O → OA → 0
we have
(20.4.5.1) 0 / L ⊗m (−B) / L ⊗(m+1) / L ⊗(m+1) |A / 0.

From 0 → O(−B) → O → OB → 0, we have

(20.4.5.2) 0 / L ⊗m (−B) / L ⊗m / L ⊗m |B / 0.

Choose m large enough so that both L ⊗(m+1) |A and L ⊗m |B have vanishing

higher cohomology (i.e., h>0 = 0 for both; use the inductive hypothesis that L is
 © 2024 Ravi Vakil. Published by Princeton University Press. 591

ample on all proper closed subvarieties, and Serre vanishing, Theorem 18.1.3(ii)).
This implies that for i ≥ 2,
∼ Hi (X, L ⊗m (−B)) (long exact sequence for (20.4.5.2))
Hi (X, L ⊗m ) =
∼ Hi (X, L ⊗m+1 ) (long exact sequence for (20.4.5.1))
=
so the higher cohomology stabilizes (is constant) for large m. From
χ(X, L ⊗m ) = h0 (X, L ⊗m ) − h1 (X, L ⊗m ) + constant,
H0 (L ⊗m ) ̸= 0 for m ≫ 0, completing Step 2.
So by replacing L by a suitably large multiple (ampleness is independent
of taking tensor powers, Theorem 16.2.2), we may assume L has a section, with
corresponding effective divisor D. We now use D as a crutch.
Step 3: L ⊗m is globally generated for m ≫ 0.
As D is effective, L ⊗m is globally generated on the complement of D: we have
a section nonvanishing on that big open set. Thus any base locus must be contained
in D. Consider the short exact sequence

(20.4.5.3) 0 / L ⊗(m−1) / L ⊗m / L ⊗m |D /0

Now L |D is ample by our inductive hypothesis. Choose m so large that H1 (X, L ⊗m |D ) =

0 (Serre vanishing, Theorem 18.1.3(ii)). From the exact sequence associated to
(20.4.5.3),
ϕm : H1 (X, L ⊗(m−1) ) / H1 (X, L ⊗m )

is surjective for m ≫ 0. Using the fact that the H1 (X, L ⊗m ) are finite-dimensional
vector spaces, as m grows, H1 (X, L ⊗m ) must eventually stabilize, so the ϕm are
isomorphisms for m ≫ 0.
Thus for large m, from the long exact sequence in cohomology for (20.4.5.3),
H0 (X, L ⊗m ) → H0 (X, L ⊗m |D ) is surjective for m ≫ 0. But H0 (X, L ⊗m |D ) has no
base points by our inductive hypothesis (applied to D), i.e., for any point p of D
there is a section of L ⊗m |D not vanishing at p, so H0 (X, L ⊗m ) has no base points
on D either, completing Step 3.
Step 4. Thus L is a base-point-free line bundle with positive degree on each
curve (by hypothesis of Theorem 20.4.1), so by Exercise 18.4.O we are done. □
The following result is the key to proving Kleiman’s criterion for ampleness,
Theorem 20.4.7.

20.4.6. Kleiman’s Theorem. — Suppose X is a projective k-scheme. If L is a nef

invertible sheaf on X, then (L k · V) ≥ 0 for every irreducible subvariety V ⊂ X of
dimension k.
As usual, this extends to the proper case, see [Kl1, Thm. IV.2.1]. And as usual,
we postpone the proof until after we appreciate the consequences.

20.4.B. E XERCISE .
(a) (limit of amples is nef) If L and H are any two invertible sheaves such that
L + ϵH is ample for all sufficiently small ϵ > 0, show that L is nef. (Hint: limϵ→0 .
This doesn’t require Kleiman’s Theorem.)
(b) (nef + ample = ample) Suppose X is a projective k-scheme, H is ample, and L is
592 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

nef. Show that L + ϵH is ample for all ϵ ∈ Q≥0 . (Hint: use the Nakai-Moishezon
criterion: ((L +ϵH )k ·V) > 0. This may help you appreciate the additive notation.)

20.4.7. Theorem (Kleiman’s criterion for ampleness). — Suppose X is a projective

k-scheme.

(a) The nef cone is the closure of the ample cone.

(b) The ample cone is the interior of the nef cone.

Side remark: of course (a) is false if “projective” is relaxed to “proper” (as

the ample cone of a proper nonprojective variety is empty, but 0 is nef so the nef
cone is nonempty). However, part (b) is true if X is proper and factorial (see [Kl1,
Thm. IV.2.2] for a proof and a more general statement). Hence if X is smooth, proper,
and nonprojective, then the interior of the nef cone is empty.

20.4.C. E XERCISE . In [Kl1, p. 326, Ex. 2], Kleiman describes Mumford’s example
of a non-ample line bundle on a smooth projective surface that meets every curve
positively. Doesn’t this contradict Theorem 20.4.7? Did Mumford and Kleiman
make a mistake?

Proof. (a) Ample invertible sheaves are nef (Exercise 18.4.V(e)), and the nef cone
is closed (Exercise 18.4.W), so the closure of the ample cone is contained in the
cone. Conversely, each nef element of N1Q (X) is the limit of ample classes by
Exercise 20.4.B(a), so the nef cone is contained in the closure of the ample cone.
(b) As the ample cone is open (Proposition 20.4.3), the ample cone is contained
in the interior of the nef cone. Conversely, suppose L is in the interior of the nef
cone, and H is any ample class. Then L − ϵH is nef for all small enough positive
ϵ. Then by Exercise 20.4.B(b), L = (L − ϵH ) + ϵH is ample. □
Suitably motivated, we prove Kleiman’s Theorem 20.4.6.

Proof. We may immediately reduce to the case where X is irreducible and reduced.
We work by induction on n := dim X. The base case n = 1 is obvious. So we
assume that (L dim V · V) ≥ 0 for all irreducible V not equal to X. We need only
show that (L n · X) ≥ 0.
Fix some very ample invertible sheaf H on X.

20.4.D. E XERCISE . Show that (L k · H n−k · X) ≥ 0 for all k < n. (Hint: use the
inductive hypothesis.)
Consider P(t) := ((L + tH )n · X) ∈ Q[t]. We wish to show that P(0) ≥ 0.
Assume otherwise that P(0) < 0. Now for t ≫ 0, L + tH is ample, so P(t) is
positive for large t. Thus P(t) has positive real roots. Let t0 be the largest positive
real root of t. (In fact there is only one positive root, as Exercise 20.4.D shows that
all the nonconstant coefficients of P(t) are nonnegative.

20.4.E. E XERCISE . Show that for (rational) t > t0 , L + tH is ample. Hint: use the
Nakai-Moishezon criterion (Theorem 20.4.1); if V ̸= X is an irreducible subvariety,
show that ((L + tH )dim V · V) > 0 by expanding (L + tH )dim V .
Let Q(t) := (L · (L + tH )n−1 · X) and R(t) := (tH · (L + tH )n−1 · X), so
P(t) = Q(t) + R(t).
 © 2024 Ravi Vakil. Published by Princeton University Press. 593

20.4.F. E XERCISE . Show that Q(t) ≥ 0 for all rational t > t0 . Hint (which
you will have to make sense of): (L + tH ) is ample by Exercise 20.4.E, so for
N sufficiently large, N(L + tH ) is very ample. Use the idea of the proof of
Proposition 20.1.6 to intersect X with n − 1 divisors in the class of N(L + tH ) so
that “((N(L + tH ))n−1 · X) is an effective curve C”. Then (L · C) ≥ 0 as L is nef.
Thus Q(t0 ) ≥ 0.

20.4.G. E XERCISE . Show that R(t0 ) > 0. (Hint: Exercise 20.4.D.)

Thus P(t0 ) > 0 as desired. □
 CHAPTER 21

Differentials

Differentials are an intuitive geometric notion, and we are going to figure out
the right description of them algebraically. The algebraic manifestation is somewhat
non-intuitive, so it is helpful to understand differentials first in terms of geometry.
Also, although the algebraic statements are odd, none of the proofs are hard or long.
You will notice that this topic could have been done as soon as we knew about
morphisms and quasicoherent sheaves. We have usually introduced new ideas
through a number of examples, but in this case we will spend a fair amount of time
discussing theory, and only then get to examples.

21.1 Motivation and game plan

Suppose X is a “smooth” k-variety. We would like to define a tangent bundle.
We will see that the right way to do this will easily apply in much more general
circumstances.
• We will see that the cotangent sheaf is more “natural” for schemes than tangent
bundle/sheaf. This is similar to the fact that the Zariski cotangent space is more
natural than the tangent space (i.e., if A is a ring and m is a maximal ideal, then
m/m2 is “more natural” than (m/m2 )∨ ), as we have repeatedly discussed since
§13.1. In both cases this is because we are understanding “spaces” via their (sheaf
of) functions on them, which is somehow dual to the geometric pictures you have
of spaces in your mind.
So we will define the cotangent sheaf first. An element of the (co)tangent space
will be called a (co)tangent vector.
• Our undergraduate intuition will continue to work — we can calculate using the
Jacobian matrix, the answers will be the same in cases where our undergraduate
calculations apply, and we will even have the advantage of avoiding δ’s and ϵ’s.
• Our construction will automatically apply for general X, even if X is not “smooth”
(or even at all nice, e.g., finite type). The cotangent sheaf will not necessarily be
locally free, but it will still be a quasicoherent sheaf.
• Better yet, this construction will naturally work “relatively”. For any π : X → Y, we
will define Ωπ = ΩX/Y , a quasicoherent sheaf on X, the sheaf of relative differentials.
The fiber of this sheaf at a point will be the cotangent vectors of the fiber of the
map. This will specialize to the earlier case by taking Y = Spec k. The idea is that
this glues together the cotangent sheaves of the fibers of the family. Figure 21.1 is

This material is published by Princeton University Press as The Rising Sea: Foundations of Algebraic
Geometry by Ravi Vakil. This version is free to view and download for personal use only. Not for
redistribution, resale, or use in derivative works. © 2024 Ravi Vakil.

595
596 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

a sketch of the relative tangent space of a map X → Y at a point p ∈ X — it is the

tangent to the fiber. (The tangent space is easier to draw than the cotangent space!)
An element of the relative (co)tangent space is called a relative (co)tangent vector.

F IGURE 21.1. The relative tangent space of a morphism X → Y at

a point p

Thus the central concept of this chapter is the cotangent sheaf Ωπ = ΩX/Y for
a morphism π : X → Y of schemes. A good picture to have in your mind is the
following. If π : X → Y is a submersion of manifolds (a map inducing a surjection on
tangent spaces), you might hope that the tangent spaces to the fibers at each point
p ∈ X might fit together to form a vector bundle. This is the relative tangent bundle
(of π), and its dual is ΩX/Y (see Figure 21.1). Even if you are not geometrically
minded, you will find this useful. (For an arithmetic example, see Exercise 21.2.H.)

21.2 Definitions and first properties

21.2.1. The affine case: three definitions.

We first study the affine case. Suppose A is a B-algebra, so we have a morphism
of rings ϕ : B → A and a morphism of schemes Spec A → Spec B. I will define an
A-module ΩA/B in three ways. This is called the module of relative differentials
or the module of Kähler differentials. The module of differentials will be defined
to be this module, as well as a map d : A → ΩA/B satisfying three properties.
(Caution: although d sends an A-module to an A-module, it is not in general
A-linear. A priori we take it as a homomorphism of abelian groups, but we will
momentarily make it a homomorphism of B-modules, Exercise 21.2.A.)
(i) additivity: da + da ′ = d(a + a ′ ).
(ii) Leibniz: d(aa ′ ) = a da ′ + a ′ da.
(iii) triviality on pullbacks: db = 0 for b ∈ ϕ(B).
These properties will not be surprising if you have seen differentials in any other
context.

21.2.A. T RIVIAL EXERCISE . Show that d is B-linear.

21.2.B. E XERCISE . Prove the quotient rule: if a ′ = as, then s2 da = s da ′ − a ′ ds.

 © 2024 Ravi Vakil. Published by Princeton University Press. 597

21.2.C. E XERCISE . State and prove the chain rule for d(f(g)) where f is a polynomial
with B-coefficients, and g ∈ A. (As motivation, think of the case B = k. So for
example, dan = nan−1 da, and more generally, if f is aPpolynomial in one P variable,
df(a) = f ′ (a) da, where f ′ is defined formally: if f = ci xi then f ′ = ci ixi−1 .)
We will now see three definitions of the module of Kähler differentials, which
will soon “sheafify” to the sheaf of relative differentials. The first definition is a
concrete hands-on definition. The second is by universal property. And the third
will globalize well, and will allow us to define ΩX/Y conveniently in general.

21.2.2. First definition of differentials: explicit description. We define ΩA/B to

be finite A-linear combinations of symbols “da” for a ∈ A, subject to the three rules
(i)–(iii) above. For example, take A = k[x, y], B = k. Then a sample differential is
3x2 dy + 4 dx ∈ ΩA/B . We have identities such as d(3xy2 ) = 3y2 dx + 6xy dy.

21.2.3. Key fact. Note that if A is generated over B (as an algebra) by xi ∈ A

(where i lies in some index set, possibly infinite), subject to some relations rj (where
j lies in some index set, and each is a polynomial in the xi ), then the A-module
ΩA/B is generated by the dxi , subject to the relations drj = 0. In short, we needn’t
take every single element of A; we can take a generating set. And we needn’t take
every single relation among these generating elements; we can take generators of
the relations.

21.2.D. E XERCISE . Verify Key fact 21.2.3. (If you wish, use the affine conormal exact
sequence, Theorem 21.2.12, to verify it; different people prefer to work through the
theory in different orders. Just take care not to make circular arguments.)
In particular:

21.2.4. Proposition. — If A is a finitely generated B-algebra, then ΩA/B is a finite type

(i.e., finitely generated) A-module. If A is a finitely presented B-algebra, then ΩA/B is a
finitely presented A-module.
Recall (§8.3.13) that a ring A is finitely presented over another ring B if it can be
expressed with finite number of generators and finite number of relations:
A = B[x1 , . . . , xn ]/(r1 (x1 , . . . , xn ), . . . , rj (x1 , . . . , xn )).
If A is Noetherian, then finitely presented is the same as finite type, as the “finite
number of relations” comes for free, so most of you will not care.
Let’s now see some examples. Among these examples are three particularly
important building blocks for ring maps: adding free variables; localizing; and
taking quotients. If we know how to deal with these, we know (at least in theory)
how to deal with any ring map. (They were similarly useful in understanding the
fibered product in practice, in §10.2.)

21.2.5. Example: taking a quotient. If A = B/I, then ΩA/B = 0: da = 0 for all

a ∈ A, as each such a is the image of an element of B. This should be believable; in
this case, there are no “relative tangent vectors”.

21.2.6. Example: adding variables. If A = B[x1 , . . . , xn ], then ΩA/B = Adx1 ⊕

· · · ⊕ Adxn . (Note that this argument applies even if we add an arbitrarily infinite
number of indeterminates.) The intuitive geometry behind this makes the answer
598 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

very reasonable. The cotangent bundle of affine n-space should indeed be free of
rank n.

21.2.7. Explicit example: an affine plane curve. Consider the plane curve y2 =
x3 − x in A2k , where the characteristic of k is not 2 (see Figure 19.7 for a picture). Let
A = k[x, y]/(y2 − x3 + x) and B = k. By Key fact 21.2.3, the module of differentials
ΩA/B is generated by dx and dy, subject to the relation

2y dy = (3x2 − 1) dx.

Thus in the locus where y ̸= 0, dx is a generator (as dy can be expressed in terms of

dx). We conclude that where y ̸= 0, Ω e A/B is isomorphic to the trivial line bundle
(invertible sheaf). Similarly, in the locus where 3x2 − 1 ̸= 0, dy is a generator. These
two loci cover the entire curve, as solving y = 0 gives x3 − x = 0, i.e., x = 0 or ±1,
and in each of these cases 3x2 − 1 ̸= 0. We have shown that Ω e A/B is an invertible
sheaf.
We can interpret dx and dy geometrically. Where does the differential dx van-
ish? The previous paragraph shows that it doesn’t vanish on the patch where 2y ̸= 0.
On the patch where 3x2 − 1 ̸= 0, where dy is a generator, dx = (2y/(3x2 − 1)) dy
from which we see that dx vanishes precisely where y = 0. You should find this
believable from the picture. We have shown that dx = 0 precisely where the curve
has a vertical tangent vector. Once we can pull back differentials (Exercise 21.2.K(a)
or Theorem 21.2.26), we can interpret dx as the pullback of a differential on the
x-axis to Spec A (pulling back along the projection to the x-axis). When we do that,
using the fact that dx doesn’t vanish on the x-axis, we can interpret the locus where
dx = 0 as the locus where the projection map branches. (Can you compute where
dy = 0, and interpret it geometrically?)
This discussion applies to plane curves more generally. Suppose A = k[x, y]/f(x, y),
where for convenience k = k. Then the same argument as the one given above
shows that Ω e A/k is free of rank 1 on the open set D(∂f/∂x), and also on D(∂f/∂y).
If Spec A is a regular curve, then these two sets cover all of Spec A. (Conversely,
Exercise 13.2.F — basically the Jacobian criterion for regularity — gives regular-
ity at the closed points. Furthermore, the curve must be reduced, or else as the
nonreduced locus is closed, it would be nonreduced at a closed point, contradicting
regularity. Finally, reducedness at a generic point is equivalent to regularity —
a scheme whose underlying set is a point is reduced if and only if it is regular.
Alternatively, we could invoke a big result, Fact 13.8.2, to get regularity at the
generic point from regularity at the closed points.)
If, on the other hand, the plane curve is singular, then Ω is not locally free of rank
one. For example, consider the plane curve Spec A where A = C[x, y]/(y2 − x3 ), so

ΩA/C = (A dx ⊕ A dy)/(2y dy − 3x2 dx).

Then the fiber of ΩA/C over the origin (computed by setting x = y = 0) is rank 2,
as it is generated by dx and dy, with no relation.
Implicit in the above discussion is the following exercise, showing that Ω can
be computed using the Jacobian matrix.
 © 2024 Ravi Vakil. Published by Princeton University Press. 599

21.2.E. I MPORTANT BUT E ASY E XERCISE (J ACOBIAN DESCRIPTION OF ΩA/B ). Sup-

pose A = B[x1 , . . . , xn ]/(f1 , . . . , fr ). Then
!
M
ΩA/B = A dxi / (A df1 + · · · + A dfr )
i

may be interpreted as the cokernel of the Jacobian matrix (13.1.7.1)

J : A⊕r / A⊕n .

(You might notice that if J has full rank everywhere, so A is smooth over B, then
ΩA/B is locally free; see Exercise 21.2.Q for more.)
The next result connects the cotangent module ΩA/B to the cotangent space at
a (rational) point.

21.2.F. E XERCISE ( THE FIBER OF Ω AT A RATIONAL POINT IS THE COTANGENT

SPACE ). Suppose B is a finite type k-algebra, and m ⊂ B is a maximal ideal
∼
with residue field k. Then there is an isomorphism of k-vector spaces δ : m/m2 −→
∼
ΩB/k ⊗B k (where the k on the right is a B-module via the isomorphism k ←→ B/m).
Hint: both are “computed” using the Jacobian matrix.
(Exercise 21.2.F is true even without the finite type hypothesis, although the hint
doesn’t apply in that case. You are welcome to prove this more general statement,
using the universal property of of ΩB/k . But we won’t need it.)
Corollary 21.2.33 will give a quite different proof, and generalize it to the case
where B/m is a finite separable extension of k. It will allow us to extend the Jacobian
description of the Zariski cotangent space to closed points whose residue field is a
separable algebraic extension of the ground field, not just the ground field itself.

21.2.G. E XERCISE . Suppose X is a finite type scheme over an algebraically closed

field. Show that the function from the closed points of X to Z≥0 given by p 7→
dim Tp X is upper semicontinuous in the Zariski topology. (Be clear on what the
Zariski topology is on the set of closed points, cf. §8.4.8.) Corollary 21.2.33 will
immediately extend this to all fields of characteristic 0, and all finite fields.

21.2.8. Example: Localization. If T is a multiplicative subset of B, and A = T −1 B,

then ΩA/B = 0. Reason: by the quotient rule (Exercise 21.2.B), if a = b/t, then
da = (t db − b dt)/t2 = 0. If A = Bf , this is intuitively believable; then Spec A is
an open subset of Spec B, so there should be no relative (co)tangent vectors.

21.2.H. I MPORTANT E XERCISE ( FIELD EXTENSIONS ). This notion of relative differ-

entials is interesting even for finite extensions of fields. In other words, even when
you map a reduced point to a reduced point, something interesting can happen
with differentials.
(a) Suppose K/k is a separable algebraic extension. Show that ΩK/k = 0. Do not
assume that K/k is a finite extension! (Hint: for any α ∈ K, there is a polynomial
f(x) such that f(α) = 0 and f ′ (α) ̸= 0.)
(b) Suppose k is a field of characteristic p, K = k(tp ), L = k(t). Compute ΩL/K
(where K ,→ L is the “obvious” inclusion).
600 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

We now delve a little deeper, and discuss two useful and geometrically moti-
vated exact sequences.

21.2.9. Theorem (cotangent exact sequence, affine version). — Suppose C → B →

A are ring morphisms. Then there is a natural exact sequence of A-modules
a⊗db7→a db da7→da
A ⊗B ΩB/C / ΩA/C / ΩA/B / 0.

F IGURE 21.2. The geometry behind the cotangent exact sequence

The proof will be quite straightforward algebraically, but the statement comes
fundamentally from geometry, and that is how best to remember it. Figure 21.2 is
a sketch of a map π : X → Y. Here X should be interpreted as Spec A, Y as Spec B,
and Spec C (we will soon call it Z) is a point. (If you would like a picture with a
higher-dimensional Spec C, just “take the product of Figure 21.2 with a curve”.) In
the Figure, Y is “smooth”, and X is “smooth over Y” — which means roughly that
all the fibers are smooth. Suppose p is a point of X. Then the tangent space of the
fiber of π at p is certainly a subspace of the tangent space of the total space of X
at p. The cokernel is naturally the pullback of the tangent space of Y at π(p). This
short exact sequence for each p should be part of a short exact sequence of “relative
tangent sheaves”
0 / TX/Y / TX/Z / π∗ TY/Z /0

on X. (We will formally define “relative tangent sheaf” in §21.2.18.) Dualizing this
yields
0 / π∗ ΩY/Z / ΩX/Z / ΩX/Y / 0.

This is precisely the statement of Theorem 21.2.9, except we also have left-
exactness. This discrepancy is because the statement of the theorem is more general;
we will see in Exercise 21.2.S that in the “smooth” case, we indeed have left-
exactness.
 © 2024 Ravi Vakil. Published by Princeton University Press. 601

21.2.10. Intriguing Remark. As always, whenever you see something right-exact,

you should suspect that there should be some sort of (co)homology theory so that
this is the end of a long exact sequence. This is indeed the case, and this exact
sequence involves André-Quillen homology (see [E, p. 386] for more). You should
expect that the next term to the left should be the first homology corresponding to
A/B, and in particular shouldn’t involve C. So if you already suspect that you have
exactness on the left in the case where A/B and B/C are “smooth” (whatever that
means), and the intuition of Figure 21.2 applies, then you should expect further that
all that is necessary is that A/B be “smooth”, and that this would imply that the
first André-Quillen homology should be zero. Even though you wouldn’t precisely
know what all the words meant, you would be completely correct! You would also
be developing a vague inkling about the cotangent complex. We will see examples
when left-exactness holds in a sufficiently “smooth” situation in Proposition 21.4.2
and Exercise 21.2.S. For a more general statement, see [Stacks, tag 06B6]. See also
[Liu, Cor. 6.3.22] and [Liu, Prop. 6.3.11] for clean discussions of the next two terms
in the sequence: “cotangent exact sequence”, “conormal exact sequence”, . . . .

21.2.11. Proof of Theorem 21.2.9 (the cotangent exact sequence, affine version). First, note
that surjectivity of ΩA/C → ΩA/B is clear, as this map is given by da 7→ da (where
a ∈ A).
Next, the composition over the middle term is clearly 0, as this composition is
given by a ⊗ db 7→ adb 7→ 0.
Finally, we wish to identify ΩA/B as the cokernel of A ⊗B ΩB/C → ΩA/C . Now
ΩA/B is exactly the same as ΩA/C , except we have extra relations: db = 0 for b ∈ B.
These are precisely the images of 1 ⊗ db on the left. □

21.2.12. Theorem (conormal exact sequence, affine version). — Suppose B is a

C-algebra, I is an ideal of B, and A = B/I. Then there is a natural exact sequence of
A-modules
a⊗db7→a db
I/I2
δ / A ⊗B ΩB/C / ΩA/C / 0.

The map δ is given by i 7→ 1 ⊗ di, or more formally, 1 ⊗ d : B/I ⊗B I → B/I ⊗B ΩB/C .

(You will recognize the map A ⊗B ΩB/C → ΩA/C from the cotangent exact
sequence, Theorem 21.2.9. A special case of the map δ secretly appeared in Exer-
cise 21.2.F.) The proof is algebraic, so the geometric discussion afterwards may help
clarify how you should really think of it.

Proof. We will identify the cokernel of δ : I/I2 → A ⊗B ΩB/C with ΩA/C . Consider
A ⊗B ΩB/C . As an A-module, it is generated by db (where b ∈ B), subject to three
types of relations: dc = 0 for c ∈ ϕ(C) (where ϕ : C → B describes B as a C-algebra),
additivity, and the Leibniz rule. Given any relation in B, d of that relation is 0.
Now ΩA/C is defined similarly, except there are more relations in A; these
are precisely the elements of I ⊂ B. Thus we obtain ΩA/C by starting out with
A ⊗B ΩB/C , and adding the additional relations di where i ∈ I. But this is precisely
the image of δ! (Be sure that you see how the identification of the cokernel of δ with
ΩA/C is precisely via the map a ⊗ db 7→ a db.) □
We now give a geometric interpretation of the conormal exact sequence, and in
particular define conormal modules/sheaves/bundles.
602 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

As with the cotangent exact sequence (Theorem 21.2.9), the conormal exact
sequence is fundamentally about geometry. To motivate it, consider the sketch
of Figure 21.3. In the sketch, everything is “smooth”, X is one-dimensional, Y is
two-dimensional, j is the inclusion j : X ,→ Y, and Z is a point. Then at a point
p ∈ X, the tangent space TX |p clearly injects into the tangent space of j(p) in Y, and
the cokernel is the normal vector space to X in Y at p. This should give an exact
sequence of bundles on X:
0 / TX / j∗ TY / NX/Y / 0.

Dualizing this should give

0 /N∨ / j∗ ΩY/Z / ΩX/Z / 0.

X/Y

This is precisely what appears in the statement of the Theorem, except (i) the exact
sequence in algebraic geometry is not necessarily exact on the left, and (ii) we see
∨
I/I2 instead of NSpec A/ Spec B .

F IGURE 21.3. The geometry behind the conormal exact sequence

21.2.13. We resolve the first issue (i) by expecting that the sequence of Theo-
rem 21.2.12 is exact on the left in appropriately “smooth” situations, and this is
indeed the case, see Theorem 21.2.31 and Remark 21.2.32. (If you enjoyed Re-
mark 21.2.10, you might correctly guess several things. The next term on the left
should be the first André-Quillen homology of A/C, so we should only need that
A/C is smooth, and B should be irrelevant. Also, if A = B/I, then we should expect
that I/I2 is the first André-Quillen homology of A/B.)

21.2.14. Conormal modules and conormal sheaves. We resolve the second issue (ii) by
declaring I/I2 to be the conormal module, and in Definition 21.2.15 we will define
the obvious analog as the conormal sheaf.
Here is some geometric intuition as to why we might want to call (the sheaf
associated to) I/I2 the conormal sheaf, which will likely confuse you, but may
offer some enlightenment. First, if Spec A is a closed point of Spec B, we expect
the conormal space to be precisely the cotangent space. And indeed if A = B/m,
the Zariski cotangent space is m/m2 . (We made this subtle connection in §13.1.) In
particular, at some point you will develop a sense of why the conormal (=cotangent)
space to the origin in A2k = Spec k[x, y] is naturally the space of linear forms
αx + βy. But then consider the z-axis in Spec k[x, y, z] = A3k , cut out by I =
 © 2024 Ravi Vakil. Published by Princeton University Press. 603

(x, y). Elements of I/I2 may be written as α(z)x + β(z)y, where α(z) and β(z) are
polynomial. This reasonably should be the conormal space to the z-axis: as z varies,
the coefficients of x and y vary. More generally, the same idea suggests that the
conormal module/sheaf to any coordinate k-plane inside n-space corresponds to
I/I2 . Now consider a k-dimensional manifold X inside an n-dimensional manifold
Y, with the classical topology. We can apply the same construction: if I is the ideal
sheaf of X in Y, then I /I 2 can be identified with the conormal sheaf (essentially
the conormal vector bundle), because analytically locally X ,→ Y can be identified
with Rk ,→ Rn . For this reason, you might hope that in algebraic geometry, if
Spec A ,→ Spec B is an inclusion of something “smooth” in something “smooth”,
I/I2 should be the conormal module (or, after applying the functor ∼, the conormal
sheaf). Motivated by this, we define the conormal module as I/I2 always, and then
notice that it has good properties (such as Theorem 21.2.12), but take care to learn
what unexpected behavior it might have when we are not in the “smooth” situation,
by working out examples such as that of §21.2.7.

21.2.15. Definition. Suppose i : X ,→ Y is a closed embedding of schemes cut out

by ideal sheaf I . Define the conormal sheaf for a closed embedding by I /I 2 ,
∨
denoted by NX/Y . (The product of quasicoherent ideal sheaves was defined in
∨
Exercise 15.6.E.) Important: We interpret NX/Y as a quasicoherent sheaf on X, not
(just) Y — be sure you understand why we may do so. (Exercise 17.1.E is one reason,
but it may confuse the issue more than help.)
∨
Define the normal sheaf as its dual NX/Y := Hom (NX/Y , OX ). This is imperfect
notation, because it suggests that the dual of N is always N ∨ . This is not always
true, as for A-modules, the natural morphism from a module to its double-dual is
not always an isomorphism. (Modules for which this is true are called reflexive,
but we won’t use this notion.)
If the normal sheaf is locally free of finite rank, then we call the associated
vector bundle the normal bundle of X in Y.
∨
21.2.I. E ASY E XERCISE . Define the conormal sheaf NX/Y (and hence the normal
sheaf NX/Y ) for a locally closed embedding i : X ,→ Y of schemes, a quasicoherent sheaf
on X. (Make sure your definition is well-defined!)
In the good situation of a regular embedding, the conormal sheaf (and hence the
normal sheaf) is locally free of finite rank (or more informally, a vector bundle). In
particular, the dual of N is indeed N ∨ . As a warm-up we deal with the important
codimension 1 case.

21.2.J. E XERCISE : N ORMAL BUNDLES TO EFFECTIVE C ARTIER DIVISORS . Suppose

∨
D ⊂ X is an effective Cartier divisor (§9.5.1). Show that the conormal sheaf ND/X
is O(−D)|D (and in particular is an invertible sheaf), and hence that the normal
sheaf is O(D)|D . This justifies our earlier definition of normal bundles to effective
Cartier divisors, in §15.6.2. It may be surprising that the normal sheaf should be
∼ A2 and D is the union of the two axes (and more generally if X
locally free if X =
is regular but D is singular), because you may be used to thinking that a “tubular
neighborhood” is isomorphic to the normal bundle.
We now treat the general case.
604 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

21.2.16. Proposition. —
(a) If the ideal I ⊂ B is generated by a regular sequence x1 , . . . , xr , then the map
γ : (B/I)⊕r → I/I2 given by

(21.2.16.1) (a1 , . . . , ar )  / a1 x1 + · · · + ar xr

describes I/I2 as a free module of rank r over B/I with basis x1 , . . . , xr .

(b) The (co)normal sheaf of a codimension r regular embedding in a locally Noetherian
scheme is locally free of rank r.
Exercise 24.6.D gives a different, short proof of (b) (and indeed (a) as well if
you pay attention) in the commonly used case where A contains a field.

Proof. (a) Clearly γ is surjective, so we must prove injectivity. We prove the case r =
1 first to highlight the main idea. We wish to show that ×x1 : B/(x1 ) → (x1 )/(x21 )
is injective. Suppose b ∈ B satisfies bx1 ∈ (x21 ), i.e., bx1 = cx21 for some c ∈ B. We
wish to show that b ∈ (x1 ). But ×x1 : B → B is injective (x1 is a non-zerodivisor),
so (b − cx1 )x1 = 0 implies b = cx1 as desired.
We now prove the general case. We wish to show (21.2.16.1) is injective. Sup-
pose we have b1 , . . . , br ∈ B with

(21.2.16.2) b1 x1 + · · · + br xr ∈ I2 ;

we wish to show that bi ∈ I. Now modulo (x1 , . . . , xr−1 ), (21.2.16.2) becomes

br xr ∈ (x2r ). Thus (as xr is a non-zerodivisor modulo (x1 , . . . , xr−1 ), by definition of
regular sequence), br is a multiple of xr modulo (x1 , . . . , xr−1 ), so br ∈ (x1 , . . . , xr )
as desired. Because the roles of the bi ’s are symmetric by Theorem 9.5.6, we are
done.
(b) follows immediately from (a). □
We will soon meet a related but harder fact, that if I is the ideal sheaf of a
regular embedding of codimension r, then I n /I n+1 is locally free, because it is
Symn (I /I 2 ) (Corollary 22.3.9, or Exercise 24.6.E).

21.2.17. Second definition: universal property. Here is a second definition that is

important philosophically, by universal property. Of course, it is a characterization
rather than a definition: by universal property nonsense, it shows that if the module
exists (with the d map), then it is unique up to unique isomorphism, and then one
still has to construct it to make sure that it exists.
Suppose A is a B-algebra, and M is a A-module. A B-linear derivation of
A into M is a map d : A → M of B-modules (not necessarily a map of A-modules)
satisfying the Leibniz rule: d(fg) = f dg + g df. As an example, suppose B = k,
and A = k[x], and M = A. Then d/dx is a k-linear derivation. As a second example,
if B = k, A = k[x], and M = k, then (d/dx)|0 (the operator “evaluate the derivative
at 0”) is a k-linear derivation.
A third example is d : A → ΩA/B , and indeed d : A → ΩA/B is the universal
B-linear derivation of A. Precisely, the map d : A → ΩA/B is defined by the following
universal property: any other B-linear derivation d ′ : A → M factors uniquely
 © 2024 Ravi Vakil. Published by Princeton University Press. 605

through d:
d′ /M
A <

d
" f

ΩA/B
Here f is a map of A-modules. (Note again that d and d ′ are not necessarily maps
of A-modules — they are only B-linear.) By universal property nonsense, if it exists,
it is unique up to unique isomorphism. The map d : A → ΩA/B clearly satisfies this
universal property, essentially by definition.
Depending on how your brain works, you may prefer using the first (construc-
tive) or second (universal property) definition to do the next two exercises.

21.2.K. E XERCISE .
(a) (pullback of differentials) If
AO ′ o AO

B′ o B
is a commutative diagram, describe a natural homomorphism of A ′ -modules
A ′ ⊗A ΩA/B → ΩA ′ /B ′ . An important special case is B = B ′ .
(b) (differentials behave well with respect to base extension, affine case) If furthermore
the above diagram is a “tensor diagram” (i.e., A ′ = ∼ B ′ ⊗B A, so the diagram is “co-
Cartesian”) then show that A ′ ⊗A ΩA/B → ΩA ′ /B ′ is an isomorphism. (Depending
on how you proceed, this may be trickier than you expect.)

21.2.L. E XERCISE : LOCALIZATION ( STRONGER FORM ; CF. E XAMPLE 21.2.8). Sup-

pose ϕ : B → A is a map of rings, S is a multiplicative subset of A, and T is a
multiplicative subset of B with ϕ(T ) ⊂ S, so we have the following commutative
diagram.
S−1O A o AO

T −1 B o B
Show that the pullback of differentials S ΩA/B → ΩS−1 A/T −1 B of Exercise 21.2.K(a)
−1

is an isomorphism. (This should be believable from the intuitive picture of “relative

tangent vectors”.) An important case is when T = {1}.

21.2.M. E XERCISE ( FIELD EXTENSIONS CONTINUED , CF. E XERCISE 21.2.H).

(a) Compute Ωk(t)/k . (Hint: §21.2.6 followed by Exercise 21.2.L.)
(b) If K/k is separably generated by t1 , . . . , tn ∈ K (i.e., t1 , . . . , tn form a transcendence
basis, and K/k(t1 , . . . , tn ) is algebraic and separable, §10.5.18), show that ΩK/k is a
free K-module (i.e., vector space) with basis dt1 , . . . , dtn . Possible approach: use
the cotangent exact sequence (Theorem 21.2.9) for k ,→ k(t1 , . . . , tn ) ,→ K to show
that the dti span ΩK/k as a K-vector space. The tricky part is showing that the
dti are linearly independent. Do this by showing that there exists a unique map
ΩK/k → K sending dt1 to 1, and dti to 0 for i > 1. Do this first for the case where
K/k(t1 , . . . , tn ) is generated by one element; then where it is finitely generated;
606 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

then the general case by defining it on all finitely generated subextensions, and
using uniqueness to show that they “all agree”.

21.2.18. Third definition: global. We now want to globalize this definition for an
arbitrary morphism of schemes π : X → Y. We could do this “affine by affine”; we
just need to make sure that the above notion behaves well with respect to “change
of affine sets”. Thus a relative differential
P on X would be the data of, for every
affine U ⊂ X, a differential of the form ai Pdbi , and on the intersection
P of two
affine open sets U ∩ U ′ , with representatives ai dbi on U and ai′ dbi′ on the
second, an equality on the overlap. Instead, we take a different approach. I will
give the (seemingly unintuitive) definition, then tell you how to think about it, and
then get back to the definition.
Let π : X → Y be any morphism of schemes. Recall that δ : X → X ×Y X is a
locally closed embedding (Proposition 11.2.1(b)). Define the relative cotangent
∨
sheaf ΩX/Y (or Ωπ ) as the conormal sheaf NX/X× YX
of the diagonal (see §21.2.14
— and if X → Y is separated you needn’t even worry about Exercise 21.2.I). (Now
is also as good a time as any to define the relative tangent sheaf TX/Y as the dual
Hom (ΩX/Y , OX ) to the relative cotangent sheaf. If we are working in the category of
k-schemes, then ΩX/k and TX/k are often called the cotangent sheaf and tangent
sheaf of X respectively.)
We now define d : OX → ΩX/Y . Let pr1 : X ×Y X → X and pr2 : X ×Y X → X be
the two projections. Then define d : OX → ΩX/Y on the open set U as follows:

df = pr∗2 f − pr∗1 f.

(How do you interpret pr∗2 f − pr∗1 f as a section of ΩX/Y ?)

Warning: d is not a morphism of quasicoherent sheaves on X, although it is
OY -linear in the only possible meaning of that phrase.
We will soon see that d is indeed a derivation of the sheaf OX (in the only
possible meaning of the phrase), and at the same time see that our new notion of
differentials agrees with our old definition on affine open sets, and hence globalizes
the definition. Note that for any open subset U ⊂ X, d induces a map

(21.2.18.1) Γ (U, OX ) / Γ (U, ΩX/Y ),

which we also call d, and interpret as “taking the derivative”.

21.2.19. Motivation. Before connecting this to our other definitions, let me try
to convince you that this is a reasonable definition to make. (This discussion is
informal and nonrigorous.) Say for example that Y is a point, and X a manifold.
Then the tangent bundle TX×X on X × X is pr∗1 TX ⊕ pr∗2 TX , where pr1 and pr2 are
the projections from X × X onto its two factors. Restrict this to the diagonal ∆, and
look at the normal bundle exact sequence:

0 / T∆ / TX×X |∆ / N∆/X / 0.

Now the left morphism sends v to (v, v), so the cokernel can be interpreted as
(v, −v). Thus N∆/X is isomorphic to TX . Thus we can turn this on its head: we
know how to find the normal bundle (or more precisely the conormal sheaf), and
we can use this to define the tangent bundle (or more precisely the cotangent sheaf).
 © 2024 Ravi Vakil. Published by Princeton University Press. 607

21.2.20. More motivation. If you work through Extended Example 15.3.4, you may
see many of the ideas here already faintly visible there, for example pr∗2 f − pr∗1 f,
and working in the ideal of the diagonal, modulo its square.

21.2.21. Testing this out in the affine case. Let’s now see how this works for the special
case Spec A → Spec B. Then the diagonal morphism δ : Spec A ,→ Spec A ⊗B A
corresponds to the ideal I of A ⊗B A that is the kernel of the ring map

P / P x i yi .
α: x i ⊗ yi

P I of A ⊗B A is generated
21.2.22. The ideal P by the elements of the form 1 ⊗ a − a ⊗ 1.
Reason: if α( xi ⊗ yi ) = 0, i.e., xi yi = 0, then

X X X
x i ⊗ yi = (xi ⊗ yi − xi yi ⊗ 1) = (xi ⊗ 1)(1 ⊗ yi − yi ⊗ 1).

The derivation is d : A → A ⊗B A, a 7→ (1 ⊗ a − a ⊗ 1) (an element of I, taken

modulo I2 ). (We shouldn’t really call this “d” until we have verified that it agrees
with our earlier definition, but we irresponsibly will anyway.)
Let’s check that d is indeed a derivation (§21.2.17). Clearly d is B-linear, so we
check the Leibniz rule:

d(aa ′ ) − a da ′ − a ′ da = 1 ⊗ aa ′ − aa ′ ⊗ 1 − a ⊗ a ′ + aa ′ ⊗ 1 − a ′ ⊗ a + a ′ a ⊗ 1
= −a ⊗ a ′ − a ′ ⊗ a + a ′ a ⊗ 1 + 1 ⊗ aa ′
= (1 ⊗ a − a ⊗ 1)(1 ⊗ a ′ − a ′ ⊗ 1)
∈ I2 .

Thus by the universal property of ΩA/B , we have a natural morphism ΩA/B → I/I2
of A-modules.

21.2.23. Theorem. — The natural morphism f : ΩA/B → I/I2 induced by the universal
property of ΩA/B is an isomorphism.

Proof. We will show this as follows. (i) We will show that f is surjective, and (ii)
we will describe g : I/I2 → ΩA/B such that g ◦ f : ΩA/B → ΩA/B is the identity
(showing that f is injective).
(i) The map f sends da to 1 ⊗ a − a ⊗ 1, and such elements generate I (§21.2.22),
so f is surjective.
(ii) Consider the map A ⊗B A → ΩA/B defined by x ⊗ y 7→ x dy. (This is a well-
defined map, by the universal property of ⊗, see §1.2.5.) Define g : I/I2 → ΩA/B as
the restriction of this map to I. We need to check that this is well-defined, i.e., that
elements of I2 are sent to 0, i.e., we need that

P P ′
P
( x i ⊗ yi ) xj ⊗ yj′ = i,j xi xj′ ⊗ yi yj′  /0
608 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

P P ′ ′
when i xi yi = xj yj = 0. But by the Leibniz rule,
X X X
xi xj′ d(yi yj′ ) = xi xj′ yi dyj′ + xi xj′ yj′ dyi
i,j i,j i,j
!  ! 
X X X X
= xi yi  xj′ dyj′  + xi dyi  xj′ yj′ 
i j i j

= 0.
Then g ◦ f is indeed the identity, as

da  / 1⊗a−a⊗1  / 1 da − a d1 = da
f g

as desired. □

21.2.N. E ASY EXERCISE . Suppose π : X → Y is a morphism of schemes, with open

subschemes Spec A ⊂ X and Spec B ⊂ Y, with Spec A ⊂ π−1 (Spec B). Identify
ΩX/Y |Spec A with Ω
e A/B , and identify d : Γ (Spec A, OX ) → Γ (Spec A, ΩX/Y ) with
d : A → ΩA/B . Thus the global construction indeed naturally “glues together” the
affine construction.
We can now use our understanding of how Ω works on affine open sets to
generalize previous statements to non-affine settings.

21.2.O. E XERCISE . If U ⊂ X is an open subset, show that the map (21.2.18.1) is a

derivation.

21.2.P. E XERCISE . Suppose π : X → Y is locally of finite type, and Y (and hence X) is

locally Noetherian. Show that ΩX/Y is a coherent sheaf on X. (Feel free to weaken
the Noetherian hypotheses for weaker conclusions.)

21.2.Q. E XERCISE . Suppose π : X → Y is smooth of relative dimension n. Prove that

ΩX/Y is locally free of rank n. Hint: the Jacobian description of Ω, Exercise 21.2.E.
The cotangent exact sequence and the conormal exact sequence for schemes
now directly follow.

21.2.24. Theorem. —
π / ρ
/ Z are morphisms of schemes.
(a) (cotangent exact sequence) Suppose X Y
Then there is an exact sequence of quasicoherent sheaves on X
π∗ ΩY/Z / ΩX/Z / ΩX/Y / 0,

globalizing Theorem 21.2.9.

(b) (conormal exact sequence) Suppose ρ : Y → Z is a morphism of schemes, and i : X ,→ Y
∨
is a closed embedding, with conormal sheaf NX/Y . Then there is an exact sequence of
sheaves on X:
δ / ∗ / ΩX/Z / 0,
N∨ X/Y i ΩY/Z
globalizing Theorem 21.2.12.

21.2.R. E XERCISE . Prove Theorem 21.2.24. (What needs to be checked?)

 © 2024 Ravi Vakil. Published by Princeton University Press. 609

You should expect these exact sequences to be left-exact as well in the presence
of appropriate smoothness (see Remark 21.2.10 and §21.2.13). The following is an
important special case for the cotangent exact sequence. (Theorem 21.2.31 will give
a similar statement for the conormal exact sequence, for smooth varieties.)

21.2.S. E XERCISE ( LEFT- EXACTNESS OF THE COTANGENT EXACT SEQUENCE WHEN

π AND ρ ARE SMOOTH ). Show that the cotangent exact sequence is exact on
the left if π and ρ are smooth. Hint: the (m + n) × (m + n) matrix you used in
Exercise 13.6.D is “block upper triangular”.

21.2.25. Pulling back relative differentials. Not surprisingly, the sheaf of relative
differentials pull back, and behave well under base change.

21.2.26. Theorem (pullback of differentials). —

(a) If

X′
µ
/X

 
Y′ /Y

is a commutative diagram of schemes, there is a natural homomorphism µ∗ ΩX/Y →

ΩX ′ /Y ′ of quasicoherent sheaves on X ′ . An important special case is Y = Y ′ .
(b) (Ω behaves well under base change) If furthermore the above diagram is a Cartesian
square (so X ′ =∼ X ×Y Y ′ ) then µ∗ ΩX/Y → ΩX ′ /Y ′ is an isomorphism.

21.2.T. E XERCISE . Derive Theorem 21.2.26 from Exercise 21.2.K. (Why does the
construction of Exercise 21.2.K(a) “glue well”?)
As a particular case of Theorem 21.2.26(b), the fiber of the sheaf of relative
differentials is indeed the sheaf of differentials of the fiber. Thus the sheaf of
differentials notion indeed “glues together” the differentials on each fiber.

21.2.U. E XERCISE . Suppose α : X → Z and β : Y → Z are two morphisms, and let

α ′ : X ×Z Y → Y and β ′ : X ×Z Y → X be the morphisms induced by fibered product.
∼
Describe an isomorphism ΩX×Z Y/Z ←→ (β ′ )∗ ΩX/Z ⊕ (α ′ )∗ ΩY/Z . (The right side
is often written as ΩX/Z ⊞ ΩY/Z .)

21.2.27. Important: Smoothness of varieties revisited.

Suppose k is a field. Since §13.2.3, we have used an awkward definition of
k-smoothness, and we now make our definition of k-smoothness more robust.

21.2.28. Redefinition. A k-scheme X is k-smooth of dimension n or smooth of

dimension n over k if it is locally of finite type, of pure dimension n, and ΩX/k is
locally free of rank n. The dimension n is often omitted, and one might (possibly)
want to call something smooth if it is the (scheme-theoretic) disjoint union of things
smooth of various dimensions.

21.2.V. E XERCISE . Show that Redefinition 21.2.28 of k-smoothness agrees with

Definition 13.2.4. (Hint: both can be described by the cokernel of the Jacobian.)
610 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

21.2.29. Remark. We will also revisit the definition of smooth morphism (Defini-
tion 13.6.2) before long, in Theorem 24.8.8.

21.2.W. ⋆⋆ E XERCISE ( FOR THOSE WITH BACKGROUND IN COMPLEX GEOMETRY ).

Suppose X is a complex algebraic variety. Show that the analytification Xan of X
(defined in Exercise 7.3.P) is smooth (in the differential-geometric sense) if and
only if X is smooth (in the algebro-geometric sense, over C). In this case, show that
complex dimension of the complex manifold Xan (half the real dimension) is dim X.
Hint: the Jacobian criterion applies in both settings.

21.2.30. Left-exactness of the conormal exact sequence for embeddings of smooth

varieties.
As described in §21.2.13, we expect the conormal exact sequence to be exact on
the left in appropriately “smooth” situations.

21.2.31. Theorem (conormal exact sequence for smooth varieties). — Suppose

i : X ,→ Y is a closed embedding of smooth varieties over a field k, with conormal sheaf
∨ ∨
NX/Y . Then NX/Y is locally free, and the conormal exact sequence (Theorem 21.2.24(b)) is
exact on the left:

(21.2.31.1) 0 /N∨ δ / i∗ ΩY/k / ΩX/k /0

X/Y

is exact.
By dualizing, i.e., applying Hom (·, OX ), we obtain the normal exact sequence

0 / TX/k / TY/k |X / NX/Y /0

(exact by Exercise 14.2.I) which is geometrically more intuitive (see Figure 21.3 and
the discussion after the proof of Theorem 21.2.12).

21.2.32. Remark. The conormal sequence is exact on the left in even more gen-
eral circumstances. Essentially all that is required is appropriate smoothness of
i ◦ g : X → Z, see [Stacks, tag 06B7], and [Stacks, tag 06BB] for related facts. In
particular, the smoothness of Y is irrelevant! As a consequence, Important Corol-
lary 21.2.33 below also holds for finite type Y in general.

Proof.

21.2.X. E XERCISE . Show that Theorem 21.2.31 is true if and only if it is true upon
base change to the algebraic closure k of k. Hint: Show that construction of all
the terms in (21.2.31.1) play well with algebraic field extensions. For the case of
differentials, see Theorem 21.2.26. For the case of the conormal sheaf, check that
the definition plays well with algebraic field extensions.
We may thus assume that k is algebraically closed, and thus (by Exercise 13.2.J)
closed points of smooth k-schemes are regular. Let I be the ideal sheaf of i : X ,→ Y.
By Exercise 13.2.M(b), i is a regular embedding in the open neighborhood of all
closed points, and hence everywhere. Thus by Proposition 21.2.16(b), I /I 2 is
locally free of rank r. By Exercise 6.6.G(a), the associated points of ker δ are a subset
of the associated points of I /I 2 , so to show that ker δ = 0, it suffices to check this
 © 2024 Ravi Vakil. Published by Princeton University Press. 611

at the associated points of I /I 2 , which are precisely the associated points of X (as
I /I 2 is locally free).
By Proposition 13.2.13 (smooth implies reduced), X is reduced. Thus we need
only check at the generic points of (the irreducible components of) X (as X has
no embedded points, see §6.6.2 or Exercise 6.6.R). But this involves checking the
left-exactness of a right-exact sequence of vector spaces (X reduced implies that the
stalk at the generic point is a vector space), and as the dimension of the left (the
codimension of X in Y by Proposition 21.2.16(b)) is precisely the difference of the
other two, we are done. □

21.2.33. Important Corollary. — Suppose Y is a smooth k-variety, and q ∈ Y is a closed

point whose residue field κ(q) is separable over k. (This is automatic if char k = 0 or if k
is a finite field.) Then the conormal exact sequence for q ,→ Y yields an isomorphism of the
Zariski cotangent space of Y at q with the fiber of ΩY/k at q.

Proof. Apply Theorem 21.2.31 with X = q. Note that q is indeed a smooth k-variety
(using separability of κ(q)/k!), and that Ωq/k = 0. □

21.2.34. Remark. This result generalizes Exercise 21.2.F to separable closed points if
Y is smooth, and as mentioned in Remark 21.2.32, even this smoothness condition
on Y can be removed. As described after the statement of Exercise 21.2.F, this has a
number of consequences. For example, it extends the Jacobian description of the
Zariski tangent space to separable closed points.

21.2.Y. E XERCISE ( CF. §13.2.15). Suppose p is a closed point of a k-variety X, with

residue field κ(p) that is separable over k. Define π : Xk := X ×k k → X by base
change from Spec k → Spec k. Suppose q ∈ π−1 (p). Show that Xk is regular at q if
and only if X is regular at p.

21.3 Examples

The examples below are organized by topic, not by difficulty.

21.3.1. The geometric genus of a curve. A regular irreducible projective curve

C (over a field k) has geometric genus h0 (C, ΩC/k ). (This will be generalized
to higher dimension in §21.5.3.) This is always finite, as ΩC/k is coherent (Exer-
cise 21.2.P), and coherent sheaves on projective k-schemes have finite-dimensional
spaces of sections (Theorem 18.1.3(i)). Sadly, this isn’t really a new invariant, by the
following.

21.3.A. E ASY E XERCISE . Assuming Miracle 18.5.2 (that ΩC/k is Serre-dualizing),

show that for a regular irreducible projective curve C, the geometric genus equals
the arithmetic genus, i.e., h0 (C, ΩC/k ) = h1 (C, OC ).
Now let’s compute some differentials!

21.3.2. The projective line. As an important first example, consider P1k , with
the usual projective coordinates x0 and x1 . As usual, the first patch corresponds
612 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

to x0 ̸= 0, and is of the form Spec k[x1/0 ] where x1/0 = x1 /x0 . The second patch
corresponds to x1 ̸= 0, and is of the form Spec k[x0/1 ] where x0/1 = x0 /x1 .
Both patches are isomorphic to A1k , and ΩA1k =∼ O 1 . (More precisely, Ωk[x]/k =
Ak
k[x] dx.) Thus ΩP1k is an invertible sheaf (a line bundle). The invertible sheaves on
P1k are of the form O(m). So which invertible sheaf is ΩP1 /k ?
Let’s take a section, dx1/0 on the first patch. It has no zeros or poles there,
so let’s check what happens on the other patch. As x1/0 = 1/x0/1 , we have
dx1/0 = −(1/x20/1 ) dx0/1 . Thus this section has a double pole where x0/1 = 0.
Hence ΩP1k /k = ∼ O(−2). As a consequence, we have shown that the geometric
1
genus of Pk is 0.
Note that the above argument works equally well if k were replaced by Z: our
theory of Weil divisors and line bundles of Chapter 15 applies (P1Z is factorial), so
the previous argument essentially without change shows that ΩP1Z /Z = ∼ O(−2). And
because Ω behaves well with respect to base change (Theorem 21.2.26(b)), and any
scheme maps to Spec Z, this implies that ΩP1B /B = ∼ O 1 (−2) for any base scheme B.
PB
(Also, as suggested by §18.5.2, this shows that ΩP1 /k is the dualizing sheaf for
P1k ; see also Example 18.5.4. But given that we haven’t yet proved Serre duality,
this isn’t so meaningful.)
Side Remark: the fact that the degree of the tangent bundle is 2 is related to the
“Hairy Ball Theorem” (the dimension 2 case of [Hat, Thm. 2.28]).

21.3.3. Hyperelliptic curves. Throughout this discussion of hyperelliptic curves,

we suppose that k = k and char k ̸= 2, so we may apply the discussion of §19.5.
Consider a double cover π : C → P1k by a regular projective curve C, branched over
2g+2 distinct points. We will use the explicit coordinate description of hyperelliptic
curves of (19.5.2.1). In particular, π is unbranched at 0. By Theorem 19.5.1, C has
genus g.

21.3.B. E XERCISE : DIFFERENTIALS ON HYPERELLIPTIC CURVES . What is the

degree of the invertible sheaf ΩC/k ? (Hint: let x be a coordinate on one of the
coordinate patches of P1k . Consider π∗ dx on C, and count poles and zeros. Use
the explicit coordinates of §19.5. You should find that π∗ dx has 2g + 2 zeros and 4
poles, counted with multiplicity, for a total of 2g − 2.) Doing this exercise will set
you up well for the Riemann-Hurwitz formula, in §21.4.

21.3.C. E XERCISE . Show that h0 (C, ΩC/k ) = g (and hence that the geometric genus
of C is g) as follows.
(a) Show that any regular differential ω on Spec k[x, y]/(y2 − f(x)) (i.e., an element
of Ω(k[x,y]/(y2 −f(x)))/k ) preserved by the involution y 7→ −y is pulled back from
a differential on Spec k[x]. (Hint: make sense of the statement “ω/dx is a rational
function on Spec k[x, y]/(y2 − f(x)) preserved by the involution y 7→ −y” and show
that ω/dx is a rational function in x.)
(b) Use (a) to show that any differential ω ∈ H0 (C, ΩC/k ) preserved by the involu-
tion i : y 7→ −y must be pulled back from P1 by π, and hence must be zero. Show
that every differential ω ∈ H0 (C, ΩC/k ) satisfies i∗ ω = −ω.
(c) Show that dx 2
y is a (regular) differential on Spec k[x, y]/(y − f(x)). Show that for
i
0 ≤ i < g, x (dx)/y extends to a global differential ωi on C (i.e., with no poles).
 © 2024 Ravi Vakil. Published by Princeton University Press. 613

(d) Show that the ωi (0 ≤ i < g) are linearly independent differentials, i.e., linearly
independent in the vector space H0 (C, ΩC/k ). (Hint: Let {p, q} = π−1 (0). Show
Pg−1
that the valuation of ωi at both p and q is i. If ω := j=s aj ωj is a nontrivial
linear combination of the ωi , with aj ∈ k, and as ̸= 0, show that the valuation of ω
at p is s, and hence ω ̸= 0.)
(e) Show that the ωi span the vector space of differentials H0 (C, ΩC/k ). (Hint:
if ω ∈ H0 (C, ΩC/k ) use (d) to show that there are unique ai such that ω ′ :=
Pg−1
ω − i=0 ai ωi vanishes at p to order ≥ g. By (b), ω ′ also vanishes at q to the
same order. Use Exercise 21.3.B to show that ω ′ must be zero.)
Hence ΩC/k is an invertible sheaf of degree 2g − 2 with g sections.

21.3.D. ⋆ E XERCISE ( TOWARD S ERRE DUALITY ). (You may later see this as an
example of Serre duality in action.)
(a) Show that h1 (C, ΩC/k ) = 1. Interpret a generator of H1 (C, ΩC/k ) as x−1 dx. (In
particular, the pullback map H1 (P1 , ΩP1 /k ) → H1 (C, ΩC/k ) is an isomorphism.)
(b) Describe a natural perfect pairing
H0 (C, ΩC/k ) × H1 (C, OC ) / H1 (C, ΩC/k ).

In terms of our explicit coordinates, you might interpret it as follows. Recall from
the proof of the hyperelliptic Riemann-Hurwitz formula (Theorem 19.5.1) that
H1 (C, OC ) can be interpreted as
Dy y yE
, 2,..., g .
x x x
Then the pairing

/ x−1 dx
D E
dx dx
y ,x y ,...,x
g−1 dx
y × yx , xy2 , . . . , xyg

is basically “multiply and read off the x−1 dx term”. Or in fancier terms: “multiply
and take the residue”. (You may want to compare this to Example 18.5.4.)

21.3.4. Discrete valuation rings. The following exercise is used in the proof of the
Riemann-Hurwitz formula, §21.4.

21.3.E. E XERCISE . Suppose that the discrete valuation ring (A, m, k) is a localization
of a finitely generated k-algebra. Let t be a uniformizer of A. Show that the
differentials are free of rank one and generated by dt, i.e., ΩA/k = A dt, as follows.
(It is also possible to show this using the ideas from §21.5.)
(a) Show that ΩA/k is a finitely generated A-module.
(b) Show that ⟨dt⟩ = ΩA/k , i.e., that ×dt : A → ΩA/k is a surjection, as follows.
Let π be the projection A → A/m = k, so for a ∈ A, a − π(a) ∈ m. Define
σ(a) = (a − π(a))/t. Show that ΩA/k = ⟨dt⟩ + mΩA/k , using the fact that for every
a ∈ A,
da = σ(a) dt + t dσ(a).
Apply Nakayama’s Lemma version 3 (Exercise 8.2.G) to ⟨dt⟩ ⊂ ΩA/k . (This
argument, with essentially no change, can be used to show that if (A, m, k) is a
localization of a finitely generated algebra over k, and t1 , . . . , tn generate m, then
dt1 , . . . , dtn generate ΩA/k .)
(c) By part (b), ΩA/k is a principal A-module. Show that ×dt is an injection as
614 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

follows. By the classification of finitely generated modules over discrete valuation

rings (Remark 6.4.4), it suffices to show that tm dt ̸= 0 for all m. The surjection
A → A/(tN ) induces a map ΩA/k → Ω(A/(tN ))/k , so it suffices to show that tm dt
is nonzero in Ω(A/(tN ))/k . Show that A/(tN ) = ∼ k[t]/(tN ). The usual differentiation
rule for polynomials gives a map k[t]/(tN ) → k[t]/(tN−1 ) which is a derivation
of k[t]/(tN ) over k, and tm dt will not map to 0 so long as N is sufficiently large.
Put the pieces together and complete the proof. (An extension of these ideas can
show that if (A, m, k) is a localization of a finitely generated algebra over k that is a
regular local ring, then ΩA/k is free of rank dim A.)

21.3.5. Projective space and the Euler exact sequence. We next examine the
differentials of projective space Pn n
k , or more generally PA where A is an arbitrary
ring. As projective space is covered by affine open sets of the form An , on which
the differentials form a rank n locally free sheaf, ΩPnA /A is also a rank n locally free
sheaf.

21.3.6. Theorem (the Euler exact sequence). — The sheaf of differentials ΩPnA /A
satisfies the following exact sequence

0 / ΩPn /A α / OPn (−1)⊕(n+1) ϕ

/ OPn / 0.
A A A

This is handy, because you can get a hold of ΩPnA /A in a concrete way. See
Exercise 21.5.Q for an application. By dualizing this exact sequence, we have an
exact sequence

ϕ∨
0 / OPn / OPn (1)⊕(n+1) / TPn /A / 0.
A A A

21.3.7. ⋆ Proof of Theorem 21.3.6. What is really going on in this proof is that we
consider those differentials on An+1
A \ {0} that are pullbacks of differentials on Pn
A.
For a different explanation, in terms of the Koszul complex (briefly discussed in
§24.4.1), see [E, §17.5].
We first describe a map ϕ : O(−1)⊕(n+1) → O, and later identify the kernel
with ΩPnA /A . The map is given by
 / x0 s0 + x1 s1 + · · · + xn sn .
ϕ : (s0 , s1 , . . . , sn )

(Do you see why?) You should think of this as a “degree 1” map, as each xi has
degree 1.

21.3.8. Remark. The dual ϕ∨ : O → O(1)⊕(n+1) of ϕ gives a map Pn → Pn via

Important Theorem 15.2.2 on maps to projective space. This map is the identity,
and this is one way of describing ϕ in a “natural” (coordinate-free) manner.

21.3.F. E ASY EXERCISE . Show that ϕ is surjective, by checking on the open set
D(xi ). (There is a one-line solution.)
Now we must identify the kernel of ϕ with the differentials, and we can do
this on each D(xi ), so long as we do it in a way that works simultaneously for each
open set. So we consider the open set U0 , where x0 ̸= 0, and we have the usual
 © 2024 Ravi Vakil. Published by Princeton University Press. 615

coordinates xj/0 = xj /x0 (1 ≤ j ≤ n). Given a differential

f1 (x1/0 , . . . , xn/0 ) dx1/0 + · · · + fn (x1/0 , . . . , xn/0 ) dxn/0
we must produce n + 1 sections of O(−1). As motivation, we just look at the first
term, and pretend that the projective coordinates are actual coordinates.
f1 dx1/0 = f1 d(x1 /x0 )
x0 dx1 − x1 dx0
= f1
x20
x1 f1
= − 2 f1 dx0 + dx1
x0 x0
Note that x0 times the “coefficient of dx0 ” plus x1 times the “coefficient of dx1 ” is
0, and also both coefficients are of homogeneous degree −1. Motivated by this, we
define
 
x1 xn f1 f2 fn
(21.3.8.1) α : f1 dx1/0 + · · · + fn dxn/0 7→ − 2 f1 − · · · − 2 fn , , , . . . ,
x0 x0 x0 x0 x0
Note that over U0 , this indeed gives an injection of ΩPnA /A to O(−1)⊕(n+1) that
surjects onto the kernel of O(−1)⊕(n+1) → OX (if (g0 , . . . , gn ) is in the kernel, take
fi = x0 gi for i > 0).
Let’s make sure this construction, applied to two different coordinate patches
(say U0 and U1 ) gives the same answer. (This verification is best ignored on a first
reading.) Note that
1 x2/1
f1 dx1/0 + f2 dx2/0 + · · · = f1 d + f2 d + ···
x0/1 x0/1
f1 f2 f2 x2/1
= − 2 dx0/1 + dx2/1 − 2 dx0/1 + · · ·
x0/1 x0/1 x0/1
f1 + f2 x2/1 + · · · f2
=− dx0/1 + dx2/1 + · · · .
x20/1 x0/1
Under this map, the dx2/1 term goes to the second factor (where the factors are
indexed 0 through n) in O(−1)⊕(n+1) , and yields f2 /x0 as desired (and similarly
for dxj/1 for j > 2). Also, the dx0/1 term goes to the “zero” factor, and yields
X X
n
! n
2
− fi (xi /x1 )/(x0 /x1 ) /x1 = − fi (xi /x1 )xi /x20
i=1 i=1
as desired. Finally, the “first” factor must be correct because the sum over i of xi
times the ith factor is 0.

21.3.G. E XERCISE . Finish the proof of Theorem 21.3.6, by verifying that this map
α : ΩPnA /A → OPnA (−1)⊕(n+1) identifies ΩPnA /A with ker ϕ.
□

21.3.H. E XERCISE ( PROMISED IN §18.5.7). Show that h1 (P2k , ΩP2k /k ) > 0. Show
that ΩP2k /k is not the direct sum of line bundles, and can’t even be filtered into line
bundles. Show that Theorem 18.5.6 (that all finite rank vector bundles on P1 split
into line bundles) cannot be extended to rank 2 vector bundles on P2k .
616 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

21.3.9. Generalizations of the Euler exact sequence. Generalizations of the Euler exact
sequence are quite useful. We won’t use them later, so no proofs will be given.
First, the argument generalizes readily if Spec A is replaced by an arbitrary base
scheme. The Euler exact sequence generalizes further in a number of ways. As
a first step, suppose V is a rank n + 1 locally free sheaf (or vector bundle) on a
scheme X. Suppose π : PV → X is the corresponding projective bundle. (See §17.2.3
for the definition of PV , including a caution about duals.) Then ΩPV /X sits in an
Euler exact sequence:
0 / ΩPV /X / OPV (−1) ⊗ π∗ V ∨ / OPV /0

It is not obvious that this is useful, but we have already implicitly seen it in the
case of P1 -bundles over curves, in Exercise 20.2.J, where the normal bundle to a
section was identified in this way.

21.3.10. ⋆⋆ Generalization to the Grassmannian. For another generalization, fix a base

field k, and let G(m, n + 1) be the space of sub-vector spaces of dimension m in an
(n + 1)-dimensional vector space V (the Grassmannian, §16.4). Over G(m, n + 1)
we have a short exact sequence of locally free sheaves
0 /S / OG(m,n+1) ⊗ V ∨ /Q /0

where OG(m,n+1) ⊗ V ∨ is the “trivial bundle whose fibers are V ∨ ” (do you under-
stand what that means?), and S is the “universal vector bundle” or “tautological
vector (sub)bundle”. Then there is a canonical isomorphism
∼
(21.3.10.1) ΩG(m,n+1)/k o / Hom (Q, S ).

21.3.I. E XERCISE . Recall that in the case of projective space, i.e., m = 1, S = O(−1)
(Exercise 17.1.H). Verify (21.3.10.1) in this case using the Euler exact sequence
(Theorem 21.3.6).

21.3.J. E XERCISE . Prove (21.3.10.1), and explain how it generalizes Exercise 20.2.I.
(The hint to Exercise 20.2.I may help.)
This Grassmannian fact generalizes further to Grassmannian bundles, and to
flag varieties, and to flag bundles.

21.4 The Riemann-Hurwitz Formula

The Riemann-Hurwitz formula generalizes our calculation of the genus g of
a double cover of P1 branched at 2g + 2 points, Theorem 19.5.1, to higher degree
covers, and to higher genus target curves.

21.4.1. Definition. A finite morphism between integral schemes X → Y is said

to be separable if it is dominant, and the induced extension of function fields
K(X)/K(Y) is a separable extension. Similarly, a generically finite morphism is
generically separable if it is dominant, and the induced extension of function fields
is a (finite) separable extension. Note that finite morphisms of integral schemes are
automatically separable in characteristic 0.
 © 2024 Ravi Vakil. Published by Princeton University Press. 617

21.4.2. Proposition. — If π : X → Y is a generically separable morphism of irreducible

varieties of the same dimension n, and Y is smooth, then the cotangent exact sequence
(Theorem 21.2.24) is exact on the left as well:

(21.4.2.1) 0 / π∗ ΩY/k ϕ
/ ΩX/k / ΩX/Y / 0.

This is an example of left-exactness of the cotangent exact sequence in the

presence of appropriate “smoothness”, see Remark 21.2.10.

Proof. We must check that ϕ is injective. Now ΩY/k is a rank n locally free sheaf on
Y, so π∗ ΩY/k is a rank n locally free sheaf on X. A locally free sheaf on an integral
scheme (such as π∗ ΩY/k ) is torsion-free (any section over any open set is nonzero
at the generic point, see §6.1.4), so if a subsheaf of it (such as ker ϕ) is nonzero, it
is nonzero at the generic point. Thus to show the injectivity of ϕ, we need only
check that ϕ is an inclusion at the generic point. We thus tensor with Oη where
η is the generic point of X. Tensoring with Oη is an exact functor (localization is
exact, Exercise 1.5.G), and Oη ⊗ ΩX/Y = 0 (as K(X)/K(Y) is a separable extension
by hypothesis, and Ω for separable field extensions is 0 by Exercise 21.2.H(a)). Also,
Oη ⊗ π∗ ΩY/k and Oη ⊗ ΩX/k are both n-dimensional Oη -vector spaces (they are
the stalks of rank n locally free sheaves at the generic point). Thus by considering

Oη ⊗ π∗ ΩY/k / Oη ⊗ ΩX/k / Oη ⊗ ΩX/Y /0

(which is Oη⊕n → Oη⊕n → 0 → 0) we see that Oη ⊗π∗ ΩY/k → Oη ⊗ΩX/k is injective,

and thus that π∗ ΩY/k → ΩX/k is injective. □
People not confined to characteristic 0 should note what goes wrong for non-
separable morphisms. For example, suppose k is a field of characteristic p, and
consider the map π : A1k = Spec k[t] → A1k = Spec k[u] given by u = tp . Then
Ωπ is the trivial invertible sheaf generated by dt. As another (similar but differ-
ent) example, if K = k(x) and K ′ = k(xp ), then the inclusion K ′ ,→ K induces
π : Spec K[t] → Spec K ′ [t]. Once again, Ωπ is an invertible sheaf, generated by dx
(which in this case is pulled back from ΩK/K ′ on Spec K). In both of these cases, we
have maps from one affine line to another, and there are nonzero relative tangent
vectors.

21.4.A. E XERCISE . If X and Y are smooth varieties of dimension n, and π : X → Y is

generically separable, show that the locus of non-smoothness is pure codimension
1, and has a natural interpretation as an effective divisor, as follows. Interpret ϕ
as an n × n Jacobian matrix (13.1.7.1) in appropriate local coordinates, and hence
interpret the locus where ϕ is not an isomorphism as (locally) the vanishing scheme
of the determinant of an n × n matrix.
We call the locus of non-smoothness (in X) the ramification locus, and its image
(in Y) the branch locus. Because of this exercise, we use the terms ramification
divisor and (assuming the image of the ramification divisor is indeed a divisor)
branch divisor.
Before getting down to our case of interest, dimension 1, we begin with some-
thing (literally) small but fun. Suppose π : X → Y is a surjective k-morphism from
a smooth k-scheme that contracts a subset of codimension greater than 1. More
precisely, suppose π is an isomorphism over an open subset of Y, from an open
618 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

subset U of X whose complement has codimension greater than 1. Then by Exer-

cise 21.4.A, Y cannot be smooth. (Small resolutions, to be defined in Exercise 22.4.N,
are examples of such π. In particular, you can find an example there.)
Suppose now that X and Y are dimension 1. Then the ramification locus is a
finite set (ramification points) of X, and the branch locus is a finite set (branch
points) of Y. (Figure 21.4 shows a morphism with two ramification points and one
branch point.) Now assume that k = k. We examine ΩX/Y near a point p ∈ X.

F IGURE 21.4. An example where the branch divisor appears with

multiplicity 2 (see Exercise 21.4.C)

As motivation for what we will see, we note that in complex geometry, noncon-
stant maps from (complex) curves to curves may be written in appropriate local
coordinates as x 7→ xm = y, from which we see that dy pulls back to mxm−1 dx, so
ΩX/Y locally looks like functions times dx modulo multiples of mxm−1 dx.
Consider now our map π : X → Y, and fix p ∈ X, and q = π(p). Because the
construction of Ω behaves well under base change (Theorem 21.2.26(b)), we may
replace Y with Spec of the local ring OY,q at q, i.e., we may assume Y = Spec B,
where B is a discrete valuation ring (as Y is a regular curve), with residue field
k = k corresponding to q. By restricting to an affine open neighborhood of p in
X, we assume that X is affine too. Similarly, as the construction of Ω behaves well
with respect to localization (Exercise 21.2.L), we may replace X by Spec OX,p , and
thus assume X = Spec A, where A is a discrete valuation ring, and π corresponds
to B → A, inducing an isomorphism of residue fields (with k = k).
Suppose their uniformizers are s and t respectively, with t 7→ usn where u is
an invertible element of A.
X AO usO n

 _
Y B t
Recall that the differentials of a discrete valuation ring over k are generated by d of
the uniformizer (Exercise 21.3.E). Then
dt = d(usn ) = unsn−1 ds + sn du.
This differential on Spec A vanishes to order at least n − 1, and precisely n − 1 if n is
not divisible by the characteristic. If the order is n − 1, we call it tame ramification,
while if the order is greater, it is called wild ramification. We define one plus this
order to be the the ramification index at this point of X.

21.4.B. E XERCISE . Show that the degree of ΩX/Y at p ∈ X is precisely the ramifi-
cation order of π at p, minus one. (The degree of a coherent sheaf on a curve was
 © 2024 Ravi Vakil. Published by Princeton University Press. 619

defined in §18.4.6. You will need to make sense of the length of a coherent sheaf at
a point, see for example Exercise 6.5.M.)

21.4.C. E XERCISE : INTERPRETING THE RAMIFICATION DIVISOR IN TERMS OF NUM -

BER OF PREIMAGES . For this exercise, we take the base field k to be algebraically
closed. Suppose all the ramification above q ∈ Y is tame (which is always true in
characteristic 0). Show that the degree of the branch divisor at q is deg π − |π−1 (q)|.
Thus the multiplicity of the branch divisor counts the extent to which the number
of preimages is less than the degree (see Figure 21.4).

21.4.3. Theorem (the Riemann-Hurwitz formula). — Suppose π : X → Y is a finite

separable morphism of projective regular curves, of pure degree n. Let R be the ramification
divisor. Then
(21.4.3.1) 2g(X) − 2 = n(2g(Y) − 2) + deg R.

21.4.D. E XERCISE . Prove the Riemann-Hurwitz formula. Hint: Apply the fact
that degree is additive in exact sequences (Exercise 18.4.I) to (21.4.2.1). Recall that
degrees of line bundles pull back well under finite morphisms of integral projective
curves, Exercise 18.4.G. A torsion coherent sheaf on a reduced curve (such as Ωπ )
is supported in dimension 0 (Exercise 14.3.F(b)), so χ(Ωπ ) = h0 (Ωπ ). Show that
the degree of R as a divisor is the same as its degree in the sense of h0 .
Here are some applications of the Riemann-Hurwitz formula.

21.4.4. Example. The degree of R is always even: any cover of a curve must be
branched over an even number of points (counted with appropriate multiplicity).

21.4.E. E ASY E XERCISE . Show that there is no nonconstant map from a smooth
projective irreducible genus 2 curve to a smooth projective irreducible genus 3
curve. (Hint: deg R ≥ 0.)

21.4.5. Definition. If the only connected finite étale cover of a connected scheme X
is an isomorphism, we say that X is simply connected in the étale topology. Do
you see why “simply connected” is a reasonable name for the concept? We won’t
use this terminology beyond the following brief discussion.

21.4.6. Example (“P1k is simply connected”). If k = k, the only connected unbranched

finite separable cover of P1k is the isomorphism, for the following reason. Suppose
X is connected and X → P1k has no ramification. Then X is a curve, and regular
by Exercise 24.8.F(a). (Feel free to take this for granted, since this exercise is in
the future. There will be no circularity.) Applying the Riemann-Hurwitz formula,
using that the ramification divisor is 0, we have 2 − 2gX = 2d with d ≥ 1 and
gX ≥ 0, from which d = 1 and gX = 0.

21.4.F. E XERCISE (“ IN CHARACTERISTIC 0, Ak1 IS SIMPLY CONNECTED ”). Show that

if k = k has characteristic 0, the only connected unbranched cover of A1k is itself.
(Aside: in characteristic p, this needn’t hold; Spec k[x, y]/(yp − y − x) → Spec k[x] is
such a map, as you can show yourself. Once the theory of the algebraic fundamental
group is developed, this translates to: “A1 is not simply connected in characteristic
p.” This cover is an example of an Artin-Schreier cover. Fun fact: the group Z/p acts
620 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

on this cover via the map y 7→ y + 1. This is an example of a Galois cover; you can
check that the extension of function fields is Galois.)

21.4.G. U NIMPORTANT E XERCISE . Extend Example 21.4.6 and Exercise 21.4.F,

by removing the k = k hypothesis, and changing “connected” to “geometrically
connected”.

21.4.7. Example: Lüroth’s Theorem. Continuing the notation of Theorem 21.4.3,

suppose g(X) = 0. Then from the Riemann-Hurwitz formula (21.4.3.1), g(Y) = 0.
(Otherwise, if g(Y) were at least 1, then the right side of the Riemann-Hurwitz
formula would be non-negative, and thus couldn’t be −2, which is the left side.)
Informally: the only maps from a genus 0 curve to a curve of positive genus are the
constant maps. This has a nonobvious algebraic consequence, by our identification
of covers of curves with field extensions (Theorem 16.3.3): all subfields of k(x)
containing k are of the form k(y) where y = f(x) for some f ∈ k(x).

k(x) P1
O

? 
K(C) C = P1

(It turns out that the hypotheses char k = 0 and k = k are not necessary.) This is
Lüroth’s Theorem.

21.4.H. E XERCISE . Use Lüroth’s Theorem to give new geometric solutions to

Exercises 7.5.K and 7.5.M. (These arguments will be less ad hoc, and more suitable
for generalization, than the algebraic solutions suggested in the hints to those
exercises.)

21.4.I. ⋆ E XERCISE ( GEOMETRIC GENUS EQUALS TOPOLOGICAL GENUS ). This

exercise is intended for those with some complex background, who know that the
Riemann-Hurwitz formula holds in the complex analytic category. Suppose C is
an irreducible regular projective complex curve. Show that there is an algebraic
nonconstant map π : C → P1C . Describe the corresponding map of Riemann surfaces.
Use the Riemann-Hurwitz formula to show that the algebraic notion of genus (as
computed using the branched cover π) agrees with the topological notion of genus
(using the same branched cover). (Recall that assuming Miracle 18.5.2 — that the
canonical bundle is Serre-dualizing — we know that the geometric genus equals
the arithmetic genus, Exercise 21.3.A.)

21.4.J. U NIMPORTANT E XERCISE . Suppose π : X → Y is a dominant morphism of

regular curves, and R is the ramification divisor of π. Show that ΩX (−R) = ∼ π∗ ΩY .
(This exercise is geometrically pleasant, but we won’t use it.) Hint: This says that
we can interpret the invertible sheaf π∗ ΩY over an open set U of X as precisely
those differentials on U vanishing along the ramification divisor.

21.4.8. Informal example: The degree of the discriminant of degree d polynomials in

one variable. You may be aware that there is a degree 2d − 2 polynomial in the
coefficients ad , . . . , a0 of the degree d polynomial
f(x) = ad xd + ad−1 xd−1 + · · · + a0 = 0
 © 2024 Ravi Vakil. Published by Princeton University Press. 621

that vanishes precisely when f(x) has a multiple root. For example, when d = 2, the
discriminant is a21 −4a0 a2 . We can “compute” this degree 2d−2 using the Riemann-
Hurwitz formula as follows. (You should try to make sense of the following
informal and imprecise discussion.) We work over an algebraically closed field
k of characteristic 0 for the sake of simplicity. If we take two general degree d
polynomials, g(x) and h(x), the degree of the discriminant “should be” the number
of λ ∈ k for which g(x) − λh(x) has a double root. Consider the morphism P1 → P1
given by x 7→ [g(x), h(x)]. (Here we use “affine coordinates” on the source P1 : by x
we mean [x, 1].) Then this morphism has degree d. A ramification point a mapping
to the branch point [λ, 1] in the target P1 corresponds to a being a double root of
g(x) − λh(x). Thus the number of branch points should be the desired degree of
the discriminant. By the Riemann-Hurwitz formula there are 2d − 2 branch points
(admittedly, with multiplicity). It is possible to turn this into a proof, and it is
interesting to do so.

21.4.9. ⋆ Bounds on automorphism groups of curves.

It is a nontrivial fact that irreducible smooth projective curves of genus g ≥ 2
have finite automorphism groups. (See [FK, §5.1] for a proof over C. See [Schm] for
the first proof in arbitrary characteristic, although better approaches are now avail-
able.) Granting this fact, we can show that in characteristic 0, the automorphism
group has order at most 84(g − 1) (Hurwitz’s Automorphisms Theorem), as follows.
Suppose C is an irreducible smooth projective curve over an algebraically
closed field k = k of characteristic 0, of genus g ≥ 2. Suppose that G is a finite group
of automorphisms of C. We now show that |G| ≤ 84(g − 1). (The case where k is
not algebraically closed is quickly dispatched by base-changing to k.)

21.4.K. E XERCISE .
(a) Let C ′ be the smooth projective curve corresponding to the field extension
K(C)G of k (via Theorem 16.3.3). (K(C)G means the G-invariants of K(C).) Describe
a morphism π : C → C ′ of degree |G|, as well as a faithful G-action on C that
commutes with π.
(b) Show that above each branch point of π, the preimages are all ramified to the
same order (as G acts transitively on them). Suppose there are n branch points and
the ith one has ramification ri (each |G|/ri times).
(c) Use the Riemann-Hurwitz formula to show that
X
n
!
ri − 1
(2g − 2) = |G| 2g(C ) − 2 +
′
ri
i=1

To maximize |G|, we wish to minimize

X
n
ri − 1
(21.4.9.1) 2g(C ′ ) − 2 +
ri
i=1

subject to (21.4.9.1) being positive. Note that 1/42 is possible: take g(C ′ ) = 0, n = 3,
and (r1 , r2 , r3 ) = (2, 3, 7).

21.4.L. E XERCISE . Show that you can’t do better than 1/42 by considering the
following cases separately:
(a) g(C ′ ) > 1,
622 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

(b) g(C ′ ) = 1,
(c) g(C ′ ) = 0 and n ≥ 5,
(d) g(C ′ ) = 0 and n = 4, and
(e) g(C ′ ) = 0 and n = 3.

21.4.M. E XERCISE . Use the fact that (21.4.9.1) is at least 1/42 to prove the result.

21.4.10. Remark. In positive characteristic, there can be many more automorphisms.

For example, [Stic], in characteristic p, if pn is a prime power that is not 2, then
n n
the completion (“compactification”) of the affine curve yp = x + xp +1 has genus
g = p (p − 1)/2 and automorphism group of order p (p + 1)(p2n − 1).
n n 3n 3n

21.5 Understanding smooth varieties using their cotangent bundles

In this section, we construct birational invariants of varieties over algebraically

closed fields (such as the geometric genus), motivate the notion of an unramified
morphism, show that varieties are “smooth almost everywhere”, and get a first
glimpse of Hodge theory.

21.5.1. The geometric genus, and other birational invariants from i-forms.
Suppose X is a projective scheme over k. Then h0 (X, ΩX/k ) is an invariant of X.
In fact it, and related invariants, are birational invariants if X is smooth, as shown
in the following Exercise 21.5.A. We first define the sheaf of (relative) i-forms
ΩiX/Y := ∧i ΩX/Y . Sections of ΩiX/Y (over some open set) are called (relative)
i-forms (over that open set).

21.5.2. Joke. Old Macdonald had a form; ei ∧ ei = 0.

21.5.A. E XERCISE (h0 (X, ΩiX/k ) ARE BIRATIONAL INVARIANTS ). Suppose X and
X ′ are birational irreducible smooth projective k-varieties. Show (for each i) that
H0 (X, ΩiX/k ) = ∼ H0 (X ′ , Ωi ′ ). Hint: fix a birational map ϕ : X 99K X ′ . By Exer-
X /k
cise 15.3.B, the complement of the domain of definition U of ϕ is codimension at
least 2. By pulling back i-forms from X ′ to U, we get a map ϕ∗ : H0 (X ′ , ΩiX ′ /k ) →
H0 (U, ΩiX/k ). Use Algebraic Hartogs’s Lemma 13.5.19 and the fact that Ωi is lo-
cally free to show the map extends to a map ϕ∗ : H0 (X ′ , ΩiX ′ /k ) → H0 (X, ΩiX/k ). If
ψ : X ′ 99K X is the inverse rational map, we similarly get a map ψ∗ : H0 (X, ΩiX/k ) →
H0 (X ′ , ΩiX ′ /k ). Show that ϕ∗ and ψ∗ are inverses by showing that each composition
is the identity on a dense open subset of X or X ′ .

21.5.3. The canonical bundle KX and the geometric genus pg (X). If X is a dimension
n smooth k-variety, the invertible sheaf (or line bundle) det ΩX/k = Ωn X/k (the
sheaf of “algebraic volume forms”) has particular importance, and is called the
canonical (invertible) sheaf, or the canonical (line) bundle. It is denoted KX (or
KX/k ). As mentioned in §18.5.2, if X is projective, then KX is the dualizing sheaf
ωX appearing in the statement of Serre duality, something we will establish in §29.4
(see Desideratum 29.1.1).
 © 2024 Ravi Vakil. Published by Princeton University Press. 623

21.5.B. E XERCISE ( THE ADJUNCTION FORMULA FOR KX ). Suppose X is a smooth

variety, and Z is a smooth subvariety of X. Show that
KZ =∼ KX |Z ⊗ det NZ/X .
(Hint: apply Exercise 14.2.P to Theorem 21.2.31.) In particular, by Exercise 21.2.J, if
Z is codimension 1, then
KZ =∼ (KX ⊗ OX (Z)) |Z ,
which is more commonly and compactly written as KX (Z)|Z . The adjunction
formula is often used inductively, for complete intersections; see Exercise 21.5.G for
an example. (See Exercise 20.2.B(a) and §29.4 for other versions of the adjunction
formula.)

21.5.4. Definition. If X is a projective (or even proper) smooth k-variety, the

birational invariant h0 (X, KX ) = h0 (X, Ωn X/k ) has particular importance. It is
called the geometric genus, and is denoted pg (X). We saw this in the case of curves
in §21.3.1. If X is an irreducible variety that is not smooth or projective, the phrase
geometric genus refers to h0 (X ′ , KX ′ ) for some smooth projective X ′ birational to X.
(By Exercise 21.5.A, this is independent of the choice of X ′ .) For example, if X is an
irreducible reduced projective curve over k, the geometric genus is the geometric
genus of the normalization of X. (But in higher dimension, it is not clear if there
exists such an X ′ . It is a nontrivial fact that this is true in characteristic 0, and it is
not yet known in positive characteristic; see Remark 22.4.6.)
It is a miracle that for a complex curve the geometric genus is the same as
the topological genus and the arithmetic genus. Exercise 18.5.A showed that
the geometric and arithmetic genus are equal (assuming Miracle 18.5.2 that the
canonical bundle is Serre-dualizing). We will connect the geometric genus to
the topological genus in our discussion of the Riemann-Hurwitz formula soon
(Exercise 21.4.I).

21.5.C. E XERCISE . Suppose Z is a regular degree d surface in Pk3 . Compute the

geometric genus pg (Z) of Z. Show that no regular quartic surface in P3k is rational
(i.e., birational to P2k , Definition 7.5.4). (Such quartic surfaces are examples of K3
surfaces, see Exercise 21.5.I.)

21.5.5. Important classes of varieties: Fano, Calabi-Yau, general type.

Suppose X is a smooth projective k-variety. Then X is said to be Fano if KX∨
is ample, and Calabi-Yau if KX = ∼ OX . (Caution: there are other definitions of
Calabi-Yau.)

21.5.D. E XERCISE . For all j ≥ 0, the jth plurigenus of a smooth projective k-variety
is h0 (X, KX⊗j ). Show that the jth plurigenus is a birational invariant.
(By contrast, h0 (X, K ⊗j ) is not a birational invariant if j < 0. For example, you
can show that h0 (X, K ∨ ) differs for X = P2k and X = Blp P2k .)
The Kodaira dimension of X, denoted κ(X), tracks the rate of growth of the
plurigenera. It is the smallest k such that h0 (X, KX⊗j )/jk is bounded (as j varies
through the positive integers), except that if all the plurigenera h0 (X, KX⊗j ) are 0,
we say κ(X) = −1. It is a nontrivial fact that the Kodaira dimension always exists,
and that it is the maximum of the dimensions of images of the jth “pluricanonical
624 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

rational maps”. The latter fact implies that the Kodaira dimension is an integer
between −1 and dim X inclusive. Exercise 21.5.D shows that the Kodaira dimension
is a birational invariant. If κ(X) = dim X, we say that X is of general type. (For
more information on the Kodaira dimension, see [Ii, §10.5].)

21.5.E. E XERCISE . Show that if KX is ample then X is of general type.

21.5.F. E XERCISE . Show that a smooth geometrically irreducible projective curve

(over a field k) is Fano (resp., Calabi-Yau, general type) if its genus is 0 (resp., 1,
greater than 1).
The “trichotomy” of Exercise 21.5.F is morally the reason that curves behave
differently depending on which of the three classes they lie in (see §19.6.1). In some
sense, important aspects of this trichotomy extend to higher dimension, which is
part of the reason for making these definitions. We now explore this trichotomy in
the case of complete intersections.

21.5.G. E XERCISE . Suppose X is a smooth complete intersection in PN k of hypersur-

faces of degree d1 , . . . , dn , where k is algebraically closed.
(a) Show that KX = ∼ O(−N − 1 + d1 + · · · + dn )|X .
(b) Show that if −N − 1 + d1 + · · · + dn is negative (resp., zero, positive) then X is
Fano (resp., Calabi-Yau, general type).
(c) Find all possible values of N and d1 , . . . , dn where X is Calabi-Yau of dimension
at most 3. Notice how small this list is.

21.5.H. E ASY EXERCISE . A smooth Fano surface is called a del Pezzo surface (part
of the Enriques-Kodaira clasification of surfaces). Show that the following surfaces
are del Pezzo: (a) P2k ; (b) P1k × P1k ; and (c) smooth surfaces in P3k of degree at most 3.
(This list has some redundancy! The remaining examples of del Pezzo surfaces are
described in Remark 27.3.7. One of them appeared earlier in §19.8.3.)

21.5.I. E XERCISE . A K3 surface over a field k is a proper smooth geometrically

connected Calabi-Yau surface X over k such that H1 (X, OX ) = 0. (Weil has written
that K3 surfaces were named in honor of Kummer, Kähler, and Kodaira, and the
mountain K2, [BPV, p. 288].) Prove that the dimension 2 (smooth) Calabi-Yau
complete intersections of Exercise 21.5.G(c) are all K3 surfaces.

21.5.6. Tantalizing side remark. If you compare the degrees of the hypersurfaces
cutting out complete intersection K3 surfaces (in Exercise 21.5.G(c)), with the
degrees of hypersurfaces cutting out complete intersection canonical curves (see
§19.8.2 and Exercise 19.8.E), you will notice a remarkable coincidence. Of course
this is not a coincidence at all.

21.5.J. E XERCISE ( NOT EVERYTHING FITS INTO THIS TRICHOTOMY ). Suppose C is a

smooth projective irreducible complex curve of genus greater than 1. Show that
C ×C P1C is neither Fano, nor Calabi-Yau, nor general type. Hint: Exercise 21.2.U.

21.5.7. ⋆ Kodaira vanishing.

The Kodaira vanishing theorem is an important tool in a number of different
areas of algebraic geometry. We state it for the sake of culture, but do not prove it
(and hence will not use it).
 © 2024 Ravi Vakil. Published by Princeton University Press. 625

21.5.8. The Kodaira Vanishing Theorem. — Suppose k is a field of characteristic 0,

and X is a smooth projective k-variety. Then for any ample invertible sheaf L , Hi (X, KX ⊗
L ) = 0 for i > 0.

21.5.9. The restriction on characteristic is necessary; Raynaud gave an example

where Kodaira vanishing fails in positive characteristic, [Ra]. The original proof is
by proving the complex-analytic version, and then transferring it into a complex
algebraic statement using Serre’s GAGA Theorem [Se3]. The general characteristic
0 case can be reduced to C — any reduction of this sort is often called (somewhat
vaguely) an application of the Lefschetz principle. (For a precise formulation of the
Lefschetz principle, with a short proof which applies in any characteristic see [Ek].
See [FR] for a generalization. See [MO90551] for more context.) Raynaud later gave
a dramatic algebraic proof of Kodaira vanishing using positive characteristic (!), see
[DeI] or [Il].

21.5.K. E XERCISE . Prove the Kodaira Vanishing Theorem in the case dim X = 1,
and for Pnk . (Neither case requires the characteristic 0 hypothesis.)

21.5.10. ⋆ A first glimpse of Hodge theory.

The invariant hj (X, ΩiX/k ) is called the Hodge number hi,j (X). By Exercise 21.5.A,
hi,0 are birational invariants. We will soon see (in Exercise 21.5.O) that this isn’t
true for all hi,j .

21.5.L. E XERCISE . Suppose X is a smooth projective variety. Assuming Mira-

cle 18.5.2 (that the canonical bundle is Serre-dualizing), show that Hodge numbers
satisfy the symmetry hp,q = hn−p,n−q . (Exercise 14.2.O will be useful.)

21.5.M. E XERCISE ( THE H ODGE NUMBERS OF PROJECTIVE SPACE ). Show that

hp,q (Pn
k ) = 1 if 0 ≤ p = q ≤ n and h
p,q n
(Pk ) = 0 otherwise. Hint: use the Euler
exact sequence (Theorem 21.3.6) and apply Exercise 14.2.N.

21.5.11. Remark: The Hodge diamond. Over k = C, further miracles occur. If X is an

irreducible smooth projective complex variety, then it turns out that there is a direct
sum decomposition

(21.5.11.1) Hm (X, C) = ⊕i+j=m Hj (X, ΩiX/C ),

P
from which hm (X, C) = i+j=m hi,j , so the Hodge numbers (purely algebraic
objects) yield the Betti numbers (a priori topological information). Moreover,
complex conjugation interchanges Hj (X, ΩiX/C ) with Hi (X, ΩjX/C ), from which

(21.5.11.2) hi,j = hj,i .

(Aside: This additional symmetry holds in characteristic 0 in general, but can fail in
positive characteristic, see for example [Ig, p. 966], [Mu1, §II], [Se4, Prop. 16].) This
is the beginning of the vast and rich subject of Hodge theory (see [GH1, §0.6] for
more, or [Vo] for much more).
If we write the Hodge numbers in a diamond, with hi,j the ith entry in the
(i + j)th row, then the diamond has the two symmetries coming from Serre duality
and complex conjugation. For example, the Hodge diamond of an irreducible
smooth projective complex surface will be of the following form:
626 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

1
q q
pg h1,1 pg
q q
1
where pg is the geometric genus of the surface, and q = h0,1 = h1,0 = h2,1 = h1,2
is called the irregularity of the surface. As another example, by Exercise 21.5.M,
the Hodge diamond of Pn is all 0 except for 1’s down the vertical axis of symmetry.
You won’t need the unproved statements (21.5.11.1) or (21.5.11.2) to solve the
following problems.

21.5.N. E XERCISE . Assuming Miracle 18.5.2 (that the canonical bundle is Serre-
dualizing), show that the Hodge diamond of a smooth projective geometrically
irreducible genus g curve over a field k is the following.
1
g g
1

21.5.O. E XERCISE . Show that the Hodge diamond of P1k × P1k is the following.
1
0 0
0 2 0
0 0
1
By comparing your answer to the Hodge diamond of P2k (Exercise 21.5.M), show
that h1,1 is not a birational invariant.
Notice that in both cases, h1,1 is the Picard number ρ (defined in §18.4.12). In
general, ρ ≤ h1,1 (see [GH1, §3.5, p. 456-7] in characteristic 0).

21.5.12. ⋆ Aside: Infinitesimal deformations and automorphisms.

It is beyond the scope of this book to make this precise, but if X is a vari-
ety, H0 (X, TX ) parametrizes infinitesimal automorphisms of X, and H1 (X, TX )
parametrizes infinitesimal deformations. As an example if X = P1 (over a field),
TP1 = ∼ O(2) (§21.3.2), so h0 (P1 , TP1 ) = 3, which is precisely the dimension of the
automorphism group of P1 (Exercise 15.5.A).

k , TPk ) using the Euler exact sequence (Theo-

21.5.P. E XERCISE . Compute h0 (Pn n

rem 21.3.6). Compare this to the dimension of the automorphism group of Pn k

(Exercise 15.5.A).

A , TPA ) = 0. (Thus projective space can’t

21.5.Q. E XERCISE . Show that H1 (Pn n

deform, and is “rigid”.)

21.5.R. E XERCISE . Assuming Miracle 18.5.2 (that the canonical bundle is Serre-
dualizing), compute hi (C, TC ) for a genus g smooth projective geometrically irre-
ducible curve over k, for i = 0 and 1. You should notice that h1 (C, TC ) for genus 0,
1, and g > 1 is 0, 1, and 3g − 3 respectively; after doing this, re-read §19.8.4.
 © 2024 Ravi Vakil. Published by Princeton University Press. 627

21.6 Generic smoothness, and consequences

We can now verify something you may already have intuited. In positive
characteristic, this is a hard theorem, in that it uses a result from commutative
algebra that we have not proved.

21.6.1. Theorem (generic smoothness of varieties). — If X is an integral finite type

k-scheme over a perfect field k of dimension n, there is a dense open subset U of X such that
U is smooth (over k) of dimension n.
Hence, by Exercise 13.2.R combined with Fact 13.8.2 (localization of regular is
regular), U is regular. Theorem 21.6.4 will generalize this to smooth morphisms, at
the expense of restricting to characteristic 0.
See §13.2.9 for an example of what goes wrong if k is not perfect.

Proof. The n = 0 case is immediate, so we assume n > 0.

We will show that the rank at the generic point is n. Then by upper semicon-
tinuity of the rank of a coherent sheaf (Exercise 14.3.J), it must be n in an open
neighborhood of the generic point, and we are done.
We thus have to check that if K is the fraction field of a dimension n integral
finite type k-scheme, i.e., (by Theorem 12.2.1) if K/k is a transcendence degree n
extension, then ΩK/k is an n-dimensional vector space. But every finitely generated
extension of a perfect field is separably generated (Theorem 10.5.17), so the result
follows by Exercise 21.2.M(b). □

21.6.2. Proof of Theorem 13.8.3, that the localization of regular local rings are
regular in the case of varieties over perfect fields. Theorem 21.6.1 allows us to
prove Theorem 13.8.3 (an important case of Fact 13.8.2).
Suppose Y is a variety over a perfect field k that is regular at its closed points
(so Y is smooth, by Exercise 13.2.R), and let η be a point of Y. We will show that Y is
regular at η. Let X = η. By Theorem 21.6.1, X contains a dense (=nonempty) open
subset of smooth points. By shrinking Y by discarding the points of X outside that
open subset, we may assume X is smooth.
Then Theorem 21.2.31 (exactness of the conormal exact sequence for smooth
varieties) implies that I /I 2 is a locally free sheaf of rank codimX/Y = dim OY,η .

21.6.A. E XERCISE . Let m be the maximal ideal of OY,η . Identify the stalk of I /I 2
at the generic point η of X with m/m2 . Conclude the proof of Theorem 13.8.3. □

21.6.B. T RICKY E XERCISE ( LOCALIZATION OF REGULAR LOCAL RINGS OF VARIETIES

ARE REGULAR , PROMISED JUST AFTER T HEOREM 13.8.3). Suppose (A, m) is a
regular local ring that is the localization of a finitely generated k-algebra, where k
is perfect. Show that the localization of A at a prime is also a regular local ring.

21.6.3. Generic smoothness for morphisms, and the Kleiman-Bertini Theorem.

We will now discuss a number of important results that fall under the rubric of
“generic smoothness for morphisms”. All require working in characteristic 0.

21.6.4. Theorem (generic smoothness on the source). — (a) Suppose π : X → Y

is a dominant finite type morphism of integral Noetherian schemes, such that char K(Y)
628 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

(= char K(X)) is zero. Then there is a nonempty (=dense) open set U ⊂ X such that π|U is
smooth of relative dimension trdeg K(X)/K(Y).
(b) Let k be a field of characteristic 0, and let π : X → Y be a dominant morphism of integral
k-varieties. Then there is a nonempty (=dense) open set U ⊂ X such that π|U is smooth of
relative dimension dim X − dim Y.

21.6.C. E XERCISE ( PROMISED IN E XERCISE 12.2.D). Suppose f1 , . . . , fn ∈ k[x1 , . . . , xn ],

where k is a field of characteristic 0. Show that f1 , . . . , fn are algebraically depen-
dent if and only if the determinant of the Jacobian matrix is identically zero. Hint:
Why is this exercise given just after Theorem 21.6.4?

21.6.5. Example. Theorem 21.6.4 fails in positive characteristic: consider the purely
inseparable extension Fp (t)/Fp (tp ). The same problem can arise even over an
algebraically closed field of characteristic p: consider A1k = Spec k[t] → Spec k[u] =
A1k , given by u 7→ tp .

Proof. Let n = trdeg K(X)/K(Y). We may replace Y by an affine open subset Spec B,
and then replace X by an affine open subset Spec A, so π corresponds to the ring
map B → A. Choose n elements x1 , . . . , xn ∈ A that form a transcendence basis for
K(A)/K(B). (Do you see why we may choose them to lie in A?) Choose additional
elements xn+1 , . . . , xN ∈ A so that x1 , . . . , xN generate A as a B-algebra.
For i = 0, . . . , n, let Fi be the subfield of K(A) generated over K(B) by x1 , . . . , xi .
For example, F0 = K(B), FN = K(A), and Fi /K(B) is a purely transcendental field
extension for i ≤ n. For i = n + 1, . . . , N, the minimal polynomial for xi ∈ K(A)
over Fi−1 (an element of Fi−1 [t]) may be interpreted as mi (x1 , . . . , xi−1 , t) for some
mi ∈ K(B)[y1 , . . . , yi−1 , t]. By multiplying mi by the products of the denominators
of its coefficients, we obtain Mi ∈ B[y1 , . . . , yi−1 , t] so that Mi (x1 , . . . , xi−1 , t) ∈
Fi−1 [t] is also a minimal polynomial for xi ∈ K(A) over Fi−1 . Thus

(21.6.5.1) K(A) = K(B)(x1 , . . . , xn )[yn+1 , . . . , yN ]/I,

where

I = (Mn+1 (x1 , . . . , xn , yn+1 ), Mn+2 (x1 , . . . , xn , yn+1 , yn+2 ), . . . , MN (x1 , . . . , xn , yn+1 , . . . , yN )).

Define A ′ := B[y1 , . . . , yN ]/(Mn+1 (y1 , . . . , yn+1 ), . . . , MN (y1 , . . . , yN )).

Let X ′ := Spec A ′ . Let Z1 , . . . , Zs be the irreducible components of X ′ . (There
are finitely many by Noetherianity of B: Exercise 3.6.T and Proposition 3.6.15.)
Precisely one of them, say Z = Z1 , dominates Spec B (because A ′ ⊗B K(B) = ∼ K(A)
by (21.6.5.1)). Let U ⊂ X be the open subset Z1 \ Z1 ∩ (Z2 ∪ · · · Zs ).
′ ′

At the generic point of U ′ , the Jacobian matrix of Mn+1 (xn+1 ), . . . , MN (xN )

with respect to x1 , . . . , xN has corank n. (Here we use the separability of the minimal
polynomials Mi .) Thus there is an open subset U ′′ ⊂ U ′ where the Jacobian has
corank n, so U ′′ → Spec B is smooth of relative dimension n. Furthermore U ′′ is
irreducible. Now U ′′ and X are birational, so by Proposition 7.5.5 there is an open
subset U ⊂ X that is isomorphic to an open subset of U ′′ . This is the open subset
we seek.

21.6.D. E XERCISE . Prove Theorem 21.6.4 (b) (using part (a)). □

If furthermore X is smooth, the situation is even better.
 © 2024 Ravi Vakil. Published by Princeton University Press. 629

21.6.6. Theorem (generic smoothness on the target). — Suppose π : X → Y is a

morphism of k-varieties, where char k = 0, and X is smooth (over k). Then there is a dense
open subset U of Y such that π|π−1 (U) is a smooth morphism.
Note that π−1 (U) may be empty! Indeed, if π is not dominant, we will have to
take such a U.
To prove Theorem 21.6.6, we use a neat trick. Suppose π : X → Y is a morphism
of schemes that are finite type over k, where char k = 0. Define

Xr = p ∈ X : rank π∗ ΩY/k → ΩX/k

p
≤r .

21.6.E. E XERCISE . Show that Xr is a constructible subset of X. Hint: If X and

Y are both smooth, show that the rank condition implies that Xr is cut out by
“determinantal equations”, or use an appropriate form of upper semicontinuity.
By Chevalley’s Theorem 8.4.2, π(Xr ) is constructible, so we can make sense of
dim π(Xr ).

21.6.7. Lemma. — If π : X → Y is a morphism of schemes that are finite type over k,

where char k = 0, then dim π(Xr ) ≤ r.

Proof. In this proof, we make repeated use of the identification of Zariski cotangent
spaces at closed points with the fibers of cotangent sheaves, using Corollary 21.2.33.
We can replace X by an irreducible component of Xr , and Y by the closure of
that component’s image in Y (with reduced subscheme structure, see §9.4.9). The
resulting map will have all of X contained in Xr , so we may as well assume X =
Xr . This boils down to the following linear algebra observation: if a linear map
ρ : V1 → V2 has rank at most r, and Vi′ is a quotient of Vi , with ρ sending V1′ to
V2′ , then the restriction of ρ to V1′ has rank at most that of ρ itself. Thus we have a
dominant morphism π : X → Y, and we wish to show that dim Y ≤ r. By generic
smoothness of varieties (Theorem 21.6.1), Y has a dense open set that is smooth,
so we may assume that Y is smooth. By generic smoothness on the source for
π : X → Y, there is a nonempty open subset U ⊂ X such that π : U → Y is smooth.
But then for any point p ∈ U, the cotangent map π∗ ΩY |π(p) → ΩX |p is injective
(Exercise 21.2.S), and has rank at most r. Taking p to be a closed point, we have
dim Y = dimπ(p) Y ≤ dim ΩY |π(p) ≤ r. □
There is not much left to do to prove the theorem.

21.6.8. Proof of Theorem 21.6.6. Reduce to the case where Y is smooth over k (by
restricting to a smaller open set, using generic smoothness of Y, Theorem 21.6.4(b)).
Say n = dim Y. Now dim π(Xn−1 ) ≤ n − 1 by Lemma 21.6.7 , so remove π(Xn−1 )
from Y as well. Then the rank of π∗ ΩY/k → ΩX/k is at least n for each closed
point of X. But as Y is regular of dimension n, we have that π∗ ΩY/k → ΩX/k is
injective for every closed point of X, and hence π∗ ΩY/k → ΩX/k is an injective map
of sheaves (Do you see why?). Thus π is smooth by Exercise 24.8.G(b). □

21.6.9. ⋆⋆ The Kleiman-Bertini Theorem. The same idea of bounding the dimen-
sion of some “bad locus” can be used to prove the Kleiman-Bertini Theorem 21.6.10
630 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

(due to Kleiman), which is useful in (for example) enumerative geometry. Through-

out this discussion k = k, although the definitions and results can be generalized.
Suppose G is a group variety (over k = k), and we have a G-action on a variety
X. We say that the action is transitive if it is transitive on closed points. A better
definition (that you can show is equivalent) is that the action morphism G × X → X
restricted to a fiber above any closed point of X is surjective. A variety X with a
transitive G-action is said to be a homogeneous space for G. For example, G acts
on itself by left-translation, and via this action G is a homogeneous space for itself.

21.6.F. E ASY E XERCISE . Suppose G is a group variety over an algebraically closed

field k of characteristic 0. Show that every homogeneous space X for G is smooth.
(In particular, taking X = G, we see that G is smooth.) Hint: X has a dense open set
U that is smooth by Theorem 21.6.4(b), and G acts transitively on the closed points
of X, so we can cover X with translates of U.

21.6.10. The Kleiman-Bertini Theorem, [Kl2, Thm. 2]. — Suppose X is homogeneous

space for a group variety G (over a field k = k of characteristic 0). Suppose α : Y → X and
β : Z → X are morphisms from smooth k-varieties Y and Z.
(a) Then there is a nonempty open subset V ⊂ G such that for every σ ∈ V(k), Y ×X Z
defined by
Y ×X Z /Z

β
 
Y
σ◦α /X
(Y is “translated by σ”) is smooth of dimension dim Y + dim Z − dim X (but possibly
empty).
(b) Furthermore, there is a nonempty open subset V ⊂ G such that

(21.6.10.1) (G ×k Y) ×X Z /G

is a smooth morphism of relative dimension dim Y + dim Z − dim X over V.

The first time you hear this, you should think of the special case where Y → X
and Z → X are locally closed embeddings (Y and Z are smooth subvarieties of X).
In this case, the Kleiman-Bertini theorem says that the second subvariety will meet
a “general translate” of the first “transversely”. (There is no such thing as a smooth
morphism of negative relative dimension. So what happens if dim Y+dim Z−dim X
is negative?)

Proof. It is more pleasant to describe this proof “backwards”, by considering how

we would prove it ourselves. We will use generic smoothness twice.
Clearly (b) implies (a), so we prove (b).
In order to show that the morphism (21.6.10.1) is smooth over a nonempty
open set V ⊂ G, it would suffice to apply generic smoothness on the target (Theo-
rem 21.6.6) to (21.6.10.1). Hence it suffices to show that (G ×k Y) ×X Z is a smooth
k-variety. Now Z is smooth over k, so it suffices to show that (G ×k Y) ×X Z → Z
is a smooth morphism (as the composition of two smooth morphisms is smooth,
Exercise 13.6.D). But this is obtained by base change from G ×k Y → X, so it suffices
to show that this latter morphism is smooth (as smoothness is preserved by base
change).
 © 2024 Ravi Vakil. Published by Princeton University Press. 631

Now G ×k Y → X is a G-equivariant morphism. (By “G-equivariant”, we mean

that the G-action on both sides respects the morphism.) By generic smoothness of
the target (Theorem 21.6.6), this is smooth over a dense open subset X. But then by
transitivity of the G-action on X, this morphism is smooth everywhere.

21.6.G. E XERCISE . Refine the above argument to show the desired statement about
relative dimension.
□

21.6.H. E XERCISE ( POOR MAN ’ S K LEIMAN -B ERTINI ). Prove Theorem 21.6.10(a)

without the hypotheses on k (on algebraic closure or characteristic), and without
the smoothness in the conclusion. Hint: This is a question about dimensions of
fibers of morphisms, so you could have solved this after reading §12.4.

21.6.I. E XERCISE ( IMPROVED CHARACTERISTIC 0 B ERTINI ). Suppose Z is a smooth

k-variety, where char k = 0 and k = k. Let V be a finite-dimensional base-point-free
linear series on Z, i.e., a finite vector space of sections of some invertible sheaf L
on Z. Show that a general element of V, considered as a closed subscheme of Z, is
regular. (More explicitly: each element s ∈ V gives a closed subscheme of Z. Then
for a general s, considered as a point of PV, the corresponding closed subscheme
is smooth over k.) Hint: figure out what this has to do with the Kleiman-Bertini
Theorem 21.6.10. Let n = dim V, G = GL(V), X = PV ∨ , take Z in Kleiman-Bertini
to be the Z of the problem, and let Y be the “universal hyperplane” over PV ∨ (the
incidence variety I ⊂ PV × PV ∨ of Definition 13.4.1).

21.6.J. E ASY E XERCISE . Interpret Bertini’s Theorem 13.4.2 over a characteristic 0

field as a corollary of Exercise 21.6.I.
In characteristic 0, Exercise 21.6.I is a good improvement on Bertini’s Theorem.
For example, we don’t need L to be very ample, or X to be projective. But unlike
Bertini’s Theorem, Exercise 21.6.I fails in positive characteristic, as shown by the
one-dimensional linear series {pQ : Q ∈ P1 }. This is essentially Example 21.6.5.
(Do you see why this does not contradict Bertini’s Theorem 13.4.2?)

21.7 Unramified morphisms

Suppose π : X → Y is a morphism of schemes. The support of the quasicoherent

sheaf Ωπ = ΩX/Y is called the ramification locus, and the image of its support,
π(Supp ΩX/Y ), is called the branch locus. If Ωπ = 0, we say that π is formally
unramified, and if π is also furthermore locally of finite type, we say π is unramified.
(Caution: there is some lack of consensus in the definition of “unramified”; “locally
of finite type” is sometimes replaced by “locally of finite presentation”, which was
the definition originally used in [Gr-EGA].)

21.7.A. E ASY E XERCISE ( EXAMPLES OF UNRAMIFIED MORPHISMS ).

(a) Show that locally closed embeddings are unramified.
(b) Show that if S is a multiplicative subset of the ring B, then Spec S−1 B → Spec B
is formally unramified. (Thus if η is the generic point of an integral scheme Y, then
632 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Spec OY,η → Y is formally unramified.)

(c) Show that finite separable field extensions (or more correctly, the corresponding
maps of schemes) are unramified.

21.7.B. E XERCISE ( PRACTICE WITH THE CONCEPT ).

(a) Show that the normalization of the node in Exercise 10.7.E (see Figure 8.4) is
unramified.
(b) Show that the normalization of the cusp in Exercise 10.7.F (see Figure 10.5) is
not unramified.

21.7.C. E ASY E XERCISE ( UNRAMIFIED MORPHISMS ARE A “ REASONABLE ” CLASS

OF MORPHISMS ). Show that the class of unramified morphisms is “reasonable” in
the sense of §8.1.

21.7.D. E XERCISE ( CHARACTERIZATIONS OF UNRAMIFIED MORPHISMS BY THEIR

FIBERS ). Suppose π : X → Y is locally of finite type.
(a) Show that π is unramified if and only if for each q ∈ Y, π−1 (q) is the (scheme-
theoretic) disjoint union of schemes of the form Spec K, where K is a finite separable
extension of κ(q).
(b) Show that π is unramified if and only if for each geometric point q, π−1 (q) :=
q ×Y X is the (scheme-theoretic) disjoint union of copies of q.

21.7.E. E XERCISE ( THE MEANING OF UNRAMIFIEDNESS IN “ VERY GEOMETRIC ”

SETTINGS ). Show that a morphism π : X → Y of finite type schemes over an
algebraically closed field is unramified if and only if π is injective on tangent
vectors at closed points. (This was mentioned in the remark immediately after
Theorem 19.1.1.)

21.7.F. E XERCISE ( FOR USE IN E XERCISE 21.7.G( B )).

(a) Suppose π : X → Y is a locally finite type morphism of locally Noetherian
schemes. Show that π is unramified if and only if δπ : X → X ×Y X is an open
embedding. Hint: Show the following. If ϕ : X → Z is a closed embedding of
Noetherian schemes, and the ideal sheaf I of ϕ satisfies I = I 2 , then ϕ is
also an open embedding. For that, show that if (A, m) is a Noetherian local ring,
and I is a proper ideal of A satisfying I = I2 , then I = 0. For that in turn, use
Nakayama version 2 (Lemma 8.2.9). Also use the fact that Supp I is closed (using
Exercise 14.3.C), so its complement is open.
⋆ (b) Adapt your proof of (a) to drop the locally Noetherian hypothesis. Hint: show
that if π is locally of finite type, then δπ is locally finitely presented.

21.7.G. E ASY E XERCISE . Suppose π : X → Y and ρ : Y → Z are locally of finite type.

Let τ = ρ ◦ π:
X
π /Y

τ ρ
 
Z
(a) Show that if τ is unramified, then so is π.
(b) Show that if ρ is unramified, and τ is smooth of relative dimension n (e.g., étale
if n = 0), then π is smooth of relative dimension n.
(Does this agree with your geometric intuition?)
 © 2024 Ravi Vakil. Published by Princeton University Press. 633

21.7.H. E XERCISE . Suppose π : X → Y is locally of finite type. Show that the locus
in X where π is unramified is open.

21.7.I. U NIMPORTANT E XERCISE ( FOR NUMBER THEORISTS ). Suppose ϕ : (A, m) →

(B, n) is a local homomorphism of local rings. In algebraic number theory, such a
ring morphism is said to be unramified if B/ϕ(m)B is a finite separable extension of
A/m. Show that if ϕ is finite type, this agrees with our definition.

21.7.1. Arithmetic side remark: the different and discriminant. If B is the ring of integers
in a number field (§10.7.1), the different ideal of B is the annihilator of ΩB/Z . It
measures how “(un)ramified” Spec B → Spec Z is, and is a ring-theoretic version
of the ramification locus. The discriminant ideal can be interpreted as the ideal
of Z corresponding to effective divisor on Spec Z that is the “push forward” (not
defined here, but defined as you might expect) of the divisor corresponding to the
different. It is a ring-theoretic version of the branch locus. If B/A is an extension of
rings of integers of number fields, the relative different ideal (of B) and relative
discriminant ideal (of A) are defined similarly. (We won’t use these ideas.)
 CHAPTER 22

⋆ Blowing up

We next discuss an important construction in algebraic geometry, the blow-up

of a scheme along a closed subscheme (cut out by a finite type ideal sheaf). The the-
ory could mostly be developed immediately after Chapter 17, but the interpretation
in terms of the conormal cone/bundle/sheaf of many classical examples makes it
natural to discuss blowing up after differentials.
We won’t use blowing up much in later chapters, so feel free to skip this
topic for now. But it is an important tool. For example, one can use it to resolve
singularities, and more generally, indeterminacy of rational maps. In particular,
blow-ups can be used to relate birational varieties to each other.
We will start with a motivational example that will give you a picture of the
construction in a particularly important (and the historically earliest) case, in §22.1.
We will then see a formal definition, in terms of a universal property, §22.2. The
definition won’t immediately have a clear connection to the motivational example.
We will deduce some consequences of the definition (assuming that the blow-up
actually exists). We then prove that the blow-up exists, by describing it quite
explicitly, in §22.3. As a consequence, we will find that the blow-up morphism is
projective, and we will deduce more consequences from this. In §22.4, we will do a
number of explicit computations, to see various sorts of applications, and to see
that many things can be computed by hand.

22.1 Motivating example: Blowing up the origin in the plane

We will generalize the following notion, which will correspond to “blowing up”
the origin of the affine plane A2k (Exercise 10.3.F). Our discussion will be informal.
Consider the subset of A2 × P1 corresponding to the following. We interpret P1 as
parametrizing the lines through the origin. Consider the subvariety

Bl(0,0) A2 := (p ∈ A2 , [ℓ] ∈ P1 ) : p ∈ ℓ) ,

which is the data of a point p in the plane, and a line ℓ containing both p and
the origin. Algebraically: let x and y be coordinates on A2 , and X and Y be
projective coordinates on P1 (“corresponding” to x and y); we will consider the
subset Bl(0,0) A2 of A2 × P1 corresponding to xY − yX = 0. We have the useful

This material is published by Princeton University Press as The Rising Sea: Foundations of Algebraic
Geometry by Ravi Vakil. This version is free to view and download for personal use only. Not for
redistribution, resale, or use in derivative works. © 2024 Ravi Vakil.

635
636 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

diagram

Bl(0,0) A2  / A2 × P1 / P1

& 2
β

You can verify that Bl(0,0) A2 is smooth over k (Definition 13.2.4 or 21.2.28)
directly (you can now make the paragraph after Exercise 10.3.F precise), but here is
an informal argument, using the projection Bl(0,0) A2 → P1 . The projective line P1
is smooth, and for each point [ℓ] in P1 , we have a smooth choice of points on the
line ℓ. Thus we are verifying smoothness by way of a fibration over P1 .
We next consider the projection to A2 , β : Bl(0,0) A2 → A2 . This is an iso-
morphism away from the origin. Loosely speaking, if p is not the origin, there is
precisely one line containing p and the origin. On the other hand, if p is the origin,
then there is a full P1 of lines containing p and the origin. Thus the preimage of
(0, 0) is a curve, and hence a divisor (an effective Cartier divisor, as the blown-up
surface is regular). This is called the exceptional divisor of the blow-up.
If we have some curve C ⊂ A2 singular at the origin, it can be potentially
partially desingularized, using the blow-up, by taking the closure of C \ {(0, 0)} in
Bl(0,0) A2 . (A desingularization or a resolution of singularities of a variety X is a
proper birational morphism X e → X from a regular scheme.) For example, consider
2 3 2
the curve y = x + x , which is regular except for a node at the origin. We can
take the preimage of the curve minus the origin, and take the closure of this locus
in the blow-up, and we will obtain a regular curve; the two branches of the node
downstairs are separated upstairs. (You can check this in Exercise 22.4.B once we
have defined things properly. The result will be called the proper transform (or strict
transform) of the curve.) We are interested in desingularizations for many reasons.
Because we understand regular curves quite well, we could hope to understand
other curves through their desingularizations. This philosophy holds true in higher
dimension as well.
More generally, we can blow up An at the origin (or more informally, “blow
up the origin”), getting a subvariety of An × Pn−1 . Algebraically, If x1 , . . . , xn
are coordinates on An , and X1 , . . . , Xn are projective coordinates on Pn−1 , then
the blow-up Bl⃗0 An is given by the equations xi Xj − xj Xi = 0. Once again, this is
smooth: Pn−1 is smooth, and for each point [ℓ] ∈ Pn−1 , we have a smooth choice of
p ∈ ℓ.
We can extend this further, by blowing up An+m along a coordinate m-plane
m
A by adding m more variables xn+1 , . . . , xn+m to the previous example; we get a
subset of An+m × Pn−1 .
Because in complex geometry, submanifolds of manifolds locally “look like”
coordinate m-planes in n-space, you might imagine that we could extend this to
blowing up a regular subvariety of a regular variety. In the course of making this
precise, we will accidentally generalize this notion greatly, defining the blow-up of
any finite type quasicoherent sheaf of ideals in a scheme. In general, blowing up
may not have such an intuitive description as in the case of blowing up something
regular inside something regular — it can do great violence to the scheme — but
even then, it is very useful.
 © 2024 Ravi Vakil. Published by Princeton University Press. 637

Our description will depend only on the closed subscheme being blown up,
and not on coordinates. That remedies a defect that was already present in the first
example, of blowing up the plane at the origin. It is not obvious that if we picked
different coordinates for the plane (preserving the origin as a closed subscheme)
that we wouldn’t have two different resulting blow-ups.
As is often the case, there are two ways of understanding the notion of blowing
up, and each is useful in different circumstances. The first is by universal property,
which lets you show some things without any work. The second is an explicit
construction, which lets you get your hands dirty and compute things (and implies
for example that the blow-up morphism is projective).
The motivating example here may seem like a very special case, but if you
understand the blow-up of the origin in n-space well enough, you will understand
blowing up in general.

22.2 Blowing up, by universal property

We now define the blow-up by a universal property. The disadvantage of starting

here is that this definition won’t obviously be the same as (or even related to) the
examples of §22.1.
Suppose X ,→ Y is a closed subscheme corresponding to a finite type quasico-
herent sheaf of ideals. (If Y is locally Noetherian, the “finite type” hypothesis is
automatic, so Noetherian readers can ignore it.)
The blow-up of X ,→ Y is a Cartesian diagram

(22.2.0.1) EX Y  / BlX Y

β
  
X /Y

such that EX Y (the scheme-theoretic preimage of X by β) is an effective Cartier

divisor (defined in §9.5.1) on BlX Y, such that any other such Cartesian diagram

(22.2.0.2) D /W

  
X / Y,

where D is an effective Cartier divisor on W, factors uniquely through it:


D /W

  
EX Y  / BlX Y

  
X / Y.
638 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

We call BlX Y the blow-up (of Y along X, or of Y with center X). (Other somewhat
archaic terms for this are monoidal transformation, σ-process, and quadratic transforma-
tion, and dilation.) We call EX Y the exceptional divisor of the blow-up. (Bl and β
stand for “blow-up”, and E stands for “exceptional”.)
(Caution: in the minimal model program, “exceptional divisor” has a slightly
different meaning; if X → Y is a birational morphism, then any Weil divisor on
X whose image in Y is smaller dimension is called an exceptional divisor in this
context.)
By a typical universal property argument, if the blow-up exists, it is unique up
to unique isomorphism. (We can even recast this more explicitly in the language of
Yoneda’s Lemma: consider the category of diagrams of the form (22.2.0.2), where
morphisms are diagrams of the form

D /W

(  (/
D′ W′

  }  }
X / Y.

Then the blow-up is a final object in this category, if one exists.)

If Z ,→ Y is any closed subscheme of Y, then the scheme-theoretic preimage
β−1 Z is called the total transform of Z. We will soon see that β is an isomorphism
away from X (Observation 22.2.2). β−1 (Z − X) is called the proper transform or
strict transform or birational transform of Z. (We will use the first terminology.
We will also define it in a more general situation.) We will soon see (in the Blow-up
Closure Lemma 22.2.7) that the proper transform is naturally isomorphic to BlZ∩X Z,
where Z ∩ X is the scheme-theoretic intersection.
We will soon show that the blow-up always exists, and describe it explicitly.
We first make a series of observations, assuming that the blow up exists.

22.2.1. Observation. If X is the empty set, then BlX Y = Y. More generally, if X is an

effective Cartier divisor, then the blow-up is an isomorphism. (Reason: idY : Y → Y
satisfies the universal property.)
∼ β−1 (U), where
22.2.A. E XERCISE . If U is an open subset of Y, then BlU∩X U =
β : BlX Y → Y is the blow-up.
Thus “we can compute the blow-up locally.”

22.2.B. E XERCISE . Show that if Yα is an open cover of Y (as α runs over some
index set), and the blow-up of Yα along X ∩ Yα exists, then the blow-up of Y along
X exists.

22.2.2. Observation. Combining Observation 22.2.1 and Exercise 22.2.A, we see that
the blow-up is an isomorphism away from the locus you are blowing up:

β|BlX Y\EX Y : BlX Y \ EX Y → Y \ X

is an isomorphism.
 © 2024 Ravi Vakil. Published by Princeton University Press. 639

22.2.3. Observation. If X = Y, then the blow-up is the empty set: the only map
W → Y such that the pullback of X is an effective Cartier divisor is ∅ ,→ Y. In this
case we have “blown Y out of existence”!

22.2.C. E XERCISE ( BLOW- UP PRESERVES IRREDUCIBILITY AND REDUCEDNESS ).

Show that if Y is irreducible, and X doesn’t contain the generic point of Y, then
BlX Y is irreducible. Show that if Y is reduced, then BlX Y is reduced.

22.2.4. Existence in a first nontrivial case: blowing up a locally principal closed

subscheme.
We next see why BlX Y exists if X ,→ Y is locally cut out by one equation. As the
question is local on Y (Exercise 22.2.B), we reduce to the affine case Spec A/(t) ,→
Spec A. (A good example to think through is A = k[x, y]/(xy) and t = x.) Let
I = ker(A → At ) = {a ∈ A : tn a = 0 for some n > 0},
and let ϕ : A → A/I be the projection.

22.2.D. E XERCISE . Show that ϕ(t) is not a zerodivisor in A/I.

22.2.E. E XERCISE . Show that β : Spec A/I → Spec A is the blow up of Spec A along
Spec A/t. In other words, show that

Spec A/(t, I) / Spec A/I

β
 
Spec A/t / Spec A

is a “blow up diagram” (22.2.0.1). Hint: In checking the universal property reduce

to the case where W (in (22.2.0.2)) is affine. Then solve the resulting problem about
rings. Depending on how you proceed, you might find Exercise 11.3.G, about the
uniqueness of extension of maps over effective Cartier divisors, helpful.

22.2.F. E XERCISE . Show that Spec A/I is the scheme-theoretic closure of D(t) in
Spec A.
Thus you might geometrically interpret Spec A/I → Spec A as “shaving off
any fuzz supported in V(t)”. In the Noetherian case, this can be interpreted as
removing those associated points lying in V(t). This is intended to be vague, and
you should think about how to make it precise only if you want to.

22.2.5. The Blow-up Closure Lemma.

Suppose we have a Cartesian diagram

W /Z
cl. emb.

  
X /Y
cl. emb.

where the bottom closed embedding corresponds to a finite type ideal sheaf (and
hence the upper closed embedding does too). The first time you read this, it may
be helpful to consider only the special case where Z → Y is a closed embedding.
640 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Then take the fibered product of this square by the blow-up β : BlX Y → Y, to
obtain the fibered cube

(22.2.5.1) W × E Y Y X
/ Z × Bl Y Y X

  
EX Y  / BlX Y
Cartier

  
W /Z
cl. emb.

    
X
cl. emb. /Y

The top square is the Cartesian diagram


(22.2.5.2) W ×Y E X Y  / Z ×Y BlX Y

  
EX Y  / BlX Y.
Cartier

22.2.G. E ASY E XERCISE . Why is (22.2.5.2) a Cartesian diagram? Show the useful
canonical isomorphisms
∼ Z ×Y E X Y =
W ×Y EX Y = ∼ W ×Y BlX Y.

22.2.6. The bottom closed embedding of (22.2.5.2) is locally cut out by one equation,
and thus the same is true of the top closed embedding as well. However, the local
equation on Z ×Y BlX Y need not be a non-zerodivisor, and thus the top closed
embedding is not necessarily an effective Cartier divisor.
Let Z be the scheme-theoretic closure of
(Z ×Y BlX Y) \ (W ×Y BlX Y)
in Z×Y BlX Y. (As W ×Y BlX Y is locally principal, we are in precisely the situation of
§22.2.4, so the scheme-theoretic closure is not mysterious.) Note that in the special
case where Z → Y is a closed embedding, Z is the proper transform, as defined in
§22.2. For this reason, it is reasonable to call Z the proper transform of Z even if
Z isn’t a closed embedding. Similarly, it is reasonable to call Z ×Y BlX Y the total
transform of Z even if Z isn’t a closed embedding.
Define EZ ,→ Z as the pullback of EX Y to Z, i.e., by the Cartesian diagram

EZ  / Z
_ proper transform
_
cl. emb. cl. emb.
  prin. 
W ×Y EX Y  loc. / Z ×Y BlX Y total transform

  
EX Y  / BlX Y.
Cartier
 © 2024 Ravi Vakil. Published by Princeton University Press. 641

Note that EZ is an effective Cartier divisor on Z. (It is locally cut out by one equation,
pulled back from a local equation of EX Y on BlX Y. Can you see why this is locally
not a zerodivisor?) It can be helpful to note that the top square of the diagram above
is a blow-up square, by Exercises 22.2.E and 22.2.F (and the fact that blow-ups can
be computed affine-locally).

22.2.7. Blow-up Closure Lemma. — Using the preceding notation: (BlW Z, EW Z)

is canonically isomorphic to (Z, EZ ). More precisely: if the blow-up BlX Y exists, then
(Z, EZ ) is the blow-up of Z along W.
This will be very useful. We make a few initial comments. The first three apply
to the special case where Z → Y is a closed embedding, and the fourth comment
basically tells us we shouldn’t have concentrated on this special case.
(1) First, note that if Z → Y is a closed embedding, then the Blow-Up Closure
Lemma states that the proper transform (as defined in §22.2) is the blow-up of Z
along the scheme-theoretic intersection W = X ∩ Z.
(2) In particular, the Blow-Up Closure Lemma lets you actually compute blow-
ups, and we will do lots of examples soon. For example, suppose C is a plane curve,
singular at a point p, and we want to blow up C at p. Then we could instead blow
up the plane at p (which we have already described how to do, even if we haven’t
yet proved that it satisfies the universal property of blowing up), and then take the
scheme-theoretic closure of C \ {p} in the blow-up.
(3) More generally, if W is some nasty subscheme of Z that we wanted to
blow-up, and Z were a finite type k-scheme, then the same trick would work. We
could work locally (Exercise 22.2.A), so we may assume that Z is affine. If W is cut
out by r equations f1 , . . . , fr ∈ Γ (Z, OZ ), then complete the f’s to a generating set
f1 , . . . , fn of Γ (Z, OZ ). This gives a closed embedding of Z into An such that W is
the scheme-theoretic intersection of Z with a coordinate linear space An−r .
(4) Most generally still, this reduces the existence of the blow-up to a specific
special case. (If you prefer to work over a fixed field k, feel free to replace Z by k
in this discussion.) Suppose that for each n, Bl(x1 ,...,xn ) Spec Z[x1 , . . . , xn ] exists.
Then I claim that the blow-up always exists. Here’s why. We may assume that
Y is affine, say Spec B, and X = Spec B/(f1 , . . . , fn ). Then we have a morphism
Y → An Z given by xi 7→ fi , such that X is the scheme-theoretic preimage of the
origin. Hence by the blow-up closure lemma, BlX Y exists. (We won’t follow this
approach, however.)

22.2.H. E XERCISE . Prove the Blow-up Closure Lemma 22.2.7. Hint: obviously,
construct maps in both directions, using the universal property. Constructing the
following diagram may or may not help. (You will recognize the bottom cube as
642 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

(22.2.5.1).)

EZ  m /Z m
eff. Cartier

v t
EW Z  / BlW Z
eff. Cartier

+   , 
Z ×Y EX Y 
loc. prin.
/ Z ×Y BlX Y

 t 
W /Zt

  
EX Y  / BlX Y
eff. Cartier

  
Xt /Yt
Hooked arrows indicate closed embeddings; and when morphisms are furthermore
locally principal or even effective Cartier, they are so indicated. Exercise 11.3.G, on
the uniqueness of extension of maps over effective Cartier divisors, may or may
not help as well. Note that if Z → Y is actually a closed embedding, then so is
Z ×Y BlX Y → BlX Y and hence also Z → BlX Y.

22.3 The blow-up exists, and is projective

22.3.1. It is now time to show that the blow-up always exists. We will see two
arguments, which are enlightening in different ways. Both will imply that the
blow-up morphism is projective, and hence quasicompact, proper, finite type,
and separated. In particular, if Y → Z is quasicompact (resp., proper, finite type,
separated), so is BlX Y → Z. (And if Y → Z is projective, and Z is quasicompact,
then BlX Y → Z is projective. See the solution to Exercise 17.3.I for the reason for
this annoying extra hypothesis.) The blow-up of a k-variety is a k-variety (using
the fact that reducedness is preserved, Exercise 22.2.C), and the blow-up of an
irreducible k-variety is an irreducible k-variety (using the fact that irreducibility is
preserved, also Exercise 22.2.C).

22.3.2. Theorem (Proj description of the blow-up). — Suppose X ,→ Y is a closed

subscheme cut out by a finite type quasicoherent sheaf of ideals I ,→ OY . Then
Proj Y OY ⊕ I ⊕ I 2 ⊕ I 3 ⊕ · · · /Y

satisfies the universal property of blowing up.

(We made sense of products of ideal sheaves, and hence I n , in Exercise 15.6.E.)
We will prove Theorem 22.3.2 soon (§22.3.3), after seeing what it tells us. Be-
cause I is finite type, the graded sheaf of algebras has degree 1 piece that is finite
 © 2024 Ravi Vakil. Published by Princeton University Press. 643

type. The graded sheaf of algebras is also clearly generated in degree 1. Thus the
sheaf of algebras satisfy Hypotheses 17.2.1 (“finite generation in degree 1”).
But first, we should make sure that the preimage of X is indeed an effective
Cartier divisor. We can work affine-locally (Exercise 22.2.A), so we may assume
that Y = Spec B, and X is cut out by the finitely generated ideal I. Then
BlX Y = ProjB B ⊕ I ⊕ I2 ⊕ · · · .

We are slightly abusing notation by using the notation BlX Y, as we haven’t yet
shown that this satisfies the universal property. Recall (§13.9.1) that the graded
ring B ⊕ I ⊕ · · · , denoted B(I)• , is called the Rees algebra of the ideal I in B.
The preimage of X isn’t just any effective Cartier divisor; it corresponds to the
invertible sheaf O(1) on this Proj . Indeed, O(1) corresponds to taking our graded
ring, chopping off the bottom piece, and sliding all the graded pieces to the left by
1 (§15.1.4); it is the invertible sheaf corresponding to the graded module
IB(I)• = I ⊕ I2 ⊕ I3 ⊕ · · ·
(where that first summand I has grading 0). But this can be interpreted as the
scheme-theoretic preimage of X, which corresponds to the ideal I of B:


IB(I)• = I B ⊕ I ⊕ I2 ⊕ · · ·  / B ⊕ I ⊕ I2 ⊕ · · · = B(I)• .

Thus the scheme-theoretic preimage of X ,→ Y to

Proj (OY (I )• ) = Proj (OY ⊕ I ⊕ I 2 ⊕ · · · ),
corresponding to the ideal sheaf corresponding to I OY (I )• = I ⊕ I 2 ⊕ I 3 ⊕
· · · , is an effective Cartier divisor in class O(1). Once we have verified that this
construction indeed yields the blow-up, this divisor will be our exceptional divisor
EX Y.
Moreover, we see that the exceptional divisor can be described beautifully as a
Proj over X:
(22.3.2.1)
EX Y = Proj X OY /I ⊕ I /I 2 ⊕ I 2 /I 3 ⊕ · · · = Proj X (OY (I )• /(I OY (I )• )) .

We will later see that in good circumstances (if X is a regular embedding in Y), this
is a projectivization of a vector bundle (the “projectivized normal bundle”), see
Exercise 22.3.D(a).

22.3.3. Proof of Theorem 22.3.2. Reduce to the case of affine target Y = Spec B
with X corresponding to ideal I ⊂ B. Reduce to the case of affine source Spec R,
with principal effective Cartier divisor V(t) (t a non-zerodivisor in R). (A principal
effective Cartier divisor is locally cut out by a single non-zerodivisor.) Thus we have
reduced to the case π : Spec R → Spec B, corresponding to ϕ : B → R. Choose x1 , . . . ,
xn with (x1 , . . . , xn ) = I, and (ϕ(x1 ), . . . , ϕ(xn )) = (t). (Here we use the hypothesis
that the scheme-theoretic pullback of V(I) is V(t), and our understanding of scheme-
theoretic pullback from Exercise 10.2.B). Hence
(22.3.3.1) (ϕ(x1 )/t, . . . , ϕ(xn )/t) = R.
(Because t is a non-zerodivisor, ϕ(xi )/t makes sense.) We will describe one map
Spec R → Proj B(I)• that will extend the map on the open set Spec Rt → Spec B. It
is then unique, by Exercise 11.3.G. Consider the ring morphism ϕ ′ : B(I)• → R
644 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

sending elements X of degree d to ϕ(X)/td . This induces a map π ′ from Spec R to

the “affine cone” Spec B(I)• .
Let Xi be the degree 1 element of B(I)• corresponding to xi . Now ϕ ′ induces
(B(I)• )Xi 0 → Rϕ(xi )/t

and hence geometrically D(ϕ(xi )/t) → D(Xi ) from a distinguished open subset
of Spec R to a distinguished open subset of the affine cone Spec B(I)• . As x runs
through the set {x1 , . . . , xn }, D(ϕ(xi )/t) runs through a distinguished open cover of
Spec R (from (22.3.3.1)). Thus the image of π ′ : Spec R → Spec B(I)• misses the van-
ishing set of the irrelevant ideal (X1 , . . . , Xn ) ⊂ B(I)• , so the map π ′ factors through
Spec B(I)• \ V(X1 , . . . , Xn ) and hence maps to Proj B(I)• (everything commuting
with morphisms to Spec B):
Spec B(I)• o ? _ Spec B(I)• \ V(X1 , . . . , Xn ) / Proj B(I)•
O 5
∴
π′
 u
Spec R
π / Spec B.

We have constructed our desired morphism Spec R → Proj B(I)• ”lifting” π. □

Here are some applications and observations arising from this construction of
the blow-up. First, we can verify that our initial motivational examples are indeed
blow-ups. For example, blowing up A2 (with coordinates x and y) at the origin
yields: B = k[x, y], I = (x, y), and Proj(B ⊕ I ⊕ I2 ⊕ · · · ) = Proj B[X, Y] where the
elements of B have degree 0, and X and Y are degree 1 and “correspond to” x and y
respectively.

22.3.4. Normal bundles to exceptional divisors. The normal bundle to an effec-

tive Cartier divisor D is the (space associated to the) invertible sheaf O(D)|D , the
invertible sheaf corresponding to the D on the total space, then restricted to D
(Exercise 21.2.J). Thus in the case of the blow-up of a point in the plane, the ex-
ceptional divisor has normal bundle O(−1). (As an aside: Castelnuovo’s Criterion,
Theorem 28.6.1, states that conversely given a smooth surface containing E = ∼ P1
with normal bundle O(−1), E can be blown-down to a point on another smooth
surface.) In the case of the blow-up of a regular subvariety of a regular variety, the
blow up turns out to be regular (see Theorem 22.3.10), the exceptional divisor is a
projective bundle over X, and the normal bundle to the exceptional divisor restricts
to O(−1) (Exercise 22.3.D).

22.3.A. H ARDER BUT ENLIGHTENING EXERCISE . If i : X ,→ Pn is a projective

scheme, identify the exceptional divisor of the blow up of the affine cone over X
(§9.3.11) at the origin with X itself, and that its normal bundle (§22.3.4) is isomorphic
to OX (−1) := i∗ OPn (−1). (In the case X = P1 , we recover the blow-up of the plane
at a point. In particular, we recover the important fact that the normal bundle to
the exceptional divisor is O(−1).)

22.3.5. The normal cone. Motivated by (22.3.2.1), as well as Exercise 22.3.D below,
we make the following definition. If X is a closed subscheme of Y cut out by I ,
then the normal cone NX Y of X in Y is defined as
NX Y := Spec X OY /I ⊕ I /I 2 ⊕ I 2 /I 3 ⊕ · · · .


(22.3.5.1)
 © 2024 Ravi Vakil. Published by Princeton University Press. 645

This can profitably be thought of as an algebro-geometric version of an infinitesimal

“tubular neighborhood”. But some cautions are in order. If Y is smooth, NX Y may
not be smooth. (You can work out the example of Y = A2k and X = V(xy).)
And even if X and Y are smooth, then although NX Y is smooth (as we will see
shortly, Exercise 22.3.D), it doesn’t “embed” in any way in Y. As a simple example,
consider a smooth point X = p of a non-rational curve Y = C. Then NX Y is rational
and cannot admit an open embedding into Y.

22.3.6. The tangent cone. If X is a closed point p, then the normal cone is called
the tangent cone to Y at p. The projectivized tangent cone is the exceptional
divisor EX Y (the Proj of the same graded sheaf of algebras). Following §9.3.12, the
tangent cone and the projectivized tangent cone can be put together in the projective
completion of the tangent cone, which contains the tangent cone as an open subset,
and the projectivized tangent cone as a complementary effective Cartier divisor.
In Exercise 22.3.D, we will see that at a regular point of Y, the tangent cone may
be identified with the tangent space, and the normal cone may often be identified
with the total space of the normal bundle.

22.3.B. E XERCISE . Suppose Y = Spec k[x, y]/(y2 − x2 − x3 ) (the nodal cubic, see
the bottom of Figure 8.4). Assume (to avoid distraction) that char k ̸= 2. Show that
the tangent cone to Y at the origin is isomorphic to Spec k[x, y]/(y2 − x2 ). Thus,
informally, the tangent cone “looks like” the original variety “infinitely magnified”.

22.3.C. E XERCISE . Suppose S• is a finitely generated graded algebra over a field

k. Exercise 22.3.A gives an isomorphism of Proj S• with the exceptional divisor to
the blow-up of Spec S• at the origin. Show that the tangent cone to Spec S• at the
origin is isomorphic to Spec S• itself. (Your geometric intuition should lead you to
find these facts believable.)

22.3.7. Blowing up regular embeddings.

The case of blow-ups of regular embeddings X ⊂ Y is particularly pleasant. For
example, the exceptional divisor is a projective bundle over X.

22.3.8. Theorem. — If I ⊂ A is generated by a regular sequence a1 , . . . , ad , then the

A (I/I ) → I /I
natural map Symn 2 n n+1
is an isomorphism.

22.3.9. Corollary. — If a closed embedding i : X ,→ Y is a regular embedding with ideal

sheaf I ⊂ OY , then the natural map Symn (I /I 2 ) → I n /I n+1 is an isomorphism.
Furthermore, in combination with Proposition 21.2.16, we see that I n /I n+1 is a locally
free sheaf.
Exercise 24.6.E gives a somewhat simpler proof of Theorem 22.3.8 (and Corol-
lary 22.3.9) in the case where A is a k-algebra.
Before starting the proof of Theorem 22.3.8 in §22.3.11, we show its utility.

22.3.D. E XERCISE .
(a) Suppose X → Y is a regular embedding with ideal sheaf I , identify the total
space (§17.1.5) of the normal sheaf (the “normal bundle”) with the normal cone
NX Y (22.3.5.1), and show that the exceptional divisor EX Y is a projective bundle
(the “projectivized normal bundle”) over X.
(b) Show that the normal bundle to EX Y in BlX Y is O(−1) (for the projective bundle
646 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

over X).
(c) Assume further that X is a reduced closed point p. Show that p is a regular point
of Y. Identify the total space of the tangent space to p with the tangent cone to Y at
p.

22.3.10. Theorem. — Suppose X ,→ Y is a closed embedding of smooth varieties over k.

Then BlX Y is also smooth.

Proof. We use the fact that smooth varieties are regular, the Smoothness-Regularity
Comparison Theorem 13.2.7(b), whose proof we still have to complete.
We may assume that k = k, because the statement of the result is preserved by
base change to the algebraic closure. Reason: Smoothness can be checked after base
change to k, by Exercise 13.2.H. The blow-up construction commutes with base
change to the algebraic closure: you can examine the construction, or wait until
you can invoke the generalization in Exercise 24.1.P(a)
We need only check smoothness of BlX Y at the points of EX Y, by Observa-
tion 22.2.2. By Exercise 13.2.M(b), X ,→ Y is a regular embedding. Then by Ex-
ercise 22.3.D(a), EX Y is a projective bundle over X, and thus smooth, and hence
regular at its closed points. But EX Y is an effective Cartier divisor on BlX Y. By the
slicing criterion for regularity (Exercise 13.2.C), it follows that BlX Y is regular at
the closed points of EX Y, hence smooth at all points of EX Y. □

22.3.11. ⋆ Proving Theorem 22.3.8.

The proof of Theorem 22.3.8 may reasonably be skipped on a first reading. We
prove Theorem 22.3.8 following [F2, A.6.1], which in turn follows [Dav]. The proof
will be completed in §22.3.13. To begin, let α be the map of graded rings

α : (A/I)[X1 , . . . , Xd ] / ⊕∞ In /In+1
n=0

which takes Xi to the image of ai in I/I2 . Clearly α is surjective.

22.3.E. E XERCISE . Show that Theorem 22.3.8 would follow from the statement that
α is an isomorphism.
Because a1 is a non-zerodivisor, we can interpret A[a2 /a1 , . . . , ad /a1 ] as a
subring of the total fraction ring (defined in §6.6.35). In particular, as A is a subring
of its total fraction ring, the map A → A[a2 /a1 , . . . , ad /a1 ] is an injection. Define

β : A[T2 , . . . , Td ] / A[a2 /a1 , . . . , ad /a1 ]

by Ti 7→ ai /a1 . Clearly, the map β is surjective, and Li := a1 Ti − ai lies in ker(β).

22.3.12. Lemma. — The kernel of β is (L2 , . . . , Ld ).

Proof. We prove the result by induction on d. We consider first the base case d = 2.
Suppose F[T2 ] ∈ ker β, so F(a2 /a1 ) = 0. Then applying the algorithm for the
deg F
Remainder Theorem, dividing a1 F(T2 ) by a1 T2 − a2 = L2 ,
deg F
(22.3.12.1) a1 F(T2 ) = G(T2 )(a1 T2 − a2 ) + R
where G(T2 ) ∈ A[T2 ], and R ∈ A is the remainder. Substituting T2 = a2 /a1
(and using the fact that A → A[a2 /a1 ] is injective), we have that R = 0. Then
 © 2024 Ravi Vakil. Published by Princeton University Press. 647

deg F
(a1 T2 − a2 )G(T2 ) ≡ 0 (mod (a1 )). Using the fact that a2 is a non-zerodivisor
deg F deg F
modulo (a1 ) (as a1 , a2 is a regular sequence by Exercise 9.5.E), a short
deg F
induction shows that the coefficients of G(T2 ) must all be divisible by a1 . Thus
F(T2 ) is divisible by a1 T2 − a2 = L2 , so the case d = 2 is proved.
We now consider the general case d > 2, assuming the result for all smaller d.
Let A ′ = A[a2 /a1 ]. Then a1 , a3 , a4 , . . . , ad is a regular sequence in A ′ . Reason: a1
is a non-zerodivisor in A ′ . A ′ /(a1 ) = (A[T2 ]/(a1 T2 − a2 ))/(a1 ) = A[T2 ]/(a1 T2 −
a2 , a1 ) = A[T2 ]/(a1 , a2 ). Then a3 is a non-zerodivisor in this ring because it
is a non-zerodivisor in A/(a1 , a2 ), a4 is a non-zerodivisor in A[T2 ]/(a1 , a2 , a3 )
because it is a non-zerodivisor in A/(a1 , a2 , a3 ), and so forth. Condition (ii) of the
Definition 9.5.4 of regular sequence also holds, as
A ′ /(a1 , a3 , a4 , . . . , ad ) = A[T2 ]/(a1 , a2 , . . . , ad ) ̸= 0.
Consider the composition
A[T2 , . . . , Td ] → A ′ [T3 , . . . , Td ] / A ′ [a3 /a1 , . . . , ad /a1 ] = A[a2 /a1 , . . . , ad /a1 ].

By the case d = 2, the kernel of the first map is L2 . By the inductive hypothesis, the
kernel of the second map is (L3 , . . . , Ld ). The result follows. □

22.3.13. Proof of Theorem 22.3.8. By Exercise 22.3.E, it suffices to prove that the
surjection α is an isomorphism. Suppose F ∈ ker(α); we wish to show that F =
0. We may assume that F is homogeneous, say of degree n. Consider the map
α ′ : A[X1 , . . . , Xd ] → ⊕∞ n n+1
n=0 I /I lifting α. Lift F to A[X1 , . . . , Xd ], so F ∈ ker(α ′ ).
We wish to show that F ∈ IA[X1 , . . . , Xd ]. Suppose F(a1 , . . . , ad ) = x ∈ In+1 . Then
we can write x as F ′ (a1 , . . . , ad ), where F ′ is a homogeneous polynomial of the
same degree as F, with coefficients in I. Then by replacing F by F − F ′ , we are
reduced to the following problem: suppose F ∈ A[X1 , . . . , Xd ] is homgeneous of
degree n, and F(a1 , . . . , ad ) = 0, we wish to show that F ∈ IA[X1 , . . . , Xn ]. But
if F(a1 , . . . , ad ) = 0, then F(1, a2 /a1 , . . . , ad /a1 ) = 0 in A[a2 /a1 , . . . , ad /a1 ]. But
Lemma 22.3.12 identifies the kernel of β, so
F(1, T2 , T3 , . . . , Td ) ∈ (a1 T2 − a2 , a1 T3 − a3 , . . . , a1 Td − ad ).
Thus the coefficients of F are in (a1 , . . . , ad ) = I as desired. □

22.4 Examples and computations

In this section we will work through a number of explicit of examples, to get a
sense of how blow-ups behave, how they are useful, and how one can work with
them explicitly. Throughout we work over a field k, and we assume throughout
that char k = 0 to avoid distraction. The examples and exercises are loosely
arranged by topic, but not in order of importance.

22.4.1. Example: Blowing up the plane along the origin.

Let’s first blow up the plane A2 along the origin, and see that the result agrees
with our discussion in §22.1. Let x and y be the coordinates on A2 . The blow-up is
Proj k[x, y, X, Y], where xY − yX = 0. (Here x and y have degree 0 and X and Y have
degree 1.) This is naturally a closed subscheme of A2 × P1 , cut out (in terms of the
648 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

projective coordinates X and Y on P1 ) by xY − yX = 0. We consider the two usual

patches on P1 : [X, Y] = [s, 1] and [1, t]. The first patch yields Spec k[x, y, s]/(sy − x),
and the second gives Spec k[x, y, t]/(y − xt). Notice that both are smooth: the first
∼ A2 , and the second is Spec k[x, t] =
is Spec k[y, s] = ∼ A2 .
We now describe the exceptional divisor. We first consider the first (s) patch.
The ideal is generated by (x, y), which in our ys-coordinates is (ys, y) = (y), which
is indeed principal. Thus on this patch the exceptional divisor is generated by y.
Similarly, in the second patch, the exceptional divisor is cut out by x. (This can be
a little confusing, but there is no contradiction!) This explicit description will be
useful in working through some of the examples below.

22.4.A. E XERCISE . Let p be a k-valued point of P2 . Exhibit an isomorphism between

Blp P2 and the Hirzebruch surface F1 = PP1 (OP1 ⊕ OP1 (1)) (Example 17.2.6). (The
map Blp P2 → P1 informally corresponds to taking a point to the line connecting
it to the origin. Do not be afraid: You can do this by explicitly working with
coordinates.)

22.4.2. Resolving singularities.

22.4.3. The proper transform of a nodal curve (Figure 22.1). (You may wish to
flip to Figure 8.4 while thinking through this exercise.) Consider next the curve
y2 = x3 + x2 inside the plane A2 (the nodal cubic, see Exercise 22.3.B). Let’s blow
up the origin, and compute the total and proper transform of the curve. (By the
Blow-up Closure Lemma 22.2.7, the latter is the blow-up of the nodal curve at the
origin.) In the first patch, we get y2 − s2 y2 − s3 y3 = 0. This factors: we get the
exceptional divisor y with multiplicity two, and the curve 1 − s2 − s3 y = 0. You
can easily check that the proper transform is regular. Also, notice that the proper
transform C e meets the exceptional divisor at two points, s = ±1. This corresponds
to the two tangent directions at the origin (as s = x/y). (You should do the second
patch yourself.)

22.4.B. E XERCISE (F IGURE 22.1). Describe both the total and proper transform of
the curve C given by y = x2 − x in Bl(0,0) A2 . Show that the proper transform of C
is isomorphic to C. Interpret the intersection of the proper transform of C with the
exceptional divisor E as the slope of C at the origin.

22.4.C. E XERCISE : BLOWING UP A CUSPIDAL PLANE CURVE ( CF. E XERCISE 10.7.F).

Describe the proper transform of the cuspidal curve C given by y2 = x3 in the
blown-up plane Bl(0,0) A2 . Show that it is regular. Show that the proper transform
of C meets the exceptional divisor E at one point, and is tangent to E there.
The previous two exercises are the first in an important sequence of singulari-
ties, which we now discuss.

22.4.D. E XERCISE : RESOLVING An CURVE SINGULARITIES . Resolve the singularity

y2 = xn+1 in A2 , by first blowing up its singular point, then considering its proper
transform and deciding what to do next.

22.4.4. Toward a definition of An curve singularities. You will notice that your
solution to Exercise 22.4.D depends only on the “power series expansion” of the
 © 2024 Ravi Vakil. Published by Princeton University Press. 649

F IGURE 22.1. Resolving curve singularities

singularity at the origin, and not on the precise equation. For example, if you
compare §22.4.3 with the n = 1 case of Exercise 22.4.D, you will see that they are
“basically the same”. We will make this precise in Definition 28.2.C.

22.4.E. E XERCISE ( WARM - UP TO E XERCISE 22.4.F). Blow up the cone point z2 =

x2 + y2 (Figure 3.4) at the origin. Show that the resulting surface is regular. Show
that the exceptional divisor is isomorphic to P1 . (Remark: You can check that
the normal bundle to this P1 is not OP1 (−1), as is the case when you blow up
a point on a smooth surface, see §22.3.4; it is OP1 (−2). Doesn’t this contradict
Exercise 22.3.D(b)?)

22.4.F. E XERCISE ( RESOLVING An SURFACE SINGULARITIES ). Resolve the singular-

ity z2 = y2 + xn+1 in A3 by first blowing up its singular point, then considering
its proper transform, and deciding what to do next. (A k-surface singularity ana-
lytically isomorphic to this is called an An surface singularity. For example, the
cone shown in Figure 3.4 is an A1 surface singularity. We make this precise in
Exercise 28.2.C.) This exercise is a bit time-consuming, but is rewarding in that it
shows that you can really resolve singularities by hand.

22.4.5. Remark: ADE-surface singularities and Dynkin diagrams (see Figure 22.2). A
k-singularity analytically isomorphic to z2 = x2 y + yn−1 (resp., z2 = x3 + y4 ,
z2 = x3 + xy3 , z2 = x3 + y5 ) is a called a Dn surface singularity (resp., E6 , E7 ,
E8 surface singularity). We will make this precise in Exercise 28.2.C, and you
will then be able to guess the definition of the corresponding curve singularity. If
you (minimally) desingularize each of these surfaces by sequentially blowing up
singular points as in Exercise 22.4.F, and look at the arrangement of exceptional
divisors (the various exceptional divisors and how they meet), you will discover the
corresponding Dynkin diagram. More precisely, if you create a graph, where the
vertices correspond to exceptional divisors, and two vertices are joined by an edge
if the two divisors meet, you will find the underlying graph of the corresponding
650 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Dynkin diagram. This is the start of several very beautiful stories; see Remark 27.3.7
for a first glimpse of one of them.

F IGURE 22.2. The exceptional divisors for resolutions of some

ADE surface singularities, and their corresponding dual graphs
(see Remark 22.4.5)

22.4.6. Remark: Resolution of singularities. Hironaka’s Theorem on resolution of

singularities implies that this idea of trying to resolve singularities by blowing
up singular loci in general can succeed in characteristic 0 (see [Hir], and [Ko2]).
(The case of dimension 1 will be shown in §28.4.4, and the case of dimension 2
will be discussed in §28.6.4.) It is not known if an analogous statement is true
in positive characteristic (except in dimension at most 3, see [CP]), but de Jong’s
Alteration Theorem [dJ] gives a result which is good enough for most applications.
Rather than producing a birational proper map X e → X from something regular, it
produces a proper map from something regular that is generically finite (and the
corresponding extension of function fields is separable).
Here are some other exercises related to resolution of singularities.

22.4.G. E XERCISE . Blowing up a nonreduced subscheme of a regular scheme can

give you something singular, as shown in this example. Describe the blow up of the
closed subscheme V(y, x2 ) in Spec k[x, y] = A2 . Show that you get an A1 surface
singularity.

22.4.H. E XERCISE . Desingularize the tacnode (see Exercise 10.7.G and Defini-
tion 28.2.2) y2 = x4 , not in two steps (as in Exercise 22.4.D), but in a single step by
blowing up (y, x2 ).

22.4.I. E XERCISE ( RESOLVING A SINGULARITY BY AN UNEXPECTED BLOW- UP ).

Suppose Y is the cone x2 + y2 = z2 , and X is the line cut out by x = 0, y = z on
Y. Show that BlX Y is regular. (In this case we are blowing up a codimension 1
locus that is not an effective Cartier divisor, see Problem 13.1.3. But it is an effective
Cartier divisor away from the cone point, so you should expect your answer to be
an isomorphism away from the cone point.)

22.4.7. Multiplicity of a function at a regular point. In order to pose Exercise 22.4.J, we

introduce a useful concept. If f is a function on a locally Noetherian scheme X, its
multiplicity at a regular point p is the largest m such that f lies in the mth power
of the maximal ideal in the local ring OX,p . For example, if f ̸= 0, V(f) is singular at
p if and only if m > 1. (Do you see why?)
 © 2024 Ravi Vakil. Published by Princeton University Press. 651

22.4.J. E XERCISE . Show that the multiplicity of the exceptional divisor in the total
transform of a subscheme Z of An when you blow up the origin is the smallest mul-
tiplicity (at the origin) of a defining equation of Z. (For example, in the case of the
nodal and cuspidal curves above, Example 22.4.3 and Exercise 22.4.C respectively,
the exceptional divisor appears with multiplicity 2.) Caution: you need to make
this question precise before solving it.

22.4.8. Resolving rational maps.

22.4.K. E XERCISE ( UNDERSTANDING THE BIRATIONAL MAP P2 o / P1 × P1 VIA

2
BLOW- UPS ). Let p and q be two distinct k-points of P , and let r be a k-point
∼
of P1 × P1 . Describe an isomorphism Bl{p,q} P2 ←→ Blr P1 × P1 . (Possible hint:
Consider lines ℓ through p and m through q; the choice of such a pair corresponds
to a point of P1 × P1 . A point s of P2 not on line pq yields a pair of lines (ps, qs)
of P1 × P1 . Conversely, a choice of lines (ℓ, m) such that neither ℓ and m is line pq
yields a point s = ℓ ∩ m ∈ P2 . This describes a birational map P2 o / P1 × P1 .
Exercise 22.4.A is related.)
Exercise 22.4.K is an example of the general phenomenon explored in the next
two exercises.

22.4.L. H ARDER BUT USEFUL EXERCISE ( BLOW- UPS RESOLVE BASE LOCI OF RATIO -
NAL MAPS TO PROJECTIVE SPACE ). Suppose we have a scheme Y, an invertible
sheaf L , and a number of sections s0 , . . . , sn of L (a linear series, Definition 15.2.1).
Then away from the closed subscheme X cut out by s0 = · · · = sn = 0 (the base
locus of the linear series), these sections give a morphism to Pn . Show that this
morphism extends uniquely to a morphism BlX Y → Pn , where this morphism cor-
responds to the invertible sheaf (β∗ L )(−EX Y), where β : BlX Y → Y is the blow-up
morphism. In other words, “blowing up the scheme-theoretic base locus resolves
this rational map”. Hint: it suffices to consider an affine open subset of Y where L
is trivial. Uniqueness might use Exercise 11.3.G.

22.4.9. Remarks. (i) Exercise 22.4.L immediately implies that blow-ups can be used
to resolve rational maps to projective schemes Y 99K Z ,→ Pn .
(ii) The following interpretation is enlightening. The linear series on Y pulls
back to a linear series on BlX Y, and the base locus of the linear series on Y pulls back
to the base locus on BlX Y. The base locus on BlX Y is EX Y, an effective Cartier divisor.
Because EX Y is not just locally principal, but also locally a non-zerodivisor, it can be
“divided out” from the β∗ si (yielding a section of (β∗ L )(−EX Y), thereby removing
the base locus, and leaving a base-point-free linear series. (In a sense that can be
made precise through the universal property, this is the smallest “modification” of
Y that can remove the base locus.) If X is already Cartier (as for example happens
with any nontrivial linear series if Y is a regular pure-dimensional curve), then we
can remove a base locus by just “dividing out X”.
(iii) You may wish to revisit Exercise 19.7.C, and interpret it in terms of Exer-
cise 22.4.L.

22.4.10. Examples. (i) The rational map Pn 99K Pn−1 given by [x0 , · · · , xn ] 99K
[x1 , · · · , xn ], defined away from p = [1, 0, · · · , 0], is resolved by blowing up p. Then
652 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

by the Blow-up Closure Lemma 22.2.7, if Y is any locally closed subscheme of Pn ,

we can project to Pn−1 once we blow up p in Y, and the invertible sheaf giving the
map to Pn−1 is (somewhat informally speaking) β∗ (OPn (1)) ⊗ O(−Ep Y).
(ii) Consider two general homogeneous cubic functions C1 and C2 in three
variables, yielding two cubic curves in P2 . They are smooth (by Bertini’s Theo-
rem 13.4.2), and meet in 9 points p1 , . . . , p9 (using our standing assumption that
we work over an algebraically closed field). Then [C1 , C2 ] gives a rational map
P2 99K P1 . To resolve the rational map, we blow up p1 , . . . , p9 . The result is (gener-
ically) an elliptic fibration Bl{p1 ,...,p9 } P2 → P1 . (This is by no means a complete
argument or even a precise statement.)
(iii) Fix six general points p1 , . . . , p6 in P2 . There is a four-dimensional vector
space of cubics vanishing at these points, and they vanish scheme-theoretically
precisely at these points. This yields a rational map P2 99K P3 , which is resolved
by blowing up the six points. The resulting morphism turns out to be a closed
embedding, and the image in P3 is a (smooth) cubic surface. This is the famous fact
that the blow-up of the plane at six general points may be represented as a (smooth)
cubic in P3 . (Again, this argument is not intended to be complete.) See §27.3.5 for a
more precise and more complete discussion.
In reasonable circumstances, Exercise 22.4.L has an interpretation in terms of
graphs of rational maps.

22.4.M. E XERCISE . Suppose s0 , . . . , sn are sections of an invertible sheaf L

on an integral scheme X, not all 0. By Remark 15.2.5, these data give a rational
map ϕ : X 99K Pn . Give an isomorphism between the graph of ϕ (§11.3.4) and
BlV(s0 ,...,sn ) X. (Your argument will not require working over a field k; it will work
in general.)
You may enjoy exploring the previous idea by working out how the Cremona
transformation P2 99K P2 (Exercise 7.5.J) can be interpreted in terms of the graph of
the rational map [x, y, z] 99K [1/x, 1/y, 1/z].

22.4.N. ⋆ E XERCISE . Resolve the rational map

[w,x]
Spec k[w, x, y, z]/(wz − xy) / P1

from the cone over the quadric surface to the projective line. Let X be the resulting
variety, and π : X → Spec k[w, x, y, z]/(wz − xy) the projection to the cone over the
quadric surface. Show that π is an isomorphism away from the cone point, and that
the preimage of the cone point is isomorphic to P1 (and thus has codimension 2, and
therefore is different from the resolution obtained by simply blowing up the cone
point). Possible hint: if Q is the quadric in P3 cut out by wz − xy = 0, then factor the
rational map as Spec k[w, x, y, z]/(wz − xy) \ {0} → Q (cf. Exercise 9.3.N), followed
∼
by the isomorphism Q −→ P1 × P1 (Example 10.6.2), followed by projection onto
one of the factors.
This is an example of a small resolution. (A small resolution X → Y is a
resolution where the locus of points of Y where the fiber has dimension r is of
codimension greater than 2r. We will not use this notion again in any essential
way. Caution: this is the definition of “small resolution” used in intersection
homology; in the minimal model program, a small resolution is a resolution that
 © 2024 Ravi Vakil. Published by Princeton University Press. 653

is an isomorphism in codimension 1. So pay attention to the context of what you

are reading.) Notice that this resolution of the morphism involves blowing up
the base locus w = x = 0, which is a cone over one of the lines on the quadric
surface wz = xy. We are blowing up an effective Weil divisor, which is necessarily
not Cartier as the blow-up is not an isomorphism. In Exercise 13.1.E, we saw that
(w, x) was not principal, while here we see that (w, x) is not even locally principal.
Essentially by Exercise 15.5.N, V(w, x) cannot even be the support of a locally
principal divisor.

22.4.11. Remark: Non-isomorphic small resolutions. If you instead resolved the map
[w, y], you would obtain a similar looking small resolution

π′ : X′ / Spec k[w, x, y, z]/(wz − xy)

(it is an isomorphism away from the origin, and the fiber over the origin is P1 ). But
it is different! More precisely, there is no morphism X → X ′ making the following
the diagram commute.

X / X′
′
π π

( v
Spec k[w, x, y, z]/(wz − xy)

The birational map X 99K X ′ is called the Atiyah flop, [At1].

22.4.12. Factorization of birational maps. We end our discussion of resolution

of rational maps by noting that just as Hironaka’s Theorem states that one may
resolve all singularities of varieties in characteristic 0 by a sequence of blow-ups
along smooth subvarieties (§22.4.6), the Weak Factorization Theorem (first proved by
Włodarczyk [Wł]) states that any two birational varieties X and Y in characteristic 0
may be related by blow-ups and “blow-downs” along smooth subvarieties. More
precisely, there are varieties X0 , . . . , Xn , X01 , . . . , X(n−1)n , with X0 = X and Xn = Y,
with morphisms Xi(i+1) → Xi and Xi(i+1) → Xi+1 (0 ≤ i < n) which are blow-ups
of smooth subvarieties.

22.4.13. Blow-ups and line bundles.

22.4.O. E XERCISE ( GENERALIZING E XERCISE 20.2.D). Suppose X is a regular

projective surface over k, and p is a k-valued point. Let β : Blp X → X be the
blow-up morphism, and let E = Ep X be the exceptional divisor. Consider the exact
sequence
γ : 17→[E]
Z / Pic(Blp X) α / Pic(Blp X \ E) /0

from (15.4.11.1). (Technically, (15.4.11.1) has Pic replaced by Cl. But because regular
local rings are unique factorization domains by §13.8.5, Pic = Cl in this case by
Proposition 15.4.15.) Note that Blp X \ E = X \ p. Show that Pic(X \ p) = Pic X. Show
that β∗ : Pic X → Pic Blp X gives a section to α. Use §22.3.4 to show that OX (E)|E =∼
OE (−1) (so in the language of Chapter 20, E is a (−1)-curve, Definition 20.2.8), and
from that show that γ is an injection. Conclude that Pic Blp X = ∼ Pic X ⊕ Z. Describe
how to find the intersection matrix on NQ (Blp X) from that of N1Q (X).
1
654 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

22.4.P. E XERCISE . Suppose D is an effective Cartier divisor (a curve) on X.

Let multp D be the multiplicity of D at p (Exercise 22.4.J), and let Dpr be the
proper transform of D. Show that β∗ D = Dpr + (multp D)E as effective Cartier
divisors. More precisely, show that the product of the local equation for Dpr
and the (multp D)th power of the local equation for E is the local equation for
∼
β∗ D, and hence that (i) β∗ D is an effective Cartier divisor, and (ii) β∗ OX (D) =
OBlp X (D ) ⊗ OBlp X (E)
pr ⊗(multp D)
. (A special case is the equation ℓ = e + m in
Hint 20.2.7.)

22.4.14. Change of the canonical line bundle under blow-ups.

As motivation for how the canonical line bundle changes under blowing up,
consider the blow-up β : Bl(0,0) A2 → A2 . Let X = Bl(0,0) A2 and Y = A2 for
convenience. We use Exercise 21.4.A to relate β∗ KY to KX .
We pick a generator for KY near (0, 0): dx ∧ dy. (This is in fact a generator for
KY everywhere on A2 , but for the sake of generalization, we point out that all that
matters is that is a generator at (0, 0), and hence near (0, 0) by geometric Nakayama,
Exercise 14.3.D.) When we pull it back to X, we can interpret it as a section of KX ,
which will generate KX away from the exceptional divisor E, but may contain E
with some multiplicity µ. Recall that X can be interpreted as the data of a point in
A2 as well as the choice of a line through the origin. We consider the open subset
U where the line is not vertical, and thus can be written as y = mx. Here we
have natural coordinates: U = Spec k[x, y, m]/(y − mx), which we can interpret as
Spec k[x, m]. The exceptional divisor E meets U, at x = 0 (in the coordinates on U),
so we can calculate µ on this open set. Pulling back dx ∧ dy to U, we get
dx ∧ dy = dx ∧ d(xm) = m(dx ∧ dx) + x(dx ∧ dm) = x(dx ∧ dm)
as dx ∧ dx = 0. Thus β∗ (dx ∧ dy) vanishes to order 1 along E.

22.4.Q. E XERCISE ( CF. U NIMPORTANT EXERCISE 21.4.J). Explain how this deter-
∼
mines a canonical isomorphism KX ←→ (β∗ KY )(E).

22.4.R. E XERCISE . Repeat the above calculation in dimension n. Show that the
exceptional divisor appears with multiplicity (n − 1).

22.4.S. ⋆ E XERCISE . Suppose k is perfect.

(a) Suppose Y is a surface over k, and p is a regular k-valued point, and let β : X → Y
be the blow-up of Y at p. Show that KX = ∼ (β∗ KY )(E). Hint: to find a generator of
KX near p, choose generators x and y of m/m2 (where m is the maximal ideal of
OY,p ), and lift them to elements of OX,p . Why does dx ∧ dy generate KX at p?
(b) Repeat part (a) in arbitrary dimension (following Exercise 22.4.R).
(c) Suppose Z is a smooth m-dimensional (closed) subvariety of a smooth n-
dimensional variety Y, and let β : X → Y be the blow-up of Y along Z. Show
that KX = ∼ (β∗ KY )((n − m − 1)E). (Recall from Theorem 22.3.10 that X = BlZ Y is
smooth.)

22.4.15. ⋆ Dimensional cohomology vanishing for quasiprojective schemes

(promised in §18.2.7).
Using the theory of blowing up, Theorem 18.2.6 (dimensional cohomology
vanishing for quasicoherent sheaves on projective k-schemes) can be extended to
quasiprojective k-schemes. Suppose X is a quasiprojective k-variety of dimension n.
 © 2024 Ravi Vakil. Published by Princeton University Press. 655

We show that X may be covered by n+1 affine open subsets. As X is quasiprojective,

there is some projective variety Y with an open embedding X ,→ Y. By replacing
Y with the closure of X in Y, we may assume that dim Y = n. Put any subscheme
structure Z on the complement of X in Y (for example the reduced subscheme
structure, §9.4.9). Let Y ′ = BlZ Y. Then Y ′ is a projective variety (§22.3.1), which can
be covered by n + 1 affine open subsets. The complement of X in Y ′ is an effective
Cartier divisor (EZ Y), so the restriction to X of each of these affine open subsets of
Y is also affine, by Exercise 8.3.E.

22.4.16. Remarks. (i) You might then hope that any dimension n variety can be
covered by n + 1 affine open subsets. This is not true. For each integer m, there is a
threefold that requires at least m affine open sets to cover it, see [RotV, Ex. 4.9]. By
the discussion above, this example is necessarily not quasiprojective. (ii) Here is a
fact useful in invariant theory, which can be proved in the same way. Suppose p1 ,
. . . , pn are closed points on a quasiprojective k-variety X. Then there is an affine
open subset of X containing all of them.

22.4.T. E XERCISE ( DIMENSIONAL COHOMOLOGY VANISHING FOR QUASIPROJECTIVE

VARIETIES ). Suppose X is a quasiprojective k-scheme of dimension d. Show that
for any quasicoherent sheaf F on X, Hi (X, F ) = 0 for i > d.

22.4.17. ⋆⋆ Deformation to the normal cone (Figure 22.3).

F IGURE 22.3. Deformation to the normal cone (§22.4.17)

The following construction is key to the modern understanding of intersection

theory in algebraic geometry, as developed by Fulton and MacPherson, [F2].
656 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

22.4.U. E XERCISE : D EFORMATION TO THE NORMAL CONE . Suppose Y is a k-variety,

and X ,→ Y is a closed subscheme.
(a) Show that the exceptional divisor of β : BlX×0 (Y × A1 ) → Y × A1 is isomorphic
to the projective completion of the normal cone to X in Y.
(b) Let π : BlX×0 (Y × A1 ) → A1 be the composition of β with the projection to A1 .
Show that π−1 (0) is the scheme-theoretic union of BlX Y with the projective com-
pletion of the normal cone to X in Y, and the intersection of these two subschemes
may be identified with EX Y, which is a closed subscheme of BlX Y in the usual way
(as the exceptional divisor of the blow-up BlX Y → Y), and a closed subscheme of
the projective completion of the normal cone as described in Exercise 9.3.O.
The map
BlX×0 (Y × A1 ) \ BlX Y / A1
is called the deformation to the normal cone (short for deformation of Y to the normal
cone of X in Y). Notice that the fiber above every k-point away from 0 ∈ A1 is
canonically isomorphic to Y, and the fiber over 0 is the normal cone. Because this
family is “nice” (more precisely, flat, the topic of Chapter 24), we can prove things
about general Y (near X) by way of this degeneration. (We will see in §24.3.10
that the deformation to the normal cone is a flat morphism, which is useful in
intersection theory.)
 Part VI

More cohomological tools

 CHAPTER 23

Derived functors

Ça me semble extrêmement plaisant de ficher comme ça beaucoup de choses, pas drôles
quand on les prend séparément, sous le grand chapeau des foncteurs dérivés.
I find it very agreeable to stick all sorts of things, which are not much fun when taken
individually, together under the heading of derived functors.
— A. Grothendieck, letter to J.-P. Serre, February 18, 1955 [GrS, p. 6]
In this chapter, we discuss derived functors, introduced by Grothendieck in his
celebrated “Tôhoku article” [Gr1], and their applications to sheaves. For quasico-
herent sheaves on quasicompact separated schemes, derived functor cohomology
will agree with Čech cohomology (§23.5). Čech cohomology will suffice for most
of our purposes, and is quite down-to-earth and computable, but derived functor
cohomology is worth seeing. First, it will apply much more generally in algebraic
geometry (e.g., étale cohomology) and elsewhere, although this is beyond the scope
of this book. Second, it will easily provide us with some useful notions, such as
the Ext functors and the Leray spectral sequence. But derived functors can be
intimidating the first time you see them, so feel free to just skim the main results,
and to return to them later.

23.1 The Tor functors

We begin with a warm-up: the case of Tor. This is a hands-on example, but if
you understand it well, you will understand derived functors in general. Tor will
be useful to prove facts about flatness, which we will discuss in §24.2. Tor is short
for “torsion” (see Remark 24.2.1).
If you have never seen this notion before, you may want to just remember its
properties. But I will prove everything anyway — it is surprisingly easy.
The idea behind Tor is as follows. Whenever we see a right-exact functor, we
always hope that it is the end of a long exact sequence. Informally, given a short
exact sequence

(23.1.0.1) 0 / N′ /N / N ′′ / 0,

This material is published by Princeton University Press as The Rising Sea: Foundations of Algebraic
Geometry by Ravi Vakil. This version is free to view and download for personal use only. Not for
redistribution, resale, or use in derivative works. © 2024 Ravi Vakil.

659
660 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

we hope M ⊗A N ′ → M ⊗A N → M ⊗A N ′′ → 0 will extend to a long exact

sequence
(23.1.0.2)
··· / TorA ′
i (M, N )
/ TorA
i (M, N)
/ TorA ′′
i (M, N )
/ ···

/ TorA ′
1 (M, N )
/ TorA
1 (M, N)
/ TorA ′′
1 (M, N )

/ M ⊗A N ′ / M ⊗A N / M ⊗A N ′′ / 0.

More precisely, we are hoping for covariant functors TorA

i (M, ·) from A-modules to
A-modules (functoriality giving 2/3 of the morphisms in (23.1.0.2)), with TorA
0 (M, N) ≡
M ⊗A N, and natural connecting homomorphisms

δ : TorA ′′
i+1 (M, N )
/ TorA ′
i (M, N )

for every short exact sequence (23.1.0.1) giving the long exact sequence (23.1.0.2).
(“Natural” means: given a morphism of short exact sequences, the natural square
you would write down involving the δ-morphism must commute.)
It turns out to be not too hard to make this work, and this will also motivate
derived functors. Let’s now define TorA i (M, N).
Take any resolution R of N by free modules:

··· / A⊕n2 / A⊕n1 / A⊕n0 /N / 0.

More precisely, build this resolution from right to left. Start by choosing generators
of N as an A-module, giving us A⊕n0 → N → 0. Then choose generators of the
kernel, and so on. Note that we are not requiring the ni to be finite (although
we could, if N is a finitely generated module and A is Noetherian). Truncate the
resolution, by stripping off the last term N (replacing → N → 0 with → 0). Then
tensor with M (which does not preserve exactness). Note that M ⊗ (A⊕ni ) = M⊕ni ,
as tensoring with M commutes with arbitrary direct sums (Exercise 1.2.M). Let
TorAi (M, N)R be the homology of this complex at the ith stage (i ≥ 0). The subscript
R reminds us that our construction depends on the resolution, although we will
soon see that it is independent of R.
We make some quick observations.
• TorA ∼
0 (M, N)R = M⊗A N, canonically. Reason: as tensoring is right-exact, and
A ⊕n1
→A ⊕n0
→ N → 0 is exact, we have that M⊕n1 → M⊕n0 → M ⊗A N → 0 is
exact, and hence that the homology of the truncated complex M⊕n1 → M⊕n0 → 0
is M ⊗A N.
• If M ⊗ · is exact (i.e., M is flat, §1.5.13), then TorA
i (M, N)R = 0 for all i > 0.
(This characterizes flatness, see Exercise 23.1.C.)
 © 2024 Ravi Vakil. Published by Princeton University Press. 661

Now given two modules N and N ′ and resolutions R and R ′ of N and N ′ , we

can “lift” any morphism N → N ′ to a morphism of the two resolutions:

··· / A⊕ni / ··· / A⊕n1 / A⊕n0 /N /0

   
··· / A⊕ni′ / ··· / A⊕n1′ / A⊕n0′ / N′ /0

We do this inductively on i. Here we use the freeness of A⊕ni : if a1 , . . . , ani are

′
generators of A⊕ni , to lift the map b : A⊕ni → A⊕ni−1 (the composition of the
′
differential A⊕ni → A⊕ni−1 with the previously constructed A⊕ni−1 → A⊕ni−1 )
′ ′ ′
to c : A⊕ni → A⊕ni , we arbitrarily lift b(aj ) from A⊕ni−1 to A⊕ni , and declare this
to be c(aj ). (Warning for people who care about such things: we are using the
Axiom of Choice here.)
Denote the choice of lifts by R → R ′ . Now truncate both complexes (remove
column N → N ′ ) and tensor with M. Maps of complexes induce maps of homology
(Exercise 1.5.D), so we have described maps (a priori depending on R → R ′ )

TorA
i (M, N)R
/ TorA ′
i (M, N )R ′ .

Recall (§1.5.7) that we say two maps of complexes f : C• → C•′ and g : C• → C•′
are homotopic if there is a sequence of maps w : Ci → Ci+1
′
such that f−g = dw+wd.
We saw that two homotopic maps give the same map on homology (Exercise 1.5.E).

23.1.A. C RUCIAL E XERCISE . Show that any two lifts R → R ′ are homotopic.
We now pull these observations together. (Be sure to digest these completely!)
(1) We get a map of A-modules TorA i (M, N)R → Tori (M, N )R ′ , independent
A ′

of the lift R → R . ′

(2) Hence for any two resolutions R and R ′ of an A-module N, we get a

∼
canonical isomorphism TorA i (M, N)R ←→ Tori (M, N)R ′ . Here’s why. Choose
A

lifts R → R ′ and R ′ → R. The composition R → R ′ → R is homotopic to the

identity (as it is a lift of the identity map N → N). Thus if fR→R ′ : TorA
i (M, N)R →
TorA ′ is the map induced by R → R , and similarly fR ′ →R is the map
′
i (M, N) R
induced by R ′ → R, then fR ′ →R ◦ fR→R ′ is the identity, and similarly fR→R ′ ◦
fR ′ →R is the identity.
(3) Hence TorA i (M, ·) doesn’t depend on the choice of resolution. It is a covari-
ant functor ModA → ModA .

23.1.1. Remark. Note that if N is a free module, then TorA

i (M, N) = 0 for all M and
all i > 0, as N has the trivial resolution 0 → N → N → 0 (it is “its own resolution”).
Finally, we get long exact sequences:

23.1.2. Proposition. — For any short exact sequence (23.1.0.1) we get a long exact
sequence of Tor’s (23.1.0.2).
662 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Proof. Given a short exact sequence (23.1.0.1), choose resolutions of N ′ and N ′′ .

Then use these to get a resolution for N as follows.
(23.1.2.1) 0 0 0

  
··· / A⊕n1′ / A⊕n0′ / N′ /0

  
··· / A⊕(n1′ +n1′′ ) / A⊕(n0′ +n0′′ ) /N /0

  
··· / A⊕n1′′ / A⊕n0′′ / N ′′ /0

  
0 0 0
′ ′′ ′ ′′ ′ ′
The map A⊕(ni+1 +ni+1 ) → A⊕(ni +ni ) is the composition A⊕ni+1 → A⊕ni ,→
′ ′′ ′′ ′′ ′ ′′
A⊕(ni +ni ) along with a lift of A⊕ni+1 → A⊕ni to A⊕(ni +ni ) ensuring that the
middle row is a complex.
′′ ′′
23.1.B. E XERCISE . Verify that it is possible to choose such a lift of A⊕ni+1 → A⊕ni
′ ′′
to A⊕(ni +ni ) .
Hence the middle row of (23.1.2.1) is exact (not just a complex), using the long
exact sequence in cohomology (Theorem 1.5.8), and the fact that the top and bottom
rows are exact. Thus the middle row is a resolution, and (23.1.2.1) is a short exact
sequence of resolutions. (This is sometimes called the horseshoe construction, as the
filling in of the middle row looks like filling in the middle of a horseshoe.) It may
be helpful to notice that the columns other than the “N-column” are all “direct sum
exact sequences”, and the horizontal maps in the middle row are “block upper
triangular”.
Then truncate (removing the right column 0 → N ′ → N → N ′′ → 0), tensor
with M (obtaining a short exact sequence of complexes) and take cohomology,
yielding the desired long exact sequence. □

23.1.C. E XERCISE ( THE Tor1 CRITERION FOR FLATNESS ). Show that the following
are equivalent conditions on an A-module M.
(i) M is flat.
(ii) TorA
i (M, N) = 0 for all i > 0 and all A-modules N.
(iii) TorA
1 (M, N) = 0 for all A-modules N.

23.1.3. Caution. Given that free modules are immediately seen to be flat, you might
think that Exercise 23.1.C implies Remark 23.1.1. This would follow if we knew
that TorA ∼ A
i (M, N) = Tori (N, M), which is clear for i = 0 (as ⊗ is symmetric), but
we won’t know this about Tori when i > 0 until Exercise 23.3.A.

23.1.D. E XERCISE . Show that the connecting homomorphism δ constructed above

is independent of all choices (of resolutions, etc.). Try to do this with as little
 © 2024 Ravi Vakil. Published by Princeton University Press. 663

annoyance as possible. (Possible hint: given two sets of choices used to build
(23.1.2.1), build a map — a three-dimensional diagram — from one version of
(23.1.2.1) to the other version.)

23.1.E. U NIMPORTANT E XERCISE . Show that TorA i (M, ·) is an additive functor

(Definition 1.5.1). (We won’t use this later, so feel free to skip it.)
We have thus established the foundations of Tor.

23.2 Derived functors in general

23.2.1. Projective resolutions. We used very little about free modules in the above
construction of Tor — in fact we used only that free modules are projective, i.e.,
those modules P such that for any surjection M / / N , it is possible to lift any
morphism P → N to P → M:

(23.2.1.1) P
exists

M //N

(As noted in §23.1, this uses the Axiom of Choice.) Equivalently, Hom(P, ·) is an
exact functor (recall that Hom(Q, ·) is always left-exact for any Q). More generally,
the same idea yields the definition of a projective object in any abelian category.
Hence by following through our entire argument with projective modules replacing
free modules throughout, (i) we can compute TorA i (M, N) by taking any projective
resolution of N, and (ii) TorA
i (M, N) = 0 for any projective A-module N.

23.2.A. E XERCISE . Show that an object P is projective if and only if every short
exact sequence 0 → Q → R → P → 0 splits. Hence show that an A-module P is
projective if and only if P is a direct summand of a free module.

23.2.B. E XERCISE . Show that projective modules are flat. Hint: Exercise 23.2.A.
Be careful if you want to use Exercise 23.1.C; see Caution 23.1.3. (In fact, finitely
generated projective modules over local rings are even free, see Remark 24.3.6.)

23.2.2. Definition: Derived functors.

The above description was low-tech, but immediately generalizes drastically.
All we are using is that M ⊗A · is a right-exact functor, and that for any A-module
N, we can find a surjection P / / N from a projective module. In general, if F is
any right-exact covariant functor from the category of A-modules to any abelian
category, this construction will define a sequence of functors Li F such that L0 F = F
and the Li F’s give a long exact sequence. We can make this more general still. We
say that an abelian category has enough projectives if for any object N there is a
surjection onto it from a projective object. Then if F is any right-exact covariant
functor from an abelian category with enough projectives to any abelian category,
then we can define the left derived functors to F, denoted Li F (i ≥ 0). You should
reread §23.1 and see that throughout we only use the fact we have a projective
664 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

resolution (repeatedly lifting maps as in (23.2.1.1)), as well as the fact that F sends
products to products (a consequence of additivity of the functor, see Remark 1.5.2)
to show that F applied to (23.1.2.1) preserves the exactness of the columns.

23.2.C. E XERCISE . The notion of an injective object in an abelian category is dual

to the notion of a projective object.
(a) State precisely the definition of an injective object.
(b) Define derived functors for (i) covariant left-exact functors (these are called right
derived functors), (ii) contravariant left-exact functors (also called right derived
functors), and (iii) contravariant right-exact functors (these are called left derived
functors), making explicit the necessary assumptions of the category having enough
injectives or projectives.

23.2.3. Notation. Recall from Definition 23.2.2 that if F is a right-exact functor, its
(left) derived functors are denoted Li F (i ≥ 0, with L0 F = F). Similarly, if F is a
left-exact functor, its (right-) derived functors are denoted Ri F. The i is a superscript,
to indicate that the long exact sequence is “ascending in i”.

23.2.4. The Ext functors.

23.2.D. E ASY E XERCISE ( AND DEFINITION ): Ext FUNCTORS FOR A- MODULES , FIRST
VERSION . As Hom(·, N) is a contravariant left-exact functor in ModA , which has
enough projectives, define ExtiA (M, N) as the ith right derived functor of Hom(·, N),
applied to M. State the corresponding long exact sequence for Ext-modules.

23.2.E. E ASY E XERCISE ( AND DEFINITION ): Ext FUNCTORS FOR A- MODULES , SEC -
OND VERSION . The category ModA has enough injectives (see §23.2.5). As
Hom(M, ·) is a covariant left-exact functor in ModA , define ExtiA (M, N) as the
ith right derived functor of Hom(M, ·), applied to N. State the corresponding long
exact sequence for Ext-modules.
We seem to have a problem with the previous two exercises: we have defined
i
Ext (M, N) twice, and we have two different long exact sequences! Fortunately,
these two definitions of Ext agree (see Exercise 23.3.B), and two long exact sequences
for Ext are better than one.

23.2.F. E ASY E XERCISE . Use the definition of Ext in Exercise 23.2.D to show that
if A is a Noetherian ring, and M and N are finitely generated A-modules, then
ExtiA (M, N) is a finitely generated A-module.
Ext-functors (for sheaves) will play a key role in the proof of Serre duality, see
§29.2.

23.2.5. ⋆ The category of A-modules has enough injectives. We will need the fact
that ModA has enough injectives, but the details of the proof won’t come up again,
so feel free to skip this discussion.

23.2.G. E XERCISE . Suppose Q is an A-module, such that for every ideal I ⊂ A,

every homomorphism I → Q extends to A → Q. Show that Q is an injective
A-module. Hint: suppose N ⊂ M is an inclusion of A-modules, and we are given
β : N → Q. We wish to show that β extends to M → Q. Use Zorn’s Lemma to
 © 2024 Ravi Vakil. Published by Princeton University Press. 665

show that among those A-modules N ′ with N ⊂ N ′ ⊂ M, such that β extends to

N ′ , there is a maximal one. If this N ′ is not M, give an extension of β to N ′ + Am,
where m ∈ M \ N ′ , obtaining a contradiction.

23.2.H. E ASY E XERCISE ( USING Z ORN ’ S L EMMA ). Show that a Z-module (i.e.,
abelian group) Q is injective if and only if it is divisible (i.e., for every q ∈ Q and
n ∈ Z̸=0 , there is q ′ ∈ Q with nq ′ = q). Hence show that any quotient of an
injective Z-module is also injective.

23.2.I. E XERCISE . Show that the category of Z-modules ModZ = Ab has enough
injectives. (Hint: if M is a Z-module, then write it as the quotient of a free Z-module
F by some K. Show that M is contained in the divisible group (F ⊗Z Q)/K.)

23.2.J. E XERCISE . Suppose Q is an injective Z-module, and A is a ring. Show that

HomZ (A, Q) is an injective A-module. Hint: First describe the A-module structure
on HomZ (A, Q). You will only use the fact that Z is a ring, and that A is an algebra
over that ring.

23.2.K. E XERCISE . Show that ModA has enough injectives. Hint: suppose M is an
A-module. By Exercise 23.2.I, we can find an inclusion of Z-modules M ,→ Q where
Q is an injective Z-module. Describe a sequence of inclusions of A-modules
 / HomZ (A, M)  
M / HomZ (A, Q).

(The A-module structure on HomZ (A, M) is via the A-action on the left argument
A, not via the A-action on the right argument M.) The right term is injective by the
previous Exercise 23.2.J.

23.2.6. ⋆ Universal δ-functors.

(This discussion is best skipped on a first reading; you should move directly to
§23.3. We won’t use the notion of δ-functors until §29.2.2.) We now describe a more
general variant of derived functors, as you may use them in the discussion of Serre
duality in Chapter 29. The advantage of the notion of universal δ-functor is that we
can apply it even in cases where A does not have enough injectives.
Abstracting key properties of derived functors, we define the data of a (coho-
mological) δ-functor, from an abelian category A to another abelian category B.
A δ-functor is a collection of additive functors T i : A → B (where T i is taken to
be 0 if i < 0), along with morphisms δi : T i (A ′′ ) → T i+1 (A ′ ) for each short exact
sequence

(23.2.6.1) 0 / A′ /A / A ′′ /0

in A , satisfying two properties:

(i) (short exact sequences yield long exact sequences) For each short exact sequence
(23.2.6.1), the sequence

··· / T i−1 (A ′′ ) δi−1 / T i (A ′ ) / T i (A) / T i (A ′′ ) δi / T i+1 (A ′ ) / ···

(where the unlabeled maps come from the covariance of the T i ) is exact.
In particular, T 0 is left-exact.
666 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

(ii) (functoriality of (i)) For each morphism of short exact sequences in A

0 / a′ /a / a ′′ /0

  
0 / A′ /A / A ′′ /0

(where the squares commute), the δi ’s give a commutative diagram

δi / T i+1 (a ′ )
T i (a ′′ )

 
δi / T i+1 (A ′ )
T i (A ′′ )

(where the vertical arrows come from the covariance of T i and T i+1 ).
Derived functor cohomology is clearly an example of a δ-functor; Čech coho-
mology of sheaves on quasicompact separated schemes is another. (You can make
these statements precise if you wish.)

23.2.7. Remark/Definition. More precisely, we should call the above a covariant

δ-functor. There is similarly a notion of a contravariant δ-functor, whose definition
you may be able to guess.

23.2.L. E XERCISE . Figure out the right definition of morphism of δ-functors

A → B. (It should then be clear that the δ-functors from A to B form a category.)

23.2.8. Definition. A (cohomological) δ-functor (T i , δi ) is universal if for any other

δ-functor (T ′i , δ ′i ), and any natural transformation of functors α : T 0 → T ′0 , there
is a unique morphism of δ-functors (T i , δi ) → (T ′i , δ ′i ) extending α. By universal
property nonsense (and Exercise 23.2.L), given any covariant left-exact functor
F : A → B, there is at most one universal δ-functor (T i , δi ) extending F (i.e., with
∼
a natural isomorphism T 0 ←→ F). The key fact about universal δ-functors is the
following.

23.2.9. Theorem. — Suppose (T i , δi ) is a covariant δ-functor from A to B, and for all

A ∈ A , there exists a monomorphism A → J with T i J = 0 for all i > 0. Then (T i , δi ) is
universal.

23.2.M. ⋆⋆ E XERCISE . Prove Theorem 23.2.9. Partial hint: motivated by Corol-

lary 23.2.11 below, follow our discussion of derived functors. Better hint (because
this exercise is hard): follow the hints in [Weib, Exercise 2.4.5], or follow the proof
of [Lan, Ch. XX, Thm. 7.1].

23.2.10. Remark. An additive functor F : A → B is said to be effaceable if for

every A ∈ A , there is a monomorphism A → J with F(J) = 0. The hypotheses of
Theorem 23.2.9 can be weakened to require only that T i is effaceable for each i > 0,
and you are welcome to prove that instead. (Indeed, [Weib, Exercise 2.4.5], [Lan,
Ch. XX, Thm. 7.1], and the original source [Gr1, II.2.2.1] deal with this case.) We
give the statement of Theorem 23.2.9 for simplicity, as we will only use this version.
 © 2024 Ravi Vakil. Published by Princeton University Press. 667

23.2.11. Corollary. — If A has enough injectives, and F is a left-exact covariant functor

A → B, then the Ri F (with the δi that accompany them) form a universal δ-functor.

Proof. Each element of A admits a monomorphism into an injective element; this is

just the definition of “enough injectives” (Exercise 23.2.C). Higher derived functors
of an injective elements I are always 0: just compute the higher derived functor by
taking the injective resolution of I “by itself”. □

23.3 Derived functors and spectral sequences

A number of useful facts can be easily proved using spectral sequences. By

doing these exercises, you will lose any fear of spectral sequence arguments in
similar situations, as you will realize they are all the same.
Before you read this section, you should read §1.6 on spectral sequences.

23.3.1. Symmetry of Tor.

23.3.A. E XERCISE ( SYMMETRY OF Tor). Show that there is an isomorphism

∼
TorA
i (M, N) ←→ Tor A
i (N, M). (Hint: take a free resolution of M and a free resolu-
tion of N. Take their “product” to somehow produce a double complex. Use both
orientations of the obvious spectral sequence and see what you get.)
On a related note:

23.3.B. E XERCISE . Show that the two definitions of Exti (M, N) given in Exer-
cises 23.2.D and 23.2.E agree.

23.3.2. Derived functors can be computed using acyclic resolutions. Suppose

F : A → B is a right-exact additive functor of abelian categories, and that A has
enough projectives. We say that A ∈ A is F-acyclic (or just acyclic if the F is clear
from context) if Li F A = 0 for i > 0. In Exercise 23.3.D, we will see that derived
functors Li F B of an object B of A can be computed by using “acyclic resolutions”.
We set the stage with a useful construction.

23.3.3. Building a “projective resolution” of an exact sequence. Suppose · · · → E2 →

E1 → E0 → 0 is an exact sequence in an abelian category with enough projectives.
668 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

We explain how to inductively build a double complex of projectives

.. .. ..
. . .

  
··· / P2,1 / P1,1 / P0,1 /0

  
··· / P2,0 / P1,0 / P0,0 /0

  
··· / E2 / E1 / E0 /0

  
0 0 0
such that the rows and columns are all exact. Suppose you have built part of the
complex, and are trying to build the Pm,n term:
(23.3.3.1) ? / Pm−1,n / Pm−2,n

  
Pm,n−1 / Pm−1,n−1 / Pm−2,n−1

 
Pm,n−2 / Pm−1,n−2

For reasons that will soon become clear, we assume that for any b ∈ Pm−1,n−1
whose image in both Pm−1,n−2 and Pm−2,n−1 is zero, there is c ∈ Pm,n−1 whose
image in Pm−1,n−1 is b, and whose image in Pm,n−2 is zero; and symmetrically
there is c ′ ∈ Pm−1,n whose image in Pm−1,n−1 is b, and whose image in Pm−2,n is
zero.
Consider
K := ker(Pm,n−1 ⊕ Pm−1,n → Pm,n−2 ⊕ Pm−1,n−1 ⊕ Pm−2,n ).
If K surjects onto both ker(Pm−1,n → Pm−2,n ) and ker(Pm,n−1 → Pm,n−2 ), then
we could take any surjection from a projective object P ↠ K, then take Pm,n to be P
(with its map to Pm,n−1 and the negative of its map to Pm−1,n ). With this choice,
we would have ensured “horizontal exactness” at Pm−1,n , and “vertical exactness”
at Pm,n−1 , and commutativity of the square in (23.3.3.1).
We now verify our two desired surjections, by “diagram-chasing” (which we
may do, see §1.5.5). Suppose a ∈ ker(Pm−1,n → Pm−2,n ), and let b be its image
in Pm−1,n−1 . We wish to find c ∈ ker(Pm,n−1 → Pm,n−2 ) so that c maps to b in
Pm−1,n−1 and 0 in Pm,n−2 . (Then (−c, a) ∈ Pm,n−1 ⊕Pm−1,n would be our desired
element of K mapping to a ∈ Pm−1,n ). But such a c exits by our assumption! And
symmetrically, we get surjectivity K ↠ ker(Pm,n−1 → Pm,n−2 ).
The final task is to ensure these assumptions hold for later stages in the building
of our double complex. We need to ensure that for any b ′ ∈ Pm−1,n that maps
to (0, 0) ∈ Pm−1,n−1 ⊕ Pm−2,n , there is an element of Pm,n mapping to (0, b ′ ) ∈
 © 2024 Ravi Vakil. Published by Princeton University Press. 669

Pm,n−1 ⊕ Pm−1,n . And we have to ensure the analogous statement with the roles
of m and n reversed. We take Q ′ = ker(Pm−1,n → Pm−1,n−1 ⊕ Pm−2,n ), and
Q ′′ = ker(Pm,n−1 → Pm,n−2 ⊕ Pm−1,n−1 ), and take surjections P ′ ↠ Q ′ and
P ′′ ↠ Q ′′ from projective objects. Then take Pm,n to be P ⊕ P ′ ⊕ P ′′ , where the
maps from P to Pm,n−1 and Pm−1,n are as described above; the map P ′ to Pm−1,n
is the above-described map (via Q ′ ); the map P ′ to Pm,n−1 is zero; and the opposite
for P ′′ (the map to Pm−1,n is zero, and the map to Pm,n−1 is as implied above,
by way of Q ′′ ). The summand P ′ ensures our first desired assumption, and the
summand P ′′ ensures our second.

23.3.C. E XERCISE . Verify that the above construction indeed gives a projective
resolution of an exact sequence. Where did you use that the sequence E• was exact?

Now let’s apply this.

23.3.D. E XERCISE . Show that you can compute the derived functors of an objects
B of A using acyclic resolutions (not just projective resolutions), i.e., by taking a
resolution
(23.3.3.2) ··· / A2 / A1 / A0 /B /0

by F-acyclic objects Ai , truncating, applying F, and taking homology. Hence

Tori (M, N) can be computed with a flat resolution of M or N. Hint: as describe
above, build a double complex of projectives “on top of” the exact sequence
(23.3.3.2). Remove the bottom row, and the right-most nonzero column, and then
apply F, to obtain a new double complex. Use a spectral sequence argument to show
that (i) the double complex has homology equal to Li F(B), and (ii) the homology
of the double complex agrees with the construction given in the statement of the
exercise. If this is too confusing, read more about the Cartan-Eilenberg resolution
below.

23.3.4. The Grothendieck composition-of-functors spectral sequence.

Suppose A , B, and C are abelian categories, F : A → B and G : B → C are a
left-exact additive covariant functors, and A and B have enough injectives. Thus
right derived functors of F, G, and G ◦ F exist. A reasonable question is: how are
they related?

23.3.5. Theorem (Grothendieck composition-of-functors spectral sequence). —

Suppose F : A → B and G : B → C are left-exact additive covariant functors, and A
and B have enough injectives. Suppose further that F sends injective elements of A
to G-acyclic elements of B. Then for each X ∈ A , there is a spectral sequence with
p,q
→ E2 = Rq G(Rp F(X)) converging to R• (G ◦ F)(X).
We will soon see the Leray spectral sequence as an application (Theorem 23.4.5).

There is more one might want to extract from the proof of Theorem 23.3.5.
For example, although E0 page of the spectral sequence will depend on some
choices (of injective resolutions), the E2 page will be independent of choice. For our
applications, we won’t need this refinement.
We will have to work to establish Theorem 23.3.5, so the proof is possibly best
skipped on a first reading.
670 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

23.3.6. ⋆ Proving Theorem 23.3.5.

Before we give the proof (in §23.3.8), we begin with some preliminaries to
motivate it. In order to discuss derived functors applied to X, we choose an
injective resolution of X:

0 /X / I0 / I1 / ··· .

To compute the derived functors Rp F(X), we apply F to the injective resolution I• :

0 / F(I0 ) / F(I1 ) / F(I2 ) / ··· .

Note that F(Ip ) is G-acyclic, by hypothesis of Theorem 23.3.5. If we were to follow

our nose, we might take simultaneous injective resolutions I•,• of the terms in
the above complex F(I• ) (the “dual” of §23.3.3 — note that only the columns are
required to be exact), and apply G, and consider the resulting double complex:

.. .. ..
(23.3.6.1) .O .O .O

0 / G(I0,2 ) / G(I1,2 ) / G(I2,2 ) / ···

O O O

0 / G(I0,1 ) / G(I1,1 ) / G(I2,1 ) / ···

O O O

0 / G(I0,0 ) / G(I1,0 ) / G(I2,0 ) / ···

O O O

0 0 0

23.3.E. E XERCISE . Consider the spectral sequence with upward orientation, starting
with (23.3.6.1) as page E0 . Show that Ep,q
2 is Rp (G ◦ F)(X) if q = 0, and 0 otherwise.
We now see half of the terms in the conclusion of Theorem 23.3.5; we are
halfway there. To complete the proof, we would want to consider another spectral
sequence, with rightward orientation, but we need to know more about (23.3.6.1);
we will build it more carefully.

23.3.7. Cartan-Eilenberg resolutions.

 © 2024 Ravi Vakil. Published by Princeton University Press. 671

Suppose · · · → Cp−1 → Cp → Cp+1 → · · · is a complex in an abelian category

B. We will build an injective resolution of C•
.. .. ..
(23.3.7.1) .O .O .O

0 / I0,2 / I1,2 / I2,2 / ···

O O O

0 / I0,1 / I1,1 / I2,1 / ···

O O O

0 / I0,0 / I1,0 / I2,0 / ···

O O O

0 / C0 / C1 / C2 / ···
O O O

0 0 0
satisfying some further properties.
We first define some notation for functions on a complex.
• Let Zp (K• ) be the kernel of the pth differential of a complex K• .
• Let Bp+1 (K• ) be the image of the pth differential of a complex K• . (The
superscript is chosen so that Bp+1 (K• ) ⊂ Kp+1 .)
• As usual, let Hp (K• ) be the cohomology at the pth step of a complex K• .
For each p, we have complexes
.. .. ..
(23.3.7.2) .O .O .O

Zp (I•,1 ) Bp (I•,1 ) Hp (I•,1 )

O O O

Zp (I•,0 ) Bp (I•,0 ) Hp (I•,0 )

O O O

Zp (C• ) Bp (C• ) Hp (C• )

O O O

0 0 0
We will construct (23.3.7.1) so that the three complexes (23.3.7.2) are all injective
resolutions (of their first nonzero terms). We begin by choosing injective resolutions
Bp,∗ of Bp (C• ) and Hp,∗ of Hp (C• ); these will eventually be the last two lines of
(23.3.7.2).
672 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

23.3.F. E XERCISE . Describe an injective resolution Zp,∗ of Zp (C• ) (the first line of
(23.3.7.2)) making the following diagram a short exact sequence of complexes.
.. .. ..
(23.3.7.3) .O .O .O

0 / Bp,1 / Zp,1 / Hp,1 /0

O O O

0 / Bp,0 / Zp,0 / Hp,0 /0

O O O

0 / Bp (C• ) / Zp (C• ) / Hp (C• ) /0

O O O

0 0 0
Hint: the “dual” problem was solved in (23.1.2.1), by a “horseshoe construction”.

23.3.G. E XERCISE . Describe an injective resolution Ip,∗ of Cp making the following

diagram a short exact sequence of complexes.
.. .. ..
(23.3.7.4) .O .O .O

0 / Zp,1 / Ip,1 / Bp+1,1 /0

O O O

0 / Zp,0 / Ip,0 / Bp+1,0 /0

O O O

0 / Zp (C• ) / Cp / Bp+1 (C• ) /0

O O O

0 0 0
(The hint for the previous problem applies again. We remark that the bottom
nonzero rows of (23.3.7.3) and (23.3.7.4) appeared in (1.5.6.3).)

23.3.H. E XERCISE /D EFINITION . Build an injective resolution (23.3.7.1) of C• such

that Zp,∗ = Zp (I•,∗ ), Bp,∗ = Bp (I•,∗ ), Hp,∗ = Hp (I•,∗ ), so the three complexes
(23.3.7.2) are injective resolutions. This is called a Cartan-Eilenberg resolution of
C• .

23.3.8. Proof of the Grothendieck spectral sequence, Theorem 23.3.5. We pick up

where we left off before our digression on Cartan-Eilenberg resolutions. Choose
an injective resolution I• of X. Apply the functor F, then take a Cartan-Eilenberg
resolution I•,• of FI• , and then apply G, to obtain (23.3.6.1).
 © 2024 Ravi Vakil. Published by Princeton University Press. 673

Exercise 23.3.E describes what happens when we take (23.3.6.1) as E0 in a

spectral sequence with upward orientation. So we now consider the rightward
orientation.
From our construction of the Cartan-Eilenberg resolution, we have injective
resolutions (23.3.7.2), and short exact sequences

(23.3.8.1) 0 / Bp (I•,q ) / Zp (I•,q ) / Hp (I•,q ) /0

(23.3.8.2) 0 / Zp (I•,q ) / Ip,q / Bp+1 (I•,q ) /0

of injective objects (from the columns of (23.3.7.3) and (23.3.7.4)). This means that
both are split exact sequences (the central term can be expressed as a direct sum of
the outer two terms), so upon application of G, both exact sequences remain exact.
Applying the left-exact functor G to

0 / Zp (I•,q ) / Ip,q / Ip+1,q ,

we find that GZp (I•,q ) = ker(GIp,q → GIp+1,q ). But this kernel is the definition
of Zp (GI•,q ), so we have an induced isomorphism GZp (I•,q ) = Zp (GI•,q ) (“G
and Zp commute”). From the exactness of (23.3.8.2) upon application of G, we see
that GBp+1 (I•,q ) = Bp+1 (GI•,q ) (both are coker(GZp (I•,q ) → GIp,q )). From the
exactness of (23.3.8.1) upon application of G, we see that GHp (I•,q ) = Hp (GI•,q )
(both are coker(GBp (I•,q ) → GZp (I•,q )) — so “G and Hp commute”).
We return to considering the rightward-oriented spectral sequence with (23.3.6.1)
as E0 . Taking cohomology in the rightward direction, we find Ep,q
1 = Hp (GI•,q ) =
p •,q p p •,q
GH (I ) (as G and H commute). Now H (I ) is an injective resolution of
(Rp F)(X) (the last resolution of (23.3.7.2)). Thus when we compute E2 by using the
vertical arrows, we find Ep,q
2 = Rq G(Rp F(X)).
You should now verify yourself that this (combined with Exercise 23.3.E) con-
cludes the proof of Theorem 23.3.5. □

23.4 Derived functor cohomology of O-modules

We wish to apply the machinery of derived functors to define cohomology of
quasicoherent sheaves on a scheme X. Rather than working in the category QCohX ,
for a number of reasons it is simpler to work in the larger category ModOX (but see
Unimportant Remark 23.5.7).

23.4.1. Theorem. — Suppose (X, OX ) is a ringed space. Then the category of OX -modules
ModOX has enough injectives.
As a side benefit (of use to others more than us), taking OX = Z, we see that
the category of sheaves of abelian groups on a fixed topological space have enough
injectives.

23.4.2. Proof. We prove Theorem 23.4.1 in a series of exercises, following Godement,

[GrS, p. 27-28]. Suppose F is an OX -module. We will exhibit an injection F ,→ Q ′
into an injective OX -module. For each p ∈ X, choose an inclusion Fp ,→ Qp into
674 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

an injective OX,p -module (which is possible as the category of OX,p -modules has
enough injectives, Exercise 23.2.K).

23.4.A. E XERCISE . Show that the skyscraper sheaf Qp := ip,∗ Qp , with module
Qp at point p ∈ X, is an injective OX -module. (You can cleverly do this by abstract
nonsense, using Exercise 23.5.B, but it is just as quick to do it by hand.)

23.4.B. E ASY E XERCISE . Show that the product (possibly infinite) of injective
objects in an abelian category is also injective.
Q
By the previous two exercises, Q ′ := p∈X Qp is an injective OX -module.

23.4.C. E ASY E XERCISE . By considering stalks, show that the natural map F → Q ′
is an injection.
This completes the proof of Theorem 23.4.1. □
We can now make a number of definitions.

23.4.3. Definition. If (X, OX ) is a ringed space, and F is an OX -module, define

Hi (X, F ) as Ri Γ (X, F ). If furthermore π : (X, OX ) → (Y, OY ) is a map of ringed
spaces, we have higher pushforwards (or derived pushforwards) Ri π∗ : ModOX →
ModOY .
We have defined these notions earlier in special cases, for quasicoherent sheaves
on quasicompact separated schemes (for Hi ), or for quasicompact separated mor-
phisms of schemes (for Ri π∗ ), in Chapter 18. We will soon (§23.5) show that these
older definitions agree with Definition 23.4.3. Thus the derived functor definition
applies much more generally than our Čech definition. But it is worthwhile to
note that almost everything we use will come out of the Čech definition. A notable
exception is the Leray spectral sequence, which we now discuss.

23.4.4. The Leray spectral sequence.

23.4.5. Theorem (Leray spectral sequence). — Suppose π : (X, OX ) → (Y, OY ) is a

morphism of ringed spaces. For any OX -module F , there is a spectral sequence with E2
term given by Hq (Y, Rp π∗ F ) abutting to Hp+q (X, F ).
This is an immediate consequence of the Grothendieck composition-of-functors
spectral sequence (Theorem 23.3.5) once we prove that the pushforward of an
injective O-module is an acyclic O-module. We do this now.

23.4.6. Definition. We make an intermediate definition that is independently

important. A sheaf F on a topological space is flasque (also sometimes called flabby)
if all restriction maps are surjective, i.e., if resU⊂V : F (V) → F (U) is surjective for
all U ⊂ V.

23.4.D. E XERCISE . Suppose (X, OX ) is a ringed space.

(a) Show that if

(23.4.6.1) 0 / F′ /F / F ′′ /0

is an exact sequence of OX -modules, and F ′ is flasque, then (23.4.6.1) is exact on sec-

tions over any open set U, i.e. 0 / F ′ (U) / F (U) / F ′′ (U) /0
 © 2024 Ravi Vakil. Published by Princeton University Press. 675

is exact.
(b) Given an exact sequence (23.4.6.1), if F ′ is flasque, show that F is flasque if
and only if F ′′ is flasque.

23.4.E. E ASY E XERCISE ( PUSHFORWARD OF FLASQUES ARE FLASQUE ).

(a) Suppose π : X → Y is a continuous map of topological spaces, and F is a flasque
sheaf of sets on X. Show that π∗ F is a flasque sheaf on Y.
(b) Suppose π : (X, OX ) → (Y, OY ) is a morphism of ringed spaces, and F is a
flasque OX -module. Show that π∗ F is a flasque OY -module.

23.4.7. Extension by zero, an occasional left adjoint to the inverse image functor. In
order to prove that injective sheaves are flasque (Exercise 23.4.H), we introduce
the “extension by zero” functor. In addition to always being a left adjoint, π−1 can
sometimes be a right adjoint, when π is an inclusion of an open subset. Suppose
i : U ,→ Y is an inclusion of an open set into Y. Define the extension of i by
zero i! : ModOU → ModOY as follows. Suppose F is an OU -module. For open
pre
W ⊂ Y, define (i! F )(W) = F (W) if W ⊂ U, and 0 otherwise (with the obvious
pre
restriction maps). This is clearly a presheaf OY -module. Define i! as (i! )sh . Note
that i! F is an OY -module, and that this defines a functor. (The symbol “!” is read
as “shriek”, presumably because people shriek when they see it. Thus “i! ” is read
as “i-lower-shriek”.)
pre
23.4.F. E XERCISE . Show that i! F satisfies the identity axiom, but need not be
a sheaf. (We won’t need this, but it may give some insight into why this is called
“extension by zero”. Possible source for an example: continuous functions on R.)

23.4.G. E XERCISE .
(a) For q ∈ Y, show that (i! F )q = Fq if q ∈ U, and 0 otherwise.
(b) Show that i! is an exact functor.
(c) If G is an OY -module, describe an inclusion i! i−1 G ,→ G . (Interesting remark
we won’t need: Let Z be the complement of U, and j : Z → Y the natural inclusion.
Then there is a short exact sequence

0 / i! i−1 G /G / j∗ j−1 G / 0.

This is best checked by describing the maps, then checking exactness at stalks.)
(d) Show that (i! , i−1 ) is an adjoint pair, so there is a natural bijection

∼
HomOY (i! F , G ) o / HomO (F , G |U )
U

for any OU -module F and OY -module G . (In particular, the sections of G over U
can be identified with HomOY (i! OU , G ).)
We are ready to use the tool of the extension by zero functor.

23.4.H. E XERCISE ( INJECTIVE SHEAVES ARE FLASQUE ). Suppose (X, OX ) is a

ringed space, and Q is an injective OX -module. Show that Q is flasque. Hint: If
U ⊂ V ⊂ X, then describe an injection of OX -modules 0 → (iU )! OU → (iV )! OV ,
where iU : U ,→ X and iV : V ,→ X are the obvious open embeddings. Apply the
exact contravariant functor Hom(·, Q).
676 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

23.4.I. E XERCISE ( FLASQUE IMPLIES Γ - ACYCLIC ). Suppose F is a flasque OX -

module. Show that F is Γ -acyclic (that Hi (X, F ) = 0 for i > 0, §23.3.2) as follows.
As ModOX has enough injectives, choose an inclusion of F into some injective I ,
and call its cokernel G :

0 /F /I /G / 0.

Then I is flasque by Exercise 23.4.H, so G is flasque by Exercise 23.4.D(b). Take the

long exact sequence in (derived functor) cohomology, and show that H1 (X, F ) = 0.
Your argument works for any flasque sheaf F , so H1 (X, G ) = 0 as well. Show that
H2 (X, F ) = 0. Turn this into an induction.
Thus if π : X → Y is a morphism of ringed spaces, and I is an injective OX -
module, then I is flasque (Exercise 23.4.H), so π∗ I is flasque (Exercise 23.4.E(b)),
so π∗ I is acyclic for the functor Γ (Exercise 23.4.I), so this completes the proof of
the Leray spectral sequence (Theorem 23.4.5). □

23.4.J. E XERCISE . Extend the Leray spectral sequence (Theorem 23.4.5) to deal with
a composition of higher pushforwards for

(X, OX )
π / (Y, OY ) ρ
/ (Z, OZ ).

23.4.8. ⋆⋆ The category of OX -modules on a ringed space need not have enough
projectives. In contrast to Theorem 23.4.1, the category of OX -modules on a ringed
space need not have enough projectives. For example, let X be P1k with the Zariski-
topology (in fact we will need very little about X — only that it is not an Alexandrov
space), but take OX to be the constant sheaf Z. We will see that ModOX — i.e., the
category of sheaves of abelian groups on X — does not have enough projectives. If
ModOX had enough projectives, then there would be a surjection ψ : P → Z from
a projective sheaf. Fix a closed point q ∈ X. We will show that the map on stalks
ψq : Pq → Zq is the zero map, contradicting the surjectivity of ψ. For each open
subset U of X, denote by ZU the extension by zero from U to X of the constant sheaf
with values in Z (§23.4.7 — ZU (V) = Z if V ⊂ U, and ZU (V) = 0 otherwise). For
each open neighborhood V of q, let W be a strictly smaller open neighborhood.
Consider the surjection ZX−q ⊕ ZW → Z. By projectivity of P, the surjection ψ lifts
to P → ZX−q ⊕ ZW . The map P(V) → Z(V) factors through ZX−q (V) ⊕ ZW (V) = 0,
and hence must be the zero map. Thus the map ψq : Pq → Zq map is zero as well
(Do you see why?) as desired.

23.5 Čech cohomology and derived functor cohomology agree

We next prove that Čech cohomology and derived functor cohomology agree,
where the former is defined.

23.5.1. Theorem. — Suppose X is a quasicompact separated scheme, and F is a

quasicoherent sheaf. Then the Čech cohomology of F agrees with the derived functor
cohomology of F .
 © 2024 Ravi Vakil. Published by Princeton University Press. 677

This statement is not as precise as it should be. We would want to know that
this isomorphism is functorial in F , and that it respects long exact sequences (so
the connecting homomorphism defined for Čech cohomology agrees with that
for derived functor cohomology). There is also an important extension to higher
pushforwards. We leave these issues for the end of this section, §23.5.5.
In case you are curious: so long as it is defined appropriately, Čech cohomology
agrees with derived functor cohomology in a wide variety of circumstances outside
of scheme theory (if the underlying topological space is paracompact), but not
always (see [Gr1, §3.8] for a counterexample). Remark 23.5.6 is also related.
The central idea in the proof (albeit with a twist) is a spectral sequence argu-
ment in the same style as those of §23.3, and uses two “cohomology-vanishing”
ingredients, one for each orientation of the spectral sequence.
(A) If (X, OX ) is a ringed space, Q is an injective OX -module, and X = ∪i Ui is a
finite open cover, then Q has no ith Čech cohomology with respect to this cover for
i > 0.
(B) If X is an affine scheme, and F is a quasicoherent sheaf on X, then (Ri Γ )(F ) =
0 for i > 0.
Translation: (A) says that building blocks of derived functor cohomology have
no Čech cohomology, and (B) says that building blocks of Čech cohomology have
no derived functor cohomology.

23.5.A. P RELIMINARY E XERCISE . Suppose (X, OX ) is a ringed space, Q is an

injective O-module, and i : U ,→ X is an open subset. Show that Q|U is injective
on U. Hint: use the fact that i−1 has an exact left adjoint i! (extension by zero), see
§23.4.7, and the following diagrams.

0 /A /B 0 / i! A / i! B

 } ?
 | ?

Q|U Q

In the course of Exercise 23.5.A, you will have proved the following fact, which
we shall use again in Exercise 29.4.A. (You can also use it to solve Exercise 23.4.A.)

23.5.B. E XERCISE ( RIGHT ADJOINTS TO EXACT FUNCTORS PRESERVE INJECTIVES ).

Show that if (F, G) is an adjoint pair of additive functors between abelian categories,
and F is exact, then G sends injective elements to injective elements.

23.5.2. Proof of Theorem 23.5.1, assuming (A) and (B). As with the facts proved in §23.3,
we take the only approach that is reasonable: we choose an injective resolution
0 → F → Q• of F and a Čech cover of X, mix these two types of information
in a double complex, and toss it into our spectral sequence machine (§1.6). More
precisely, choose a finite affine open cover X = ∪i Ui and an injective resolution

0 /F / Q0 / Q1 / ··· .
678 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Consider the double complex

.. .. .. ..
(23.5.2.1) .O .O .O .O

0O / ⊕i Q2 (Ui ) / ⊕i,j Q2 (Uij ) / ⊕i,j,k Q2 (Uijk ) / ···

O O O

0O / ⊕i Q1 (Ui ) / ⊕i,j Q1 (Uij ) / ⊕i,j,k Q1 (Uijk ) / ···

O O O

0O / ⊕i Q0 (Ui ) / ⊕i,j Q0 (Uij ) / ⊕i,j,k Q0 (Uijk ) / ···

O O O

0 /0 /0 /0 / ···

We take this as the E0 term in a spectral sequence. First, we use the rightward
filtration. As higher Čech cohomology of injective O-modules is 0 (assumption
(A)), we get 0’s everywhere except in “column 0”, where we get Qi (X) in row i:

.. .. .. ..
.O .O .O .O

0O Q2 (X) 0O 0O ···
O

0O Q1 (X) 0O 0O ···
O

0O Q0 (X) 0O 0O ···
O

0 0 0 0
 © 2024 Ravi Vakil. Published by Princeton University Press. 679

Then we take cohomology in the vertical direction, and we get derived functor
cohomology of F on X on the E2 page:

.. .. .. ..
. . . .

0Z R2 Γ (X, F ) 0V 0 ···
Z

0Z R1 Γ (X, F ) 0V 0 ···
Z

0 Γ (X, F ) 0 0 ···

0 0 0

We then start over on the E0 page, and this time use the upward orientation
(corresponding to choosing the upward arrow first). By Proposition 23.5.A, Qi |UJ
is injective on UJ , so we are computing the derived functor cohomology of F on
UJ . Then the higher derived functor cohomology of F on UJ is 0 (assumption (B)),
so all entries are 0 except possibly on row 0. Thus the E1 term is:
(23.5.2.2)
0 /0 /0 /0 / ···

0 /0 /0 /0 / ···

0 / ⊕i Γ (Ui , F ) / ⊕i,j Γ (Uij , F ) / ⊕i,j,k Γ (Uijk , F ) / ···

0 /0 /0 /0 / ···

Row 0 is precisely the Čech complex of F , so the spectral sequence converges at

the E2 term, yielding the Čech cohomology. Since one orientation yields derived
functor cohomology and one yields Čech cohomology, we are done. □
So it remains to show (A) and (B).

23.5.3. Ingredient (A): injectives have no Čech cohomology.

23.5.C. E XERCISE . Suppose X = ∪j Uj is a finite cover of X by open sets, and

F is a flasque sheaf (Definition 23.4.6) on X. Show that the Čech complex for F
with respect to ∪n
j=1 Uj has no cohomology in positive degree, i.e., that it is exact
except in degree 0 (where it has cohomology F (X), by the sheaf axioms). Hint:
use induction on n. Consider the short exact sequence of complexes (18.2.4.2) (see
680 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

also (18.2.3.1)). The corresponding long exact sequence will immediately give the
desired result for i > 1, and flasqueness will be used for i = 1.
Thus flasque sheaves have no Čech cohomology, so injective O-modules in
particular (Exercise 23.4.H) have none. This is all we need for our algebro-geometric
applications, but to show you how general this machinery is, we give an entertain-
ing application.

23.5.D. U NIMPORTANT E XERCISE ( PERVERSE PROOF OF INCLUSION - EXCLUSION

THROUGH COHOMOLOGY OF SHEAVES ). The inclusion-exclusion principle is
(equivalent to) the following: suppose that X is a finite set, and Ui (1 ≤ i ≤ n) are
finite sets covering X. As usual, define UI = ∩i∈I Ui for I ⊂ {1, . . . , n}. Then
X X X X
|X| = |Ui | − |U|I| | + |U|I| | − |U|I| | + · · · .
|I|=2 |I|=3 |I|=4

Prove this by endowing X with the discrete topology, showing that the constant
sheaf Q is flasque, considering the Čech complex computing Hi (X, Q) using the
cover {Ui }, and using Exercise 1.5.B.

23.5.4. ⋆ Ingredient (B): quasicoherent sheaves on affine schemes have no derived

functor cohomology.
We show the following statement by induction on k. Suppose X is an affine
scheme, and F is a quasicoherent sheaf on X. Then Ri Γ (X, F ) = 0 for 0 < i ≤ k.
The result is vacuously true for k = 0; so suppose we know the result for all
0 < k ′ < k (with X replaced by any affine scheme). Suppose α ∈ Rk Γ (X, F ). We
wish to show that α = 0. Choose an injective resolution by OX -modules
..
.O

d1

QO 1
d0

QO 0

FO

0
Then α has a representative α in Qk (X) = Γ (X, Qk ), such that dα ′ = 0. Because
′

the injective resolution is exact, α ′ is locally a boundary. In other words, in the open
neighborhood of any point p ∈ X, there is an open set Vp such that α ′ |Vp = dα ′′ for
some α ′′ ∈ Qk−1 (Vp ). (Be sure you see why this is true! Recall that taking cokernel
of a map of sheaves requires sheafification, see Proposition 2.6.1.) By shrinking Vp
if necessary, we can assume Vp is affine. By the quasicompactness of X, we can
choose a finite number of the Vp ’s that cover X. Rename these as Ui , so we have
an affine cover X. Consider the Čech cover of X with respect to this affine cover
 © 2024 Ravi Vakil. Published by Princeton University Press. 681

(not the affine cover you might have thought we would use — that of X by itself
— but instead an affine cover tailored for our particular α). Consider the double
complex (23.5.2.1), as the E0 term in a spectral sequence.
First consider the rightward orientation. As in the argument in §23.5.2, the
spectral sequence converges at E2 , where we get 0 everywhere, except that the
derived functor cohomology appears in the 0th column.
Next, start over again, choosing the upward orientation. On the E1 page, row
0 is the Čech complex, as in (23.5.2.2). All the rows between 1 and k − 1 are 0
by our inductive hypothesis, but we don’t yet know anything about the higher
rows. Because we are interested in the kth derived functor, we focus on the kth
antidiagonal (Ep,k−p
• ). The only possibly nonzero terms in this antidiagonal
Q are
Ek,0
1 and E0,k
1 . We look first at the term on the bottom row Ek,0
1 = |I|=k Γ (UI , F ),
which is part of the Čech complex:
/Q /Q /Q / ··· .
··· |I|=k−1 Γ (UI , F ) |I|=k Γ (UI , F ) |I|=k+1 Γ (UI , F )

But we have already verified that the Čech cohomology of a quasicoherent sheaf on
an affine scheme vanishes — this is the one spot where we use the quasicoherence
of F . Thus this term vanishes by the E2 page (i.e., Ek,0
i = 0 for i ≥ 2).
So the only term of interest in the kth antidiagonal of E1 is E0,k
1 , which is the
homology of
Q Q Q
/ /
(23.5.4.1) i Qk−1 (Ui ) i Qk (Ui ) i Qk+1 (Ui ),
Q
which is i Rk Γ (Ui , F ) (using Preliminary Exercise 23.5.A which stated that the
Qj |Ui are injective on Ui , so they can be used to compute Rk (Γ (Ui , F )). So E0,k
2 is
the homology of

0 / Q Rk Γ (Ui , F ) /Q Rk Γ (Uij , F )
i i,j

and thereafter all differentials to and from the E0,k

• terms will be 0, as the sources
and targets of those arrows
Q k will be 0. Consider now our lift of α ′ of our original
class α ∈ R Γ (X, F ) to i R Γ (Ui , F ). Its image in the homology of (23.5.4.1) is
k

zero — this was how we chose our cover Ui to begin with! Thus α = 0 as desired,
completing our proof. □

23.5.E. ⋆⋆ E XERCISE . The proof is not quite complete. We have a class α ∈

Rk Γ (X, F ), and we have interpreted Rk Γ (X, F ) as
Q
/ Q Rk Γ (Uij , F ) .

ker i R k
Γ (Ui , F ) i,j

We have two maps Rk Γ (X, F ) → Rk Γ (Ui , F ), one coming from the natural re-
striction (under which we can see that the image of α is zero), and one coming
from the actual spectral sequence machinery. Verify that they are the same map.
(Possible hint: with the spectral sequence orientation used, the E0,k
∞ term is indeed
the quotient of the homology of the double complex, so the map goes the right
way.)

23.5.5. ⋆ Tying up loose ends.

682 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

23.5.F. I MPORTANT E XERCISE . State and prove the generalization of Theorem 23.5.1
to higher pushforwards Ri π∗ , where π : X → Y is a quasicompact separated mor-
phism of schemes.

23.5.G. E XERCISE . Show that the isomorphism of Theorem 23.5.1 is functorial in

F , i.e., given a morphism F → G , the diagram

Hi (X, F ) o / Ri Γ (X, F )

 
Hi (X, G ) o / Ri Γ (X, G )

commutes, where the horizontal arrows are the isomorphisms of Theorem 23.5.1,
and the vertical arrows come from functoriality of Hi and Ri Γ . (Hint: “spectral
sequences are functorial in E0 ”, which can be easily seen from the construction,
although we haven’t said it explicitly. See §1.6.7, which remarks on the functoriality
of spectral sequences.)

23.5.H. E XERCISE . Show that the isomorphisms of Theorem 23.5.1 induce isomor-
phisms of long exact sequences.

23.5.6. Remark. If you wish, you can use the above argument to prove the following
theorem of Leray. Suppose we have a sheaf of abelian groups F on a topological
space X, and some covering {Ui } of X such that the (derived functor) cohomology
of F in positive degree vanishes on every finite intersection of the Ui . Then the
cohomology of F can be calculated by the Čech cohomology of the cover {Ui }; there
is no need to pass to the colimit of all covers, as is the case for Čech cohomology in
general.

23.5.7. Unimportant Remark: Working in QCohX rather than ModOX . In our definition
of derived functors of quasicoherent sheaves on X, we could have tried to work
in the category of quasicoherent sheaves QCohX itself, rather than in the larger
category ModOX . There are several reasons why this would require more effort.
It is not hard to show that QCohX has enough injectives if X is Noetherian (see
for example [Ha1, Exer. III.3.6(a)]). Because we don’t have “extension by zero”
(§23.4.7) in QCoh, the proofs that injective quasicoherent sheaves on an open set U
restrict to injective quasicoherent sheaves on smaller open subsets V (the analog of
Exercise 23.5.A) and that injective quasicoherent sheaves are flasque (the analog of
Exercise 23.4.H) are harder. You can use this to show that Hi and Ri π∗ computed
in QCoh are the same as those computed in ModO (once you make the statements
precise). It is true that injective elements of QCohX (X Noetherian) are injective in
ModOX , but this requires work (see [Mur, Prop. 68]).
It is true that QCohX has enough injectives for any scheme X, but this is much
harder, see [EE]. And as is clear from the previous paragraph, “enough injectives”
is only the beginning of what we want.

23.5.8. Unimportant Remark. Theorem 23.5.1 implies that if π : X → Y is quasicom-

pact and separated, then Ri π∗ sends QCohX to QCohY (by showing an isomorphism
with Čech cohomology). If X and Y are Noetherian, the hypothesis “separated” can
be relaxed to “quasiseparated”, using the ideas of Unimportant Remark 23.5.7. But
 © 2024 Ravi Vakil. Published by Princeton University Press. 683

because morphisms from a Noetherian scheme are automatically quasiseparated

(Easy Exercise 8.3.B(b)), we can then remove the redundant “quasiseparated” as-
sumption. This sounds exciting, but it is not nearly as useful as the separated case,
because without Čech cohomology, it is hard to compute anything.
 CHAPTER 24

Flatness

The concept of flatness is a riddle that comes out of algebra, but which technically is
the answer to many prayers.
— D. Mumford [Mu7, III.10]
We return to the important concept of flatness. We could have discussed flatness
at length as soon as we had discussed quasicoherent sheaves and morphisms. But
it is an unexpected idea, and the algebra and geometry are not obviously connected,
so we have left it for relatively late. Serre has stated that he introduced flatness
purely for reasons of algebra in his landmark “GAGA” paper [Se3], and that it
was Grothendieck who recognized its geometric significance. “Exact” seems more
descriptive terminology than “flat” (so we would have “exact modules” and “exact
morphisms”), but this has not caught on.

24.0.1. A flat morphism π : X → Y is the right notion of a “nice”, or “nicely varying”

family over Y. For example, if π is a projective flat family over a connected Noether-
ian base (translation: π : X → Y is a projective flat morphism, with Y connected and
Noetherian), we will see that various numerical invariants of fibers are constant,
including the dimension (§24.5.5), and numbers interpretable in terms of an Euler
characteristic (see §24.7):
(a) the Hilbert polynomial (Corollary 24.7.2),
(b) the degree (in projective space) (Exercise 24.7.A(a)),
(b’) (as a special case of (b)) if π is finite, the degree of π (recovering and
extending the fact that the degree of a projective map between regular
curves is constant, §16.3.4, see Exercise 24.3.G and §24.3.12),
(c) the arithmetic genus (Exercise 24.7.A(b)),
(d) the degree of a line bundle if the fiber is a curve (Corollary 24.7.3), and
(e) intersections of divisors and line bundles (Exercise 24.7.D).
One might think that the right hypothesis might be smoothness, or more generally
some sort of equisingularity, but we only need something weaker. And this is a
good thing: branched covers are not fibrations in any traditional sense, yet they
still behave well — the double cover A1 → A1 given by y 7→ x2 has constant
degree 2 (§10.3.4, revisited in §16.3.10). Another key example is that of a family
of smooth curves degenerating to a nodal curve (Figure 24.1) — the topology of
the (underlying analytic) curve changes, but the arithmetic genus remains con-
stant. One can prove things about regular curves by first proving them about a

This material is published by Princeton University Press as The Rising Sea: Foundations of Algebraic
Geometry by Ravi Vakil. This version is free to view and download for personal use only. Not for
redistribution, resale, or use in derivative works. © 2024 Ravi Vakil.

685
686 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

nodal degeneration, and then showing that the result behaves well in flat families.
Degeneration techniques such as this are ubiquitous in algebraic geometry.

F IGURE 24.1. A flat family of smooth curves degenerating to a

nodal curve: y2 = x3 + x2 + tx.

Given the cohomological nature of the constancy of Euler characteristic result,

you should not be surprised that the hypothesis needed (flatness) is cohomological
in nature — it can be characterized by vanishing of Tor (Exercise 23.1.C), which we
use to great effect in §24.2.
But flatness is important for other reasons too. As a start: as this is the right
notion of a “nice family”, it allows us to correctly define the notion of moduli space.
For example, the Hilbert scheme of Pn “parametrizes closed subschemes of Pn ”.
Maps from a scheme B to the Hilbert scheme correspond to (finitely presented)
closed subschemes of Pn B flat over B. By universal property nonsense, this defines
the Hilbert scheme up to unique isomorphism (although we of course must show
that it exists, which takes some effort — see Theorem 25.3.1 for more). The moduli
space of smooth projective curves is defined by the universal property that maps to
the moduli space correspond to projective flat (finitely presented) families whose
geometric fibers are smooth curves. (Sadly, this moduli space does not exist...) On
a related note, flatness is central in deformation theory: it is key to understanding
how schemes (and other geometric objects, such as vector bundles) can deform
(cf. §21.5.12). Finally, the notion of Galois descent generalizes to (faithfully) “flat
descent”, which allows us to “glue” in more exotic Grothendieck topologies in the
same way we do in the Zariski topology (or more classical topologies); but this is
beyond the scope of our current discussion.
Flatness has many aspects of different flavors, and it is easy to lose sight of
the forest for the trees. Because the algebra of flatness seems so unrelated to
the geometry, it can be nonintuitive. We will necessarily begin with algebraic
foundations, but you should focus on the following points: methods of showing
things are flat (both general criteria and explicit examples), and classification of flat
modules over particular kinds of rings. You should try every exercise dealing with
explicit examples such as these.
Here is an outline of the chapter, to help focus your attention.
 © 2024 Ravi Vakil. Published by Princeton University Press. 687

• In §24.1, we discuss some of the easier facts, which are algebraic in nature.
• §24.2, §24.3, and §24.6 give ideal-theoretic criteria for flatness. §24.2 and
§24.3 should be read together. The first uses Tor to understand flatness,
and the second uses these insights to develop ideal-theoretic criteria for
flatness. §24.6, on local criteria for flatness, is harder.
• §24.5 is relatively free-standing, and could be read immediately after
§24.1. It deals with topological aspects of flatness, such as the fact that
flat morphisms are open maps in good situations. (Recall that a map of
topological spaces is an open map if the image of every open set is an
open set.)
• In §24.7, we explain the fact that “the Euler characteristic of quasicoherent
sheaves is constant in flat families” (with appropriate hypotheses), and
its many happy consequences. This section is surprisingly easy given its
utility.
You should focus on what flatness implies and how to “picture” it, but also on
explicit criteria for flatness in different situations, such as for integral domains
(Observation 24.1.2), principal ideal domains (Exercise 24.3.B), discrete valuation
rings (Exercise 24.3.C), the dual numbers (Exercise 24.3.D), and local rings (Theo-
rem 24.3.5).
Happy families are all alike; every unhappy family is unhappy in its own way. (See,
for example, Figure 24.2, and Exercises 24.1.G(d), 24.1.L, and 24.3.H.)
— L. Tolstoy, Anna Karenina, [To, l. 1]

24.1 Easier facts

Many facts about flatness are easy or immediate, although a number are tricky.
As always, I will try to make clear which is which, to help you remember the easy
facts and the key ideas of proofs of the harder facts. We will pick the low-hanging
fruit first.
We recall the definition of a flat A-module (§1.5.13). If M ∈ ModA , M ⊗A · is
always right-exact (Exercise 1.2.H). We say that M is a flat A-module (or flat over A
or A-flat) if M ⊗A · is an exact functor. We say that a ring morphism B → A is flat if
A is flat as a B-module. (In particular, the algebra structure of A is irrelevant.)

24.1.1. Two key examples.

(i) Free A-modules (even of infinite rank) are clearly flat. More generally,
projective modules are flat (Exercise 23.2.B).
(ii) Localizations are flat: Suppose S is a multiplicative subset of B. Then
B → S−1 B is a flat ring morphism because localization is exact (Exercise 1.5.G(a)).

24.1.A. E ASY E XERCISE : FIRST EXAMPLES .

(a) (trick question) Classify flat modules over a field k.
(b) Show that A[x1 , . . . , xn ] is a flat A-module.
(c) Show that the ring morphism Q[x] → Q[y], with x 7→ y2 , is flat. (This will help
us understand Example 10.3.4 better, see §24.3.12.)
We make some quick but important observations.
688 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

24.1.2. Important Observation. If x is a non-zerodivisor of A, and M is a flat

A-module, then

M
×x
/M

is injective. (Apply the exact functor M⊗A · to the exact sequence 0 / A ×x / A .)

In particular, flat modules are torsion-free. (Torsion-freeness was defined in §6.1.4.)
This observation gives an easy way of recognizing when a module is not flat. We
will use it many times.

24.1.B. E XERCISE : ANOTHER EXAMPLE . Show that a finitely generated module

over a principal ideal domain is flat if and only if it is torsion-free if and only if it is
free. Hint: Remark 6.4.4 classifies finitely generated modules over a principal ideal
domain. (Exercise 24.3.B sheds more light on flatness over a principal ideal domain.
Proposition 14.3.2 is also related.)

24.1.C. E XERCISE ( FLATNESS IS PRESERVED BY CHANGE OF BASE RING ). Show

that if M is a flat B-module, B → A is a homomorphism, then M ⊗B A is a flat
A-module.

24.1.D. E XERCISE ( FLATNESS IS PRESERVED BY COMPOSITION ). Show that if A is a

flat B-algebra, and M is A-flat, then M is also B-flat.

24.1.3. Proposition (flatness is a stalk/prime-local property). — An A-module M is

flat if and only if Mp is a flat Ap -module for all prime ideals p ⊂ A.
You are welcome to prove that it suffices to check only maximal ideals, but we
don’t need this extension.

Proof. Suppose first that M is a flat A-module. Given any exact sequence of Ap -
modules

(24.1.3.1) 0 / N′ /N / N ′′ / 0,

we wish to show that

(24.1.3.2) 0 / Mp ⊗Ap N ′ / Mp ⊗ A p N / Mp ⊗Ap N ′′ /0

is also exact. Now (24.1.3.1) is also an exact sequence of A-modules, so

(24.1.3.3) 0 / M ⊗A N ′ / M ⊗A N / M ⊗A N ′′ /0

is exact (by A-flatness of M). But M ⊗A N is canonically isomorphic to M ⊗A

Ap ⊗Ap N = Mp ⊗Ap N (and similarly for N ′ , N ′′ ), so we can rewrite (24.1.3.3) as
(24.1.3.2) (make sure you see why the maps in (24.1.3.3) and (24.1.3.2) are the same)
as desired.
Assume next that Mp is a flat Ap -module for all p. Given any short exact
sequence (24.1.3.1), we seek to show that (24.1.3.3) is exact. By right-exactness of
tensor product (Exercise 1.2.H), (24.1.3.3) is exact except possibly on the left. Let K
be ker(M ⊗A N ′ → M ⊗A N), so

(24.1.3.4) 0 /K / M ⊗A N ′ / M ⊗A N / M ⊗A N ′′ /0
 © 2024 Ravi Vakil. Published by Princeton University Press. 689

is exact; we wish to show that K = 0. Given any p, localizing (24.1.3.1) at p and

tensoring with the exact Ap -module Mp yields

(24.1.3.5) 0 / Mp ⊗A Np′ / Mp ⊗A Np / Mp ⊗A Np′′ / 0.

p p p

∼
Localizing (24.1.3.4) at p and using the isomorphisms Mp ⊗Ap Np ←→ (M ⊗A N)p ,
we obtain the exact sequence

0 / Kp / Mp ⊗Ap Np′ / M p ⊗ A p Np / Mp ⊗Ap Np′′ / 0,

which is the same as the exact sequence (24.1.3.5) except for the Kp . Hence Kp = 0
for all primes p ⊂ A, so K = 0 (by Exercise 4.1.F) as desired. □

24.1.4. Flatness for schemes.

Motivated by Proposition 24.1.3, the extension of the notion of flatness to
schemes is straightforward.

24.1.5. Definition: flat quasicoherent sheaves. We say that a quasicoherent sheaf

F on a scheme X is flat at p ∈ X if Fp is a flat OX,p -module. We say that a
quasicoherent sheaf F on a scheme X is flat (over X) if it is flat at all p ∈ X. In light
of Proposition 24.1.3, we can check this notion on an affine open cover of X.

24.1.6. Definition: flat morphism. Similarly, we say that a morphism of schemes

π : X → Y is flat at p ∈ X if OX,p is a flat OY,π(p) -module. We say that a morphism
of schemes π : X → Y is flat if it is flat at all p ∈ X. We can check flatness locally on
the source and target.
We can combine these two definitions into a single fancy definition.

24.1.7. Definition: flat quasicoherent sheaf over a base. Suppose π : X → Y is a

morphism of schemes, and F is a quasicoherent sheaf on X. We say that F is flat
(over Y) at p ∈ X if Fp is a flat OY,π(p) -module. We say that F is flat (over Y) if it is
flat at all p ∈ X.
Definitions 24.1.5 and 24.1.6 correspond to the cases X = Y and F = OX
respectively. (Definition 24.1.7 applies without change to the category of ringed
spaces, but we won’t use this.)

24.1.E. E ASY E XERCISE ( REALITY CHECK ). Show that open embeddings are flat.
Our results about flatness over rings above carry over easily to schemes.

24.1.F. E XERCISE . Show that a map of rings B → A is flat if and only if the
corresponding morphism of schemes Spec A → Spec B is flat. More generally, if
B → A is a map of rings, and M is an A-module, show that M is B-flat if and
only if M
f is flat over Spec B. (Be careful: this requires more than merely invoking
Proposition 24.1.3.) Hint: working indirectly in both directions is a good way to
proceed. “Suppose Mp is not Bq -flat. Then there is an exact sequence of Bp -modules
(hence also B-modules) 0 → N ′ → N → N ′′ → 0 such that...” “Suppose M is not
B-flat. Then there is an exact sequence of B-modules ...”
Thus if π : X → Y is an affine morphism, and F is a quasicoherent sheaf on X,
then F is flat over Y if and only if π∗ F is flat over Y.
690 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

24.1.G. E ASY E XERCISE ( EXAMPLES AND REALITY CHECKS ).

(a) If X is a scheme, and p ∈ X, show that the natural morphism Spec OX,p → X is
flat. (Hint: localization is flat, §24.1.1.)
(b) Show that An A → Spec A is flat.
(c) If F is a locally free sheaf on a scheme X, show that PF → X (Definition 17.2.4)
is flat.
(d) Show that Spec k → Spec k[t]/(t2 ) is not flat. (Draw a picture to try to see what
is not “nice” about this morphism. Some more insight about flatness of the dual
numbers will be given in the criterion of Exercise 24.3.D.)

24.1.H. E XERCISE . Show that if π : X → Y is a flat morphism, then π∗ : QCohY →

QCohX is an exact functor.

24.1.I. E XERCISE ( FLATNESS IS PRESERVED BY COMPOSITION ). Suppose π : X → Y

and F is a quasicoherent sheaf on X, flat over Y. Suppose also that ψ : Y → Z is a
flat morphism. Show that F is flat over Z.

24.1.J. E XERCISE ( FLATNESS IS PRESERVED BY BASE CHANGE ). Suppose π : X → Y

is a morphism, and F is a quasicoherent sheaf on X, flat over Y. If ρ : Y ′ → Y is any
morphism, and ρ ′ : X ×Y Y ′ → X is the induced morphism, show that (ρ ′ )∗ F is flat
over Y ′ .
(ρ ′ )∗ F F

ρ′
X ×Y Y ′ /X

π
 
Y′
ρ
/Y

In particular, using Exercise 24.1.A(a), if Y = Spec k (and X and Y ′ are any k-

schemes, and F is any quasicoherent sheaf on X), then (ρ ′ )∗ F is flat over Y ′ . For
example, X ×k Y ′ is always flat over Y ′ — “products over a field are flat over their
factors”. (Feel free to immediately generalize this further; for example, F can be a
quasicoherent sheaf on a scheme Z over X, flat over Y.)

24.1.8. Flat morphisms behave reasonably. Since flatness is clearly local on the target
from the definition, the previous two exercises show that the class of flat morphisms
is “reasonable” in the sense of §8.1.
The following exercise is very useful for visualizing flatness and non-flatness
(see for example Figure 24.2).

24.1.K. E XERCISE ( FLAT MAPS SEND ASSOCIATED POINTS TO ASSOCIATED POINTS ).

Suppose π : X → Y is a flat morphism of locally Noetherian schemes. Show that any
associated point of X must map to an associated point of Y. (Feel free to immediately
generalize this to a coherent sheaf F on X, flat over Y, without π itself needing to
be flat.) Hint: suppose π♯ : (B, n) → (A, m) is a local morphism of Noetherian local
rings (i.e., π♯ (n) ⊂ m, §7.3.1). Suppose n is not an associated prime of B. Show that
there is an element f ∈ n not in any associated prime of B (perhaps using prime
avoidance, Proposition 12.2.13), and hence is a non-zerodivisor. Show that π♯ f ∈ m
 © 2024 Ravi Vakil. Published by Princeton University Press. 691

is a non-zerodivisor of A using Observation 24.1.2, and thus show that m is not an

associated prime of A.

24.1.L. E XERCISE . Use Exercise 24.1.K to show that the following morphisms are
not flat (see Figure 24.2):
(a) Spec k[x, y]/(xy) → Spec k[x],
(b) Spec k[x, y]/(y2 , xy) → Spec k[x],
(c) Bl(0,0) A2k → A2k .
Hint for (c): first pull back to a line through the origin to obtain a something akin
to (a). (This foreshadows the statement and proof of Proposition 24.5.6, which says
that for flat morphisms “there is no jumping of fiber dimension”.)

F IGURE 24.2. Morphisms that are not flat (Exercise 24.1.L)

24.1.9. Theorem (cohomology commutes with flat base change). — Suppose

ρ′
X′ /X

π′ π
 
Y′
ρ
/Y
is a Cartesian diagram, and π (and thus π ′ ) is quasicompact and separated (so higher
pushforwards of quasicoherent sheaves by π and π ′ exist, as described in §18.7). Suppose
also that ρ is flat, and F is a quasicoherent sheaf on X. Then the natural “push-pull”
∗
morphisms (Exercise 18.7.B(a)) ρ∗ (Ri π∗ F ) → Ri π∗′ (ρ ′ F ) are isomorphisms.

24.1.M. E XERCISE . Prove Theorem 24.1.9. Hint: Exercise 18.7.B(b) is the special
case where ρ is affine. Extend it to the quasicompact and separated case using
the same idea as the solution to Exercise 14.4.C (which used Exercise 6.2.G). Your
proof of the case i = 0 will only need a quasiseparated hypothesis in place of the
separated hypothesis.
A useful special case is where Y ′ is the generic point of a reduced component of
Y. In other words, in light of Exercise 24.1.G(a), the stalk of the higher pushforward
of F at the generic point is the cohomology of F on the fiber over the generic
692 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

point. This is a first example of something important: understanding cohomology

of (quasicoherent sheaves on) fibers in terms of higher pushforwards. (We would
certainly hope that higher pushforwards would tell us something about higher
cohomology of fibers, but this is certainly not a priori clear!) In comparison to this
result, which shows that cohomology of any quasicoherent sheaf commutes with
flat base change, §24.7 and Chapter 25 deal with when and how cohomology of a
flat quasicoherent sheaf commutes with any base change.

24.1.10. Pulling back closed subschemes (and ideal sheaves) by flat morphisms.
Closed subschemes pull back particularly well under flat morphisms, and this
can be helpful to keep in mind. As pointed out in Remark 14.5.9, in the case of flat
morphisms, pullback of ideal sheaves as quasicoherent sheaves agrees with pullback
in terms of the pullback of the corresponding closed subschemes. In other words,
closed subscheme exact sequences pull back (remain exact) under flat pullbacks.
This is in fact not just a necessary condition for flatness; it is also sufficient, which
can be shown using the ideal-theoretic criterion for flatness (Theorem 24.3.1). There
is an analogous fact about pulling ideal sheaves of flat subschemes by arbitrary
pullbacks, see §24.2.2.

24.1.N. E XERCISE .
(a) Suppose D is an effective Cartier divisor on Y and π : X → Y is a flat morphism.
Show that the pullback of D to X (by π) is also an effective Cartier divisor.
(b) Use part (a) to show that under a flat morphism, regular embeddings pull back
to regular embeddings.

24.1.O. E XERCISE .
(a) Suppose π : X → Y is a morphism, and Z ,→ Y is a closed embedding cut out
by an ideal sheaf I ⊂ OY . For this exercise only, for any ideal sheaf J ⊂ OY
corresponding to a closed subscheme W ⊂ Y, let π♭ J ⊂ OX be the ideal sheaf
corresponding to W ×Y X ⊂ X (so π∗ J → π♭ J is surjective but is not always
an isomorphism, as discussed in Remark 14.5.9). Show that (π♭ I )n = π♭ (I n ).
(Products of quasicoherent ideal sheaves were defined in Exercise 15.6.E.)
k , and Z is the origin. Let J = π I be the
∗
(b) Suppose further that π is flat, Y = An
quasicoherent sheaf of algebras on X cutting out the pullback W of Z. Prove that the
graded sheaf of algebras ⊕n≥0 J n /J n+1 (do you understand the multiplication?)
is isomorphic to OW [x1 , . . . , xn ] (interpreted as a graded sheaf of algebras). (Hint:
first show that J n /J n+1 = ∼ Symn (J /J 2 ), Corollary 22.3.9. This approach will
be revisited in Exercise 24.6.E.)

24.1.P. E XERCISE .
(a) Show that blowing up commutes with flat base change. More precisely, if
π : X → Y is any flat morphism, and Z ,→ Y is any closed embedding, give a
∼
canonical isomorphism (BlZ Y) ×Y X ←→ BlZ×Y X X. (You can proceed by universal
property, using Exercise 24.1.N(a), or by using the Proj construction of the blow up
and Exercise 24.1.O.)
(b) Give an example to show that blowing up does not commute with base change
in general.
 © 2024 Ravi Vakil. Published by Princeton University Press. 693

24.2 Flatness through Tor

We defined the Tor (bi-)functor in §23.1: TorA i (M, N) is obtained by taking a

free resolution of N, removing the N, tensoring it with M, and taking homology.
Exercise 23.1.C characterized flatness in terms of Tor: M is A-flat if TorA 1 (M, N) = 0
for all N. In this section, we reap the easier benefits of this characterization, recalling
key properties of Tor when needed. In §24.3, we work harder to extract more from
Tor.
It is sometimes possible to compute Tor from its definition, as shown in the
following exercise that we will use repeatedly.

24.2.A. E XERCISE . Define (M : x) as {m ∈ M : xm = 0} ⊂ M — it consists of the

elements of M annihilated by x. If x is not a zerodivisor, show that


M/xM if i = 0;
TorA (M, A/(x)) = (M : x) if i = 1;
i


0 if i > 1.
Hint: use the resolution

0 /A ×x
/A / A/(x) /0

of A/(x).

24.2.1. Remark. As a corollary of Exercise 24.2.A, we see again that flat modules
over an integral domain are torsion-free (and more generally, Observation 24.1.2).
Also, Exercise 24.2.A gives the reason for the notation Tor — it is short for torsion.

24.2.B. E XERCISE . If B is A-flat, use the FHHF Theorem (Exercise 1.5.I(c)) to give
∼
an isomorphism B ⊗A TorA i (M, N) ←→ Tori (B ⊗ M, B ⊗ N).
B

Recall that the Tor functor is symmetric in its entries (there is an isomorphism
∼
i (M, N) ←→ Tori (N, M), Exercise 23.3.A). This gives us a quick but very
TorA A

useful result.

24.2.C. E ASY E XERCISE . If 0 → N ′ → N → N ′′ → 0 is an exact sequence of

A-modules, and N ′′ is flat (e.g., free), show that 0 → M ⊗A N ′ → M ⊗A N →
M ⊗A N ′′ → 0 is exact for any A-module M.
We would have cared about this result long before seeing Tor, so it gives some
motivation for learning about Tor. (Unimportant side question: Can you prove this
without Tor, using a diagram chase?)

24.2.D. E XERCISE ( IMPORTANT CONSEQUENCE OF E XERCISE 24.2.C). Suppose

0 → F ′ → F → F ′′ → 0 is a short exact sequence of quasicoherent sheaves on a
scheme Y, and F ′′ is flat (e.g., locally free). Show that if π : X → Y is any morphism
of schemes, the pulled back sequence 0 → π∗ F ′ → π∗ F → π∗ F ′′ → 0 remains
exact.

24.2.E. E XERCISE ( CF. E XERCISE 14.2.H FOR THE ANALOGOUS FACTS ABOUT VEC -
TOR BUNDLES ). Suppose 0 → M ′ → M → M ′′ → 0 is an exact sequence of
A-modules.
694 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

(a) If M and M ′′ are both flat, show that M ′ is too. (Hint: Recall the long ex-
act sequence for Tor, Proposition 23.1.2. Also, use that N is flat if and only if
Tori (N, N ′ ) = 0 for all i > 0 and all N ′ , Exercise 23.1.C.)
(b) If M ′ and M ′′ are both flat, show that M is too. (Same hint.)
(c) If M ′ and M are both flat, show that M ′′ need not be flat.

24.2.F. E XERCISE . If · · · → M−2 → M−1 → M0 → 0 is an exact sequence of flat

A-modules, show that it remains exact upon tensoring with any other A-module.
(One possible approach: break the exact sequence into short exact sequences.)

24.2.G. E ASY E XERCISE . If 0 → M0 → M1 → · · · → Mn → 0 is an exact sequence,

and Mi is flat for i > 0, show that M0 is flat too. (Hint: break the exact sequence
into short exact sequences.)
Exercises 24.2.F and 24.2.G will be useful.

24.2.2. Pulling back quasicoherent ideal sheaves of flat closed subschemes by

arbitrary morphisms (promised in §24.1.10). Suppose

W
α /X

 
Y
β
/Z

is a fibered product, and V ,→ X is a closed subscheme. Then Y ×Z V is a closed

subscheme of W (§10.2.2). There are two possible senses in which IV/X can be
“pulled back” to W: as a quasicoherent sheaf α∗ IV/X , and as the ideal of the “pulled
back” closed subscheme IY×Z V/W . As pointed out in Remark 14.5.9, these are
not necessarily the same, but they are the same if β is flat. We now give another
important case in which they are the same.

24.2.H. E XERCISE . If V is flat over Z (with no hypotheses on β), show that

α∗ IV/X =∼ IY× V/W . Hint: Exercise 24.2.D.
Z

24.2.3. ⋆ The Künneth formula for quasicoherent sheaves.

It is worthwhile to know that the Künneth formula holds for quasicoherent
sheaves, and that the proof is reasonable. But we will not use it, so you can read
this section only if you particularly feel like it.

24.2.I. E XERCISE . Suppose X is a separated scheme, and j : Spec A ,→ X is an

open embedding. Show that j∗ OSpec A is a quasicoherent sheaf on X, flat (over OX ).
(Partial hint: the Cancellation Theorem 11.1.1 for affine morphisms.)
Suppose further that X = ∪n i=1 Ui is a cover of X by a finite number of affine
open sets, so X is quasicompact. (We still assume X separated.) Denote the cover by
U := {Ui }. As usual with Cech covers, for I ⊂ {1, . . . , n}, define UI = ∩i∈I Ui , and
let jI : UI ,→ X be the corresponding open embedding.
For the purposes of this discussion only, define the Cech complex of sheaves for
the cover U of X, denoted C•X (U), by

0 / ⊕|I|=1 jI∗ OU / ⊕|I|=2 jI∗ OU / ··· / ⊕|I|=n jI∗ OU / 0.

I I I
 © 2024 Ravi Vakil. Published by Princeton University Press. 695

Define the augmented Cech complex of sheaves for X = {Ui }, denoted C•X,aug (F ), by
prepending OX = ⊕|I|=0 jI∗ OUI :

0 / OX / ⊕|I|=1 jI∗ OU / ⊕|I|=2 jI∗ OU / ··· / ⊕|I|=n jI∗ OU / 0.

I I I

(Be sure you understand the definition of the maps in these complexes.)

24.2.J. E XERCISE . Show that the augmented Cech complex of sheaves C•X,aug (U) is
an exact sequence of flat quasicoherent sheaves on X. (Hint: check on any affine
open subset of X.)
Before proceeding further, we prove a useful homological statement. Recall
that a finite exact sequence of flat modules remains exact upon tensoring with
any other module (Exercise 24.2.F). Recall also that an exact sequence of modules
remains exact upon tensoring with any flat module (a version of the FHHF theorem,
Exercise 1.5.I(c)). The following statement generalizes both of these naturally.

24.2.K. E XERCISE . Suppose C• is an finite exact sequence of A-modules, and F•

is an finite exact sequence of flat A-modules. Show that the total complex of the
double complex C• ⊗A F• is also an exact sequence of A-modules. (If you wish, you
might show that if C• is merely a complex, not necessarily exact, then C• ⊗A F• has
“the same cohomology” as C• , further extending the FHHF theorem. You may also
wish to remove the finiteness assumptions irrelevant to your solution.)

24.2.L. E XERCISE . We return to the situation of Exercise 24.2.J. Suppose F is a qua-

sicoherent sheaf (not necessarily flat) on X (which is quasicompact and separated).
Show that F ⊗ C•X,aug (U) is an exact sequence sequence of quasicoherent sheaves
on X. (Equivalently, F ⊗ C•X (U) is a complex, exact except at the first step, where it
has kernel/cohmology sheaf canonically identified with F .) Hint: check on each
affine open subset of X.
Following Serre, you might interpret Exercises 24.2.J and 24.2.L in terms of
partitions of unity.
We use these pleasant exact sequences to prove a form of the Künneth formula.
Suppose that X and Y are both quasicompact separated k-schemes (e.g., varieties
over k). Name the projection maps πX : X ×k Y → X and πY : X ×k Y → Y. Let UX
and UY be finite covers of X and Y (respectively) by affine open sets.

24.2.M. E XERCISE . Show that π∗X C•X,aug (UX ) is an exact complex of flat quasicoher-
ent sheaves on X ×k Y.
Define C•X ⊠ C•Y := π∗X C•X (UX ) ⊗ π∗Y C•Y (UY ), interpreted as the total complex
associated to the double complex of the right side.

24.2.N. E XERCISE . Show that C•X ⊠ C•Y is a complex of flat quasicoherent sheaves on
X ×k Y, exact except at the first step, where the cohomology/kernel is canonically
identified with OX×Y . (You may or may not find it helpful to prove a similarly
statement for a similarly defined C•X,aug ⊠ C•Y,aug .)
696 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

24.2.O. E XERCISE . Suppose now that F is a quasicoherent sheaf on X, and G is a

quasicoherent sheaf on Y. Show that (F ⊠G )⊗(C•X ⊠C•Y ) is a complex of quasicoher-
ent sheaves on X ×k Y, exact except at the first step, where the cohomology/kernel
is canonically identified with F ⊠ G .

24.2.P. E XERCISE ( THE K ÜNNETH FORMULA FOR QUASICOHERENT SHEAVES ). By

suitably identifying (F ⊠ G ) ⊗ (C•X ⊠ C•Y ) with (F ⊗ C•X ) ⊠ (G ⊗ C•Y ), show that for
all n, Hn (X ×k Y, F ⊠ G ) = ⊕i+j=n Hi (X, F ) ⊗k Hj (Y, G ). (Hint: Don’t use spectral
sequences; work directly with the complexes. Show by direct calculation that if you
have two complexes of k-modules, the cohomology of the total complex of their
tensor product is the “direct sum of the tensor product of their cohomologies”.)

24.2.4. Remarks on hypotheses. Why did we require quasicoherence? Where did we

use the fact that we were working over a field k?

24.2.5. ⋆⋆ Extensions. Given quasicompact separated morphisms of schemes

ρX : X → Z and ρY : Y → Z, the resulting ρX×Y : X ×Z Y → Z, and quasicoherent
sheaves F and X and G on Y, what is the relationship between Rn ρX×Y
∗ F ⊠Z G ,
R ρ∗ F and R ρ∗ G ? By following through your proof, you may be able to extend
i X j Y

the statement of Exercise 24.2.P considerably.

24.3 Ideal-theoretic criteria for flatness

The following theorem will allow us to classify flat modules over a number of
rings. It is a refined version of Exercise 23.1.C, that M is a flat A-module if and only
if TorA
1 (M, N) = 0 for all A-modules N.

24.3.1. Theorem (ideal-theoretic criterion for flatness). — The A-module M is flat if

and only if TorA
1 (M, A/I) = 0 for every ideal I.

24.3.2. Remarks. Before getting to the proof, we make some side remarks that may
give some insight into how to think about flatness. Theorem 24.3.1 is profitably
stated without the theory of Tor. It is equivalent to the statement that M is flat if
and only if for all ideals I ⊂ A, I ⊗A M → M is an injection, and you can reinterpret
the proof in this guise. Perhaps better, M is flat if and only if I ⊗A M → IM is an
isomorphism for every ideal I.
Flatness is often informally described as “continuously varying fibers”, and
this can be made more precise as follows. An A-module M is flat if and only if
it restricts nicely to closed subschemes of Spec A. More precisely, what we lose
in this restriction, the submodule IM of elements which “vanish on Z”, is easy to
understand: it consists of formal linear combinations of elements i ⊗ m, with no
surprise relations among them — i.e., the tensor product I ⊗A M. This is the content
of the following exercise, in which you may use Theorem 24.3.1.

24.3.A. ⋆ E XERCISE ( THE EQUATIONAL CRITERION FORP FLATNESS ). Show that

an A-module M is flat if and only if for every relation Pai mi = 0 with ai ∈ A
and mi ∈ M, there exist mj′ ∈ M and aij ∈ A such that j aij mj′ = mi for all i
 © 2024 Ravi Vakil. Published by Princeton University Press. 697

P
and i ai aij = 0 in A for all j. (Translation: whenever elements of M satisfy an
A-linear relation, this is “because” of linear equations holding in A.)

24.3.3. Unimportant remark. In the statement of Theorem 24.3.1, it suffices to

check only finitely generated ideals. This is essentially the content of the following
statement, which you can prove if you wish: Show that an A-module M is flat if and
only if for all finitely generated ideals I, the natural map I ⊗A M → M is an injection.
Hint: if there is a counterexample for an ideal J that is not finitely generated, use it
to find another counterexample for an ideal I that is finitely generated.

24.3.4. Proof of the ideal-theoretic criterion for flatness, Theorem 24.3.1. By Exer-
cise 23.1.C, we need only show that TorA A
1 (M, A/I) = 0 for all I implies Tor1 (M, N) =
0 for all A-modules N, and hence that M is flat.
We first prove that TorA 1 (M, N) = 0 for all finitely generated modules N, by
induction on the number n of generators a1 , . . . , an of N. The base case (if n = 1,
so N =∼ A/ Ann(a1 ) — our first use of the annihilator ideal in a long time) is our
assumption. If n > 1, then Aan = ∼ A/ Ann(an ) is a submodule of N, and the
quotient Q is generated by the images of a1 , . . . , an−1 , so the result follows by
considering the Tor1 portion of the Tor long exact sequence for

0 / A/ Ann(an ) /N /Q / 0.

We deal with the case of general N by abstract nonsense. Notice that N is the
union of its finitely generated submodules {Nα }. In fancy language, this union
is a filtered colimit — any two finitely generated submodules are contained in a
finitely generated submodule (specifically, the submodule they generate). Filtered
colimits of modules commute with cohomology (Exercise 1.5.M), so Tor1 (M, N) is
the colimit over α of Tor1 (M, Nα ) = 0, and is thus 0. □
We now use Theorem 24.3.1 to get explicit characterizations of flat modules
over three (types of) rings: principal ideal domains, dual numbers, and some local
rings.
Recall Observation 24.1.2, that flatness implies torsion-free. The converse is
true for principal ideal domains:

24.3.B. E XERCISE ( FLAT = TORSION - FREE FOR A PID). Show that a module over a
principal ideal domain is flat if and only if it is torsion-free.

24.3.C. E XERCISE ( FLATNESS OVER A DVR). Suppose M is a module over a discrete

valuation ring A with uniformizer t. Show that M is flat if and only if t is not a
zerodivisor on M, i.e., (using the notation defined in Exercise 24.2.A) (M : t) = 0.
(See Exercise 24.1.B for the case of finitely generated modules.) This yields a simple
and very important geometric interpretation of flatness over a regular curve, which
we discuss in §24.3.9.

24.3.D. E XERCISE ( FLATNESS OVER THE DUAL NUMBERS ). Show that M is flat
over k[t]/(t2 ) if and only if the “multiplication by t” map M/tM → tM is an
isomorphism. (This fact is important in deformation theory and elsewhere.) Hint:
k[t]/(t2 ) has only three ideals.
698 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

24.3.5. Important Theorem (flat = free = projective for finitely presented modules
over local rings). — Suppose (A, m) is a local ring (not necessarily Noetherian), and M
is a finitely presented A-module. Then the following are equivalent.
(a) M is free.
(b) M is projective.
(c) M is flat.
(d) TorA
1 (M, A/m) = 0.

24.3.6. Remarks. Warning: modules over local rings can be flat without being free:
Q is a flat Z(p) -algebra (Z(p) is the localization of Z at p, not the p-adics), as all
localizations are flat (§24.1.1), but it is not free (Do you see why?).
Also, non-Noetherian people may be pleased to know that with a little work,
“finitely presented” can be weakened to “finitely generated”: use [Mat2, Thm. 7.10]
in the proof below, where finite presentation comes up.
Finally, Proposition 24.4.5 is a close variant of Theorem 24.3.5 for graded rings.

Proof. For any ring, free modules are projective (§23.2.1), and projective modules
are flat (Exercise 23.2.B), and flat modules M satisfy TorA 1 (M, N) = 0 for any N
(Exercise 23.1.C), so we need only show that if TorA
1 (M, A/m) = 0, then M is free.
(At this point, you should see Nakayama coming from a mile away.) Now
M/mM is a finite-dimensional vector space over the field A/m. Choose a basis of
M/mM, and lift it to elements m1 , . . . , mn ∈ M. Consider A⊕n → M given by
ei 7→ mi . We will show this is an isomorphism. It is surjective by Nakayama’s
Lemma (see Exercise 8.2.H): the image is all of M modulo the maximal ideal, hence
is everything. As M is finitely presented, by Lemma 6.4.6 (“finitely presented
implies always finitely presented”), the kernel K is finitely generated. Tensor 0 →
K → A⊕n → M → 0 with A/m. As M is flat, the result is still exact (Exercise 24.2.C):

0 / K/mK / (A/m)⊕n / M/mM / 0.

But (A/m)⊕n → M/mM is an isomorphism by construction, so K/mK = 0. As K is

finitely generated, K = 0 by Nakayama’s Lemma 8.2.9. □
Here is an immediate and useful corollary — really just a geometric interpreta-
tion.

24.3.7. Corollary (flat = locally free for finitely presented sheaves). — A finitely
presented sheaf F on X is flat (over X) if and only if it is locally free.

Proof. Local freeness of a finitely presented sheaf can be checked at the stalks,
Exercise 14.3.E. □

24.3.E. E XERCISE . Suppose π : X → Y is a finite flat morphism of locally Noetherian

schemes, and F is a finite rank locally free sheaf on X.
(a) Show that π∗ F is a finite rank locally free sheaf on Y.
(b) If Y is irreducible with generic point η, the degree of π above η is n, and
F is locally free of rank r, show that π∗ F is locally free of rank nr. (This
may be trickier than it looks.)
 © 2024 Ravi Vakil. Published by Princeton University Press. 699

24.3.8. Remark. In particular, taking F = OX , we see that in the locally Noetherian

setting, finite flat morphisms have locally constant degree.

24.3.F. ⋆ E XERCISE ( INTERESTING VARIANT OF T HEOREM 24.3.5, BUT UNIMPOR -

TANT FOR US ). Suppose A is a ring (not necessarily local), and M is a finitely
presented A-module. Show that M is flat if and only if it is projective. Hint: show
that M is projective if and only if Mm is free for every maximal ideal m. The harder
direction of this implication uses the fact that HomAm (Mm , Nm ) = HomA (M, N)m ,
which follows from Exercise 1.5.H. (Remark: There exist finitely generated flat mod-
ules that are not projective.
Q∞ They are necessarily not finitely presented. Example
without proof: let A = i=1 F2 , interpreted as functions Z≥0 → Z/2, and let M be
the module of functions modulo those of proper support, i.e., those vanishing at
almost all points of Z≥0 .)

24.3.G. E XERCISE . Prove the following useful criterion for flatness: Suppose
π : X → Y is a finite morphism, and Y is reduced and locally Noetherian. Then π
is flat if and only if π∗ OX is locally free, if and only if the rank of π∗ OX is locally
constant (dimκ(q) (π∗ OX )q ⊗ κ(q) is a locally constant function of q ∈ Y). Partial
hint: Exercise 14.3.K.

24.3.H. E XERCISE . Show that the normalization of the node (see Figure 8.4) is
not flat. Hint: use Exercise 24.3.G. (This exercise can be strengthened to show that
nontrivial normalizations are never flat.)

24.3.I. E XERCISE . In A4k = Spec k[w, x, y, z], let X be the union of the wx-plane with
the yz-plane:
(24.3.8.1) X = Spec k[w, x, y, z]/(wy, wz, xy, xz).
The projection A4k → A2k given by k[a, b] → k[w, x, y, z] with a 7→ w − y, b 7→ x − z
restricts to a morphism X → A2k . Show that this morphism is not flat.

24.3.9. Flat families over regular curves. Exercise 24.3.C gives an elegant geometric
criterion for when morphisms to regular curves are flat.

24.3.J. E XERCISE ( CRITERION FOR FLATNESS OVER A REGULAR CURVE ). Suppose

π : X → Y is a morphism from a locally Noetherian scheme to a regular (locally
Noetherian) curve. (The local Noetherian hypothesis on X is so we can discuss its
associated points.) Show that π is flat if and only if all associated points of X map
to a generic point of Y. (This is a partial converse to Exercise 24.1.K, that flat maps
always send associated points to associated points. As with Exercise 24.1.K, feel
free to immediately generalize your argument to a coherent sheaf F on X.)

24.3.10. For example, a nonconstant map from an integral (locally Noetherian)

scheme to a regular curve must be flat. (As another example, the deformation to
the normal cone, discussed in the double-starred section §22.4.17 and depicted
in Figure 22.3, is flat.) Exercise 24.3.H (and the comment after it) shows that the
regular condition is necessary. The example of two planes meeting at a point in
Exercise 24.3.I shows that the dimension 1 condition is necessary.

24.3.11. ⋆ Unimportant remark: A valuative criterion for flatness. Exercise 24.3.J shows
that flatness over a regular curves is geometrically intuitive (and is “visualizable”).
700 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

It gives a criterion for flatness in general: suppose π : X → Y is finitely presented

morphism. If π is flat, then for every morphism Y ′ → Y where Y ′ is the Spec of a
discrete valuation ring, π ′ : X ×Y Y ′ → Y ′ is flat, so no associated points of X ×Y Y ′
map to the closed point of Y ′ . If Y is reduced and locally Noetherian, then this is
a sufficient condition; this can reasonably be called a valuative criterion for flatness.
(Reducedness is necessary: consider Exercise 24.1.G(d).) This gives an excellent
way to visualize flatness, which you should try to put into words (perhaps after
learning about flat limits below). See [Gr-EGA, IV3 .11.8.1] for a proof (and an
extension without Noetherian hypotheses).

24.3.12. Revisiting the degree of a projective morphism from a curve to a regular curve.
As hinted after the statement of Proposition 16.3.5, we can now better understand
why nonconstant projective morphisms from a curve to a regular curve have a
well-defined degree, which can be determined by taking the preimage of any point
(§16.3.4). (Example 10.3.4 was particularly enlightening.) This is because such maps
are flat by Exercise 24.3.J, and then the degree is constant by Remark 24.3.8 (see
also Exercise 24.3.G). Also, Exercise 24.3.G yields a new proof of Proposition 16.3.5.

24.3.13. ⋆ Flat limits. Here is an important consequence of Exercise 24.3.J, which

we can informally state as: we can take flat limits over one-parameter families.
More precisely: suppose A is a discrete valuation ring, and let 0 be the closed point
of Spec A and η the generic point. Suppose X is a locally Noetherian scheme over
A, and Y is a closed subscheme of X|η . Let Y ′ be the scheme-theoretic closure of Y
in X. Then π : Y ′ → Spec A has no associated points over 0 by Exercise 9.4.D, so
Y ′ is flat over Spec A by Exercise 24.3.J. Similarly, suppose Z is a one-dimensional
Noetherian scheme, 0 is a regular point of Z, and π : X → Z is a morphism from a
locally Noetherian scheme to Z. If Y is a closed subscheme of π−1 (Z \ {0}), flat over
Z \ {0}, and Y ′ is the scheme-theoretic closure of Y in X, then Y ′ is flat over Z. In
both cases, the closure Y ′ |0 is often called the flat limit of Y. (Feel free to weaken
the Noetherian hypotheses on X.)

24.3.K. E XERCISE . Suppose (with the language of the previous paragraph) that A
is a discrete valuation ring, X is a locally Noetherian A-scheme, and Y is a closed
subscheme of the generic fiber X|η . Show that there is only one closed subscheme
Y ′ of X such that Y ′ |η = Y, and Y ′ is flat over A.

24.3.L. E XERCISE . Consider the family of pairs of points in A2C with coordinates
parametrized by A1C with coordinate t, given by {(0, 0), (3t, 4t)}.

(a) Find the flat limit as t → 0. (It will be a length 2 subscheme. The main
point of this exercise is for you to work through the algebra.)
(b) Find the flat limit “as t → ∞”.

24.3.M. E XERCISE . Suppose C = V(f(x, y)) is a smooth plane curve in A2C , and p
is a closed point of C, so OC,p is a discrete valuation ring. Consider the family of
pairs of points of C, parametrized by q ∈ C \ {p}, given by {p, q}. Explain why the
flat limit over p (the “limit as q 7→ p”) can be identified with the tangent direction
to C at p. Translation: the tangent line is indeed the limit of secants, without deltas
and epsilons.
 © 2024 Ravi Vakil. Published by Princeton University Press. 701

24.3.N. H ARDER E XERCISE ( AN EXPLICIT FLAT LIMIT ). Let X = A3 × A1 → Z = A1

over a field k, where the coordinates on A3 are x, y, and z, and the coordinates on
A1 are t. Define Y away from t = 0 as the union of the two lines y = z = 0 (the
x-axis) and x = z − t = 0 (the y-axis translated by t). Find the flat limit at t = 0.
(Hints: (i) it is not the union of the two axes, although it includes this union. The
flat limit is nonreduced at the node, and the “fuzz” points out of the plane they
are contained in. (ii) (y, z)(x, z) ̸= (xy, z). (iii) Once you have a candidate flat limit,
be sure to check that it is the flat limit. (iv) If you get stuck, read Example 24.3.14
below.)
Consider a projective version of the previous example, where two lines in P3
degenerate to meet. The limit consists of two lines meeting at a node, with some
nonreduced structure at the node. Before the two lines come together, their space of
global sections is two-dimensional. When they come together, it is not immediately
obvious that their flat limit also has two-dimensional space of global sections as
well. The reduced version (the union of the two lines meeting at a point) has a
one-dimensional space of global sections, but the effect of the nonreduced structure
on the space of global sections may not be immediately clear. However, we will
see that “cohomology groups can only jump up in flat limits”, as a consequence
(indeed the main moral) of the Semicontinuity Theorem 25.1.1.

24.3.14. ⋆⋆ Example of variation of cohomology groups in flat families. We can

use a variant of Exercise 24.3.N to see an example of a cohomology group actually
jumping. We work over an algebraically closed field to avoid distractions. Before
we get down to explicit algebra, here is the general idea. Consider a twisted cubic
C in P3 . A projection prp from a random point p ∈ P3 will take C to a nodal plane
cubic curve. Picture this projection “dynamically”, by choosing coordinates so
p is at [1, 0, 0, 0], and considering the map ϕt : [w, x, y, z] 7→ [tw, x, y, z]; ϕ1 is the
identity on P3 , ϕt is an automorphism of P3 for t ̸= 0, and ϕ0 is the projection.
The limit of ϕt (C) as t → 0 will be a nodal cubic, with nonreduced structure at
the node “analytically the same” as what we saw when two lines came together in
Exercise 24.3.N. (The phrase “analytically the same” can be made precise once we
define completions in §28.)
Let’s now see this in practice. Rather than working directly with the twisted
cubic, we use another example where we saw a similar picture. Consider the
nodal (affine) plane cubic y2 = x3 + x2 . Its normalization (see Figure 8.4, Example
(3) of §8.3.5, Exercise 10.7.E, . . . ) was obtained by adding an extra variable m
corresponding to y/x (which can be interpreted as blowing up the origin, see
§22.4.3). We use the variable m rather than t (used in §8.3.5) in order to reserve t
for the parameter for the flat family.
We picture the nodal cubic C as lying in the xy-plane in A3 = Spec k[x, y, m],
and the normalization C e projecting to it, with m = y/x. What are the equations
e Clearly, they include the equations y2 = x3 + x2 and y = mx, but these
for C?
are not enough — the m-axis (i.e., x = y = 0) is also in V(y2 − x3 − x2 , y − mx).
A little thought (and the algebra we have seen earlier in this example) will make
clear that we have a third equation x = m2 − 1, which along with y = mx implies
y2 = x2 +x3 . Now we have enough equations: k[x, y, m]/(x−(m2 −1), y−mx) is an
integral domain, as it is clearly isomorphic to k[m]. Indeed, you should recognize
702 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

this as the algebra appearing in Exercise 10.7.E, so we have described C̃ as

(24.3.14.1) Spec k[x, y, m]/(x − (m2 − 1), y − mx).
Next, we want to formalize our intuition of the dynamic projection to the xy-
e ⊂ A3 . We picture it as follows. Given a point (x, y, z) at time 1, at time t
plane of C
we want it to be at (x, y, zt). At time t = 1, we “start with” C,
e and at time t = 0 we
have (set-theoretically) C. Thus at time t ̸= 0, the curve C is sent to the curve
e

Spec k[x, y, m]/(x−((m/t)2 −1), y−(m/t)x) = Spec k[x, y, m]/(m2 −t2 (x+1), ty−mx).
The family over Spec k[t, t−1 ] is thus

Spec k[x, y, m, t, t−1 ]/(m2 − t2 (x + 1), ty − mx) / Spec k[t, t−1 ].

We have inverted t because we are so far dealing only with nonzero t. For t ̸= 0,
this is certainly a “nice” family, and so surely flat.

24.3.O. E XERCISE . Check this, as painlessly as possible! Hint: by a change of

coordinates, show that the family is constant “over Spec k[t, t−1 ]”, and hence pulled
back (in some way you must figure out) via Spec k[t, t−1 ] → Spec k from

Spec k[x, y, M]/(M2 − (x + 1), y − Mx) / Spec k,

which is flat by Trick Question 24.1.A(a).

We now figure out the flat limit of this family over t = 0, in Spec k[x, y, m, t] →
A1 = Spec k[t]. We first hope that our flat family is given by the equations we have
already written down:
Spec k[x, y, m, t]/(m2 − t2 (x + 1), ty − mx).
But this is not flat over A1 = Spec k[t], as the fiber dimension jumps (§24.5.5):
substituting t = 0 into the equations (obtaining the fiber over 0 ∈ A1 ), we find
Spec k[x, y, m]/(m2 , mx). This is set-theoretically the entire xy-plane (m = 0),
which of course has dimension 2. Notice for later reference that this “false limit” is
scheme-theoretically the xy-plane, with some nonreduced structure along the y-axis.
(This may remind you of Figure 4.4.)
So we are missing at least one equation. One clue as to what equation is
missing: the equation y2 = x3 + x2 clearly holds for t ̸= 0, and does not hold for our
naive attempt at a limit scheme m2 = mx = 0. In retrospect, we should not have
discarded this original equation when writing (24.3.14.1). So we put this equation
back in, and have a second hope for describing the flat family over A1 :

Spec k[x, y, m, t]/(m2 − t2 (x + 1), ty − mx, y2 − x3 − x2 ) / Spec k[t].

When t = 0, we get Spec k[x, y, m]/(m2 , mx, y2 − x3 − x2 ), so it passes the “no-

dimension-jumping” criterion for flatness (§24.5.5 again), which is necessary but
not sufficient. But unfortunately, it is not flat, and this not so easy to see.

24.3.P. E XERCISE . Show that in the ring

k[x, y, m, t]/(m2 − t2 (x + 1), ty − mx, y2 − x3 − x2 ),
my − tx(x + 1) ̸= 0, but t(my − tx(x + 1)) = 0.
 © 2024 Ravi Vakil. Published by Princeton University Press. 703

This suggests what equation we need to add: my − tx(x + 1) = 0. Let

A = k[x, y, m, t]/(ty − mx, m2 − t2 (x + 1), y2 − x2 (x + 1), my − tx(x + 1))
for convenience. We next hope that Spec A → Spec k[t] is flat at t = 0. We can show
this using Exercise 24.3.B, by showing that t is not a zerodivisor on A. We do this
by giving a “normal form” for elements of A.

24.3.Q. E XERCISE . Show that each element of A can be written uniquely in the
form f(t)m + g(x, t)y + h(x, t), where f, g, and h are polynomials. Then show that
t is not a zerodivisor on A, and conclude that Spec A → A1 is indeed flat.
In particular, the flat limit when t = 0 is given by
Spec A/(t) = Spec k[x, y, m]/(m2 , mx, my, y2 − x2 − x3 ).

24.3.R. E XERCISE . Show that the flat limit is nonreduced, and the “nonreducedness
has length 1 and supported at the origin”. More precisely, if X = Spec A/(t), show
that IXred is a skyscraper sheaf, with value k, supported at the origin. Sketch this
flat limit X.

24.3.15. Note that we have a nonzero global function on X, given by m, which is

supported at the origin (i.e., 0 away from the origin).
We now use this example to get a projective example with interesting behavior.
We take the projective completion of this example, to get a family of cubic curves in
P3 degenerating to a nodal cubic C with a nonreduced point.

24.3.S. E XERCISE . Do this: describe this family (in P3 × A1 ) precisely.

Take the long exact sequence corresponding to
0 / ICred / OC / OCred / 0,
to get
0 / H0 (C, ICred ) α / H0 (C, OC ) / H0 (C, OCred )

/ H1 (C, ICred ) / H1 (C, OC ) / H1 (C, OCred )

/ H2 (C, ICred )
We have H1 (C, ICred ) = H2 (C, ICred ) = 0 as ICred is supported in dimension 0
(by dimensional cohomology vanishing, Theorem 18.2.6). Also, Hi (Cred , OCred ) =
Hi (C, OCred ) (property (v) of cohomology, see §18.1). The (reduced) nodal cubic
Cred has h0 (O) = 1 (§11.4.7) and h1 (O) = 1 (cubic plane curves have genus 1,
(18.6.7.1)). Also, h0 (C, ICred ) = 1 as observed above. Finally, α is not 0, as there
exists a nonzero function on C vanishing on Cred (§24.3.15 — convince yourself that
this function extends from the affine patch Spec A to the projective completion).
Using the long exact sequence, we conclude h0 (C, OC ) = 2 and h1 (C, OC ) = 1.
Thus in this example we see that (h0 (O), h1 (O)) = (1, 0) for the general member
of the family (twisted cubics are isomorphic to P1 ), and the special member (the
704 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

flat limit) has (h0 (O), h1 (O)) = (2, 1). Notice that both cohomology groups have
jumped, yet the Euler characteristic has remained the same. The first behavior, as
stated after Exercise 24.3.N, is an example of the Semicontinuity Theorem 25.1.1.
The second, constancy of Euler characteristics in flat families, is the subject of §24.7.
(It is no coincidence that the example had a singular limit, see §25.1.3.)

24.4 ⋆⋆ Aside: The Koszul complex and the Hilbert Syzygy

Theorem
To show that the theory we have developed so far is already very useful, we
use it to prove the Hilbert Syzygy Theorem 16.1.3. This in turn gives us an excuse
to introduce an important tool in commutative algebra, the Koszul complex. We
will not use the contents of this section, so you should read it only if you feel like it.

24.4.1. The Koszul complex. Suppose E is a rank r free A-module. Given an element
s ∈ E∨ , i.e., an A-linear map s : E → A, the Koszul complex K• (s) associated to s is
the chain complex:

0 / ∧r E dr
/ ∧r−1 E dr−1
/ ··· d2
/ ∧1 E d1
/A / 0.
The differential is defined by
X
k
dk (e1 ∧ · · · ∧ ek ) = (−1)i+1 s(ei ) e1 ∧ · · · ∧ e^i ∧ · · · ∧ ek
i=1

where e1 , . . . , er ∈ E (and e^i as usual means “omit the term ei ”). It is not hard
to see that this is a complex (and it also follows from Exercise 24.4.B). It is a little
harder to see that it is well-defined.
For convenience, choose a basis of E, and let suppose s : A⊕r → A sends the
ith basis element of E = A⊕r to fi . In this case we denote Koszul complex by
K• (f1 , . . . , fr ).

24.4.A. E XERCISE . If A = k[x1 , . . . , xr ], show by hand that the Koszul com-

plex K• (x1 , . . . , xr ) is exact except the last term, where the cohomology is k =
A/(x1 , . . . , xr ). (One possible method: watch what happens to each monomial, as
in §18.3.)
Thus the Koszul complex yields a free resolution of k[x1 , · · · , xr ]/(x1 , · · · , xr ).
(Exercise 24.4.C will generalize this.)

24.4.B. E XERCISE ( USEFUL INTERPRETATION OF THE K OSZUL COMPLEX ). Show

that the Koszul complex K• (f1 , . . . , fr ) is the complex
 
⊗ri=1 0 / A ×fi / A /0 .

First figure out what this means!

24.4.C. E XERCISE . Show that if f1 , . . . , fr is a regular sequence, then K• (f1 , . . . , fr )

is exact everywhere but at the last term, where the cohomology is A/(f1 , . . . , fr ).
Hint: use induction on r, and Exercise 24.4.B.
 © 2024 Ravi Vakil. Published by Princeton University Press. 705

Hence if f1 , . . . , fr is a regular sequence, the Koszul complex gives a free reso-

lution of A/(f1 , . . . , fr ) of length r, called the Koszul resolution of A/(f1 , . . . , fr ).

24.4.2. Remarks. (a) The n = 2 case of the Koszul complex implicitly came
up in the proof of Theorem 9.5.6 — it is the total complex of the tiny double
complex (9.5.6.1). (b) One can similarly take the Koszul complex of an A-module
M: K• (s, M) := K• (s) ⊗A M. (c) The cohomology of the Koszul complex is called
Koszul homology. It measures the failure of a sequence to be regular.

24.4.3. Finitely generated modules over regular local rings have finite free resolu-
tions.
For the next four exercises, suppose (A, m) is a (Noetherian) regular local ring
of dimension n, and let f1 , . . . , fn be generators of m.

24.4.D. E XERCISE . Show that for any A-module M, TorA

i (M, A/m) = 0 for i > n.
Hint: take a suitable free resolution of A/m.

24.4.E. E XERCISE . If TorAi (M, A/m) = 0 for i > b ≥ 1 and α : A

⊕a
→ M is a
A
surjection, show that Tori (ker α, A/m) = 0 for i > b − 1.

24.4.F. E XERCISE . Show that every finitely generated A-module M has a free
resolution of length at most n. Hint: By Theorem 24.3.5, if M is a finitely generated
A-module such that TorA i (M, A/m) = 0, then M is free.

24.4.G. E XERCISE . Give an example to show that this bound cannot be improved.

24.4.4. Proof of the Hilbert Syzygy Theorem 16.1.3.

We now prove the Hilbert Syzygy Theorem following the outline suggested
by §24.4.3. Let S• be the graded ring k[x0 , . . . , xn ], with the usual grading, where k
is a field. Recall the graded module S(e)• defined in §15.1.4 by S(e)d = Se+d , so
]• = OPn (e).
S(e) k

24.4.H. WARM - UP E XERCISE . Guided by the Koszul resolution of Exercise 24.4.A,

describe an exact sequence on Pn
A:

/ O(−n − 1) / ··· / O(−j)⊕( / ··· /

n+1
0 j )

O(−1)⊕(n+1) /O / 0.

Suppose M• is finitely generated graded S• -module. Choose homogeneous

elements a1 , . . . , ar of degree e1 , . . . , er which generate M• , so the map

×(a1 , . . . , ar ) : ⊕ri=1 S• / M•

is a surjection. We rewrite it as a degree-preserving map of graded modules:

×(a1 , . . . , ar ) : ⊕ri=1 S(−ei )• / / M• .

706 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

24.4.5. Proposition (analog of Theorem 24.3.5). — Suppose M• is a finitely generated

∼ ⊕n S(−di )•
graded S• -module, and TorS1 • (M• , k) = 0. (Here k = S• /S+ .) Then M• = i=1
as graded modules for some choice of di .
In other words, M is “free in the graded sense”. To prove this, we start with a
graded variant of Nakayama’s Lemma.

24.4.I. E XERCISE ( GRADED N AKAYAMA ). Suppose C• is a finitely generated S• -

module such that C• ⊗S• k = 0. Show that C• = 0. Hint: if C• ̸= 0, look at
the nonzero element of C of least degree. (You won’t need to invoke the “usual”
Nakayama’s Lemma.)

24.4.6. Proof of Proposition 24.4.5. Suppose TorS1 • (M• , k) = 0. Let y0 , . . . , yr be homo-

geneous elements of M• such that their images form a basis for M• /(x0 , . . . , xn ) =
M• ⊗S• k. We have a map α : ⊕ri=1 S(− deg yi )• → M• that becomes an isomor-
phism upon tensoring with k. If Q• = coker α, then by applying ⊗S• k to

⊕ri=1 S(− deg yi )•

α / M• / Q• /0

we see that Q• ⊗S• k = 0 and hence Q• = 0 by Exercise 24.4.I, so α is surjective.

Next, let K• = ker α. Applying ⊗S• k to
/ K• / ⊕r α / M• /0
0 i=1 S(− deg yi )•

and using the hypothesis TorS1 • (M• , k) = 0, we see that K• ⊗S• k = 0 and thus
K• = 0 as well, so α gives us our desired isomorphism. □
We are now ready to completely follow the template of §24.4.3.

24.4.J. E XERCISE . Prove that every finite-generated graded S• -module M• has a

“graded free resolution” of length at most n + 1:

0 / Fn+1
•
/ ··· / F1• / M• /0

j
where Fq q
• = ⊕i=1 S(−ai,q )• .

24.4.K. E XERCISE . Prove the Hilbert Syzygy Theorem 16.1.3.

24.5 Topological implications of flatness

We now discuss some topological aspects and consequences of flatness, that
boil down to the Going-Down Theorem for flat morphisms (§24.5.3), which in
turn comes from faithful flatness. Because dimension in algebraic geometry is
a topological notion, we will show that dimensions of fibers behave well in flat
families (§24.5.5).

24.5.1. Faithful flatness. The notion of faithful flatness is handy for many reasons,
and we describe only a few. A B-module M is faithfully flat if for all complexes of
B-modules
(24.5.1.1) N′ /N / N ′′ ,
 © 2024 Ravi Vakil. Published by Princeton University Press. 707

(24.5.1.1) is exact if and only if (24.5.1.1)⊗B M is exact. A B-algebra A is faithfully

flat if it is faithfully flat as a B-module.

24.5.A. E XERCISE . Show that a flat B-module M is faithfully flat if and only if for
all B-modules N, M ⊗B N = 0 implies that N = 0.

24.5.B. E XERCISE . Suppose M is a flat B-module. Show that the following are
equivalent.

(a) M is faithfully flat;

(b) for all prime ideals p ⊂ B, M ⊗B κ(p) is nonzero;
(c) for all maximal ideals m ⊂ B, M ⊗B κ(m) = M/mM is nonzero.

Suppose π : X → Y is a morphism of schemes. We say that π is faithfully flat if

it is flat and surjective. (Unlike flatness, faithful flatness is not particularly useful
for quasicoherent sheaves, so we do not define faithfully flat quasicoherent sheaves
over a base.)

24.5.C. E XERCISE ( CF. E XERCISE 24.5.B( B )). Suppose B → A is a ring morphism.

Show that A is faithfully flat over B if and only if Spec A → Spec B is faithfully flat.

24.5.2. Faithfully flat morphisms behave “reasonably”. Faithfully flat morphisms

behave “reasonably” in the sense of §8.1 (preserved by composition and base
change, and local on the target) because both flatness and surjectivity have this
property (§24.1.8 and Exercise 10.4.G respectively).

24.5.D. E XERCISE . Suppose π : Spec A → Spec B is flat.

(a) Show that π is faithfully flat if and only if every closed point q ∈ Spec B is in the
image of π. (Hint: Exercise 24.5.B(c).)
(b) Hence show that every flat (local) morphism of local rings (Definition 7.3.1) is
faithfully flat. (Morphisms of local rings are assumed to be local, i.e., the maximal
ideal pulls back to the maximal ideal.)

24.5.3. Going-Down for flat morphisms. A consequence of Exercise 24.5.D is

the following useful result, whose statement makes no mention of faithful flatness.
(The statement is not coincidentally reminiscent of the Going-Down Theorem for
finite extensions of integrally closed domains, Theorem 12.2.12.)

24.5.E. E XERCISE (G OING -D OWN T HEOREM FOR FLAT MORPHISMS ).

(a) Suppose that B → A is a flat morphism of rings, corresponding to a map
π : Spec A → Spec B. Suppose q ⊂ q ′ are prime ideals of B, and p ′ is a prime ideal
of A with π([p ′ ]) = [q ′ ]. Show that there exists a prime p ⊂ p ′ of A with π([p]) = [q].
Hint: show that Bq ′ → Ap ′ is a flat local ring homomorphism, and hence faithfully
flat by the Exercise 24.5.D(b).
(b) Part (a) gives a geometric consequence of flatness. Draw a picture illustrating
this.
(c) Recall the Going-Up Theorem, described in §8.2.4. State the Going-Down
Theorem for flat morphisms in a way parallel to Exercise 8.2.F, and prove it.

24.5.F. E XERCISE . Suppose π : X → Y is an integral (e.g., finite) flat morphism, and

Y has pure dimension n. Show that X has pure dimension n. (This generalizes
708 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Exercise 12.1.H(a).) Hint: π satisfies both Going-Up (see Exercise 8.2.F) and Going-
Down (Exercise 24.5.E.

24.5.G. I MPORTANT E XERCISE : FLAT MORPHISMS ARE OPEN MAPS ( IN REASONABLE

SITUATIONS ). Suppose π : X → Y is locally of finite type and flat, and Y (and hence
X) is locally Noetherian. Show that π is an open map (i.e., sends open sets to open
sets). Hint: reduce to showing that π(X) is open for all such π. Reduce to the case
where X and Y are both affine. Use Chevalley’s Theorem 8.4.2 to show that π(X) is
constructible. Use the Going-Down Theorem for flat morphisms, Exercise 24.5.E, to
show that π(X) is stable under generization. Conclude using Exercise 8.4.C.

24.5.4. Remarks.
(i) Of course, not all open morphisms are flat: witness Spec k[t]/(t) → Spec k[t]/(t2 ).
(ii) Also, in quite reasonable circumstances, flat morphisms are not open: witness
Spec k(t) → Spec k[t] (flat by Example 24.1.1(ii)).

24.5.H. ⋆ E XERCISE . Weaken the hypotheses of “locally of finite type” and “locally
Noetherian” to just “locally finitely presented” Hint: as with the similar general-
ization in Exercise 10.3.I of Chevalley’s Theorem 8.4.2, use the fact that any such
morphism is “locally” pulled back from a Noetherian situation.

24.5.5. Dimensions of fibers are well-behaved for flat morphisms.

24.5.6. Proposition. — Suppose π : X → Y is a flat morphism of locally Noetherian

schemes, with p ∈ X and q ∈ Y such that π(p) = q. Then
codimX p = codimY q + codimπ−1 (q) p
(see Figure 12.4).
Informal translation: the dimension of the fibers is the difference of the dimen-
sions of X and Y (at least locally). Compare this to Exercise 12.4.A, which stated
that without the flatness hypothesis, we would only have inequality (≤).

24.5.I. E XERCISE . Prove Proposition 24.5.6 as follows. As just mentioned, Exer-

cise 12.4.A gives one inequality, so show the other. Given a chain of irreducible
closed subsets in Y containing q, and a chain of irreducible closed subsets in
π−1 (q) ⊂ X containing p, construct a chain of irreducible closed subsets in X con-
taining p, using the Going-Down Theorem for flat morphisms (Exercise 24.5.E).

As a consequence of Proposition 24.5.6, if π : X → Y is a flat map of irreducible

varieties, then the fibers of π all have pure dimension dim X − dim Y. (Warning:
Spec k[t]/(t) → Spec k[t]/(t2 ) does not exhibit dimensional jumping of fibers, is
open, and sends associated points to associated points, cf. Exercise 24.1.K, but is not
flat. If you prefer a reduced example, the normalization of the cuspidal plane cubic
Spec k[t] → Spec k[x, y]/(y2 − x3 ), shown in Figure 10.5, also has these properties.)
This leads us to the following useful definition.

24.5.7. Definition. If π : X → Y is a flat morphism that is locally of finite type, and

all fibers of π have pure dimension n, we say that π is flat of relative dimension n.
(In particular, when one says a morphism is flat of relative dimension n, the locally
finite type hypotheses are implied. Remark 24.5.9 motivates this hypothesis.)
 © 2024 Ravi Vakil. Published by Princeton University Press. 709

24.5.J. E XERCISE . Suppose π : X → Y is a flat morphism of finite type k-schemes,

and Y is pure dimensional (so “codimension is the difference of dimensions”, cf.
Theorem 12.2.9). Show that the following are equivalent.

(i) The scheme X has pure dimension dim Y + n.

(ii) The morphism π is flat of relative dimension n.

24.5.K. E XERCISE . Show that the notion of a morphism being “flat of relative
dimension n” is preserved by arbitrary base change. Hint to show that fiber
dimension is preserved: Exercise 12.2.M.

24.5.L. E XERCISE . Suppose π : X → Y and ρ : Y → Z are flat of relative dimension

m and n respectively (hence locally of finite type). Show that ρ ◦ π is flat of relative
dimension m + n.

24.5.8. Remark. For practice, you can now give a second solution to Exercise 12.2.F,
that dimension is additive for products of varieties.

24.5.9. Remark. The reason for the “locally finite type” assumption in the definition
of “flat of relative dimension n” is that we want this to be a “reasonable” class of
morphisms. In particular, we want our notion of “flatness of relative dimension n”
to be preserved by base change. Consider the Cartesian diagram

Spec k(x) ⊗k k(y) / Spec k(y)

π′ π
 
Spec k(x) / Spec k.

Both π and π ′ are trivially flat, because they are morphisms to spectra of fields
(Exercise 24.1.A(a)). But the dimension of the fiber of π is 0, while (as described in
Remark 12.2.17) the dimension of the fiber of π ′ is 1.

24.5.10. Morphisms to Spec k are universally open. We now deliver on a promise

made in §10.5.6.

24.5.11. Theorem. — Every morphism X → Spec k is universally open, i.e., remains

open after any base change.
⋆ Proof. We wish to show that for any k-scheme Y, X ×k Y → Y is an open map of
topological spaces, that is, sends open sets to open sets. By considering an affine
cover of Y, we see that we may reduce to the case when Y is affine, say Spec S. Then
by considering an affine cover of X, we see that we may reduce to the case where X
is affine, say Spec R. So we wish to show that π : Spec R ⊗k S → Spec S is an open
map.
Because distinguished affine open sets form a base of the Zariski topology of
an affine scheme, itP suffices to show that for all f ∈ R ⊗k S, π(D(f)) is open. Write
n
f as the finite sum i=1 ri ⊗ si . Let R ′ be the subring of R generated by the ri , so
′ ∼
R = k[r1 , . . . , rn ]/I for some (finitely generated) ideal I. To avoid confusion, we
denote the element f when considered as an element of R ′ ⊗k S by f ′ . Consider the
710 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

diagram

Spec R ⊗k S
ρ
/ Spec R ′ ⊗k S

π π′
& x
Spec S

By Exercise 24.5.H, π ′ is an open map (as R ′ is finitely presented over k), so

π (D(f ′ )) is an open subset of Spec S. We conclude by showing that π(D(f)) =
′

π ′ (D(f ′ )). Notice that ρ−1 (D(f ′ )) = D(f) (as ρ♯ (f ′ ) = f), so we have the inclusion
π(D(f)) ⊂ π ′ (D(f ′ )). To show the opposite inclusion, suppose p ∈ π ′ (D(f ′ )). Then
the fiber of π ′ |D(f ′ ) over p is nonempty, i.e., the following ring is not the zero ring:

(R ′ ⊗k S)f ′ ⊗S κ(p) = (R ′ ⊗k κ(p))f ′ .

Thus R ′ ⊗k κ(p) is not the zero ring, and f ′ is not nilpotent in this ring. But R ′ ⊗k κ(p)
is a subring of R ⊗k κ(p) (· ⊗k κ(p) is an exact functor; everything is flat over a
field), so R ⊗k κ(p) is also not the zero ring, and f is not nilpotent in this ring.
Hence (R ⊗k κ(p))f ̸= 0, so the fiber of π|D(f) over p, Spec(R ⊗k S)f ⊗S κ(p), is also
nonempty. □

24.5.12. Fancier flatness facts without proof.

We conclude with a couple of important facts, although we won’t use them, so
we will not prove them.
”Generic flatness” would be better called “general flatness” (cf. the discussion
after Definition 3.6.11). Because we’re not proving it, we may as well state it in
some generality. For a slick proof of generic flatness under Noetherian hypotheses,
see [Mu2, p. 57-58].

24.5.13. Theorem (generic flatness, [Stacks, tag 052B]). — If π : X → Y is a finite

type morphism of schemes, Y is reduced, and F is a finite type quasicoherent sheaf on X,
then there is a dense open subscheme U ⊂ Y such that F |π−1 (U) is of finite presentation
over π−1 (U), and flat over U.

24.5.14. Interpretation of the degree of a generically finite morphism. We use this to

interpret the degree of a rational map of varieties in terms of counting preimages,
fulfilling a promise made in §12.2.2. Suppose π : X 99K Y is a generically finite
rational map of k-varieties of degree d. For simplicity we assume that X and Y are
irreducible. (Feel free to relax this.) Replace X by an open subset on which π is
a morphism. By Proposition 12.4.4 (“generically finite implies generally finite”),
there is a dense open subset V of Y over which π is finite. By Theorem 24.5.13, there
is a dense open subset V ′ of V over which π is finite and flat. Then by Remark 24.3.8
(“finite flat morphisms have locally constant degree”), over this open subset V ′ , π
has “locally constant degree”, which is necessarily d .

24.5.15. The flat locus is open (flatness is an open condition). Generic flatness can be
used to show one last topological aspect of flatness: in reasonable circumstances,
the locus where a quasicoherent sheaf is flat over a base is an open subset. More
precisely:
 © 2024 Ravi Vakil. Published by Princeton University Press. 711

24.5.16. Theorem (the flat locus is open). — Suppose π : X → Y is a locally finite type
morphism of locally Noetherian schemes, and F is a coherent sheaf on X.
(a) The locus of points of X at which F is Y-flat is an open subset of X.
(b) If π is closed (e.g., proper), then the locus of points of Y over which F is flat (i.e.
those q ∈ Y for which F is flat at all p ∈ π−1 (q)) is an open subset of Y.
Part (b) follows immediately from part (a). Part (a) reduces to a nontrivial
statement in commutative algebra, see for example [Mat2, Thm. 24.3] or [Gr-EGA,
IV3 .11.1.1]. As is often the case, Noetherian hypotheses can be dropped in exchange
for local finite presentation hypotheses on the morphism π, see [Gr-EGA, IV3 .11.3.1]
or [Stacks, tag 00RC].

24.6 Local criteria for flatness

For a Noetherian local ring, there is a greatly improved version of the ideal-
theoretic criterion of Theorem 24.3.1: we need check only one ideal — the maximal
ideal.

24.6.1. Theorem (local criterion for flatness). — Suppose (B, n) → (A, m) is a local
morphism of Noetherian local rings (§7.3.1), and that M is a finitely generated A-module.
Then M is B-flat if and only if TorB
1 (M, B/n) = 0.

This is a miracle: flatness over all of Spec B is determined by what happens

over the closed point.
We have already seen Theorem 24.6.1 in a special case: if (B, n) → (A, m) is an
isomorphism, it is an immediate consequence of Theorem 24.3.5.
The following proof is due to Ogus, [OB]. To highlight what makes it work,
we make a temporary and unneeded definition.

24.6.2. Definition. Suppose F : B → A is an additive functor of abelian categories.

For the sake of this section only, we say F is half-exact if for every short exact
sequence 0 → M ′ → M → M ′′ → 0 in B, FM ′ → FM → FM ′′ is exact in A .
Examples are left-exact functors, right-exact functors, and derived functors of
either.

24.6.3. Theorem (Nakayama’s Lemma for half-exact functors). — Suppose T :

(B, n) → (A, m) is a local morphism of Noetherian local rings, and the additive functor
T : f.g.ModB → f.g.ModA is B-linear and half-exact. If T (B/n) = 0 then T (N) = 0 for all
finitely generated B-modules N (i.e., T is the zero functor).

Proof. We will use Nakayama’s Lemma for A, in the form that if M ∈ f.g.ModA ,
a ∈ m, and ×a : M → M is a surjection, then M = 0. (By now you should see how
this is one of many versions of Nakayama’s Lemma from §8.2.7.)
Let k = B/n for convenience. We wish to show that T (M) = 0 for all M ∈
f.g.ModB . If not, by Noetherianity of M, there is a maximal M ′ ⊂ M with T (M/M ′ ) ̸=
0. Chose such a maximal M ′ .
Let N = M/M ′ , so T (N) ̸= 0, but T (N ′′ ) = 0 for every nontrivial quotient N ′′
of N. We thus wish to prove that if N ∈ f.g.ModB , and T (N ′′ ) = 0 for every proper
quotient N ′′ of N, then T (N) = 0.
712 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Let I = Ann(N) = {b ∈ B : bN = 0}. If I = B, then N = 0, so T (N) = 0. If I =

n, then N = (B/n)⊕m (N Noetherian implies m is finite), so T (N) = T (B/n)⊕m = 0.
Otherwise, there exists b ∈ n \ I.
Consider the short exact sequence

0 / bN /N / N/bN /0.

As bN ̸= 0, N/bN is a nontrivial quotient of N, so T (N/bN) = 0. By the half-

exactness of T , as T (N) ̸= 0, we have that T (bN) ̸= 0.
Next, consider the short exact sequence

0 / ker(N −→
×b
N) /N ×b
/ bN /0.

×b
As T (bN) ̸= 0, bN is not a nontrivial quotient of N, so ker(N −→ N) = 0. Thus we
have a third short exact sequence

0 /N ×b
/N / N/bN /0

from which
×T (b)
T (N) / T (N) / T (N/bN) = 0

is exact. (Here we are using the B-linearity of T to identify the map T (N) → T (N)
as multiplication by T (b) ∈ m.) Then by Nakayama’s Lemma for A, T (N) = 0,
contradicting T (N) ̸= 0. □

24.6.4. Proof of Theorem 24.6.1. Suppose M is finitely generated A-module. One

direction is easy: if M is flat over B, then TorB
1 (M, B/n) = 0 by the Tor criterion
(Exercise 23.1.C).
For the harder direction, suppose now that TorB 1 (M, B/n) = 0. Define T :
f.g.ModB → f.g.ModA by N 7→ Tor1 (M, N). Then T is A-linear (do you see why?)
B

and half-exact. By Nakayama’s Lemma for half-exact functors, T (B/n) = 0 implies

that T (N) = 0 for all finitely generated B-modules, and in particular all B/I, from
which we see that M is flat over B by the ideal-theoretic criterion for flatness
(Theorem 24.3.1). □

24.6.5. Slicing criteria for flatness.

A useful variant of the local criterion is the following. Suppose t is a non-
zerodivisor of B in n (geometrically: an effective Cartier divisor on the target
passing through the closed point). If M is flat over B, then t is not a zerodivisor
of M (Observation 24.1.2). Also, M/tM is a flat B/tB-module (flatness commutes
with base change, Exercise 24.1.J). The next result says that this is a characterization
of flatness, at least when M is finitely generated, or somewhat more generally.

24.6.6. Theorem (slicing criterion for flatness on the target, see Figure 24.3). —
Suppose (B, n) → (A, m) is a local morphism of Noetherian local rings, M is a finitely
generated A-module, and t ∈ n is a non-zerodivisor on B. Then M is B-flat if and only if
(i) t is not a zerodivisor on M, and
(ii) M/tM is flat over B/(t).
 © 2024 Ravi Vakil. Published by Princeton University Press. 713

F IGURE 24.3. The (necessary and sufficient) slicing criterion for

flatness on the target, Theorem 24.6.6

(For slicing criteria for other properties, see Exercises 13.2.C, Theorem 24.6.9,
and Theorem 26.2.3.)
To prove this, we show a lemma which we will also use to prove Theo-
rem 24.6.14. (Notice that it has no hypotheses of local rings or finite generation.)

24.6.7. Lemma. — Suppose I ⊂ B is an ideal, and M is a B-module. Suppose that (i)

TorB B
1 (B/I, M) = 0, and (ii) M/IM is flat over B/I. Then Tor1 (B/J, M) = 0 for every
ideal J containing I.

Proof. The functor ⊗B B/J is the composition of the functor ⊗B B/I with the functor
⊗B/I B/J. From the Grothendieck composition-of-functors spectral sequence 23.3.5,
there is a spectral sequence with E2 given by TorB/I B
q (B/J, Torp (B/I, M)) converging
B
to Torp+q (B/J, M). The hypotheses of the theorem guarantee that

B/I
Tor0 (B/J, TorB
1 (B/I, M)) = 0

by (i), and

B/I B/I
Tor1 (B/J, TorB
0 (B/I, M)) = Tor1 (B/J, M/IM) = 0

by (ii) (using the ideal-theoretic criterion for flatness for B/I, Theorem 24.3.1). Thus
TorB
1 (B/J, M) = 0 as desired. (This can also be proved readily by hand without the
machinery of spectral sequences.) □

24.6.8. Proof of the slicing criterion for flatness on the target, Theorem 24.6.6. Theo-
rem 24.6.6 then follows by applying the previous lemma with I = (t) and J = n,
×t
using TorB1 (B/tB, M) = ker(M −→ M) from Exercise 24.2.A, which is zero by
hypothesis. □
714 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

24.6.A. E XERCISE . Use the slicing criterion to give a second solution to Exer-
cise 24.3.I, on two planes in P4 meeting at a point. (This will use only the “easy
direction” of the slicing criterion.)
Here is a similar slicing criterion for flatness, but in the source. Note that the
criterion only goes one way.

24.6.9. Theorem (slicing criterion for flatness on the source, see Figure 24.4). —
Suppose (B, n) → (A, m) is a local morphism of Noetherian local rings, M is a finitely
generated A-module, and t ∈ m ⊂ A is a non-zerodivisor on M/nM. If M is flat over B
then M/tM is flat over B.

F IGURE 24.4. The (sufficient) slicing criterion for flatness on the

source, Theorem 24.6.9

Proof. We have an exact sequence of A-modules (and hence B-modules)

M
×t
/M / M/tM / 0.

Extend this to a flat resolution of B-modules (for example by building it leftwards

using free B-modules):

(24.6.9.1) ··· /M ×t
/M / M/tM / 0.

We can compute TorB 1 (B/tM, B/n) using any flat resolution (Exercise 23.3.D), so we
use (24.6.9.1). We apply ⊗B B/n to (24.6.9.1) and truncate the right term to get the
complex whose cohomology will give us TorB i (B/tM, B/n):

··· / M/n ×t
/ M/n / 0.

But ×t : M/n → M/n has kernel zero by hypothesis, so the cohomology at the first
step is zero, giving us that TorB
1 (M/tM, B/n) = 0. Then by the local criterion for
flatness (Theorem 24.6.1), M/tN is flat over B as desired. □
 © 2024 Ravi Vakil. Published by Princeton University Press. 715

24.6.10. ⋆⋆ Relative effective Cartier divisors. Theorem 24.6.9 has an immediate

geometric interpretation: “Suppose π : X → Y is a morphism of Noetherian schemes,
F is a coherent sheaf on X, and Z ,→ X is a locally principal subscheme ...” In the
special case where F = OX , this leads to the notion of a relative effective Cartier
divisor: a locally principal subscheme of X that is an effective Cartier divisor on all
the fibers of π. Theorem 24.6.9 implies that if π is flat, then any relative effective
Cartier divisor is also flat. (See Exercise 25.3.H for an important application.)

24.6.11. Over a field, regular embeddings are locally flat covers of the coordinate
axes in affine space.

24.6.B. U SEFUL E XERCISE . Let (A, m) be a Noetherian local ring, where A contains
a field k (i.e., A is a k-algebra). Suppose f1 , . . . , fn ∈ m. These n functions on
Spec A determine a morphism π : Spec A → An k . Show that π is flat if and only
if f1 , . . . , fn form a regular sequence. More generally, if M is a finitely generated
A-module, show that M is flat over An k if and only if f1 , . . . , fn form an M-regular
sequence. Hint: the slicing criterion for flatness on the target, Theorem 24.6.6.
Exercise 24.6.B makes a number of sophisticated facts surprisingly easy (so
long as we are working over a field k).

24.6.C. E XERCISE . Use Exercise 24.6.B to give an immediate proof of Theorem 9.5.6,
that an M-regular sequence in m remains regular upon any reordering, so long as
A is a k-algebra for some k. (The case of “mixed characteristic” is not covered.) Our
original proof was notably harder (although more general).

24.6.D. E XERCISE . Use Exercise 24.6.B to reprove that the (co)normal sheaf of a
codimension r regular embedding in a Noetherian k-scheme is locally free of rank
r (Proposition 21.2.16(b) for k-schemes).

24.6.E. E XERCISE ( CF. E XERCISE 24.1.O). Use Exercise 24.6.B to reprove that if
I is the ideal sheaf of a codimension r regular embedding in a Noetherian k-
schemes, then the natural map Symn (I /I 2 ) → I n /I n+1 is an isomorphism
(Corollary 22.3.9 for k-schemes).

24.6.F. ⋆⋆ E XERCISE . For this exercise, Recall Exercise 24.4.A. If (A, m) is a Noe-
therian local ring containing a field k, and f1 , . . . , fr is a regular sequence in m,
and s sends the ith basis element of E = A⊕r to fi , use Exercise 24.6.B to show
that the Koszul complex K• (s) (§24.4.1) is exact except at the last term, where the
cohomology is A/(f1 , . . . , fr ). (For a more general statement, see Exercise 24.4.C.)

24.6.12. ⋆⋆ Fibral flatness. We conclude with a criterion for flatness that can be
very useful, but that we will not need in the rest of the book. It applies when
we have the following commutative diagram of locally Noetherian schemes, and
where F is a coherent sheaf on X.

X
π /Y

τ ρ
 
Z
For each point r ∈ Z, let πr : Xr → Yr be the base-change of π.
716 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

24.6.13. Fibral flatness theorem, version one. — Suppose F is flat over Z, and for
every point r ∈ Z, F |Xr is flat over Yr . Then F is flat over Y.

24.6.G. E XERCISE . Show that to prove Theorem 24.6.13, it suffices to prove the
following.

24.6.14. Fibral flatness theorem, commutative algebra version one. — Suppose

(24.6.14.1) (A, m) o (B, n)
c ;

(R, r)
is a commutative diagram of Noetherian local rings (sending maximal ideals to maximal
ideals as usual), and M is a finitely generated A-module. Suppose M is flat over R, and
M/rM is flat over B/rB. Then M is flat over B.

Proof. We will use Lemma 24.6.7. We wish to show that TorB 1 (M, B/n) = 0 (to
invoke the local criterion for flatness, Theorem 24.6.1), and we know that M/rM
is flat over B/rB, so we wish to apply Lemma 24.6.7 for the module M, the ring B,
and the ideals J = n and I = rB. The only hypothesis of Lemma 24.6.7 remaining to
show is that TorB1 (M, B/rB) = 0.
Applying the right-exact functor ·⊗B M to the surjection of B-modules r⊗R B ↠
rB yields a surjection
α : r ⊗R M = r ⊗R B ⊗B M / / (rB) ⊗B M.

Consider the composition γ of maps of B-modules

r ⊗R M
α / (rB) ⊗B M β
/ M.

Now α is surjective, and the composition β ◦ α is injective (apply · ⊗R M to 0 →

r → R, and use that M is R-flat). Thus α must be injective and thus an isomorphism,
from which β must be injective.
Now apply M ⊗B · to the short exact sequence
0 / rB /B / B/rB / 0.

Since B is flat over B, TorB

1 (M, B) = 0, so we have an exact sequence

0 = TorB
1 (M, B)
/ TorB
1 (M, B/rB)
/

M ⊗B rB
β
/ M ⊗B B / M ⊗B B/rB / 0.

From the injectivity of β we have TorR

1 (M, B/rB) = 0 as desired. □
Theorem 24.6.14 can be extended to the following (which can also lead to a
geometric statement similar to Theorem 24.6.13).

24.6.15. Fibral flatness theorem, commutative algebra version two. — Suppose

(24.6.14.1) is a commutative diagram of Noetherian local rings, and M is a nonzero
finitely generated A-module. Then (a) is true if and only if (b) is true.
 © 2024 Ravi Vakil. Published by Princeton University Press. 717

(a) (i) M is flat over R, and (ii) M/rM is flat over B/rB.
(b) (i) M is flat over B, and (ii) B is flat over R

24.6.H. E ASY EXERCISE . Show that (b) implies (a).

Proof. Most of the result was shown in Theorem 24.6.14 and Exercise 24.6.H. It only
remains to show that (a)(i), (a)(ii), and (b)(i) imply (b)(ii).

24.6.I. E XERCISE . Show that (a)(i), (a)(ii), and (b)(i) imply that M is faithfully flat
over B.
We now show directly that B is flat over R. Let N• be an exact sequence of
R-modules. We wish to show that B ⊗R N• is exact. But by (a)(i), M ⊗R N• is exact,
and M ⊗R N• = M ⊗B (B ⊗R N• ) is exact. As M is faithfully flat over B, we have
that B ⊗R N• is exact as desired. □

24.7 Flatness implies constant Euler characteristic

We come to an important consequence of flatness promised in §24.0.1. We will

see that this result implies many answers and examples to questions that we would
have asked before we even knew about flatness.

24.7.1. Important Theorem (Euler characteristic is constant in flat families). —

Suppose π : X → Y is a projective morphism of locally Noetherian
P schemes, and F is a
coherent sheaf on X, flat over Y. Then χ(Xq , F |Xq ) = i≥0 (−1)i hi (Xq , F |Xq ) is a
locally constant function of q ∈ Y (where Xq = π−1 (q)).
This is the first sign that “cohomology behaves well in flat families.” (We
will soon see a second: the Semicontinuity Theorem 25.1.1. A different proof of
Theorem 24.7.1, giving an extension to the proper case, will be given in §25.2.5.)
The Noetherian hypotheses are used to ensure that π∗ F (m) is a coherent sheaf.
Theorem 24.7.1 gives a necessary condition for flatness. Converses (yielding a
sufficient condition) are given in Exercise 24.7.C.

Proof. We make three quick reductions. (i) The question is local on the target
Y, so we may reduce to case Y is affine, say Y = Spec B, so π factors through a
closed embedding X ,→ Pn n
B for some n. (ii) We may reduce to the case X = PB , by
considering F as a sheaf on PB . (iii) We may reduce to showing that for m ≫ 0,
n

h0 (Xq , F (m)|Xq ) is a locally constant function of q ∈ Y, and hi (Xq , F (m)|Xq ) =

0 for i > 0. (Serre vanishing ensures that hi (Xq , F (m)|Xq ) = 0 for i > 0 for
m ≫ 0, but a priori there could be a different m for each q, and we need m to be
independent of q.)
Twist by O(m) for m ≫ 0, so that all the higher cohomology Hi>0 (X, F (m))
of F vanishes (by Serre vanishing, Theorem 18.1.3(ii)). Now consider the Čech
complex C • for F (m). Note that all the terms in the Čech complex C • are flat,
because F is flat. (Do you see why?) As all higher cohomology groups (higher
pushforwards) vanish, C • is exact except at the first term, where the cohomology
is Γ (π∗ F (m)). We add the module Γ (π∗ F (m)) to the front of the complex, so it is
718 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

once again exact:

(24.7.1.1)
0 / Γ (π∗ F (m)) / C1 / C2 / ··· / C n+1 / 0.

(We have done this trick of tacking on a module before, for example in (18.2.4.1).)
Thus by Exercise 24.2.G, as we have an exact sequence in which all but the first
terms are flat, the first term is flat as well. Thus π∗ F (m) is a flat coherent sheaf on
Y, and hence locally free (Corollary 24.3.7), and thus has locally constant rank.
Suppose q ∈ Y. We wish to show that the Hilbert function hF |Xq (m) is a locally
constant function of q. To compute hF |Xq (m), we tensor the Čech resolution with
κ(q) and take cohomology. Now the extended Čech resolution (with Γ (π∗ F (m))
tacked on the front), (24.7.1.1), is an exact sequence of flat modules, and hence
remains exact upon tensoring with κ(q) (Exercise 24.2.F). Thus
∼ Γ (π∗ F (m)|q ),
Γ (π∗ F (m)) ⊗ κ(q) =
so the Hilbert function hF |Xq (m) is the rank at q of a locally free sheaf, which is a
locally constant function of q ∈ Y. □
We now give some ridiculously useful consequences of Theorem 24.7.1.

24.7.2. Corollary. — Assume the same hypotheses and notation as in Theorem 24.7.1.
Then the Hilbert polynomial of F |Xq is locally constant as a function of q ∈ Y.

24.7.A. C RUCIAL E XERCISE . Suppose X → Y is a projective flat morphism, where

Y is connected. Show that the following functions of q ∈ Y are constant: (a) the
degree of the fiber, (b) the dimension of the fiber, (c) the arithmetic genus of the
fiber.

24.7.B. E XERCISE . Use §24.3.9 and Exercise 24.7.A(a) to give another solution to
Exercise 16.3.D(a) (“the degree of a finite morphism from a curve with no embedded
points to an irreducible regular curve is constant”).

24.7.C. U NIMPORTANT E XERCISE ( CONVERSES TO T HEOREM 24.7.1). (We won’t

use this exercise for anything.)
(a) Suppose A is a ring, and S• is a finitely generated A-algebra that is flat over A.
Show that Proj S• is flat over A.
(b) Suppose π : X → Y is a projective morphism of locally Noetherian schemes
(which as always includes the data of an invertible sheaf OX (1) on X), such that
π∗ OX (m) is locally free for all m ≥ m0 for some m0 . Show that π is flat. Hint:
describe X as  M 
Proj OY (⊕m≥m0 π∗ OX (m)) .
(c) More generally, suppose π : X → Y is a projective morphism of locally Noetherian
schemes, and F is a coherent sheaf on X, such that π∗ F (m) is locally free for all
m ≥ m0 for some m0 . Show that F is flat over Y.
(d) Suppose π : X → Y is a projective morphism of locally Noetherian schemes, and
F is a coherent sheaf on X, such that the Hilbert polynomial of F |Xq is a locally
constant function of q ∈ Y. If Y is reduced, show that F must be flat over Y. (Hint:
Exercise 14.3.K shows that constant rank implies local freeness in particularly nice
circumstances.)
 © 2024 Ravi Vakil. Published by Princeton University Press. 719

Another consequence of Corollary 24.7.2 is something remarkably useful.

24.7.3. Corollary. — An invertible sheaf L on a flat projective family of curves X has

locally constant degree on the fibers.
(Recall that the definition of the degree of a line bundle on a projective curve
requires no hypotheses on the curve such as regularity, see (18.4.7.1).)

Proof. An invertible sheaf L on a flat family of curves is always flat (as locally it is
isomorphic to the structure sheaf). Hence χ(Xq , Lq ) is a constant function of q. By
the definition of degree given in (18.4.7.1), deg(Xq , Lq ) = χ(Xq , Lq ) − χ(Xq , OXq ).
The result follows from the local constancy of χ(Xq , Lq ) and χ(Xq , OXq ) (Theo-
rem 24.7.1). □
The following exercise is a serious generalization of Corollary 24.7.3.

24.7.D. ⋆ E XERCISE FOR THOSE WHO HAVE READ STARRED C HAPTER 20: INTERSEC -
TION NUMBERS ARE LOCALLY CONSTANT IN FLAT FAMILIES . Suppose π : X → Y
is a flat projective morphism to a connected scheme; L1 , . . . , Ln are line bundles
on X; and F is a coherent sheaf on X, flat over Y, such that the support of F when
restricted to any fiber of π has dimension at most n. If q is any point of Y, define
(the temporary notation) (L1 · L2 · · · Ln · F )q to be the intersection on the fiber
Xq of L1 , . . . , Ln with F |Xq (Definition 20.1.1). Show that (L1 · L2 · · · Ln · F )q is
independent of q. (The projective hypothesis can be relaxed to properness using
§25.2.5.)
Corollary 24.7.3 motivates the following discussion.

24.7.4. Definition. Suppose L1 and L2 are line bundles on a k-variety X. We say

that L1 and L2 are algebraically equivalent if there exists a connected (but not
necessarily irreducible) k-variety Y with two k-valued points q1 and q2 , and a line
bundle L on X × Y such that the restriction of L to the fibers over q1 and q2 are
isomorphic to L1 and L2 respectively.

24.7.E. ⋆ E XERCISE . Show that “algebraic equivalence” really is an equivalence

relation. Show that the line bundles algebraically equivalent to O form a subgroup
of Pic X. This subgroup is denoted Pic0 X. (You will use the notion of gluing
two schemes together along isomorphic closed subschemes, §15.2.12.) Warning:
despite the unfortunately identical notation for degree 0 line bundles on curves
(Definition 18.4.2), this is not the same concept.

24.7.5. The Néron-Severi group is the group of line bundles Pic X modulo alge-
braic equivalence, Pic X/ Pic0 X. (This definition was promised in §18.4.12.) By
Corollary 24.7.3, algebraic equivalence implies numerical equivalence (§18.4.12):
Pic0 X ⊂ Picτ X. (Side Remark: a line bundle L on a proper k-scheme X is numer-
ically trivial if and only if there exists an integer m ̸= 0 with L ⊗m algebraically
trivial. Thus Picτ X/ Pic0 X is torsion. See [SGA6, XIII, Thm. 4.6] for a proof, or
[Laz, Cor. 1.4.38] for the projective case.)

24.7.6. ⋆ Hironaka’s example of a proper nonprojective smooth threefold.

720 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

In §15.2.11, we produced a proper nonprojective variety, but it was singular.

We can use Corollary 24.7.3 to give a smooth example, due to Hironaka. It is in in
some sense “locally a blow-up of P3 ”.
Inside P3k , fix two conics C1 and C2 , which meet in two (k-valued) points, p1
and p2 . We construct a proper map π : X → P3k as follows. For i = 1 and 2, let
Ui = P3 \ {pi }. For both i, consider the projective morphism πi : Xi → Ui obtained
by blowing up Ci , and then blowing up the proper transform of C3−i . The two πi
glue together over P3 \ {p1 , p2 } to give a morphism π : X → P3 . (Away from p1 and
p2 , C1 and C2 are disjoint, so blowing up one and then the other is the same as
blowing up their union.)

F IGURE 24.5. Hironaka’s example of a proper nonprojective

smooth threefold

Note that π is proper, as it is proper over U1 and U2 , and the notion of

properness is local on the target (Proposition 11.4.4(b)). As P3k is proper (over
k, Theorem 8.4.10), and compositions of proper morphisms are proper (Proposi-
tion 11.4.4(c)), X is proper.

24.7.F. E XERCISE . Show that X is smooth. (Hint: Theorem 22.3.10.) Let Ei be the
preimage of Ci \ {p1 , p2 }. Show that π|Ei : Ei → Ci \ {p1 , p2 } is a P1 -bundle (and
flat).

24.7.G. E XERCISE . Let Ei be the closure of Ei in X. Show that Ei → Ci is flat. (Hint:

Exercise 24.3.J.)

24.7.H. E XERCISE . Show that π−1 (pi ) is the union of two P1 ’s, say Yi and Zi ,
meeting at a point, such that Yi , Y3−i , Z3−i ∈ Ei but Zi ∈
/ Ei .

24.7.I. E XERCISE . Show that X is not projective as follows. Suppose otherwise

L is a very ample line bundle on X, so L has positive degree on every curve
(including the Yi and Zi ). Using flatness of Ei → Ci , and constancy of degree in flat
families (Exercise 24.7.D), show that degYi L = degY3−i L + degZ3−i L . Obtain a
contradiction. (This argument will remind you of the argument of §15.2.11.)

24.7.7. The notion of “projective morphism” is not local on the target. Note that π : X →
P3 is not projective, as otherwise X would be projective (as the composition of
projective morphisms is projective if the final target is quasicompact, Exercise 17.3.I).
But away from each pi , π is projective (as it is a composition of blow-ups, which are
 © 2024 Ravi Vakil. Published by Princeton University Press. 721

projective by construction, and the final target is quasicompact, so Exercise 17.3.I

applies). Thus the notion of “projective morphism” is not local on the target. (A
simpler but non-smooth example can be constructed using §15.2.11.)

24.8 Smooth and étale morphisms, and flatness

We defined smooth and étale morphisms earlier (Definition 13.6.2), but we are
now in a position to understand them much better. We will see that the notion of
unramified morphism (§21.7) is a natural companion to them.
Our three algebro-geometric definitions won’t be so obviously a natural triplet,
but we will discuss the definitions given in [Gr-EGA] (§24.8.13), and in this con-
text the three types of morphisms look very similar. (We briefly mention other
approaches and definitions in §24.8.14.)
The three classes of morphisms we will discuss in this chapter are the analogs
of the following types of maps of manifolds, in differential geometry (cf. §13.6.1).
• Submersions are maps inducing surjections of tangent spaces everywhere.
They are useful in the notion of a fibration. (Perhaps a more relevant
notion from differential geometry, allowing singularities, is: “locally on
the source a smooth fibration”.)
• Isomorphisms locally on the source (or local isomorphisms) are maps inducing
isomorphisms of tangent spaces.
• Immersions are maps inducing injections of tangent spaces.
(Recall our warning from §9.2: “immersion” is often used in algebraic geometry
with a different meaning.)

F IGURE 24.6. Sketches of notions from differential geometry: (a)

maps “locally on the source a smooth fibration”, (b) local isomor-
phisms, and (c) immersions. (In algebraic geometry: (a) smooth
morphisms, (b) étale morphisms, and (c) unramified morphisms.)

In order to better understand smooth and étale morphisms, we temporarily

forget our earlier definitions, and consider some examples of things we want to
be analogs of “local isomorphism” (or “locally on the source an isomorphism”)
and “locally on the source a smooth fibration”, and see if they help us make good
definitions.
722 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

24.8.1. “Local isomorphisms” (étale morphisms). Consider the parabola x = y2

projecting to the x-axis, over the complex numbers. (This example has come
up repeatedly, in one form or another, including in Exercise 13.6.F.) We might
reasonably want this to be a local isomorphism (on the source) away from the
origin. We might also want the notion of local isomorphism space to be an open
condition: the locus where a morphism is a local isomorphism should be open
on the source. This is true for the differential geometric definition. But then this
morphism should be a local isomorphism over the generic point, and here we
get a non-trivial residue field extension (C(y)/C(y2 )), not an isomorphism. Thus
we are forced to consider (the Spec’s of) certain finite extensions of fields to be
“isomorphisms locally on the source” — in this case, even “covering spaces”. (We
will see in Exercise 24.8.D that we want precisely the finite separable extensions;
you could have shown this earlier.)
Note also in this example there are no (nonempty) Zariski-open subsets U ⊂
Spec C[x] \ {0} and V ⊂ Spec C[x, y]/(x − y2 ) \ {(0, 0)} where the map sends U into
V isomorphically (Exercise 13.6.F), so this is not a local isomorphism in a way you
may have seen before. This leads to the notion of the étale topology, which is not
even a topology in the usual sense, but a “Grothendieck topology” (§6.2.10). The
étale topology is beyond the scope of this book.

24.8.2. Submersions (smooth morphisms).

24.8.3. Fibers are smooth varieties. As a first approximation of the algebro-geometric

version of submersion, we will want the fibers to be smooth varieties (over the
residue field). So the very first thing we need is to generalize the notion of “variety”
over a base. It is reasonable to do this by having a locally finite type hypothesis. For
somewhat subtle reasons, we will require the stronger condition that the morphism
be locally of finite presentation. (If you really care, you can see where it comes up
in our discussion. But of course there is no difference for Noetherian readers.)
The fibers are not just varieties; they should be smooth varieties (of dimension
n, say). From the case of smoothness over a field, or from our intuition of what
smooth varieties should look like, we expect the sheaf of differentials on the fibers
to be locally free of rank n, or even better, that the sheaf of relative differentials be
locally free of rank n.

24.8.4. Flatness. At this point, our first approximation of “smooth morphism” is

some version of “locally finitely presented, and fibers are smooth varieties”. But
that isn’t quite enough. For example, a horrible map from a scheme X to a curve
Y that maps a different regular variety to each point of Y (X is the infinite disjoint
union of these) should not be considered a smooth fibration in any reasonable sense.
Also, we might not want to consider Spec k → Spec k[ϵ]/(ϵ2 ) to be a submersion;
for example, this isn’t surjective on tangent spaces, and more generally the picture
“doesn’t look like a fibration”.
Both problems are failures of π : X → Y to be a nice, “’continuous” family.
Whenever we are looking for some vague notion of “niceness” we know that
“flatness” will be in the definition.
For comparison, note that “unramified” has no flatness hypothesis, and indeed
we didn’t expect it, as we would want the inclusion (closed embedding) of the
origin into A1 to be unramified. But then weird things may be unramified. For
 © 2024 Ravi Vakil. Published by Princeton University Press. 723

F IGURE 24.7. We don’t want this to be a “smooth morphism”

`
z∈C Spec C, then the morphism X → AC sending the point
1
example, if X =
1
corresponding to z to the point z ∈ AC is unramified. Such is life.

24.8.5. Desired Alternate Definitions. We might hope that a morphism π : X → Y is

smooth of relative dimension n if and only if
(i) π is locally of finite presentation,
(ii) π is flat of relative dimension n, and
(iii) ΩX/Y is locally free of rank n.
We might similarly hope that a morphism π : X → Y is étale if and only if
(i) π is locally of finite presentation,
(ii) π is flat, and
(iii) ΩX/Y = 0.
We will shortly show (in Theorem 24.8.8 and Exercise 24.8.C respectively) that these
desired definitions are equivalent to Definition 13.6.2.

24.8.6. Different characterizations of smooth and étale morphisms.

The main result in this section is a description of equivalent characterizations
of smooth morphisms, Theorem 24.8.8. We will state the theorem, and then give
consequences, and then finally give a proof. As an important precursor:

24.8.7. Proposition. — Smooth morphisms are flat.

Proof. We prove this first in the Noetherian setting. Suppose π : X → Y is a

morphism of locally Noetherian schemes, that is smooth of relative dimension
n. We wish to show that it is flat at a point p ∈ X. By the definition of smooth
morphisms (§13.6.2), we may replace X and Y with affine neighborhoods of p and
π(p) respectively, so that π is the morphism

(24.8.7.1) Spec B[x1 , . . . , xn+r ]/(f1 , . . . , fr ) / Spec B,

and p is a point where the Jacobian matrix is full rank.

24.8.A. E XERCISE . Show inductively, using the slicing criterion for flatness on the
source (Theorem 24.6.9), that
An+r
B ⊃ V(f1 ) ⊃ V(f1 , f2 ) ⊃ · · · ⊃ V(f1 , . . . , fr ) = X
724 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

are all smooth and flat over B at p, thereby proving Proposition 24.8.7 in the
Noetherian case. (You will need the fact that smooth k-schemes are reduced,
Proposition 13.2.13.)
(Noetherian readers can smugly skip to the end of the proof.) We prove the
general case by a sneaky trick. We still write π locally as (24.8.7.1). By rearranging
the x-variables, we may assume that the Jacobian of f1 , . . . , fr by the first r variables
x1 , . . . , xr is invertible. Suppose
X ai ain+r
fi (x1 , . . . , xn+r ) = biai ,...,ai x1 1 · · · xn+r ,
1 n+r
ai1 ,...,ain+r

where bij,...,k ∈ B. (These are finite sums!) Consider the polynomial algebra
P = Z[cij,...,k ] where the indices on the c-variables are precisely those that appear
on the indices on the b-variables. Define gi ∈ P[x1 , . . . , xn+r ] by
X ai ain+r
gi (x1 , . . . , xn+r ) = ciai ,...,ai x1 1 · · · xn+r .
1 n+r
ai1 ,...,ain+r

Let J ∈ P[x1 , . . . , xn+r ] be the determinant of the Jacobian of g1 , . . . , gr by x1 , . . . , xr .

Finally, let R := PJ . We have a Cartesian diagram

Spec B[x1 , . . . , xn+r ]/(f1 , . . . , fr ) / Spec R[x1 , . . . , xn+r ]/(g1 , . . . , gr )

π α
 
Spec B / Spec R

sending cij,...,k to bij,...,k , and gi to fi . Because J is invertible in R, α is smooth of

relative dimension n (basically by definition). From the Noetherian case of the
result already proved, α is flat. Because flatness is preserved by base change, we
have that π is flat as desired. □
We now come to the central result of this section.

24.8.8. Theorem. — Suppose π : X → Y is a morphism of schemes. Then the following

are equivalent.
(i) The morphism π is smooth of relative dimension n (Definition 13.6.2).
(ii) The morphism π is locally finitely presented and flat of relative dimension n; and
Ωπ is locally free of rank n (Desired Alternate Definition 24.8.5).
(iii) The morphism π is locally finitely presented and flat, and the fibers are smooth
k-schemes of pure dimension n.
(iv) The morphism π is locally finitely presented and flat, and the geometric fibers are
smooth k-schemes of pure dimension n.
We first dispatch with the parts of this theorem we have basically already
proved.

24.8.B. E XERCISE . Prove that (i) implies (ii) implies (iii). Show that (iii) is equivalent
to (iv). Hints: smooth implies flat by Proposition 24.8.7; the definition of smooth
k-varieties in terms of differentials in Redefinition 21.2.28; and smoothness is
independent of field extension by Exercise 13.2.H, respectively.
 © 2024 Ravi Vakil. Published by Princeton University Press. 725

Before finishing the proof of Theorem 24.8.8 in §24.8.11, we motivate the result
by giving a number of exercises and results assuming it.

24.8.C. E XERCISE . Show that a morphism π is étale (Definition 13.6.2) if and

only if π is locally of finite presentation, flat, and Ωπ = 0 (Desired Alternate
Definition 24.8.5).

24.8.D. E ASY E XERCISE . Suppose π : X → Y is a morphism. Show that the following

are equivalent.
(a) π is étale.
(b) π is smooth and unramified.
(c)π is locally finitely presented, flat, and unramified.
(d) π is locally finitely presented, flat, and for each q ∈ Y, π−1 (q) is the
disjoint union of schemes of the form Spec K, where K is a finite separable
extension of κ(q).
(e) π is locally finitely presented, flat, and for each geometric point q of Y,
π−1 (q) is the disjoint union of copies of q.

24.8.E. I MPORTANT E XERCISE . If π : X → Y is étale, show that any preimage

p ∈ X of any regular point q ∈ Y whose local ring has dimension n is also a
regular point whose local ring has dimension n. Hint: Prove the result by induction
on dim OY,q . “Slice” by an element of mY,q \ m2Y,q . Use the slicing criterion for
regularity (Exercise 13.2.C).

24.8.9. Proof of the Smoothness-Regularity Comparison Theorem 13.2.7(b) (every smooth

k-scheme is regular). By Observation 13.6.4, any dimension n smooth k-scheme X
can locally be expressed as an étale cover of An
k , which is regular by Exercise 13.3.P.
Then by Exercise 24.8.E, X is regular. □

24.8.F. E XERCISE . Suppose π : X → Y is a morphism of pure dimensional k-varieties,

and Y is smooth (over k). Use the cotangent exact sequence (Theorem 21.2.24) to
show the following.
(a) Suppose that dim X = dim Y = n, and π is unramified. Show that X is smooth.
(b) Suppose dim X = m > dim Y = n, and the (scheme-theoretic) fibers of π over
closed points are smooth of dimension m − n. Show that X is smooth.

24.8.G. E XERCISE . Suppose π : X → Y is a morphism (over k) of smooth pure

dimensional k-varieties, where dim X = m and dim Y = n. (The two parts are
clearly essentially the same geometric situation.)
(a) Show that if the fibers of π are all smooth of dimension m − n, then π is smooth
of relative dimension m − n.
(b) Show that if π∗ ΩY → ΩX is an “injection of vector bundles” (see Defini-
tion 14.2.4), or equivalently, “Tπ is surjective”, then π is smooth of relative di-
mension m − n.
Possible approach: for each point q ∈ Y, OY,q is a regular local ring. Work by
induction on dim OY,q , and “slice” by an element of mY,q \ m2Y,q .

24.8.10. Remark. Suppose π : X → Y is locally finitely presented, ρ : Y → Z is étale,

and τ = ρ ◦ π. Then π is smooth of dimension n (e.g., étale, taking n = 0) if and
726 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

only if τ is. One direction is Exercise 13.6.D, and the other is Exercise 21.7.G(b).
Similarly, π is unramified if and only if τ is (Exercises 21.7.C and 21.7.G(a)).

24.8.11. ⋆⋆ Proof of Theorem 24.8.8.

By Exercise 24.8.B, it remains only to show that (iii) implies (i). (This argument
is reminiscent of the proof of Theorem 13.6.6.)
Fix a closed point p ∈ X. Let q = π(p) ∈ Y. We will show that there is an open
neighborhood Ui of p and Vi of q of the form stated in Definition 13.6.2. It suffices
to deal with the case that p is a closed point in π−1 (q), because closed points are
dense in finite type schemes (Exercise 5.3.F). (However, we cannot assume that q is
a closed point of Y.)
As π is finitely presented, there are affine open neighborhoods U ⊂ X of p and
V ⊂ Y of q, with π(U) ⊂ V, such that the morphism π|U : U → V can be written as

π|U
U = Spec B[x1 , . . . , xN ]/I / Spec B = V,

where I is finitely generated. Choose generators f1 , . . . , fs of I. Consider the

Cartesian diagram

 /U
U


Spec κ(q)[x1 , . . . , xN ]/I  / Spec B[x1 , . . . , xN ]/I

π|U
  
Spec κ(q)  / Spec B

 /V
q

where U is the fiber of U over q, and I is the ideal cutting out U in AN κ(q) .
Recall from §24.2.2 that for a closed subscheme flat a base, the pullback of
the defining ideal as a quasicoherent sheaf is the same as the ideal of the pulled
back closed subscheme. Hence because B[x1 , . . . , xN ]/I is flat over B, we have that
I = I ⊗B κ(q). Then f1 := f1 ⊗ 1, . . . , fs := fs ⊗ 1 generate I ⊗B κ(q) = I.
By hypothesis (iii), the fiber U ⊂ AN κ(q) is smooth over κ(q), so OU,p is an
integral domain of dimension n (Exercise 13.2.O). Also by hypothesis (iii), the
Jacobian matrix of f1 , . . . , fs (with respect to x1 , . . . , xN ) has corank n at p, i.e.,
the Jacobian matrix has rank r := N − n. So we can rename the fi so that the
′
Jacobian matrix of the first r of them, f1 , . . . , fr , has corank n at p. Then U :=
Spec κ(q)[x1 , . . . , xN ]/(f1 , . . . , fr ) is smooth of relative dimension n over κ(q) (in a
neighborhood of p), so OU ′ ,p is an integral domain of dimension n (Exercise 13.2.O
again).
Then

OU ′ ,p / / O ′ /(fr+1 , . . . , fs ) = O
U ,p U,p
 © 2024 Ravi Vakil. Published by Princeton University Press. 727

F IGURE 24.8. Geometry behind part of the proof of Theorem 24.8.8

is a surjection of integral domains of the same dimension n, and thus an isomor-

phism by Easy Exercise 13.2.M. Hence U = U ′ near p in U (or more precisely, there
is a neighborhood of p in U that is also a neighborhood of p in U ′ ).
Let U ′ ⊂ Spec B[x1 , . . . , xN ]/(f1 , . . . , fr ) be the open subset where the Jacobian
matrix of f1 , . . . , fr (with respect to x1 , . . . , xN ) has corank n (i.e., full rank r). Then
U ′ → Spec B is smooth of relative dimension n by definition. We will show that
there is a neighborhood of p in U that is also a neighborhood of p in U ′ , thereby
completing the proof.
Let R := B[x1 , . . . , xn ]/(f1 , . . . , fr ) and J := (fr+1 , . . . , fs ) ⊂ R for convenience
(see Figure 24.8). As J is a finitely generated ideal, it is supported on a closed subset
of Spec R (Exercise 6.6.B(b)). We wish to show that p is not in the support of J in
Spec R. It suffices to show that J ⊗R κ(p) = 0 by Nakayama’s Lemma.
Because U = Spec R/J ⊂ X is smooth over B, R/J is flat over B (Proposi-
tion 24.8.7). Now J ⊗B κ(q) is the ideal cutting out U ′ in U (§24.2.2 again, using
flatness of U = Spec R/J), and U = U ′ near p in U. Hence J ⊗B κ(q) = 0 near p (i.e.,
J ⊗^B κ(q) = 0 in an open neighborhood of p), so (J ⊗B κ(q)) ⊗R κ(p) = 0. Then

J ⊗R κ(p) = J ⊗R (κ(p) ⊗B κ(q)) = (J ⊗B κ(q)) ⊗R κ(p) = 0

as desired. □

24.8.12. ⋆⋆ Formally unramified, smooth, and étale.

[Gr-EGA] takes a different starting point for the definition of unramified,
smooth, and étale morphisms. The definitions there make clear that these three
definitions form a family.
The cost of these definitions are that they are perhaps less immediately moti-
vated by geometry, and it is harder to show some basic properties. The benefit is
that it is possible to show more (for example, left-exactness of the cotangent and
conormal exact sequences, and good interpretations in terms of completions of local
rings). But we simply introduce these ideas here, and do not explore them. See
728 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

[BLR, §2.2] for an excellent discussion. (You should largely ignore what follows,
unless you find later in life that you really care.)

24.8.13. Definition. We say that π : X → Y is formally smooth (resp., formally

étale, formally unramified) if for all affine schemes Z, and every closed subscheme
Z0 ⊂ Z defined by a nilpotent ideal, and every morphism Z → Y, the canonical
map HomY (Z, X) → HomY (Z0 , X) is surjective (resp., bijective, injective). This is
summarized in the following diagram, which is reminiscent of the valuative criteria
for separatedness and properness.
Z0 _ /X
>
?
nilpotent ideal π
 
Z /Y
(You can check that this is the same as the definition we would get by replacing
“nilpotent” by “square-zero”. This is sometimes an easier formulation to work
with.)
Then [Gr-EGA] defines smooth as morphisms that are formally smooth and
locally of finite presentation, and similarly for unramified and étale.
One can show that [Gr-EGA]’s definitions of formally unramified, and smooth
agree with the definitions we give. For “formally unramified” (where our defini-
tion given in §21.7 is not obviously the same as the definition of [Gr-EGA] given
here), see [Gr-EGA, IV4 .17.2.1] or [Stacks, tag 00UO]. For “smooth”, see [Gr-EGA,
IV4 .17.5.1] or [Stacks, tag 00TN]. (Our characterization of étale as smooth of relative
dimension 0 then agrees with [Gr-EGA]. Our definition of unramified as formally
unramified plus locally of finite type disagrees with [Gr-EGA], as mentioned in
§21.7.)

24.8.14. Other starting points. There are a number of quite different ways of defin-
ing smooth, étale, and unramified morphisms, and it is nontrivial to relate them.
We have just described the approach of [Gr-EGA]. Another common approach is
the characterization of smooth morphisms as locally finitely presented, flat, and
with regular geometric fibers (see Theorem 24.8.8). Yet another definition is via a
naive version of the cotangent complex; this is the approach taken by [Stacks, tag
00T2], and is less frightening than it sounds. Finally, the different characterizations
of Exercise 24.8.D give alternate initial definitions of étaleness.
 CHAPTER 25

Cohomology and base change theorems

Cohomology and base change theorems are powerful results relating the higher
pushforwards of coherent sheaves with the cohomology of those sheaves on fibers,
and more generally explaining when higher pushforward commutes with base
change. We first describe some of their statements, and immediate applications, in
§25.1. We give their unexpectedly elegant proofs in §25.2. Then in §25.3 we use
them to introduce a number of important ideas in moduli theory.

25.1 Statements and applications

Higher pushforwards are easy to define, but it is hard to get a geometric sense
of what they are, or how they behave. For example, given a morphism π : X → Y,
and a quasicoherent sheaf F on X, you might reasonably hope that the fibers of
Ri π∗ F are the cohomologies of F along the fibers. More precisely, given ψ : q → Y
corresponding to the inclusion of a point (better: ψ : Spec κ(q) → Y), yielding the
Cartesian diagram
ψ′
(25.1.0.1) Xq /X

π′ π
 
q
ψ
/ Y,
one might hope that the morphism

ϕp
q : ψ (R π∗ F )
∗ p / Hp (Xq , F |Xq )

(the push-pull map in cohomology given in Exercise 18.7.C) is an isomorphism.

(Note: F |Xq and (ψ ′ )∗ F are symbols for the same thing. The first is often preferred,
but we sometimes use the second because we will consider more general ψ and
ψ ′ .) We could then picture Ri π∗ F as somehow fitting together the cohomology
groups of fibers into a coherent sheaf. (Warning: this is too much to hope for, see
Exercise 25.1.A.)
It would also be wonderful if hp (Xq , (ψ ′ )∗ F ) was constant, and ϕp
q put them
together into a nice locally free sheaf (vector bundle) Rp π∗ F .
The base change ψ : q → Y should not be special. As long as we are dream-
ing, we may as well hope that in good circumstances, given a Cartesian diagram

This material is published by Princeton University Press as The Rising Sea: Foundations of Algebraic
Geometry by Ravi Vakil. This version is free to view and download for personal use only. Not for
redistribution, resale, or use in derivative works. © 2024 Ravi Vakil.

729
730 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

(18.7.4.1)
ψ′
(25.1.0.2) W /X

π′ π
 
Z
ψ
/ Y,
the natural push-pull morphism

ϕp / Rp π∗′ (ψ ′ )∗ F
Z : ψ (R π∗ F )
∗ p
(25.1.0.3)

of sheaves on Z (Exercise 18.7.B(a)) is an isomorphism. (In some cases, we can

already address this question. For example, cohomology commutes with flat base
change, Theorem 24.1.9, so the result holds if ψ is flat.
We formalize our dreams into three nice properties that we might wish in this
situation. We will see that they are closely related. Suppose F is a coherent sheaf
on X, π : X → Y is proper, Y (hence X) is Noetherian, and F is flat over Y.
(a) Given a Cartesian square (25.1.0.1), is ϕp
q : R π∗ F ⊗κ(q) → H (Xq , F |Xq )
p p

an isomorphism?
(b) Given a Cartesian square (25.1.0.2), is ϕp Z : ψ (R π∗ F ) → R π∗ (ψ ) F
∗ p p ′ ′ ∗

an isomorphism?
(c) Is Rp π∗ F locally free?
We turn first to property (a). The dimension of the left side Rp π∗ F ⊗ κ(q) is an
upper semicontinuous function of q ∈ Y by upper semicontinuity of rank of finite
type quasicoherent sheaves (Exercise 14.3.J). The Semicontinuity Theorem states
that the dimension of the right is also upper semicontinuous. More formally:

25.1.1. Semicontinuity Theorem. — Suppose π : X → Y is a proper morphism of

Noetherian schemes, and F is a coherent sheaf on X flat over Y. Then for each p ≥ 0,
the function Y → Z given by q 7→ dimκ(q) Hp (Xq , F |Xq ) is an upper semicontinuous
function of q ∈ Y.
Translation: ranks of cohomology groups are upper semicontinuous in proper
flat families. (A proof will be given in §25.2.4.)

25.1.2. Example. You may already have seen an example of cohomology groups
jumping, in §24.3.14. Here is a simpler example, albeit not of the structure sheaf. Let
(E, p0 ) be an elliptic curve over a field k, and consider the projection π : E×E → E to
the second factor. Let L be the invertible sheaf (line bundle) on E×E corresponding
to the divisor that is the diagonal, minus the section of p0 × E of π (where p0 ∈ E).
Then L |p0 (i.e., L |E×p0 ) is trivial, but L |p is non-trivial for any p ̸= p0 (as we
showed in our study of genus 1 curves, in §19.9). Thus h0 (E, L |p ) is 0 in general,
but jumps to 1 for p = p0 .

25.1.A. E XERCISE . Show that π∗ L = 0. Thus we cannot picture π∗ L as “gluing

together” h0 of the fibers; in this example, cohomology does not commute with
“base change” or “taking fibers”.

25.1.3. Side Remark. In characteristic 0, the cohomology of O doesn’t jump in

smooth families. Over C, this is because Betti numbers are constant in connected
families, and (21.5.11.1) (from Hodge theory) expresses the Betti constants hkBetti
 © 2024 Ravi Vakil. Published by Princeton University Press. 731

as sums (over i + j = k) of upper semicontinuous functions hj (Ωi ), so the Hodge

numbers hj (Ωi ) must in fact be constant. The general characteristic 0 case can
be reduced to C by an application of the Lefschetz principle (which also arose in
§21.5.9). But ranks of cohomology groups of O for smooth families of varieties can
jump in positive characteristic (see for example [MO70920]). Also, the example of
§24.3.14 shows that the “smoothness” hypothesis cannot be removed.

25.1.4. Grauert’s Theorem. If Rp π∗ F is locally free (property (c)) and ϕp q is an

isomorphism (property (a)), then hp (Xq , F |Xq ) is clearly locally constant. The
following is a partial converse.

25.1.5. Grauert’s Theorem. — If π : X → Y is proper, Y is reduced and locally Noetherian,

F is a coherent sheaf on X flat over Y, and hp (Xq , F |Xq ) is a locally constant function of
q ∈ Y, then Rp π∗ F is locally free, and ϕp
Z is an isomorphism for all ψ : Z → Y.

In other words, if cohomology groups of fibers have locally constant dimension

(over a reduced base), then they can be fit together to form a vector bundle, and the
fiber of the pushforward is identified with the cohomology of the fiber. Our dreams
at the start of this chapter have come true!
(See §25.2.10 to remove Noetherian hypotheses.)
We further note that if Y is integral, π is proper, and F is a coherent sheaf on
X flat over Y, then by the Semicontinuity Theorem 25.1.1 there is a dense open
subset of Y on which Rp π∗ F is locally free (and on which the fiber of the pth higher
pushforward is the pth cohomology of the fiber).
The following statement is even more magical than Grauert’s Theorem 25.1.5.

25.1.6. Cohomology and Base Change Theorem. — Suppose π is proper, Y is locally

Noetherian, F is coherent over X and flat over Y, and ϕp
q is surjective. Then the following
hold.

(i) There is an open neighborhood U of q such that for any ψ : Z → U, ϕp Z is an

isomorphism. In particular, ϕp q is an isomorphism.
(ii) Furthermore,
(a) ϕp−1
q is surjective (hence an isomorphism by (i)) if and only if
(b) Rp π∗ F is locally free in some open neighborhood of q (or equivalently,
(Rp π∗ F )q is a free OY,q -module, Exercise 14.3.E).
(These then imply that hp (Xr , F |Xr ) is constant for r in an open neighborhood
of q.)

(Proofs of Theorems 25.1.5 and 25.1.6 will be given in §25.2. Note in (ii) that
if p = 0, ϕp−1
q is automatically surjective, as ϕ−1
q is the zero map. See §25.2.10 to
remove Noetherian hypotheses.)
This is amazing: the hypothesis that ϕp q is surjective involves what happens
only over points q of X, with reduced structure, yet it has implications over the
(possibly nonreduced) scheme as a whole! This might remind you of the local
criterion for flatness (Theorem 24.6.1), which indeed is the key technical ingredient
of the proof.
Here are some consequences.
732 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

25.1.B. E XERCISE . Use Theorem 25.1.6 to give a second solution to Exercise 24.3.E.
(This is a big weapon to bring to bear on this problem, but it is still enlighten-
ing; your original solution to Exercise 24.3.E may foreshadow the proof of the
Cohomology and Base Change Theorem 25.1.6.)

25.1.C. E XERCISE . Suppose π : X → Y is proper, Y is locally Noetherian, and F is a

coherent sheaf on X, flat over Y. Suppose further that Hp (Xq , F |Xq ) = 0 for some
q ∈ Y. Show that there is an open neighborhood U of q such that (Rp π∗ F )|U = 0.

25.1.D. E XERCISE . Suppose π : X → Y is proper, Y is locally Noetherian, and F is

a coherent sheaf on X, flat over Y. Suppose further that Hp (Xq , F |Xq ) = 0 for all
q ∈ Y. Show that the (p − 1)st cohomology commutes with arbitrary base change:
ϕp−1
Z is an isomorphism for all ψ : Z → Y.

25.1.E. E XERCISE . Suppose π is proper, Y is locally Noetherian, and F is a coherent

sheaf on X flat over Y. Suppose further that Rp π∗ F = 0 for p ≥ p0 . Show that
Hp (Xq , F |Xq ) = 0 for all q ∈ Y, p ≥ p0 .

25.1.F. E XERCISE . Suppose π is proper, Y is locally Noetherian, and F is a coherent

sheaf on X, flat over Y. Suppose further that Y is reduced. Show that there exists
a dense open subset U of Y such that ϕp Z is an isomorphism for all ψ : Z → U
and all p. (Hint: find suitable open neighborhoods of the generic points of Y. See
Exercise 24.1.M and the paragraph following it.)

25.1.G. E XERCISE . Suppose π : C → S is a proper smooth morphism whose fibers

are genus g > 1 curves (a “flat family of smooth genus g curves”). Suppose further
that S is locally Noetherian.
(a) If n > 1, show that π∗ Ω⊗n
C/S is a vector bundle of rank (2n − 1)(g − 1), and that
this construction commutes with any base change S ′ → S (with no Noetherian
hypotheses on S ′ ). More precisely, if

C′ /C

π′ π
 
S′
ρ
/S

is a Cartesian diagram, show that ρ∗ π∗ Ω⊗n ∼ ′ ⊗n

C/S = π∗ ΩC ′ /S ′ .
(b) If S is reduced, show that π∗ ΩC/S is a vector bundle of rank g, and that this
construction commutes with any base change S ′ → S (with no hypotheses on S).
Remark: In fact, the reducedness hypothesis can be removed. One bizarre way of
doing this is by showing that there is a moduli space Mg of smooth genus g > 1
curves, which is smooth (albeit a Deligne-Mumford stack, not a scheme), and then
applying this argument to the universal curve over Mg , from which all families of
genus g curves are pulled back (cf. Exercise 25.3.A).

25.1.7. An important class of morphisms: Proper, O-connected morphisms π : X →

Y of locally Noetherian schemes.
If a morphism π : X → Y satisfies the property that the natural map OY →
π∗ OX is an isomorphism, we say that π is O-connected. Clearly the notion of
O-connectedness is local on the target, and preserved by composition.
 © 2024 Ravi Vakil. Published by Princeton University Press. 733

25.1.H. E ASY E XERCISE . Show that proper O-connected morphisms of locally

Noetherian schemes are surjective.

25.1.8. We will soon meet Zariski’s Connectedness Lemma 28.4.1, which shows that
proper, O-connected morphisms of locally Noetherian schemes have connected
fibers. In some sense, this class of morphisms is really the right class of morphisms
capturing what we might want by “connected fibers”; this is the motivation for the
terminology. The following result gives some evidence for this point of view, in the
flat context.

25.1.I. I MPORTANT E XERCISE . Suppose π : X → Y is a proper flat morphism of

locally Noetherian schemes, whose fibers satisfy h0 (Xq , OXq ) = 1. (Important
remark: this is satisfied if π has geometrically connected and geometrically reduced
fibers, by §11.4.7.) Show that π is O-connected. Hint: consider

ϕ0 ∼
(25.1.8.1) OY ⊗ κ(q) / (π∗ OX ) ⊗ κ(q) q
/ H0 (Xq , OX ) o / κ(q).
q

The composition is surjective, hence ϕ0q is surjective, hence it is an isomorphism

by the Cohomology and Base Change Theorem 25.1.6(i). Then by the Cohomology
and Base Change Theorem 25.1.6(ii), π∗ OX is locally free, thus of rank 1. Perhaps
use Nakayama’s Lemma to show that a map of invertible sheaves OY → π∗ OX
that is a surjection on fibers over points (with reduced structure) is necessarily an
isomorphism of sheaves. Then extract “surjection on fibers” from (25.1.8.1).

25.1.9. ⋆ Unimportant remark. This class of proper, O-connected morphisms

is not preserved by arbitrary base change, and thus is not “reasonable” in the
sense of §8.1. But you can show that they are preserved by flat base change,
using the fact that cohomology commutes with flat base change, Theorem 24.1.9.
Furthermore, the conditions of Exercise 25.1.I behave well under base change, and
Noetherian hypotheses can be removed from the Cohomology and Base Change
Theorem 25.1.6 (at the expense of finitely presented hypotheses, see §25.2.10), so
the class of morphisms π : X → Y that are proper, finitely presented, and flat, with
geometrically connected and geometrically reduced fibers, is “reasonable” (and
useful).

25.1.10. We next address the following question. Suppose π : X → Y is a morphism

of schemes. Given an invertible sheaf L on X, we ask when it is the pullback of
an invertible sheaf M on Y. For this to be true, we certainly need that L is trivial
on the fibers. We will see that if π is a proper O-connected morphism of locally
Noetherian schemes, then this often suffices. Given L , we recover M as π∗ L ; the
fibers of M are one-dimensional, and glue together to form a line bundle. We now
begin to make this precise.

25.1.J. E XERCISE . Suppose π : X → Y is a proper, O-connected morphism of locally

Noetherian schemes. Show that if M is any invertible sheaf on Y, then the natural
morphism M → π∗ π∗ M is an isomorphism. In particular, we can recover M from
π∗ M by applying the pushforward π∗ . (Your proof will work more generally when
M is locally free of finite rank.)
734 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

25.1.11. Proposition. — Suppose π : X → Y is a flat, proper morphism of locally

Noetherian schemes with geometrically connected and geometrically reduced fibers (hence
O-connected, by Exercise 25.1.I). Suppose also that Y is reduced, and L is an invertible
sheaf on X that is trivial on the fibers of π (i.e., L |Xq is a trivial invertible sheaf on Xq
for all q ∈ Y). Then π∗ L is an invertible sheaf on Y (call it M ), and the natural map
π∗ M → L is an isomorphism.

Proof. By Grauert’s Theorem 25.1.5, π∗ L is locally free of rank 1 (again, call it M ),

and M ⊗OY κ(q) → H0 (Xq , L |Xq ) is an isomorphism. We have a natural map of
invertible sheaves π∗ M = π∗ π∗ L → L . To show that it is an isomorphism, we
need only show that it is surjective. (Do you see why? If A is a ring, and ϕ : A → A
is a surjection of A-modules, why is ϕ an isomorphism?) For this, it suffices to
show that it is surjective on the fibers of π. (Do you see why? Hint: if the cokernel
of the map is not 0, then it is not 0 above some point of Y.) But this follows from the
first line of the proof (using for example that M = ∼ O in a neighborhood of q). □

Proposition 25.1.11 has some pleasant consequences. For example, if you have
two invertible sheaves A and B on X that are isomorphic on every fiber of π, then
they differ by a pullback of an invertible sheaf on Y: just apply Proposition 25.1.11
to A ⊗ B ∨ .

25.1.12. Projective bundles.

25.1.K. E XERCISE . Let X be a locally Noetherian scheme, and let pr1 : X × Pn → X

be the projection onto the first factor. Suppose L is a line bundle on X × Pn ,
whose degree on every fiber of pr1 is zero. Use the Cohomology and Base Change
Theorem 25.1.6 to show that (pr1 )∗ L is an invertible sheaf on X. Show that the
natural map pr∗1 ((pr1 )∗ L ) → L is an isomorphism, as at the end of proof of
Proposition 25.1.11.
Your argument will apply just as well to the situation where pr1 : X × Pn → X
is replaced by a Pn -bundle over X, pr1 : Z → X; or by pr1 : Z → X which is a proper
smooth morphism whose geometric fibers are integral curves of genus 0.
Furthermore, the locally Noetherian hypotheses can be removed, see §25.2.10.

25.1.L. E XERCISE . Suppose X is a connected Noetherian scheme. Show that

∼ Pic X × Z. Hint: the map Pic X × Pic Pn → Pic(X × Pn ) is given by
Pic(X × Pn ) =
(L , O(m)) 7→ pr∗1 L ⊗ pr∗2 O(m), where pr1 : X × Pn → X and pr2 : X × Pn → Pn
are the projections from X × Pn to its factors. (The notation ⊠ is often used for this
construction, see §15.2.10.)
A very similar argument will show that if Z is a Pn -bundle over X, then Pic Z =∼
Pic X × Z. You will undoubtedly also be able to figure out the right statement if X is
not connected.

25.1.13. Remark. As mentioned in §19.10.1, the Picard group of a scheme often

“wants to be a scheme”. You may be able to make this precise in the case of Pic Pn Z.
In this case, the scheme Pic PnZ is “Z copies of Spec Z”, with the “obvious” group
scheme structure. Can you figure out what functor it represents? Can you show
that it represents this functor? This will require extending Exercise 25.1.L out of the
Noetherian setting, using §25.2.10.
 © 2024 Ravi Vakil. Published by Princeton University Press. 735

25.1.M. E XERCISE . Suppose π : X → Y is a projective flat morphism over a

Noetherian integral scheme, all of whose geometric fibers are isomorphic to Pn
(over the appropriate field). Show that π is a projective bundle if and only if there
is an invertible sheaf L on X that restricts to O(1) on all the geometric fibers. (One
direction is clear: if it is a projective bundle, then it has a O(1) which comes from
the projectivization, see Exercise 17.2.D. In the other direction, the candidate vector
bundle is π∗ L . Show that it is indeed a locally free sheaf of the desired rank. Show
that its projectivization is indeed π : X → Y.)
Caution: the map π : Proj R[x, y, z]/(x2 + y2 + z2 ) → Spec R shows that not
every projective flat morphism over a Noetherian integral scheme, all of whose
geometric fibers are isomorphic to Pn , is necessarily a Pn -bundle. However, Tsen’s
Theorem implies that if the target is a smooth curve over an algebraically closed field, then
the morphism is a Pn -bundle (see [GS, Thm. 6.2.8]). Example 18.4.10 shows that
“curve” cannot be replaced by “5-fold” in this statement — the “universal smooth
plane conic” is not a P1 -bundle over the parameter space U ⊂ P5 of smooth plane
conics. If you wish, you can extend Example 18.4.10 to show that “curve” cannot
even be replaced by “surface”. (Just replace the P5 of all conics with a generally
chosen P2 of conics — but then figure out what goes wrong if you try to replace it
with a generally chosen P1 of conics.)

25.1.N. E XERCISE . Suppose π : X → Y is the projectivization of a finite rank locally

free sheaf F over a locally Noetherian scheme (i.e., X = ∼ Proj (Sym• F ∨ )) (§17.2.4).
∼ Proj (Sym• (F ∨ ⊗ L ))).
Recall from §17.2.4 that for any invertible sheaf L on Y, X =
Show that these are the only ways in which it is the projectivization of a vector
bundle. (Hint: recover F by pushing forward O(1).)

25.2 Proofs of cohomology and base change theorems

The key to proving the Semicontinuity Theorem 25.1.1, Grauert’s Theorem 25.1.5,
and the Cohomology and Base Change Theorem 25.1.6 is the following wonderful
idea of Mumford (see [Mu3, p. 47 Lem. 1]). It turns questions of pushforwards (and
how they behave under arbitrary base change) into something computable with
vector bundles (hence questions of linear algebra). After stating it, we will interpret
it.

25.2.1. Key Theorem. — Suppose π : X → Spec B is a proper morphism, and F is a

coherent sheaf on X, flat over Spec B, where B is Noetherian. Then there is a complex
δ−2 / K−1 δ−1 / K0 δ0 / K1 δ1 / ··· / Kn δn /0
(25.2.1.1) ···
of finitely generated free B-modules and an isomorphism of functors
∼
(25.2.1.2) Hp (X ×B A, F ⊗B A) ←→ Hp (K• ⊗B A)
for all p, for all ring maps B → A. (Here A needn’t be Noetherian.)
The complex (25.2.1.1) is sometimes called the Mumford complex.
Because (25.2.1.1) is a complex of free B-modules, all of the information is
contained in the maps, which are matrices with entries in B. This will turn questions
736 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

about cohomology (and base change) into questions about linear algebra. For
example, semicontinuity will turn into the fact that ranks of matrices (with functions
as entries) drop on closed subsets (§12.2.16(ii)).
Although the complex (25.2.1.1) is infinite, by (25.2.1.2) it has no cohomology
in negative degree, even after any ring extension B → A (as the left side of (25.2.1.2)
is 0 for p < 0).
The idea behind the proof is as follows: take the Čech complex, produce a
complex of finite rank free modules mapping to it “with the same cohomology” (a
quasiisomorphic complex, §18.2.3). We first construct the complex so that (25.2.1.2)
holds for B = A in the next lemma, and then show the same complex works for
general A, in Lemma 25.2.3 immediately thereafter.

25.2.2. Lemma. — Let B be a Noetherian ring. Suppose C• is a complex of B-modules

such that Hi (C• ) are finitely generated B-modules, and such that Cp = 0 for p > n.
Then there exists a complex K• of finite rank free B-modules such that Kp = 0 for p > n,
and a homomorphism of complexes α : K• → C• such that α induces isomorphisms
∼
Hi (K• ) −→ Hi (C• ) for all i.

Proof. We build this complex inductively. (This may remind you of §23.3.3.) Assume
we have defined (Kp , αp , δp ) for p ≥ m + 1 (as in (25.2.2.1)) such that the squares
commute, and the top row is a complex, and αq defines an isomorphism of coho-
mology Hq (K• ) → Hq (C• ) for q ≥ m + 2 and a surjection ker(δm+1 ) → Hm+1 (C• ),
and the Kp are finite rank free B-modules. (Our base case is m = p: take Kn = 0 for
n > p.)
δm+1 / Km+2 δm+2 / ···
(25.2.2.1) Km+1
αm+1 αm+2
 
··· / Cm−1 / Cm / Cm+1 / Cm+2 / ··· .
ϵm−1 ϵm ϵm+1

We construct (Km , δm , αm ). Choose generators of Hm (C• ), say c1 , . . . , cM . Let

 
m+1
m+1 m+1 α m+1 •
D := ker ker(δ ) −→ H (C ) .

Choose generators of Dm+1 , say d1 , . . . , dN . (This is where we use the Noetherian

hypotheses — to ensure this kernel Dm+1 is finitely generated.) Let Km = B⊕(M+N) .
Define δm : Km → Km+1 by sending the last N generators to d1 , . . . , dN , and the
first M generators to 0. Define αm by sending the first M generators of B⊕(M+N)
to (lifts of) c1 , . . . , cM , and sending the last N generators to arbitrarily chosen lifts
of the αm+1 (di ) (as the αm+1 (di ) are 0 in Hm+1 (C• ), and thus lie in the image of
ϵm ), so the square (with upper left corner Km ) commutes. Then by construction,
we have completed our inductive step:

δm / Km+1 δm+1 / Km+2 / ···

Km
αm αm+1 αm+2
  
··· / Cm−1 / Cm / Cm+1 / Cm+2 / ··· .
ϵm−1 ϵm ϵm+1
 © 2024 Ravi Vakil. Published by Princeton University Press. 737

25.2.3. Lemma. — Suppose α : K• → C• is a morphism of complexes of flat B-modules,

bounded on the right (i.e., Kn = Cn = 0 for n ≫ 0), inducing isomorphisms of cohomology
(a quasiisomorphism, §18.2.3). Then “this quasiisomorphism commutes with arbitrary
change of base ring”: for every B-algebra A, the maps Hp (K• ⊗B A) → Hp (C• ⊗B A) are
isomorphisms.

Proof. The mapping cone M• of α : K• → C• is exact by Exercise 1.6.E. Then M• ⊗B A

is still exact, by Exercise 24.2.F. But M• ⊗B A is the mapping cone of

α ⊗B A : K• ⊗B A / C• ⊗B A,

so by Exercise 1.6.E, α ⊗B A induces an isomorphism of cohomology (i.e., is a

quasiisomorphism) too. □
Proof of Key Theorem 25.2.1. Choose a finite affine covering of X. Take the Čech
complex C• for F with respect to this cover. By Grothendieck’s Coherence Theo-
rem 18.9.1 (which had Noetherian hypotheses), the cohomology of F is coherent.
(Theorem 18.9.1 required serious work. If you need Theorem 25.2.1 only in the pro-
jective case, the analogous statement with projective hypotheses, Theorem 18.7.1(d),
was much easier.) Apply Lemma 25.2.2 to get the nicer variant K• of the same
complex C• . By Lemma 25.2.3, if we tensor with A and take cohomology, we get
the same answer whether we use K• or C• . □
We now use Theorem 25.2.1 to prove some of the fundamental results stated
earlier: the Semicontinuity Theorem 25.1.1, Grauert’s Theorem 25.1.5, and the
Cohomology and Base Change Theorem 25.1.6. We will also (§25.2.5) give a new
proof of Theorem 24.7.1, that Euler characteristics are locally constant in flat families
(that applies more generally in proper situations).

25.2.4. Proof of the Semicontinuity Theorem 25.1.1. The result is local on Y, so we

may assume Y is affine. Let K• be a complex as in Key Theorem 25.2.1.
Then for q ∈ Y,
dimκ(q) Hp (Xq , F |Xq ) = dimκ(q) ker(δp ⊗B κ(q)) − dimκ(q) im(δp−1 ⊗B κ(q))
= dimκ(q) (Kp ⊗B κ(q)) − dimκ(q) im(δp ⊗B κ(q))
(25.2.4.1) − dimκ(q) im(δp−1 ⊗B κ(q))
Now dimκ(q) im(δp ⊗B κ(q)) is a lower semicontinuous function on Y. (Reason:
the locus where the dimension is less than some number N is obtained by setting
all N × N minors of the matrix Kp → Kp+1 to 0; cf. §12.2.16(ii)).) The same is true
for dimκ(q) im(δp−1 ⊗B κ(q)). The result follows. □

25.2.5. A new proof (and extension to the proper case) of Theorem 24.7.1 that
Euler characteristics of flat sheaves are locally constant.
If K• were finite “on the left” as well — if Kp = 0 for p ≪ 0 — then we
would have a short proof of Theorem 24.7.1. By taking alternating sums (over p) of
(25.2.4.1), we would have that
X X
χ(Xq , F |Xq ) = (−1)p hp (Xq , F |Xq ) = (−1)p rank Kp ,
738 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

which is locally constant. The only problem is that the sums are infinite. We
patch this problem by truncating the complex K• below where there is cohomology.
Define J• by Jp = Kp for p ≥ 0, Jp = 0 for p < −1, and J−1 := ker(K−1 → K0 ). Then
J• is a complex in the obvious way, and the map of complexes K• → C• clearly
factors through J• :

··· / J−3 / J−2 / J−1 / J0 / J1 / J2 / ···

··· /0 /0 / ker(K−1 → K0 ) / K0 / K1 / K2 / ···

∼ ∼
     
··· /0 /0 /0 / C0 / C1 / C2 / ···

Clearly J• → C• induces an isomorphism on cohomology (recall both have 0

cohomology for p < 0).
Now J−1 (the kernel of a map of coherent modules) is coherent. Consider the
mapping cone M• of β : J• → C• :

0 → J−1 → C−1 ⊕ J0 → C0 ⊕ J1 → · · · → Cn−1 ⊕ Jn → Cn → 0.

From Exercise 1.6.E, as J• → C• induces an isomorphism on cohomology, the map-

ping cone has no cohomology — it is exact. All terms in it are flat except possibly
J−1 (the Cp are flat by assumption, and Ji is free for i ̸= −1). Hence J−1 is flat too,
by Exercise 24.2.G. But flat coherent sheaves are locally free (Corollary 24.3.7). Then
Theorem 24.7.1 follows from
X X
χ(Xq , F |Xq ) = (−1)p hp (Xq , F |Xq ) = (−1)p rank Jp .

25.2.6. Proof of Grauert’s Theorem 25.1.5 and the Cohomology and Base Change
Theorem 25.1.6 (following E. Larson).
(⋆⋆ Experts: You might see in the proof that what makes pth cohomology
commute with base change is that the complex of Key Theorem 25.2.1 can be
“broken” into two complexes at the pth step. You might even want to interpret this
in terms of the Čech complex as an object of the derived category of B-modules.)
Thanks to Theorem 25.2.1, Theorems 25.1.5 and 25.1.6 are now statements about
a complex of free modules over a Noetherian ring.

25.2.7. Proof of Grauert’s Theorem 25.1.5. By hypothesis, hp (Xq , F |Xq ) is a

locally constant function of q ∈ Y. From (25.2.4.1), hp (Xq , F |Xq ) = rank Kp −
rank im(δp |q ) − rank im(δp−1 |q ). But rank Kp is constant, and rank im(δp |q ) and
rank im(δp−1 |q ) are lower semicontinuous, so in fact rank im(δp |q ) and rank im(δp−1 |q )
must be locally constant. By Exercise 14.3.L, both δp−1 and δp are maps of vector
bundles (in the sense of Definition 14.2.4). Then by Observation 14.2.5(ii) coker δp−1
and im δp are both locally free (of finite rank). In the short exact sequence

(25.2.7.1) 0 / Hp (K• ) / coker δp−1 / im δp /0

 © 2024 Ravi Vakil. Published by Princeton University Press. 739

(Exercise 1.5.6.4, the “dual” definition of cohomology), both coker δp−1 and im δp
correspond to finite rank locally free sheaves. Thus Hp (K• ) does as well, by Exer-
cise 14.2.H(a).

25.2.A. E XERCISE . Show (perhaps using (25.2.7.1)) that the construction of Hp (K• )
commutes with any base change, thereby completing the proof of Grauert’s Theo-
rem 25.1.5. □
In order to prove the Cohomology and Base Change Theorem 25.1.6, we need a
preliminary result.

25.2.8. Lemma. — Suppose

p−1
δK δp
Kp−1 / Kp K
/ Kp+1

  
δJp−1 δp
Jp−1 / Jp J
/ Jp+1

is a map of complexes, with the left vertical arrow surjective. Then Hp (K• ) → Hp (J• ) is
surjective if and only if ker δp p
K → ker δJ is surjective.

Proof. The map im δp−1 K → im δp−1

J is surjective: any element α of im δp−1
J lifts to
p−1 p−1 p
J , then can lift to K , which then can map to K , which maps to α. Then apply
the Snake Lemma 1.6.5 to

0 / im δp−1 / ker δp / Hp (K• ) /0

K K

  
0 / im δp−1 / ker δp / Hp (J• ) / 0.
J J

25.2.9. Proof of the Cohomology and Base Change Theorem 25.1.6. We focus on
the complex near the pth step, and its restriction to the point q ∈ X:

Kp−1 / Kp δp / Kp+1 free B-modules

  p 
Kp−1 ⊗B κ(q) / Kp ⊗B κ(q) δ ⊗κ(q)
/ Kp+1 ⊗B κ(q)

• By hypothesis, ϕp q : (R π∗ F )|q → H (Xq , F |Xq ) is surjective.

p p

• By Lemma 25.2.8, this is equivalent to (ker δp ) ⊗ κ(q) → ker(δp ⊗ κ(q))

being surjective.
• By Exercise 14.2.K, this is equivalent to δp being a map of vector bundles
near q.
• This implies (by Observation 14.2.5(ii)) that ker δp is finite rank locally
free, and the construction of ker δp commutes with any base change in a
neighborhood of q.
740 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Now Hp (K• ) = coker(Kp−1 → ker δp ), i.e., we have the exact sequence

Kp−1 / ker δp / Hp (K• ) / 0.

Thus Hp (K• ) commutes with any base change (as tensor product is right exact).
This completes the proof of part (i) of the Theorem.
For part (ii), consider again the map Kp−1 → ker δp of finite rank locally free
sheaves, whose cokernel is Hp (K• ). Now Hp (K• ) is locally free if and only if
Kp−1 → ker δp is a map of vector bundles, if and only if (since δp is a map of vector
bundles) δp−1 is a map of vector bundles, if and only if Hp−1 (K• ) → Hp−1 (K• |q ) is
surjective. This completes the proof of part (ii). □

25.2.10. ⋆ Removing Noetherian conditions.

It can be helpful to have versions of the theorems of §25.1 without Noetherian
conditions; important examples come from moduli theory, and will be discussed in
the next section. Noetherian conditions can often be exchanged for finite presenta-
tion conditions. We begin with an extension of Exercise 10.3.H.

25.2.B. E XERCISE . Suppose π : X → Spec B is a finitely presented morphism, and

F is a finitely presented sheaf on X. Show that there exists a base change diagram
of the form
(25.2.10.1) F F′

X
σ / X′

π π′
 
Spec B
ρ
/ Spec Z[x1 , . . . , xN ]/I

where N is some integer, I ⊂ Z[x1 , . . . , xN ], and π ′ is finitely presented (= finite

type as the target is Noetherian, see §8.3.13), and a finitely presented (= coherent)
quasicoherent sheaf F ′ on X ′ with F = ∼ σ∗ F ′ .

25.2.11. Properties of π ′ . (The ideal I appears in the statement of Exercise 25.2.B

not because it is needed there, but to make the statement of this remark correct.)
If π is proper, then diagram (25.2.10.1) can be constructed so that π ′ is also proper
(using [Gr-EGA, IV3 .8.10.5]). Furthermore, if F is flat over Spec B, then (25.2.10.1)
can be constructed so that F ′ is flat over Spec Z[x1 , . . . , xN ]/I (using [Gr-EGA,
IV3 .11.2.6]). This requires significantly more work.

25.2.C. E XERCISE . Assuming the results stated in §25.2.11, prove the following
results, with the “locally Noetherian” hypotheses removed, and “finite presentation”
hypotheses added:
(a) the constancy of Euler characteristic in flat families (Theorem 24.7.1, ex-
tended to the proper case as in §25.2.5);
(b) the Semicontinuity Theorem 25.1.1;
(c) Grauert’s Theorem 25.1.5 (you will have to show that Z[x1 , . . . , xN ]/I in
(25.2.10.1) can be taken to be reduced); and
(d) the Cohomology and Base Change Theorem 25.1.6.
 © 2024 Ravi Vakil. Published by Princeton University Press. 741

25.2.12. Necessity of finite presentation conditions. The finite presentation conditions

are necessary. There is a projective flat morphism to a connected target where the
fiber dimension jumps. There is a finite flat morphism where the degree of the fiber
is not locally constant. There is a projective flat morphism to a connected target
where the fibers are curves, and the arithmetic genus is not constant. See [Stacks,
tag 05LB] for the first example; the other two use the same idea.

25.3 Applying cohomology and base change to moduli problems

The theory of moduli relies on ideas of cohomology and base change. We

explore this by examining two special cases of one of the primordial moduli spaces,
the Hilbert scheme: the Grassmannian, and the fact that degree d hypersurfaces in
projective space are “parametrized” by another projective space (corresponding to
degree d polynomials, see Remark 4.5.3).
As suggested in §24.0.1, the Hilbert functor HilbY Pn of Pn Y parametrizes flat
finitely presented closed subschemes of Pn Y , where Y is an arbitrary scheme. More
precisely, it is a contravariant functor sending the Y-scheme X to the set of finitely
presented closed subschemes of X ×Y Pn n
Y = PX , flat over X (and sending morphisms
X1 → X2 to pullbacks of flat families). An early achievement of Grothendieck was
the construction of the Hilbert scheme, which can then be cleverly used to construct
many other moduli spaces.

25.3.1. Theorem (Grothendieck). — HilbZ Pn is representable by a scheme locally of

finite type.
(Grothendieck’s original argument is in [Gr4]. A readable construction is given
in [Mu2], and in [FGIKNV, Ch. 5].)
We won’t use the following exercise, but it is conceptually very important, and
explains why “moduli spaces tautologically always have universal families” (with
the “correct” definition of “moduli space”). The proof will remind you of Yoneda’s
Lemma. (You have already seen universal families of moduli spaces under other
names, such as the universal hyperplane, Definition 13.4.1, or tautological bundles,
§16.4.1.)

25.3.A. E XERCISE ( THE UNIVERSAL FAMILY OVER THE H ILBERT SCHEME ). Assume
Theorem 25.3.1. Show that there exists a closed subscheme Z of HilbZ Pn × Pn
Z

 closed / HilbZ Pn × Pn
Z Z

flat

HilbZ Pn

(which we call the universal family over the Hilbert scheme) such that a flat family
 closed /
Y Pn
B

flat

B
742 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

corresponds to a map π : B → HilbZ Pn if and only if Y is the pullback of Z by π,

i.e., diagram

 closed /
Y Pn
B

flat
  *
(
Z / HilbZ Pn × Pn
closed
B Z

flat
( 
HilbZ Pn

is a Cartesian diagram. Hint: The Hilbert functor applied to the Hilbert scheme by
definition corresponds to the set Maps(HilbZ Pn , HilbZ Pn ). The universal family
will correspond to the identity map.

25.3.B. E ASY E XERCISE . Assuming Theorem 25.3.1, show that HilbY Pn is repre-
sentable, by showing that it is represented by HilbZ Pn ×Z Y. Thus the general case
follows from the “universal” case of Y = Z.

25.3.C. E XERCISE . Assuming Theorem 25.3.1, show that HilbZ Pn is the disjoint
p(m) n
union of schemes HilbZ P , each one corresponding to finitely presented closed
subschemes of PnZ whose fibers have fixed Hilbert polynomial p(m). Hint: Corol-
lary 24.7.2.
Motivated by Exercise 25.3.C, what Grothendieck actually proved was the
following.
p(m)
25.3.2. Theorem (Grothendieck). — Each HilbZ Pn is projective over Z.

25.3.D. E XERCISE . Prove that Theorem 25.3.2 implies Theorem 25.3.1.

In order to get some feeling for the Hilbert scheme, we discuss two important
examples, without relying on Theorem 25.3.1.

25.3.3. The Grassmannian.

We have defined the Grassmannian G(k, n) twice before, in §7.7 and §16.4. The
second time involved showing the representability of a (contravariant) functor
(from Sheaves to Sets), of rank k locally free quotient sheaves of a rank n free sheaf.
We now consider a parameter space for a more geometric problem. The space
will again be G(k, n), but because we won’t immediately know this, we invent
some temporary notation. Let G ′ (k, n) be the contravariant functor (from Schemes
to Sets) which assigns to a scheme B the set of finitely presented closed subschemes
of Pn−1
B , flat over B, whose fiber over any point b ∈ B is a (linearly embedded)
Pκ(b) in Pn−1
k−1
κ(b) :


X / Pn−1
cl. subscheme
(25.3.3.1) B

flat, f. pr.
π
 w
B
 © 2024 Ravi Vakil. Published by Princeton University Press. 743

(This describes the map to Sets; you should think through how pullback makes this
into a contravariant functor.)

25.3.4. Theorem. — The functor G ′ (k, n) is represented by G(k, n).

Translation: there is a natural (i.e., functorial) bijection between diagrams of
the form (25.3.3.1) (where the fibers are Pk−1 ’s) and diagrams of the form (16.4.0.1)
(the diagrams that G(k, n) parametrizes, or represents).

Proof. One direction is notably easier. Suppose we are given a diagram of the form
(16.4.0.1) over a scheme B,

(25.3.4.1) OB⊕n / / Q,

where Q is locally free of rank k. Applying Proj B to the Sym• construction on both
OB⊕n and Q, we obtain a closed embedding
 / Proj Sym• O ⊕n
(25.3.4.2) Proj B (Sym• Q)


B B = Pn−1 × B

& x
B
(as, for example, in Exercise 17.2.H).
The fibers are linearly embedded Pk−1 ’s (as base change, in this case to a point
of B, commutes with the Proj construction, Exercise 17.2.E). Note that Proj (Sym• Q)
is flat and finitely presented over B, as it is a projective bundle. We have constructed
a diagram of the form (25.3.3.1).
We now need to reverse this. The trick is to produce (25.3.4.1) from our geo-
metric situation (25.3.3.1), and this is where cohomology and base change will be
used.
Given a diagram of the form (25.3.3.1) (where the fibers are Pk−1 ’s), consider
the closed subscheme exact sequence for X:

0 / IX /On−1
/ OX / 0.
PB

Tensor this with OPn−1 (1):

B

(25.3.4.3) 0 / IX (1) / O n−1 (1) / OX (1) / 0.

PB

Note that OX (1) restricted to each fiber of π is O(1) on Pk−1 (over the residue
field), for which all higher cohomology vanishes (§18.3).

25.3.E. E XERCISE . Show that Ri π∗ OX (1) = 0 for i > 0, and π∗ OX (1) is locally free
of rank k. Hint: use the Cohomology and Base Change Theorem 25.1.6. Either use
the non-Noetherian discussion of §25.2.10 (which we haven’t proved), or else just
assume B is locally Noetherian.

25.3.F. E XERCISE . Show that the long exact sequence obtained by applying π∗ to
(25.3.4.3) is just a short exact sequence of locally free sheaves

0 / π∗ IX (1) / π∗ O n−1 (1) / π∗ OX (1) / 0.

P B
744 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

of ranks n−k, n, and k respectively, where the middle term is canonically identified
with OB⊕n .

The surjection OB⊕n / / π∗ OX (1) is precisely a diagram of the sort we

wished to construct, (16.4.0.1).

25.3.G. E XERCISE . Close the loop, by using these two “inverse” constructions to
show that G(k, n) represents the functor G ′ (k, n). □

25.3.5. Hypersurfaces.
Ages ago (in Remark 4.5.3), we informally said that hypersurfaces of degree d
n+d
in Pn are parametrized by a P( d )−1 . We now make this precise. We work over
a base Z for suitable generality. You are welcome to replace Z by a field of your
choice, but by the same argument as in Easy Exercise 25.3.B, all other cases are
obtained from this one by base change.
Define the contravariant functor Hd,n : Sch → Sets from schemes to sets as
follows. To a scheme B, we associated the set of all closed subschemes X ,→ Pn B,
flat and finitely presented over B, all of whose fibers are degree d hypersurfaces in
Pn (over the appropriate residue field). To a morphism B1 → B2 , we obtain a map
Hd,n (B2 ) → Hd,n (B1 ) by pullback.
n+d
25.3.6. Proposition. — The functor Hd,n is represented by P( d )−1 .

As with the case of the Grassmannian, one direction is easy, and the other
requires cohomology and base change.
n+d
25.3.H. E ASY E XERCISE . Over P( d )−1 , described a closed subscheme X ,→
n+d
Pn × P( d )−1 that will be the universal hypersurface (the universal family over
n+d
the moduli space). Show that X is flat and finitely presented over P( d )−1 . (For
flatness, you can use the local criterion of flatness on the source, Theorem 24.6.9,
but it is possible to deal with it easily by working by hand.)
n+d
Thus given any morphism B → P( d )−1 , by pullback, we have a degree d
hypersurface X over B (an element of Hd,n (B)).
Our goal is to reverse this process: from a degree d hypersurface π : X → B
n+d
over B (an element of Hd,n (B)), we want to describe a morphism B → P( d )−1 .
Consider the closed subscheme exact sequence for X ,→ PnB , twisted by OPB (d):
n

(25.3.6.1) 0 / IX (d) / OPn (d) / OX (d) / 0.

B

25.3.I. E XERCISE ( CF. E XERCISE 25.3.E). Show that the higher pushforwards (by π)
of each term of (25.3.6.1) is 0, and that the long exact sequence of pushforwards of
(25.3.6.1) is

0 / π∗ IX (d) / π∗ OPn (d) / π∗ OX (d) / 0.

B

where the middle term is free of rank n+d

d (whose summands can be identified
with degree d monomials in the projective variables x1 , . . . , xn (see Exercise 9.3.J),
and the left term π∗ IX (d) is locally free of rank 1 (basically, a line bundle).
 © 2024 Ravi Vakil. Published by Princeton University Press. 745

⊕(n+d)
(It is helpful to interpret the middle term OB d as parametrizing homoge-
neous degree d polynomials in n + 1 variables, and the rank 1 subsheaf of π∗ IX (d)
as “the equation of X”. This will motivate what comes next.)
⊕(n+d)
Taking the dual of the injection π∗ IX (d) ,→ OB d , we have a surjection
⊕(n+d
d )
OB //L

from a free sheaf onto an invertible sheaf L = (π∗ IX (d))∨ , which (by the functorial
n+d
description of projective space, Theorem 15.2.2) yields a morphism B → P( d )−1 .

25.3.J. E XERCISE . Close the loop: show that these two constructions are inverses,
thereby proving Proposition 25.3.6.

25.3.7. Remark. The proof of the representability of the Hilbert scheme shares a
number of features of our arguments about the Grassmannian and the parameter
space of hypersurfaces.
 CHAPTER 26

Depth and Cohen-Macaulayness

We now introduce the notion of depth. Depth is an algebraic rather than geo-
metric concept, so we concentrate on developing some geometric sense of what it
means. Most important is the geometric bound on depth by dimension of (closures
of) associated points (Theorem 26.1.3). A central tool to understanding depth is the
Koszul complex (briefly discussed in §24.4.1), but we avoid this approach, as we
can prove what we need directly.
When the depth of a local ring equals its dimension, the ring is said to be
Cohen-Macaulay, and this is an important way in which schemes can be “nice”.
For example, regular local rings are Cohen-Macaulay (§26.2.5), as are regular
embeddings in smooth varieties (Proposition 26.2.6). Cohen-Macaulayness will
be key to the proof of Serre duality in Chapter 29, through the Miracle Flatness
Theorem 26.2.11.
Another application of depth is Serre’s R1 + S2 criterion for normality (Theo-
rem 26.3.2), which we will use to prove that regular schemes are normal (§26.3.5)
without having to show that regular local rings are factorial (the Auslander-
Buchsbaum Theorem, Fact 13.8.5), and to prove that regular embeddings in smooth
schemes are normal if they are regular in codimension 1 (§26.3.3).

26.1 Depth

26.1.1. Informal introduction. Given a point p on a scheme X, sometimes you can

locally “slice” through the point — there is an effective Cartier divisor through p (in
a neighborhood of p, say). You can do this if and only if the point is not an associated
point of X. Sometimes you can slice again, if the first slice still doesn’t have p as an
associated point. The depth of (X at) the point p is how deep you can slice — how
many slices you can make before you are blocked by the point becoming associated.
Obviously you can’t make more than dim X slices, since at each stage you lose a
dimension, and after dim X slices, you will be left (set-theoretically) with just the
point p (which is then an associated point). Thus the depth is bounded by the
dimension. In figuring out the depth, you might think you would have to slice with
care in order to maximize the number of slices you can take. You needn’t worry;
you can slice haphazardly, and will always end at the same step.

This material is published by Princeton University Press as The Rising Sea: Foundations of Algebraic
Geometry by Ravi Vakil. This version is free to view and download for personal use only. Not for
redistribution, resale, or use in derivative works. © 2024 Ravi Vakil.

747
748 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

26.1.2. Definition. Suppose (A, m) is a Noetherian local ring, and M is a finitely

generated A-module. The depth of M (denoted depth M) is the length of the
longest M-regular sequence with elements in m. (The theory of regular sequences
was discussed in §9.5.3.)
(More generally, if R is a Noetherian ring, I ⊂ R is an ideal, and M is a finitely
generated R-module, then the I-depth of M, denoted depthI M, is the length of the
longest M-regular sequence with elements in I. We won’t need this.)

26.1.A. I MPORTANT E XERCISE . Suppose M is a nonzero finitely generated module

over a Noetherian local ring (A, m). Show that depth M = 0 if and only if every
element of m is a zerodivisor of M if and only if m is an associated prime of M.

26.1.B. E XERCISE . Suppose M is a finitely generated module over a Noetherian

local ring A. Show that depth M ≤ dim Supp M. In particular,
(26.1.2.1) depth A ≤ dim A.
(We will improve this result in Theorem 26.1.7.)
At this point, it is hard to determine the depth of an A-module M. You can
start trying to build an M-regular sequence by successively choosing x1 , x2 , . . . ,
but how do you know you have made the right choices to find the longest one? The
happy answer is that you can’t go wrong; this is the content of the next result. (We
then describe how to find the depth of a module M in practice, in §26.1.5.)

26.1.3. Theorem. — Suppose M is a finitely generated module over a Noetherian local

ring (A, m). Then all maximal M-regular sequences contained in m (M-regular sequences
x1 , . . . , xn in m that cannot be extended to x1 , . . . , xn+1 ) have the same length. Thus the
depth of M is the length of any maximal M-regular sequence.
We prove Theorem 26.1.3 by giving a cohomological criterion, Theorem 26.1.4,
for a regular sequence to be maximal. (You will then prove Theorem 26.1.3 in
Exercise 26.1.D.) We will also use this criterion to give a better bound on the depth
in Theorem 26.1.7. Theorem 26.1.3 is the key technical result of this Chapter. An
important moral is that depth should be understood as a “cohomological” property.

26.1.4. Theorem (cohomological criterion for existence of regular sequences). —

Suppose (A, m) is a Noetherian local ring, and M is a nonzero finitely generated A-module.
The following are equivalent.
(i) For every finitely generated A-module N with Supp N = {[m]} (i.e., N has finite
length, Exercise 6.6.X(a)), ExtiA (N, M) = 0 for all i < n.
(ii) ExtiA (A/m, M) = 0 for all i < n.
(iii) There exists an M-regular sequence in m of length n.
This result can be extended in various ways, see for example [Mat1, Thm. 28].

Proof. Clearly (i) implies (ii).

26.1.C. E XERCISE . Prove that (ii) implies (i). Hint: Use induction on the length of N,
and the long exact sequence for ExtiA (·, M).
Proof that (iii) implies (ii). The case n = 0 is vacuous. We inductively prove the result
for all n. Suppose (iii) is satisfied, where n ≥ 1, and assume that we know (ii) for
 © 2024 Ravi Vakil. Published by Princeton University Press. 749

“all smaller n”. Choose a regular sequence x1 , . . . , xn of length n. Then x1 is a

non-zerodivisor on M, so we have an exact sequence

(26.1.4.1) 0 /M ×x1
/M / M/x1 M / 0.

Then M/x1 M has a regular sequence x2 , . . . , xn−1 of length n − 1, so by the

inductive hypothesis, ExtiA (A/m, M/x1 M) = 0 for i < n − 1. Taking the Ext long
exact sequence for ExtiA (A/m, ·) for (26.1.4.1), we find that

ExtiA (A/m, M)
×x1
/ ExtiA (A/m, M)

is an injection for i < n. Now ExtiA (A/m, M) can be computed by taking an

injective resolution of M, and applying HomA (A/m, ·). Hence as x1 lies in m (and
thus annihilates A/m), multiplication by x1 is the zero map. Thus (ii) holds for n as
well.
Proof that (ii) implies (iii). The case n = 0 is vacuous.
We deal next with the case n = 1, by showing the contrapositive. Assume
that there are no non-zerodivisors in m on M, so by Exercise 26.1.A, m is an as-
sociated prime of M. Thus (§6.6.7) we have an injection A/m ,→ M, yielding
HomA (A/m, M) ̸= 0 as desired.
We now inductively prove the result for all n > 1. Suppose (ii) is satisfied,
where n ≥ 2, and assume that we know (iii) for “all smaller n” . Then by the case
n = 1, there exists a non-zerodivisor x1 on M, so we have a short exact sequence
(26.1.4.1). A portion of the Ext long exact sequence for ExtiA (A/m, ·) for (26.1.4.1) is

ExtiA (A/m, M) / ExtiA (A/m, M/x1 M) / Exti+1 (A/m, M).

A

By assumption, both ExtiA (A/m, M) and Exti+1 A (A/m, M) are 0 for i < n − 1, so
ExtiA (A/m, M/x1 M) = 0 for i < n − 1, so by the inductive hypothesis, we have an
(M/x1 M)-regular sequence x2 , . . . , xn of length n − 1 in m. Adding x1 to the front
of this sequence, we are done. □

26.1.D. E XERCISE . Prove Theorem 26.1.3. Hint: by Theorem 26.1.4 (notably, the
equivalence of (ii) and (iii)), you have control of how long an M-regular sequence
in m can be. Use the criterion, and the long exact sequence used in the proof of
Theorem 26.1.4, to show that any M-regular sequence in m can be extended to this
length.

26.1.E. E XERCISE . Suppose M is a finitely generated module over a Noether-

ian local ring (A, m). If x is a non-zerodivisor in m, show that depth(M/xM) =
depth M − 1. Hint: Theorem 26.1.3.

26.1.5. Finding the depth of a module. We can now compute the depth of M by
successively finding non-zerodivisors, as follows. Is there a non-zerodivisor x on
M in m?
(a) If not, then depth M = 0 (Exercise 26.1.A).
(b) If so, then choose any such x, and (using the previous exercise) repeat the
process with M/xM.
The process must terminate by Exercise 26.1.B.
 750 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

26.1.F. I MPORTANT E XERCISE . Suppose (A, m) is a dimension d regular local ring.

Show that depth A = d. Hint: Exercise 13.2.B.

26.1.G. E XERCISE ( CF. E XERCISE 24.3.I). Suppose X = Spec R, where

R = k[w, x, y, z]/(wy, wz, xy, xz),
the union of two coordinate two-planes in A4k meeting at the origin. Show that
the depth of the local ring of X at the origin is 1. Hint: Show that w − y is not a
zerodivisor, and that R/(w − y) has an embedded point at the origin.

26.1.6. Depth is bounded by the dimension of associated prime ideals.

Theorem 26.1.4 can be used to give an important improvement of the bound
(26.1.2.1) on depth by the dimension:

26.1.7. Theorem. — The depth of a module M is at most the smallest dim A/p as p runs
over the associated prime ideals of M.
(The example of two planes meeting at a point in Exercise 26.1.G shows that this
bound is not sharp.) The key step in the proof of Theorem 26.1.7 is the following
result of Ischebeck.

26.1.8. Lemma. — Suppose (A, m) is a Noetherian local ring, and M and N are nonzero
finitely generated A-modules. Then ExtiA (N, M) = 0 for i < depth M − dim Supp N.

Proof. Consider the following statements.

(generalr ) Lemma 26.1.8 holds for dim Supp N ≤ r.
(primer ) Lemma 26.1.8 holds for dim Supp N ≤ r and N = A/p for some prime p.
Note that (general0 ) (and hence (prime0 )) is true, as in this case Supp N = {[m]},
and the result follows from Theorem 26.1.4 (from “(iii) implies (i)”).

26.1.H. E XERCISE . Show that (primer ) implies (generalr ). Hint: For a short exact se-
quence 0 → M ′ → M → M ′′ → 0, Supp M = Supp M ′ ∪ Supp M ′′ (Exercise 6.6.B).

We conclude the proof by showing that (generalr−1 ) implies (primer ) for r ≥ 1.

Fix a prime p ⊂ A with dim A/p = r. Since dim A/p > 0, we can choose x ∈ m \ p.
Consider the exact sequence

(26.1.8.1) 0 / A/p ×x
/ A/p / A/(p + (x)) / 0,

noting that dim Supp A/(p+(x)) ≤ r−1 (Do you see why?). Then the Ext long exact
sequence obtained by applying HomA (·, M) to (26.1.8.1), along with the vanishing
of ExtiA (A/(p + (x))), M) for i < depth M − r + 1 (by (generalr−1 )), implies that

ExtiA (A/p, M)
×x
/ ExtiA (A/p, M)

is an isomorphism for i < depth M − r. But ExtiA (A/p, M) is a finitely generated

A-module (Exercise 23.2.F), so by Nakayama’s Lemma 8.2.8, ExtiA (A/p, M) = 0 for
i < depth M − r. □

26.1.I. E ASY E XERCISE . Prove Theorem 26.1.7. Hint: if p ∈ Ass(M), then

HomA (A/p, M) ̸= 0 (§6.6.7).
 © 2024 Ravi Vakil. Published by Princeton University Press. 751

26.2 Cohen-Macaulay rings and schemes

26.2.1. Definition. A Noetherian local ring (A, m) is Cohen-Macaulay if depth A =

dim A, i.e., if equality holds in (26.1.2.1). (One may define Cohen-Macaulay mod-
ule similarly, but we won’t need this concept.) A locally Noetherian scheme is
Cohen-Macaulay if all of its local rings are Cohen-Macaulay. We won’t use this
language, but a ring A is a Cohen-Macaulay ring of all of its localizations Ap at
all primes p are Cohen-Macaulay, or equivalently, if Spec A is Cohen-Macaulay.
(Fact 26.2.14 below, that Cohen-Maculayness of a local ring is preserved by lo-
calization, ensures that there is no contradiction between this definition, and the
definition of a Cohen-Macaulay local ring.)

26.2.A. E XERCISE . Show that every locally Noetherian scheme of dimension 0

is Cohen-Macaulay. Show that a locally Noetherian scheme of dimension 1 is
Cohen-Macaulay if and only if it has no embedded points.

26.2.2. (Counter)example. Let X be the example of Exercise 26.1.G — two planes

meeting at a point. By Exercise 26.1.G, X is not Cohen-Macaulay.

26.2.B. E XERCISE . Suppose A is Cohen-Macaulay Noetherian local ring. Use

Theorem 26.1.7 to show that Spec A is pure dimensional, and has no embedded
points. (It is not true that Noetherian local rings of pure dimension having no
embedded prime ideals are Cohen-Macaulay, see Example 26.2.2.)

26.2.3. Theorem (slicing criterion for Cohen-Macaulayness). — Suppose (A, m) is

a Noetherian local ring, and x ∈ m is a non-zerodivisor. Then (A, m) is Cohen-Macaulay if
and only if (A/x, m) is Cohen-Macaulay.
Compare this to the slicing criteria for regularity and flatness (Exercise 13.2.C
and Theorem 24.6.6 respectively).

26.2.4. Geometric interpretation of the slicing criterion. Suppose X is a locally Noether-

ian scheme, and D is an effective Cartier divisor. If X is Cohen-Macaulay, then so is
D. If D is Cohen-Macaulay, then X is Cohen-Macaulay at the points of D.

26.2.C. E XERCISE . Prove Theorem 26.2.3, using Theorem 26.1.3, the fact that maxi-
mal regular sequences (in m) all have the same length. (Hint: Recall Exercise 12.3.I,
that dim A/(f) ≥ dim A − 1 for any f ∈ A.)

26.2.D. E XERCISE . Show that if (A, m) is Cohen-Macaulay, then a set of elements

x1 , . . . , xr ∈ m is a regular sequence (for A) if and only if dim A/(x1 , . . . , xr ) =
dim A − r.

26.2.5. By Exercise 26.1.F, regular local rings (Noetherian by definition) are Cohen-
Macaulay. In particular, as smooth schemes over a field k are regular (Theo-
rem 13.2.7(b), proved in §24.8.9), we see that smooth k-schemes are Cohen-Macaulay.
(In combination with Exercise 26.2.B, this explains why effective Cartier divisors
on smooth varieties have no embedded points, justifying a comment made in
Aside 20.2.1.) Combining this with Exercise 26.2.D or §26.2.4, we have the follow-
ing.
752 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

26.2.6. Proposition. — Regular embeddings in smooth k-schemes are Cohen-Macaulay.

26.2.7. As a consequence of Proposition 26.2.6 and the fact that Cohen-Macaulay

schemes have no embedded points (Exercise 26.2.B), we see that regular embed-
dings in smooth k-schemes (in An n
k or Pk for example) have no embedded points,
generalizing Exercise 6.6.F (the hypersurface in An
k case). This is not clear without
the theory of Cohen-Macaulayness!

26.2.8. Alternate definition of Cohen-Macaulayness. The slicing criterion (Theo-

rem 26.2.3) gives an enlightening alternative inductive definition of Cohen-Macaulayness
in terms of effective Cartier divisors, in the spirit of the method of §26.1.5 for com-
puting depth. Suppose as before that (A, m) is a Noetherian local ring.

(i) If dim A = 0, then A is Cohen-Macaulay (Exercise 26.2.A).

(iia) If dim A > 0, and every element of m is a zerodivisor, then A is not Cohen-
Macaulay (by Exercise 26.1.A).
(iib) Otherwise, choose any non-zerodivisor x in m. Then A is Cohen-Macaulay
if and only if A/(x) (necessarily of dimension dim A−1 by Krull’s Principal
Ideal Theorem 12.3.3) is Cohen-Macaulay.

The following example could have been stated (but not proved) before we
knew any algebraic geometry at all. (We work over C rather than over an arbitrary
field only to ensure that the statement requires as little background as possible.)

26.2.E. F UN E XERCISE (M AX N OETHER ’ S AF + BG T HEOREM ). Suppose f, g ∈

C[x0 , x1 , x2 ] are two homogeneous polynomials, cutting out two curves in P2C
that meet “transversely”, i.e., at a finite number of reduced points. Suppose h ∈
C[x0 , x1 , x2 ] is a homogeneous polynomial vanishing at these points. Show that
h ∈ (f, g). Hint: show that the intersection of the affine cones V(f) and V(g) in A3
has no embedded points. (This problem is quite nontrivial to do without the theory
developed in this chapter! As a sign that this is subtle: you can easily construct
three quadratics e, f, g ∈ C[x0 , x1 , x2 ] cutting out precisely the two points [1, 0, 0]
and [0, 1, 0] ∈ P2C , yet the line z = 0 is not in the ideal (e, f, g) for degree reasons.)
As additional motivation for Cohen-Macaulayness, the next exercise is seem-
ingly basic, but we need Cohen-Macaulayness to prove it cleanly.

26.2.F. E XERCISE . Suppose H1 , . . . , Hn are n hypersurfaces in Pn C of degrees d1 ,

. . . , dn respectively, and suppose their scheme-theoretic intersection Z is a finite
number of points. Show that the sum (over the points of Z) of the lengths of Z at the
points is d1 · · · dn . Equivalently: show that the length of Z is d1 · · · dn . Informally:
The hypersurfaces intersect at d1 · · · dn points, “counted correctly”. Hint: Bézout’s
Theorem, in the form of Exercise 18.6.J.

26.2.9. Miracle flatness.

We conclude with a remarkably simple and useful criterion for flatness, which
we shall use in the proof of Serre duality. The main content is the following algebraic
result.

26.2.10. Miracle Flatness Theorem (algebraic version). — Suppose ϕ : (B, n) →

(A, m) is a (local) homomorphism of Noetherian local rings, such that A is Cohen-Macaulay,
 © 2024 Ravi Vakil. Published by Princeton University Press. 753

and B is regular, and A/nA = A ⊗B (B/n) (the ring corresponding to the fiber) has pure
dimension dim A − dim B. Then ϕ is flat.

Proof. We prove Theorem 26.2.10 by induction on dim B. If dim B = 0, then B is a

field (Exercise 13.2.A), so the result is immediate, as everything is flat over a field
(Exercise 24.1.A(a)). Assume next that dim B > 0, and we have proved the result
for all “B of smaller dimension”. Choose x ∈ n \ n2 , so B/x is a regular local ring of
dimension dim B − 1 (Exercise 12.3.I). Then
dim A/xA ≤ dim B/(x) + dim A/nA (Key Exercise 12.4.A)
= dim B − 1 + dim A/nA (Exercise 12.3.I)
= dim A − 1 (by hypothesis of Theorem 26.2.10).
By (Exercise 12.3.I), dim A/xA ≥ dim A − 1, so we have dim A/xA = dim B/(x) +
dim A/nA = dim A − 1. By Exercise 26.2.D, A/xA is a Cohen-Macaulay ring, and x
is a non-zerodivisor on A. The inductive hypothesis then applies to (B/(x), n) →
(A/xA, m), so A/xA is flat over B/(x). Then by the local slicing criterion for flatness
(on the target, Theorem 24.6.6), B → A is flat, as desired. □

26.2.11. Miracle Flatness Theorem. — Suppose π : X → Y is a morphism of pure

dimensional finite type k-schemes, where X is Cohen-Macaulay, Y is regular, and the fibers
of π have dimension dim X − dim Y. Then π is flat.

26.2.G. E XERCISE . Prove the Miracle Flatness Theorem 26.2.11. (Do not forget that
schemes usually have non-closed points!)
The geometric situation in the Miracle Flatness Theorem 26.2.11 is part of the
following pretty package.

26.2.H. E XERCISE ( THE M IRACLE F LATNESS PACKAGE ). Suppose π : X → Y is a

map of locally finite type k-schemes, where both X and Y are pure dimensional,
and Y is regular. Show that if any two of the following hold, then the third does as
well:
(i) π is flat of relative dimension dim X − dim Y.
(ii) X is Cohen-Macaulay.
(iii) Every fiber Xy is Cohen-Macaulay of pure dimension dim X − dim Y.
Hint: if ϕ : B → A is a flat ring map, then ϕ sends non-zerodivisors to non-
zerodivisors (Observation 24.1.2).
The statement of Exercise 26.2.H can be improved, at the expense of killing the
symmetry. In the implication (ii) and (iii) imply (i), the Cohen-Macaulay hypotheses
in (iii) are not needed.

26.2.12. Example. As an example of Exercise 26.2.H in action, we consider the

example of two planes meeting a point, continuing the notation of Exercise 26.1.G.
Consider the morphism π : X → Y := A2 given by (w − y, x − z). Note that Y is
regular (§13.3.6). But X is not Cohen-Macaulay (Exercise 26.1.G) and π is not flat
(Exercise 24.3.I), so by Exercise 26.2.H, you can use either of these to prove the
other.

26.2.13. Fancy properties of Cohen-Macaulayness.

754 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

We mention a few additional properties of Cohen-Macaulayness without proof.

They are worth seeing, but we will not use them.
If A is a Noetherian ring that is the quotient of a Cohen-Macaulay ring, the
locus of Cohen-Macaulay points in Spec A is open, see [Mat2, Ex. 24.2]. Related:

26.2.14. Fact (cf. Fact 13.8.5). — Any localization of a Cohen-Macaulay Noetherian local
ring at a prime ideal is also Cohen-Macaulay.
(See [Stacks, tag 00NB], [E, Prop. 18.8], [Mat2, Thm. 17.3(iii)], or [Mat1, Thm. 30]
for a direct proof.) In particular, for varieties, Cohen-Macaulayness may be checked
at closed points (Exercise 5.3.F). The fact that Cohen-Macaulayness is preserved
by localization can be used to quickly show that Cohen-Macaulay local rings are
catenary (see [E, Cor. 18.10], [Mat1, Thm. 31(ii)], or [Mat2, Thm. 17.9]). A ring A
is Cohen-Macaulay if and only if A[x] is Cohen-Macaulay ([E, Prop 18.9], [Mat2,
Thm 17.7], [Mat1, Thm. 33]), if and only if A[[x]] is [Mat2, p. 137]. A local ring is
Cohen-Macaulay if and only if its completion is Cohen-Macaulay ([E, Prop. 18.8],
[Mat2, Thm 17.5]).

26.3 Serre’s R1 + S2 criterion for normality

The notion of depth yields a useful criterion for normality, due to Serre.

26.3.1. Definition. Suppose A is a Noetherian ring, and k ∈ Z≥0 . We say A has

property Rk (A is regular in codimension ≤ k, or more sloppily, A is regular in
codimension k) if for every prime p ⊂ A of codimension at most k, Ap is regular.
(We have repeatedly seen the importance of regularity in codimension 1.) Thus a
Noetherian ring is regular and only if it has property Rk for all k.
We say A has property Sk if for every prime p ⊂ A, the local ring Ap has
depth at least min(k, dim Ap ) — “the local rings are Cohen-Macaulay up until
codimension k, and have depth at least k thereafter”. Thus a Noetherian ring is
Cohen-Macaulay if and only if it has property Sk for all k.

26.3.A. E ASY EXERCISE . Note that a Noetherian ring A trivially always has prop-
erty S0 . Show that a Noetherian ring A has property R0 if Spec A is “generically
reduced”: it is reduced at the generic point of each of its irreducible components.
Show that a Noetherian ring A has property S1 if and only if Spec A has no embed-
ded points. (Possible hint: Exercise 26.1.A.)

26.3.B. E XERCISE ( IMPORTANT ELEMENTARY INTERPRETATION OF S2 ). Suppose A

is a Noetherian integral domain. Show that A has property S2 if and only if for all
nonzero g ∈ A, A/(g) has no embedded primes.

26.3.C. E XERCISE . Show that a Noetherian ring A has no nilpotents (Spec A is

reduced) if and only if it has properties R0 and S1 . (Hint: show that a Noetherian
scheme is reduced if and only if each irreducible component is generically reduced,
and it has no embedded points. )
Incrementing the subscripts in Exercise 26.3.C yields Serre’s criterion.
 © 2024 Ravi Vakil. Published by Princeton University Press. 755

26.3.2. Theorem (Serre’s criterion for normality). — A Noetherian ring A is normal

if and only if it has properties R1 and S2 .
(Recall from §5.4.1 that a Noetherian ring A is normal if its localizations Ap
at all primes p are integrally closed domains.) Thus failure of normality can have
two possible causes: it can be failure of R1 (something we already knew, from the
equivalence of (a) and (g) in Theorem 13.5.8), or it can be the more subtle failure of
S2 . Examples of varieties satisfying R1 but not S2 are given in Example 26.3.4 (two
planes meeting at a point) and Exercise 26.3.F (the pinched plane).

26.3.D. E ASY E XERCISE . Prove Serre’s criterion for normality (Theorem 26.3.2).
Hint: look closely at Proposition 13.5.24 and Proposition 13.5.26.

26.3.3. Applications. First, Serre’s Criterion for normality implies that Cohen-
Macaulay schemes are normal if and only if they are regular in codimension 1. Thus
to check normality of hypersurfaces (or more generally, regular embeddings) in Pn k
(or more generally, in any smooth variety), it suffices to check that their singular
locus has codimension greater than one. (You should think through the details of
why these statements are true.) In particular, this gives a new (more complicated)
proof of Exercise 5.4.I(b) and (c).

26.3.E. E XERCISE ( PRACTICE WITH THE CONCEPT ). Show that two-dimensional

normal varieties are Cohen-Macaulay.

26.3.4. Example: two planes meeting at a point. The variety X of Exercise 26.1.G (two
planes meeting at a point) is not normal (why?), but it is regular away from the
origin. This implies that X does not have property S2 (and hence is not Cohen-
Macaulay), without the algebraic manipulations of Exercise 26.1.G.
We already knew that this example was not Cohen-Macaulay, but the same idea
can show that the pinched plane Spec k[x3 , x2 , xy, y] (appearing in Exercise 13.5.H)
is not Cohen-Macaulay. Because of the “extrinsic” description of the ring, it is
difficult to do this in another way.

26.3.F. E XERCISE : THE PINCHED PLANE IS NOT C OHEN -M ACAULAY. Let A be

the subring k[x3 , x2 , xy, y] of k[x, y]. Show that A is not Cohen-Macaulay. Hint:
Exercise 13.5.H showed that A is not integrally closed.

26.3.5. Regular local rings are integrally closed. Serre’s criterion for normality
can be used to show that regular local rings are integrally closed without going
through the hard Fact 13.8.5 (the Auslander-Buchsbaum Theorem) that they are
unique factorization domains. Regular local rings are Cohen-Macaulay (§26.2.5),
and (by Exercise 26.3.C) regular in codimension 0 as they are integral domains
(Theorem 13.2.11), so we need only show that regular local rings are regular in
codimension 1. We can invoke a different hard Fact 13.8.2 that localizations of
regular local rings are again regular, but at least we have shown this for localiza-
tions of finitely generated algebras over a perfect field (see Theorem 13.8.3 and
Exercise 21.6.B).

26.3.6. Caution. As is made clear by the following exercise, the condition S2 is a

condition on all prime ideals, not just those of codimension at most 2.
756 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

26.3.G. E XERCISE . Give an example of a variety satisfying R1 , and Cohen-Macaulay

at all points of codimension at most 2, which is not normal.

26.3.7. ⋆⋆ Projective normality. We will not use the notion of projective normality,
but mention it because many readers may come across it, and it is easy to introduce.
k . Let L = π O(1) be the
∗
It is a property of a closed embedding of π : X ,→ Pn
line bundle giving the closed embedding (§15.2), and for convenience let R(L )• :=
⊕m≥0 Γ (X, L ⊗m ) be the graded ring of “sections of all powers of L ”. Then we
say (Definition 1) X ,→ Pn k is projectively normal if X is normal and for all m ≥ 0,
H0 (Pnk , O(m)) → H 0
(X, OX (m)) is surjective. In other words, all sections of L
⊗m

for all m ≥ 0 “come from” polynomials; there are no “new” sections not “arising”
from H0 (X, L ). Equivalently (Definition 2), via L = π∗ O(1), R(L )• is a quotient
of the polynomial ring k[x0 , . . . , xn ]. You will quickly be able to show that this is
equivalent to the statement (Definition 3) that (X is normal and) H1 (Pn k , IX (m)) = 0
for all m ≥ 0, and also to the statement (Definition 4) that X is normal, R(L )• is
generated in degree 1, and X ,→ Pn k is an embedding by a complete linear series
(assuming it is nondegenerate).

26.3.H. E XERCISE . Show that complete intersections in Pn k are projectively normal

if and only if they are normal if and only if they are regular in codimension 1.
It is not too hard to show that X ,→ Pn
k is normal if and only if its affine cone is
normal, [Ha1, Ex. II.5.14].
 CHAPTER 27

The twenty-seven lines on a cubic surface

Wake an algebraic geometer in the dead of night, whispering: “27”. Chances are, he
will respond: “lines on a cubic surface”.
— R. Donagi and R. Smith, [DS] (on page 27, of course)

F IGURE 27.1. The twenty-seven lines on the Clebsch cubic surface

(from Greg Egan)

Since the middle of the nineteenth century, geometers have been entranced by
the fact that there are 27 lines on every smooth cubic surface, and by the remarkable
structure of the configuration of the lines. Their discovery by Cayley and Salmon
in 1849 has been called the beginning of modern algebraic geometry, [Do, p. 55].
The reason so many people are bewitched by this fact is because it requires
some magic, and this magic connects to many other things, including fundamental
ideas we have discussed, other beautiful classical constructions (such as Pascal’s
Mystical Hexagon Theorem 19.4.4, the fact that most smooth quartic plane curves
have 28 bitangents, exceptional Lie groups, . . . ), and many themes in modern
algebraic geometry (deformation theory, intersection theory, enumerative geometry,
arithmetic and diophantine questions, ...). It will be particularly pleasant for us, as
it takes advantage of many of the things we have learned.
You are now ready to be initiated into the secret fellowship of the twenty-seven
lines.
This material is published by Princeton University Press as The Rising Sea: Foundations of Algebraic
Geometry by Ravi Vakil. This version is free to view and download for personal use only. Not for
redistribution, resale, or use in derivative works. © 2024 Ravi Vakil.

757
758 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

27.0.1. Theorem. — Every smooth cubic surface in P3k contains exactly 27 lines.
Theorem 27.0.1 is closely related to the following.

27.0.2. Theorem. — Every smooth cubic surface over k is isomorphic to P2 blown up at 6

points.
There are many reasons why people consider these facts magical. First, there
is the fact that there are always 27 lines. Unlike most questions in enumerative
geometry, there are no weasel words such as “a general cubic surface” or “most
cubic surfaces” or “counted correctly” — as in, “every monic degree d polynomial
has d roots — counted correctly”. And somehow (and we will see how) it is
precisely the smoothness of the surface that makes it work.
Second, there is the magic that you always get the blow-up of the plane at six
points (§27.3).
Third, there is the magical incidence structure of the 27 lines, which relates to
E6 in Lie theory. The Weyl group of E6 is the symmetry group of the incidence
structure (see Remark 27.2.5). In a natural way, the 27 lines form a basis of the
27-dimensional fundamental representation of E6 .

27.0.3. On the structure of this chapter. Throughout this chapter, X will be a

smooth cubic surface over an algebraically closed field k. In §27.1, we establish
some preliminary facts. In §27.2, we prove Theorem 27.0.1, away from characteristic
3. In §27.3, we prove Theorem 27.0.2. We remark here that the only input that
§27.3 needs from §27.2 is Exercise 27.2.J. This can be done directly by hand (see for
example [Rei1, §7], [Be, Thm. IV.13], or [Sh, p. 246-7]), and Theorem 27.0.2 readily
implies Theorem 27.0.1, using Exercise 27.3.F. We would thus have another, shorter,
proof of Theorem 27.0.1. The reason for giving the argument of §27.2 is that it is
natural given what we have done so far, it gives you some glimpse of some ideas
used more broadly in the subject (the key idea is that a map from one moduli space
to another is finite and flat), and it may help you further appreciate and digest the
tools we have developed.

27.0.4. On the location of this chapter.

This topic is Chapter 27 for purely numerological reasons; it could certainly
be read after Chapter 24. This is a feature rather than a bug: you needn’t read the
details of the later chapters in order to enjoy this one. In fact, you should probably
read this many chapters earlier, taking key facts as “black boxes”, to appreciate the
value of seemingly technical definitions and theorems before you learn about them
in detail. This topic is a fitting conclusion to an energetic introduction to algebraic
geometry, even if you do not earlier cover everything needed to justify all details.

27.1 Preliminary facts

By Essential Exercise 15.1.C, there is a 20-dimensional vector space of cubic

forms in four variables, so the cubic surfaces in P3 are parametrized by P19 . Let
∆ ⊂ P19 be the locus corresponding to singular cubic hypersurfaces. Bertini’s
 © 2024 Ravi Vakil. Published by Princeton University Press. 759

Theorem 13.4.2 shows that the locus of smooth cubics P19 \ ∆ is dense and open in
the space P19 of all cubics.

27.1.A. E XERCISE . Show that any smooth cubic surface X is “anticanonically

embedded” — it is embedded by the complete anticanonical linear series KX∨ .
Hint: the adjunction formula, Exercise 21.5.B.

27.1.B. E XERCISE . Suppose X ⊂ P3k is a smooth cubic surface. Suppose C is a curve

on X. Show that C is a line if and only if C is a (−1)-curve (Definition 20.2.8). Hint:
the adjunction formula again, perhaps in the guise of Exercise 20.2.B(a); and also
Exercise 18.6.K.
It will be useful to find a single cubic surface over a single field with exactly 27
lines:

27.1.C. H ANDS - ON E XERCISE . Show that the Fermat cubic surface

(27.1.0.1) x30 + x31 + x32 + x33 = 0
in P3C has precisely 27 lines, each of the form
x0 + ωxi = xj + ω ′ xk = 0,
where {1, 2, 3} = {i, j, k}, j < k, and ω and ω ′ are cube roots of 1 (possibly the same).
This will require some brute force. Possible hint: up to a permutation of coordinates,
show that every line in P3 can be written x0 = ax2 + bx3 , x1 = cx2 + dx3 . Show
that this line is on (27.1.0.1) if and only if
(27.1.0.2) a3 + c3 + 1 = b3 + d3 + 1 = a2 b + c2 d = ab2 + cd2 = 0
Show that if a, b, c, and d are all nonzero, then (27.1.0.2) has no solutions.

27.1.1. ⋆ The singular cubic surfaces give an irreducible divisor ∆ in the parameter
space P19 of all cubic surfaces. This discussion is starred not because it is hard,
but because it is not needed in the rest of the chapter. Nonetheless, it is a pretty
application of what we have learned, and it foreshadows key parts of the argument
in §27.2.
Bertini’s Theorem 13.4.2 shows that the locus of smooth cubics in the P19 of all
cubics is dense and open, but the proof suggests more: that the complement ∆ of
this locus is of pure codimension 1. We now show that this is the case, and even
that this “discriminant hypersurface” ∆ is irreducible.

27.1.D. E XERCISE . (Hint for both: recall the solution to Exercise 12.2.N.)
(a) Define the incidence correspondence Y ⊂ P19 × P3 corresponding to the data of
a cubic surface X, along with a singular point p ∈ X. (This is part of the exercise! We
need Y as a scheme, not just as a set.) Let µ be the projection Y → P19 .
Y
µ P15 -bundle
~ 
P19 P3

(b) Show that Y is an irreducible smooth variety of dimension 18, by describing it

as a P15 -bundle over P3 .
760 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

27.1.E. E XERCISE . Show that there exists a cubic surface with a single singular
point. Feel free to assume k is your favorite field; the main point is to familiarize
yourself with geometric ideas, not the peculiarities of positive characteristic fields.
(Hint if char k ̸= 2: Exercise 13.3.C.)

27.1.F. E XERCISE . Show that µ(Y) = ∆ is a closed irreducible subset of codimension

1. Hint for the codimension 1 statement: Exercise 12.4.A or Theorem 12.4.1, in
combination with Exercise 27.1.E.
Your argument will generalize with essentially no change to deal with degree
d hypersurfaces in Pn .

27.2 Every smooth cubic surface (over k) contains 27 lines

We are now ready to prove Theorem 27.0.1. Until Exercise 27.2.K, to avoid
distraction, we assume char k = 0. However, the following argument carries
through without change if char k ̸= 3. The one required check — and the reason for
the restriction on the characteristic — is that Exercise 27.1.C works with C replaced
by any such k.

27.2.A. E XERCISE . (Hint for both: recall the solution to Exercise 12.2.N.)
(a) Define the incidence correspondence Z ⊂ P19 × G(1, 3) corresponding to the
data of a line ℓ in P3 contained in a cubic surface X. (As in Exercise 27.1.D, this is
part of the exercise.) Let π be the projection Z → P19 .

Z
π P15 -bundle
~ "
P19 G(1, 3)

(b) Show that Z is an irreducible smooth variety of dimension 19, by describing it

as a P15 -bundle over the Grassmannian G(1, 3).

27.2.B. E XERCISE . Use the fact that there exists a cubic surface with a finite number
of lines (Exercise 27.1.C), and the behavior of dimensions of fibers of morphisms
(Exercise 12.4.A or Theorem 12.4.1) to show the following.
(a) Every cubic surface contains a line, i.e., π is surjective. (Hint: show that π
is projective.)
(b) “Most cubic surfaces have a finite number of lines”: there is a dense open
subset U ⊂ P19 such that the cubic surfaces parametrized by closed points
of U have a positive finite number of lines. (Hint: upper semicontinuity of
fiber dimension, Theorem 12.4.3.)

The following fact is the key result in the proof of Theorem 27.0.1, and one
of the main miracles of the 27 lines, that ensures that the lines stay distinct on a
smooth surface. It states, informally, that two lines can’t come together without
damaging the surface. This is really a result in deformation theory: we are explicitly
showing that a line in a smooth cubic surface has no first-order deformations.
 © 2024 Ravi Vakil. Published by Princeton University Press. 761

27.2.1. Theorem. — If ℓ is a line in a regular cubic surface X, then {ℓ ⊂ X} is a reduced

isolated point of the fiber of π : Z → P19 .
Before proving Theorem 27.2.1, we use it to prove Theorem 27.0.1.

27.2.2. Proof of Theorem 27.0.1, assuming Theorem 27.2.1. Now π is a projective

morphism, and by Theorem 27.2.1, over P19 \ ∆, π has relative dimension 0, and
hence has finite fibers. Hence by Theorem 18.1.6 (projective + finite fibers = finite),
π is finite over P19 \ ∆.
Furthermore, Z and P19 are both smooth of dimension 19 (over Z), and the
fibers of π : Z → P19 are (geometrically) reduced points above P19 \ ∆, again by
Theorem 27.2.1. Then by Exercise 24.8.G, π is étale over P19 \ ∆ (and in particular
flat).
Thus, over P19 \ ∆, π is a finite flat morphism, and so the fibers of π (again,
away from ∆) always have the same number of points, “counted correctly” (Re-
mark 24.3.8). But by Theorem 27.2.1, above each closed point of P19 \ ∆, each
point of the fiber of π counts with multiplicity one (here using k = k). Finally, by
Exercise 27.1.C, the Fermat cubic surface gives an example of one regular cubic
surface with precisely 27 lines, so (as P19 \ ∆ is connected) we are done. □
We have actually shown that away from ∆, Z → P19 is a finite étale morphism
of degree 27.

27.2.3. ⋆ Proof of Theorem 27.2.1. Choose projective coordinates so that the line ℓ
is given, in a distinguished affine subset (with coordinates named x, y, z), by the
z-axis. (We use affine coordinates to help visualize what we are doing, although
this argument is better done in projective coordinates. On a second reading, you
should translate this to a fully projective argument.)

27.2.C. E XERCISE . Consider the lines of the form (x, y, z) = (a, b, 0) + t(a ′ , b ′ , 1)
(where (a, b, a ′ , b ′ ) ∈ A4 is fixed, and t varies in A1 ). Show that a, b, a ′ , b ′ can be
interpreted as the “usual” coordinates on one of the standard open subsets of the
Grassmannian (see §7.7), with [ℓ] as the origin.
Having set up local coordinates on the moduli space, we can get down to
business. Suppose f(x, y, z) = 0 is the (affine version) of the equation for the cubic
surface X. Because X contains the z-axis ℓ, f(x, y, z) ∈ (x, y). More generally, the
line
(27.2.3.1) (x, y, z) = (a, b, 0) + t(a ′ , b ′ , 1)
lies in X precisely when f(a + ta ′ , b + tb ′ , t) is 0 as a cubic polynomial in t. This is
equivalent to four equations in a, a ′ , b, and b ′ , corresponding to the coefficients of
t3 , t2 , t, and 1. This is better than just a set-theoretic statement:

27.2.D. E XERCISE . Verify that these four equations are local equations for the
scheme-theoretic fiber (or scheme-theoretic preimage) π−1 ([X]).
Now we come to the crux of the argument, where we use the regularity of
X (along ℓ). We have a precise question in algebra. We are given a cubic surface
X given by f(x, y, z) = 0, containing ℓ, and we know that X is regular (including
“at ∞”, i.e., in P3 ). To show that [ℓ] = V(a, a ′ , b, b ′ ) is a reduced isolated point in
the fiber of π, we work in the ring k[a, a ′ , b, b ′ ]/(a, a ′ , b, b ′ )2 , i.e., we impose the
762 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

F IGURE 27.2. Parameters for the space of “lines near ℓ”, in terms
of where they meet the z = 0 plane and the z = 1 plane

equations
(27.2.3.2) a2 = aa ′ = · · · = (b ′ )2 = 0,
and try to show that a = a ′ = b = b ′ = 0. (It is essential that you understand
why we are setting (a, a ′ , b, b ′ )2 = 0. You can also interpret this argument in terms
of the derivatives of the functions involved — which after all can be interpreted
as forgetting higher-order information and remembering only linear terms in the
relevant variables, cf. Exercise 13.1.I. See [Mu6, §8D] for a description of this
calculation in terms of derivatives.)
Suppose f(x, y, z) = cx3 x3 + cx2 y x2 y + · · · + c1 1 = 0, where cx3 , cx2 y , · · · ∈ k.
Because ℓ ⊂ X, i.e., f ∈ (x, y), we have c1 = cz = cz2 = cz3 = 0. We now substitute
(27.2.3.1) into f, and then apply (27.2.3.2). Only the coefficients of f of monomials
involving precisely one x or y survive:
cx (a + a ′ t) + cxz (a + a ′ t)t + cxz2 (a + a ′ t)t2
+cy (b + b ′ t) + cyz (b + b ′ t)t + cyz2 (b + b ′ t)t2
= (a + a ′ t)(cx + cxz t + cxz2 t2 ) + (b + b ′ t)(cy + cyz t + cyz2 t2 )
is required to be 0 as a polynomial in t. (Recall that cx , . . . , cyz2 are fixed elements
of k.) Let Cx (t) = cx + cxz t + cxz2 t2 and Cy (t) = cy + cyz t + cyz2 t2 for convenience.
Now X is regular at (0, 0, 0) precisely when cx and cy are not both 0 (as cz = 0).
More generally, X is regular at (0, 0, t0 ) precisely if cx + cxz t0 + cxz2 t20 = Cx (t0 )
and cy + cyz t0 + cyz2 t20 = Cy (t0 ) are not both zero. You should be able to quickly
check that X is regular at the point of ℓ “at ∞” precisely if cxz2 and cyz2 are not
both zero. We summarize this as follows: X is regular at every point of ℓ precisely if
the two quadratics Cx (t) and Cy (t) have no common roots, including “at ∞”.
We now use this to force a = a ′ = b = b ′ = 0 using (a + a ′ t)Cx (t) + (b +
′
b t)Cy (t) ≡ 0.
 © 2024 Ravi Vakil. Published by Princeton University Press. 763

We deal first with the special case where Cx and Cy have two distinct roots,
both finite (i.e., cxz2 and cyz2 are nonzero). If t0 and t1 are the roots of Cx (t), then
substituting t0 and t1 into (a + a ′ t)Cx (t) + (b + b ′ t)Cy (t), we obtain b + b ′ t0 = 0,
and b + b ′ t1 = 0, from which b = b ′ = 0. Similarly, a = a ′ = 0.

27.2.E. E XERCISE . Deal with the remaining cases to conclude the proof of Theo-
rem 27.2.1. (It is possible to do this quite cleverly. For example, you may be able to
re-choose coordinates to ensure that Cx and Cy have finite roots.)
□

27.2.4. The configuration of lines.

By the “configuration of lines” on a cubic surface, we mean the data of which
pairs of the 27 lines intersect. We can readily work this out in the special case
of the Fermat cubic surface (Exercise 27.1.C). (It can be more enlightening to use
the description of X as a blow-up of P2 , see Exercise 27.3.F.) We now show that
the configuration is the “same” (interpreted appropriately) for all smooth cubic
surfaces.

27.2.F. E XERCISE . Construct a degree 27! finite étale map W → P19 \ ∆, that
parametrizes a cubic surface along with an ordered list of 27 distinct lines. Hint:
let W ′ be the 27th fibered power of Z over P19 \ ∆, interpreted as parametrizing
a cubic surface with an ordered list of 27 lines, not necessarily distinct. Let W be
the subset corresponding to where the lines are distinct, and show that W is open
and closed in W ′ , and thus a union of connected components of W ′ . An alternative
way of identifying W within W ′ , in a way that might generalize to Exercise 27.2.H:
make sense of the fact for each i, j, “ℓi · ℓj ” is a locally constant function on W ′ , so
the locus where these functions are positive for all i ̸= j is both open and closed;
show that this is the W we seek.
We now make sense of the statement that the configuration of lines on the
Fermat cubic surface (call it X0 ) is the “same” as the configuration on some other
smooth cubic surface (call it X1 ). Lift the point [X0 ] to a point w0 ∈ W. Let W ′′ be
the connected component of W containing w0 .

27.2.G. E XERCISE . Show that the morphism W ′′ → P19 \ ∆ is finite and étale (and
surjective).
Choose a point w1 ∈ W ′′ mapping to [X1 ]. Because W parametrizes a “labeling”
or ordering of the 27 lines on a surface, we now have chosen an identification of
the lines on X0 with those of X1 . Let the lines be ℓ1 , . . . , ℓ27 on X0 , and let the
corresponding lines on X1 be m1 , . . . , m27 .

27.2.H. E XERCISE ( USING STARRED E XERCISE 24.7.D). Show that ℓi · ℓj = mi · mj

for all i and j.

27.2.I. E XERCISE . Show that for each smooth cubic surface X ⊂ P3k , each line on X
meets exactly 10 other lines ℓ1 , ℓ1′ , . . . , ℓ5 , ℓ5′ on X, where ℓi and ℓi′ meet for each i,
and no other pair of the lines meet.

27.2.J. E XERCISE . Show that every smooth cubic surface contains two disjoint
lines ℓ and ℓ ′ , such that there are precisely five other lines ℓ1 , . . . , ℓ5 meeting both ℓ
764 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

and ℓ ′ . (In fact, for any two disjoint lines ℓ and ℓ ′ in a smooth cubic surface, there
are precise five other lines meeting them; but this is strictly harder than what the
problem asks.)

27.2.5. Remark (cf. Remark 27.3.7): The Weyl group W(E6 ). The symmetry group of
the configuration of lines — i.e., the subgroup of the permutations of the 27 lines
preserving the intersection data — magically turns out to be the Weyl group of E6 ,
a group of order 51840. (You know enough to at least verify that the size of the
group is 51840, using the Fermat cubic surface of Exercise 27.1.C, but this takes
some work.) It is no coincidence that the degree of W ′′ over P19 \ ∆ is 51840, and
the Galois group of the Galois closure of K(Z)/K(P19 \ ∆) is isomorphic to W(E6 )
(see [H1, III.3]).

27.2.K. ⋆⋆ E XERCISE . Prove Theorem 27.2.1 in arbitrary characteristic. Begin by

figuring out the right statement of Exercise 27.2.A over Z, and proving it. Then
follow the argument given in this section, making changes when necessary.

27.2.6. ⋆ Fano varieties of lines, and Hilbert schemes. In Exercises 27.2.A

and 27.2.K, you constructed a moduli space of lines contained in a X, as a scheme.
Your argument can be generalized to any X ⊂ PN . This construction is called the
Fano variety of lines of X (not to be confused with the notion of a Fano variety, §21.5.5),
and is an example of a Hilbert scheme.

27.3 Every smooth cubic surface (over k) is a blown up plane

We now prove Theorem 27.0.2. As stated in §27.0.3, this section is remarkably

independent from the previous one; all we will need is Exercise 27.2.J, and it is
possible to prove that in other ways.
Suppose X is a smooth cubic surface (over k). Suppose ℓ is a line on X, and
choose coordinates on the ambient P3 so that ℓ is cut out by x0 and x1 . Projection
from ℓ gives a rational map P3 99K P1 (given by [x0 , x1 , x2 , x3 ] 7→ [x0 , x1 ]), which
extends to a morphism on X. The reason is that this rational map is resolved by
blowing up the closed subscheme V(x0 , x1 ) (Exercise 22.4.L). But (x0 , x1 ) cuts out
the effective Cartier divisor ℓ on X, and blowing up an effective Cartier divisor does
not change X (Observation 22.2.1).
Now choose two disjoint lines ℓ and ℓ ′ as in Exercise 27.2.J, and consider the
morphism ρ : X → P1 × P1 , where the map to the first P1 is projection from ℓ, and
the map to the second P1 is the projection from ℓ ′ . The first P1 can then be identified
with ℓ ′ , and the second with ℓ.

27.3.A. E XERCISE . Show that the morphism ρ is birational. Hint: given a general
point of (p, q) ∈ ℓ ′ × ℓ, we obtain a point of X as follows: the line pq in P3 meets
the cubic X at three points by Bézout’s Theorem 9.3.D: p, q, and some third point
x ∈ X; send (p, q) to x. (This idea appeared earlier in the development of the group
law on the cubic curve, see Proposition 19.10.3.) Given a general point x ∈ X, we
obtain a point (p, q) ∈ ℓ ′ × ℓ by projecting from ℓ ′ and ℓ.
 © 2024 Ravi Vakil. Published by Princeton University Press. 765

In particular, we have shown for the first time that every smooth cubic surface
over k is rational.

27.3.B. E XERCISE : ρ CONTRACTS PRECISELY ℓ1 , . . . , ℓ5 . Show that ρ is an isomor-

phism away from the ℓi mentioned in Exercise 27.2.J, and that each ρ(ℓi ) is a point
pi ∈ ℓ ′ × ℓ.

27.3.1. Proposition. — The morphism ρ : X → P1 × P1 is the blow-up of P1 × P1 at the

five pi .
We give two proofs. The first demonstrates the utility of a number of technical
tools we have developed (or will develop in the future — there is one “forward
reference” here). The second is a down-to-earth explicit “calculation”. Different
people prefer different approaches, but there are advantages in seeing them side by
side. The first shows how to apply tools in quite general situations where we don’t
have explicit control over the geometry. The second gives us rather complete and
direct control of the geometry in question, without needing formal functions.

27.3.2. Proof 1 of Proposition 27.3.1. By Castelnuovo’s Criterion (Theorem 28.6.1), as

the lines ℓi are (−1)-curves (Exercise 27.1.B), they can be contracted. More precisely,
there is a morphism β : X → X ′ that is the blow-up of X ′ at five closed points p1′ , . . . ,
p5′ , such that ℓi is the exceptional divisor at pi′ . We wish to show that X ′ is P1 × P1 .
The morphism ρ : X → P1 ×P1 yields a morphism ρ ′ : X ′ \{p1′ , . . . p5′ } → P1 ×P1 .
We now show that ρ ′ extends over pi′ for each i, sending pi′ to pi . Choose an open
neighborhood of pi ∈ P1 × P1 isomorphic to A2 , with coordinates x and y. Then
both x and y pull back to functions on a punctured open neighborhood of pi′ (i.e.,
there is some open neighborhood U of pi′ such that x and y are functions on U\ {pi′ }).
By Algebraic Hartogs’s Lemma 13.5.19, they extend over pi′ , and this extension is
unique as P1 × P1 is separated — use the Reduced-to-Separated Theorem 11.3.2 if
you really need to. Thus ρ ′ extends over pi′ . (Do you see why ρ ′ (pi′ ) = pi ?)

27.3.C. E XERCISE . Show that the birational morphism ρ ′ : X ′ → P1 ×P1 is invertible.

Hint: you can use Zariski’s Main Theorem (in the guise of Exercise 28.5.E), but you
needn’t use something so powerful. Instead, note that the birational map ρ ′−1 is a
morphism away from p1 , . . . , p5 . Use essentially the same argument as in the last
paragraph to extend ρ ′−1 over each pi .
□

27.3.3. Proof 2 of Proposition 27.3.1. (I learned this slick version of a very classical
argument from J. Kollár, who is famous for explaining complicated things in simple
ways.) Choose coordinates on P3 so that ℓ is given by x0 = x1 = 0, and ℓ ′ is
given by x2 = x3 = 0. Let the projective coordinates on ℓ be u0 and u1 , and let
the projective coordinates on ℓ ′ be v0 and v1 (so for example ℓ ,→ P3 is given by
[u0 , u1 ] 7→ [0, 0, u0 , u1 ]). Consider the rational map

α : P1u0 ,u1 × P1v0 ,v1 × P1s,t → P3

(where the subscripts indicate the names of the projective coordinates) given by

(27.3.3.1) ([u0 , u1 ], [v0 , v1 ], [s, t]) 7→ [sv0 , sv1 , tu0 , tu1 ].

766 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Because X contains ℓ = V(x0 , x1 ) and ℓ ′ = V(x2 , x3 ), the equation of X ⊂ P3 is of

the form
F(x0 , x1 , x2 , x3 ) = L02 x0 x2 + L03 x0 x3 + L12 x1 x2 + L13 x1 x3
where the Lij are linear forms in x0 , x1 , x2 , x3 . Thus
α∗ F = st (sG(u0 , u1 , v0 , v1 ) + tH(u0 , u1 , v0 , v1 )) .
From (27.3.3.1), G has degree 1 in the ui and degree 2 in the vi , and H has degree 2
in the ui and degree 1 in the vi .

27.3.D. E XERCISE . Interpret the morphism X → P1 ×P1 as a rational map as pulling

back F = 0 by α, then projecting to P1u0 ,u1 × P1v0 ,v1 . Show that the five lines ℓ1 , . . . ,
ℓ5 contracted (to p1 , . . . , p5 ) by X → P1 × P1 must correspond set-theoretically to
the locus G = H = 0 in some sense. Show that G = H = 0 set-theoretically cuts
out the five points p1 , . . . , p5 in P1 × P1 . Show that (in the intersection theory of
P1 × P1 , Exercise 20.2.C) V(G) · V(H) = 5. Show that near any of the five points
p1 , . . . , p5 of P1 × P1 , the effective Cartier divisors V(G) and V(H) must intersect
trasnversely (with “multiplicity 1”) — you’ll have to make this precise. Show that
the projection near earch pi must be the blow-up of P1 × P1 at pi . □

27.3.4. Conclusion of the proof of Theorem 27.0.2. As a consequence we see that X is

the blow-up of P1 × P1 at 5 points. Because the blow-up of P1 × P1 at one point
is isomorphic to the blow-up of P2 at two points (Exercise 22.4.K), Theorem 27.0.2
then follows. □

27.3.5. Reversing the process.

(This is a more precise version of 22.4.10(iii).) The process can be reversed: we
can blow-up P2 at six points, and embed it in P3 . We first explain why we can’t
blow up P2 at just any six points and hope to embed the result in P3 . Because the
cubic surface is embedded anticanonically (Exercise 27.1.A), we see that any curve
C in X must satisfy degC KX < 0.

27.3.E. E XERCISE . Suppose P2 is sequentially blown up at p1 , . . . , p6 , resulting in a

smooth surface X.
(a) If pi lies on the exceptional divisor of the blow-up at pj (i > j), then show that
there is a curve C ⊂ X isomorphic to P1 , with degC KX ≥ 0.
(b) If the pi are distinct points on P2 , and three of them are collinear, show that
there is a curve C ⊂ X isomorphic to P1 , with degC KX ≥ 0.
(c) If the six pi are distinct points on a smooth conic, show that there is a curve
C ⊂ X isomorphic to P1 , with degC KX ≥ 0.
Thus the only chance we have of obtaining a smooth cubic surface by blowing
up six points on P2 is by blowing up six distinct points, no three on a line and not
all on a conic.

27.3.6. Proposition. — The anticanonical map of P2 blown up at six distinct points, no

three on a line and not all on a conic, gives a closed embedding into P3 , as a cubic surface.
Because we won’t use this, we only describe the main steps of the proof: first
count sections of the anticanonical bundle KP∨ ∼
2 = OP2 (3) (there is a 4-dimensional
 © 2024 Ravi Vakil. Published by Princeton University Press. 767

vector space of cubics on P2 vanishing at the six points, and these correspond to
sections of the anticanonical bundle of the blow-up via Exercise 22.4.S(a)). Then
show that these sections separate points and tangent vectors of X, thus showing that
the anticanonical linear series gives a closed embedding, Theorem 19.1.1. Judicious
use of the Cremona transformation (Exercise 7.5.J) can reduce the amount of tedious
case-checking in this step.

27.3.F. E XERCISE . Suppose X is the blow-up of P2k at six distinct points p1 , . . . ,

p6 , no three on a line and not all on a conic. Verify that the only (−1)-curves on X
are the six exceptional divisors, the proper transforms of the 15 lines pi pj , and the
proper transforms of the six conics through five of the six points, for a total of 27.

27.3.G. E XERCISE . Solve Exercises 27.2.I and 27.2.J again, this time using the
description of X as a blow-up of P2 .

27.3.7. Remark. If you blow up 4 ≤ n ≤ 8 points on P2 , with no three on a line and

no six on a conic, then the symmetry group of the configuration of lines is a Weyl
group, as shown in the following table.

n 4 5 6 7 8
W(A4 ) W(D5 ) W(E6 ) W(E7 ) W(E8 )
(If you know about Dynkin diagrams, you may see the pattern, and may be able to
interpret what happens for n = 3 and n = 9. The surface corresponding to n = 4
appeared in §19.8.3.) This generalizes part of Remark 27.2.5, and the rest of it can
similarly be generalized.

27.3.H. E XERCISE . Show that the surfaces described in Remark 27.3.7 are del Pezzo
surfaces (defined in Exercise 21.5.H). (You don’t know need to know about Dynkin
diagrams to do this. It turns out that every del Pezzo is of the sort described here
or in Exercise 21.5.H.)
 CHAPTER 28

Power series and the Theorem on Formal Functions

Power series are a central tool in analytic geometry. Their analog in algebraic
geometry, completion, is similarly useful. We will only touch on some aspects of
the subject.
In §28.1, we deal with some algebraic preliminaries. In §28.2, we use comple-
tions to (finally) give a definition of various singularities, such as nodes. (We won’t
use these definitions in what follows.) In §28.3, we state the main technical result of
the chapter, the Theorem on Formal Functions 28.3.2. The subsequent three sections
give applications. In §28.4, we prove Zariski’s Connectedness Lemma 28.4.1 and
the Stein Factorization Theorem 28.4.3. In §28.5, we prove a commonly used version
of (Grothendieck’s version of) Zariski’s Main Theorem 28.5.1; we rely on §28.4. In
§28.6, we prove Castelnuovo’s Criterion for contracting (−1)-curves, which we
used in Chapter 27. The proof of Castelnuovo’s Criterion also uses §28.4. Finally, in
§28.7, we prove the Theorem on Formal Functions 28.3.2.
There are deliberately many small sections in this chapter, so you can see
that they tend not to be as hard at they look, with the exception of the proof of
the Theorem on Formal Functions itself, and possibly Theorem 28.1.6 relating
completion to exactness and flatness.

28.1 Algebraic preliminaries

Suppose A is a ring, and J1 ⊃ J2 ⊃ · · · is a decreasing sequence of ideals.
The limit lim A/Jn is often denoted A, ^ and is called the completion of A (for this
sequence of ideals); the sequence of ideals is left implicit. The most important case
of completion is if Jn = In , where I is an ideal of A. The limit lim A/In is called the
I-adic completion of A, or the completion of A along I or at I.

28.1.1. Example. We define k[[x1 , . . . , xn ]] as the completion of k[x1 , . . . , xn ] at the

maximal ideal (x1 , . . . , xn ). This is the ring of formal power series in n variables over
k (Example 1.3.3).
The p-adic numbers (Example 1.3.4) are another example.

28.1.A. E XERCISE . Suppose that J1′ ⊃ J2′ ⊃ · · · is a decreasing series of ideals that is
′ ′
cofinal with Jn . In other words, for every Jn , there is some JN with Jn ⊃ JN , and for
′ ′
every Jn , there is some JN with Jn ⊃ JN . Show that there is a canonical isomorphism

This material is published by Princeton University Press as The Rising Sea: Foundations of Algebraic
Geometry by Ravi Vakil. This version is free to view and download for personal use only. Not for
redistribution, resale, or use in derivative works. © 2024 Ravi Vakil.

769
770 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

∼
lim A/Jn′ ←→ lim A/Jn . Thus what matters is less the specific sequence of ideals
than the “cofinal” equivalence class.

28.1.2. Preliminary remarks. We have an obvious morphism A → A. ^

^
If we put the discrete topology on A/Jn , then A naturally has the structure of a
topological ring (a ring that is a topological space, where all the ring operations are
continuous, or equivalently, a ring object in the category of topological spaces). We
then can have the notion of a topological module M over a topological ring A ′ — a
module over the underlying ring A ′ , with a topology, such that the action of A ′ on
M is continuous. This is a useful point of view, but we won’t use it.

28.1.B. E XERCISE . Suppose m is a maximal ideal of a ring A. Show that the

completion of A at m is canonically isomorphic to the completion of Am at m.
If (A, m) is a Noetherian local ring, then the natural map A → A ^ is an injection:
anything in the kernel must lie in ∩mi , which is 0 by the Krull Intersection Theorem
(Exercise 13.9.A(b)). Thus “no information is lost by completing”, just as analytic
functions are (locally) determined by their power series expansion.
(In the case of completion at an ideal I, we say that A is I-adically separated if
A→A ^ is an injection, and complete with respect to I if A → A ^ is an isomorphism,
although we won’t need these phrases. For example, the Krull Intersection Theorem
implies that if I is a proper ideal of a Noetherian integral domain or a Noetherian
local ring, then A is I-adically separated.)

28.1.C. E XERCISE . Suppose that (A, m) is Noetherian local ring containing its
residue field k (i.e., it is a k-algebra). Let x1 , . . . , xn be elements of A whose images
are a basis for m/m2 . Show that the map of k-algebras

(28.1.2.1) k[[t1 , . . . , tn ]] /A
^

defined by ti 7→ xi is a surjection. (First explain why there is such a map!) As usual,

for local rings, the completion is assumed to be at the maximal ideal.
Exercise 28.1.C is a special case of the Cohen Structure Theorem. (See [E, §7.4] for
more on this topic.)

28.1.D. E XERCISE . Let X be a locally Noetherian scheme over k, let p ∈ X be

a rational (k-valued) point. Suppose p ∈ X is regular, of codimension d (i.e.,
∼
dim OX,p = d). Describe an isomorphism O^X,p ←→ k[[x1 , . . . , xd ]] as topological
rings. (Hint: As in Exercise 28.1.C, choose d elements of m that restrict to a basis of
m/m2 ; these will be your x1 , . . . , xd . Show that the map (28.1.2.1) has no kernel. It
may help to identify mn /mn+1 with Symn (m/m2 ) using Theorem 22.3.8.)
The converse also holds: if (A, m, k) is a Noetherian local ring that is a k-algebra,
^ =
and A ∼ k[[x1 , . . . , xd ]], then A is a regular local ring of dimension d; see [AtM,
Prop. 11.24] for the key step.

28.1.3. Remark. Suppose p is a smooth rational (k-valued) point of a k-variety of

dimension d, and f ∈ OX,p (f is a “local function”). By way of the isomorphism of
Exercise 28.1.D, we can interpret f as an element of k[[x1 , . . . , xd ]]. This should be
interpreted as the “power series expansion of f at p”, in the “local coordinates x1 ,
. . . , xd ”.
 © 2024 Ravi Vakil. Published by Princeton University Press. 771

28.1.4. Some more geometric motivation.

Exercise 28.1.D may give some motivation for completion: “in the completion,
regular schemes look like affine space”. This is often stated in the suggestive
language of “formally locally, regular schemes are isomorphic to affine space”; this
will be made somewhat more precise in §28.3.1.
We now give a little more geometric motivation for completion, that we will
not use later on. Recall from §13.6.1 that étale morphisms are designed to look like
“local isomorphisms” in differential geometry. But Exercise 13.6.F showed that this
metaphor badly fails in one important way. More precisely, let Y = Spec k[t], and
X = Spec k[u, 1/u], and let p ∈ X be given by u = 1, and q ∈ Y be given by t = 1.
Suppose char k ̸= 2. The morphism π : X → Y induced by t 7→ u2 is étale, and
π(p) = q. But there is no open neighborhood of p that π maps isomorphically onto
an open neighborhood of q. However, the following Exercise shows that π induces
an isomorphism of completions (an isomorphism of “formal neighborhoods”).

28.1.E. E XERCISE . Continuing the notation of the previous paragraph, show that π
induces an isomorphism of completions O^Y,q → O^X,p .

28.1.F. E XERCISE . Suppose Y = Spec A, and X = Spec A[t]/(f(t)), and π : X → Y

is the morphism induced by A → A[t]/(f(t)). Suppose p ∈ X is a closed point,
q ∈ Y is a closed point, π(p) = q, and π is étale at p and induces an isomorphism of
residue fields. Show that π induces an isomorphism of completions O^Y,q → O^X,p .
(If you are not familiar with Hensel’s Lemma, you might rediscover its central idea
in the course of solving this exercise.)
With a little more care, you can show more generally that if π : X → Y is an
étale morphism, π(p) = q, and π induces an isomorphism of residue fields at p,
then π induces an isomorphism of completions O^Y,q → O^X,p . (You may even wish
to think about how to remove the hypothesis of isomorphism of residue fields.)
These ideas are close to the definition of “formal étaleness” discussed in
§24.8.13.
You can interpret these results as a statement that “the implicit function theorem
works formally locally”, even though it doesn’t work Zariski-locally. (The étale
topology is somewhere between these two; it is partially designed so that the
implicit function theorem in this sense always holds, essentially by fiat. But this is
not the place to discuss the étale topology.)
Here are two more fun problems secretly about Hensel’s Lemma.

28.1.G. E XERCISE . Show that x2 + y2 = −2 has solutions modulo 7100 .

28.1.H. E XERCISE . Suppose we have a plane curve f(x, y) = 0 in A2Q , such that (0, a)
is a smooth point whose tangent line is nonvertical. Show that there is precisely
one power series solution y = p(x) such that f(x, p(x)) = 0 (as a power series), and
p(0) = a.

28.1.5. Completion and exactness.

We conclude this section with an interesting and useful statement.

28.1.6. Theorem. — Suppose A is a Noetherian ring, and I ⊂ A is an ideal. For any

^ = lim M/Ij M be the completion of M with respect to I.
A-module M, let M
772 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

(a) The completion A ^ (of A with respect to I) is flat over A.

(b) If M is finitely generated, then the natural map A ^ ⊗A M → M
^ is an isomorphism.
(c) If 0 → M → N → P → 0 is a short exact sequence of finitely generated A-modules,
then 0 → M ^ →N ^ →P ^ → 0 is exact. (Thus completion preserves exact sequences of
finitely generated modules, by Exercise 1.5.F.)
We will use (a) in §28.2, and (b) in §28.7.

28.1.7. Remark. Before proving Theorem 28.1.6, we make some remarks. Parts
(a) and (b) together clearly imply part (c), but we will use (c) to prove (a) and
(b). Also, note a delicate distinction (which helps me remember the statement): if
0 → M → N → P → 0 is an exact sequence of A-modules, not necessarily finitely
generated, then

(28.1.7.1) 0 /A
^ ⊗A M /A
^ ⊗A N /A
^ ⊗A P /0

is always exact, but

(28.1.7.2) 0 /M
^ /N
^ /P
^ /0

need not be exact — and when it is exact, it is often because the modules are finitely
generated, and thus (28.1.7.2) is really (28.1.7.1).
Caution: completion is not always exact. Consider the exact sequence of k[t]-
modules
×(t,t2 ,t3 ,... )
0 / ⊕∞ k[t] / ⊕∞ k[t] / ⊕∞ k[t]/(tn ) / 0.
n=1 n=1 n=1

After completion with respect to the ideal ((t, t, t, . . . )), the sequence is no longer ex-
act in the middle: (t2 , t3 , t4 , . . .) maps to 0, but is not in the image of the completion
of the previous term.
⋆⋆ Proof. The key step is to prove (c), which we do through a series of exercises.
Suppose that 0 → M → N → P → 0 is a short exact sequence of finitely generated
A-modules.
^ →P
28.1.I. E XERCISE . Show that N ^ is surjective. Hint: consider an element of P ^ as
j
a sequence (pj ∈ P/I P)j≥0 , where the image of pj+1 is pj , cf. Exercise 1.3.C. Build
a preimage (nj ∈ N/Ij N)j≥0 by induction on j.
^ → P)
We now wish to identify ker(N ^ with M.
^

28.1.J. E XERCISE . Show that for each j ≥ 0,

(28.1.7.3) 0 / M/(M ∩ Ij N) / N/Ij N / P/Ij P /0

is exact. (Possible hint: show that 0 → M ∩ Ij N → M → N/Ij N → P/Ij P → 0 is

exact.)
The short exact sequences (28.1.7.3) form an inverse system as j varies. Its limit
is left-exact (because limits always are), but it is also right-exact by Exercise 1.5.N,
as the “transition maps on the left” M/(M ∩ Ij+1 N) → M/(M ∩ Ij N) are clearly
surjective. Thus

(28.1.7.4) 0 / lim M/(M ∩ Ij N) /N

^ /P
^ /0
 © 2024 Ravi Vakil. Published by Princeton University Press. 773

is exact. To complete the proof of (c), it suffices (by Exercise 28.1.A) to show that
the sequence of submodules Ij M is cofinal with the sequence M ∩ Ij N, so that the
^
term lim M/(M ∩ Ij N) on the left of (28.1.7.4) is naturally identified with M.

28.1.K. E XERCISE . Prove this. Hint: clearly Ij M ⊂ M ∩ Ij N. By Corollary 13.9.4 to

the Artin-Rees Lemma 13.9.3, for some integer s, M ∩ Ij+s N = Ij (M ∩ Is N) for all
j ≥ 0, and clearly Ij (M ∩ Is N) ⊂ Ij M.
This completes the proof of part (c) of Theorem 28.1.6.
For part (b), present M as

(28.1.7.5) A⊕m
α / A⊕n /M /0

where α is an m × n matrix with coefficients in A. Completion is exact in this case

by part (c), and clearly commutes with finite direct sums, so

^ ⊕m
A /A
^ ⊕n /M
^ /0

is exact. Tensor product is right-exact, and commutes with direct sums (Exer-
cise 1.2.M), so
^ ⊕m
A /A
^ ⊕n /A
^ ⊗A M /0

is exact as well. Notice that the maps from A ^ ⊕m to A^ ⊕n in both right-exact se-
quences are the same; they are both α. Thus their cokernels are identified, and (b)
follows.
Finally, to prove (a), we need to extend the ideal-theoretic criterion for flatness
(Theorem 24.3.1) slightly. Recall (§24.3.2) that it is equivalent to the fact that an
A-module M is flat if and only if for all ideals I, the natural map I ⊗A M → M is
an injection.

28.1.L. E XERCISE ( STRONGER FORM OF THE IDEAL - THEORETIC CRITERION FOR

FLATNESS ). Show that an A-module M is flat if and only if for all finitely gen-
erated ideals I, the natural map I ⊗A M → M is an injection. (Hint: if there is a
counterexample for an ideal J that is not finitely generated, use it to find another
counterexample for an ideal I that is finitely generated.)
By this criterion, to prove (a) it suffices to prove that the multiplication map
I ⊗A A^ →A ^ is an injection for all finitely generated ideals I. But by part (b), this is
the same showing that ^I → A ^ is an injection; and this follows from part (c). □

28.2 Types of singularities

Singularities are often most easily defined in terms of completions.

28.2.1. Definition. Suppose X is a dimension 1 variety over k, and p ∈ X is a closed

point. We say that X has a node at p if the completion of OX,p at mX,p is isomorphic
(as topological rings) to k[[x, y]]/(xy). (How does this compare to the definition of
node in §13.3.2?)
774 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

28.2.A. E XERCISE . Suppose k = k and char k ̸= 2. Show that the curve in A2k cut
out by y2 = x2 + x3 , which we have been studying repeatedly since Figure 8.4, has
a node at the origin.

28.2.B. E XERCISE . Suppose k = k and char k ̸= 2, and we have f(x, y) ∈ k[x, y].
(a) Show that Spec k[x, y]/(f(x, y)) has a node at the origin if and only if f has no
terms of degree 0 or 1, and the degree 2 terms are not a perfect square. (This
generalizes Exercise 28.2.A.)
(b) Show that Spec k[x, y]/(f(x, y)) has a node at the point (a, b) if and only if
 2   2   2 2
∂f ∂f ∂ f ∂ f ∂ f
= = 0 and 2 2
̸=
∂x ∂y ∂x ∂y ∂x∂y
at (a, b).
The definition of node outside the case of varieties over algebraically closed
fields is more problematic, and we give some possible ways forward. For varieties
over a non-algebraically closed field k, one can always base change to the closure
k. As an alternative approach, if p is a k-valued point of a variety over k (not
necessarily algebraically closed), then we could take the same Definition 28.2.1; this
might reasonably be called a split node, because the branches (or more precisely,
the tangent directions) are distinguished. Those singularities that are not split
nodes, but which become nodes after base change to k (such as the origin in
Spec R[x, y]/(x2 + y2 ), or Spec Q[x, y]/(x2 − 2y2 )) might reasonably be called non-
split nodes.

28.2.2. Definition. We may define other singularities similarly. To avoid complica-

tions, we do so over an algebraically closed field. Suppose X is a variety over k, and
p is a closed point of X, where char k ̸= 2, 3. We say that X has a cusp (resp., tacn-
ode, triple point) at p if O^X,p is isomorphic to the completion of k[x, y]/(y2 − x3 )
(resp., k[x, y]/(y2 − x4 ), k[x, y]/(y3 − x3 )). (See Figure 13.3 for pictures of the first
two.) We say that X has an ordinary multiple point of multiplicity m, or ordinary
m-fold point, if O^X,p is isomorphic to the completion of k[x, y]/(f(x, y)) where f
is a homogeneous polynomial of degree m with no repeated roots. (You can see
quickly that an ordinary 2-fold point — or ordinary double point — is precisely a
node, and an ordinary 3-fold point is a triple point.)

28.2.C. T RIVIAL E XERCISE . (For this exercise, work over an algebraically closed
field for simplicity.) Define An curve singularity (see §22.4.4). Define An surface
singularity (see Exercise 22.4.F). Define Dn , E6 , E7 , and E8 surface and curve
singularities (see Remark 22.4.5).

28.2.3. Using this definition. We now give an example of how this definition can
be used.

28.2.D. E XERCISE . Suppose X is a k-variety with a node at a closed point p. Show

that the blow up of X at p yields a morphism β : X̃ → X, where the exceptional
divisor β−1 (p) consists of two reduced smooth points. Hint: use the fact that
completion is flat (Theorem 28.1.6(a)), and that blowing up commutes with flat base
change (Exercise 24.1.P(a)), to turn this into a calculation on the “formal model” of
the node, Spec k[[x, y]]/(xy).
 © 2024 Ravi Vakil. Published by Princeton University Press. 775

28.2.E. E XERCISE . Continuing the terminology of the previous exercise, describe

an exact sequence
(28.2.3.1) 0 / OX / β∗ O / Op / 0.
X̃

Hint: faithful flatness.

28.2.F. E XERCISE . We continue the terminology of the previous two exercises. If X

is a pure-dimensional reduced projective curve, show that pa (X) = pa (X̃) + 1.
Thus “resolving a node” of a curve reduces the arithmetic genus of the curve
by 1. If you wish, you can readily show that “resolving a cusp” reduces the genus
by 1, and “resolving a tacnode” reduces the genus by 2. In general, for each type of
curve singularity, the contribution it makes to the genus — the difference between
the genus of the curve and that of its “partial normalization at the singular point”,
when the curve is projective — is called the δ-invariant. Thus δ for a node or cusp
is 1, and δ for a tacnode is 2.

28.2.G. E XERCISE . Show that for any singularity type (other than a smooth point),
δ > 0.

28.2.H. E XERCISE . Show that a reduced irreducible degree d plane curve can have
at most d−1


2 singularities.

28.2.I. E XERCISE . Here is an informal trick for speedily working out the genus of
a nodal (projective reduced) curve when it comes up in examples. Assuming it is
possible, draw a sketch on a piece of paper (i.e., the real plane) where nodes “look
like nodes”, and a curve of genus g is drawn with g holes, as in Figure 28.1. (In
Figure 28.1, we see four irreducible components, of geometric genus 0, 0, 1, and 3
respectively. The first three components are smooth, while the fourth meets itself
once. There are six additional nodes.) Then if the curve is connected, count the
number of holes visible in the picture. Prove that this recipe works. (Feel free to
figure out what to do if the curve is not connected.)

F IGURE 28.1. This curve has genus 8 (count the “holes”!)

28.2.4. This can be quite useful. For example, the sketches in Figure 28.2 can remind
you that a degree d plane curve has arithmetic genus d−1

1 1
2 , the curve on P × P
776 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

coming from a section of O(a, b) (§15.2.10) has genus (a − 1)(b − 1), and a curve in
class 2C + 2F on the Hirzebruch surface F2 (using the language of Exercise 20.2.M)
has arithmetic genus 3. In each case, we are taking a suitable curve in the class,
taking advantage of the fact that arithmetic genus is constant in flat families (Crucial
Exercise 24.7.A).

F IGURE 28.2. These curves have arithmetic genus 6, 6, and 3 respectively

28.2.5. Other definitions. If k = C, then definitions of this sort agree with the
analytic definitions (see [Ar1, §1]). For example, a complex algebraic curve singu-
larity is a node if and only if it is analytically isomorphic to an open neighborhood
of xy = 0 in C2 . There is also a notion of isomorphism “étale-locally”, which we
do not define here. Once again, this leads to the same definition of these types of
singularities (see [Ar2, §2]).

28.3 The Theorem on Formal Functions

Suppose π : X → Y is proper morphism of locally Noetherian schemes, and F is

a coherent sheaf on X, so Ri π∗ F is a coherent sheaf (Grothendieck’s Coherence
Theorem 18.9.1). Fix a point q ∈ Y. We already have a sense that there is an
imperfect relationship between the fiber of Ri π∗ F at q and the cohomology of F
“on the fiber Xq ”, and much of Chapter 25 was devoted to making this precise.
The Theorem on Formal Functions deals with this issue in a different way.
Rather than comparing the fiber of Ri π∗ F with the cohomology of F restricted to
 © 2024 Ravi Vakil. Published by Princeton University Press. 777

the fiber, it gives an isomorphism between information on “infinitesimal thicken-

ings” of the fiber. Informally:

Ri π∗ F / Hi (Xq , F |Xq )

+
Hi (0th order nbhd of Xq , F )
O

&
Hi (1st order nbhd of Xq , F )
O

&
Hi (2nd order nbhd of Xq , F )
O

..
.O

 
completion of Ri π∗ F at q
Thm. 28.3.2: iso / limit

28.3.1. We now make this more precise. If Z is a closed subscheme of W cut out by
ideal sheaf I , we say that the closed subscheme cut out by I n+1 is the nth order
formal (or infinitesimal) neighborhood (or thickening) of Z (in W). The phrase
“formal neighborhood” without mention of an “order” refers to the information
contained in all of these neighborhoods at once, often in the form of a limit of the
sort we will soon describe. (This also leads us to the notion of formal schemes, an
important notion we will nonetheless not need.)
We turn now to our situation of interest. Rather than the fiber of Ri π∗ F at q, we
consider its completion: we take the stalk at q, and complete it at the maximal ideal
mq . Rather than the cohomology of F along the fiber, we consider the cohomology
of F when restricted to the nth formal neighborhood Xn of the fiber, and take the
limit. The Theorem of Formal Functions says that these are canonically identified.
Before we state the Theorem on Formal Functions, we need to be precise about
what these limits are, and how they are defined. To concentrate on the essential,
we do this in the special case where q is a closed point of Y, and leave the (mild)
extension to the general case to you (Exercise 28.3.A).
We deal first with Ri π∗ F . It is helpful to note that by Exercise 28.1.B, we can
compute the restriction to the nth formal neighborhood either by pulling back to
Spec OY,q (“the germ of Y at q”, i.e., tensoring with OY,q ); or in any affine open
neighborhood of q ∈ Y. For convenience, we pick an affine open neighborhood
Spec B ⊂ Y of q. (Again, by Exercise 28.1.B, it won’t matter which affine open subset
we take.) Define XB := π−1 (Spec B) for convenience. Let m ⊂ B be the maximal
ideal corresponding to q. Then ((Ri π∗ F )q )^ is canonically the completion of the
coherent B-module Hi (XB , F |XB ) at m:

^
(Ri π∗ F )q = lim Hi (XB , F |XB )/mn Hi (XB , F |XB ).
778 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

We turn next to the cohomology of F on thickenings of the fiber. We have

closed embeddings Xn ,→ Xn+1 , and thus maps of cohomology groups

Hi (Xn+1 , F |Xn+1 ) / Hi (Xn , F |Xn ).

We have base change maps

(28.3.1.1) Hi (XB , F |XB )/mn Hi (XB , F |XB ) / Hi (Xn , F |Xn )

(see (25.1.0.3)) such that the square

Hi (XB , F |XB )/mn+1 Hi (XB , F |XB ) / Hi (Xn+1 , F |X )

n+1

 
Hi (XB , F |XB )/mn Hi (XB , F |XB ) / Hi (Xn , F |Xn )

commutes. (Do you see why? Basically, this is again because they are base change
maps.) Thus we have an induced map of limits:

/ limn→∞ Hi (Xn , F |X ).

^
(28.3.1.2) (Ri π∗ F )q n

28.3.A. E XERCISE . Extend the previous discussion to the case where q is not a
closed point of Y.
The Theorem on Formal Functions states that this is an isomorphism.

28.3.2. Theorem on Formal Functions. — Suppose π : X → Y is a proper morphism of

locally Noetherian schemes, F is a coherent sheaf on X, and q ∈ Y. Then (28.3.1.2) is an
isomorphism for all i ≥ 0.
Warning: the Theorem on Formal Functions does not imply anything about the
maps “at finite level”, i.e., about (28.3.1.1) for “finite n”.
The proof of Theorem 28.3.2 is subtle, and is postponed to the double-starred
§28.7. We first give four important applications to keep you motivated and excited:
Zariski’s Connectedness Lemma 28.4.1, Stein factorization 28.4.3, Zariski’s Main
Theorem 28.5.1, and Castelnuovo’s Criterion 28.6.1 for contracting (−1)-curves on
surfaces.

28.4 Zariski’s Connectedness Lemma and Stein factorization

We now state and prove Zariski’s Connectedness Lemma, which was men-
tioned in §25.1.8.

28.4.1. Zariski’s Connectedness Lemma. — If a proper morphism π : X → Y of locally

Noetherian schemes is O-connected (see §25.1.7), then π−1 (q) is connected for every q ∈ Y.

The proof requires the following result.

28.4.A. E ASY E XERCISE ( THE COMPLETION OF A LOCAL RING IS A LOCAL RING ).

Suppose (A, m) is a Noetherian local ring. Show that the completion of A along
 © 2024 Ravi Vakil. Published by Princeton University Press. 779

^ := lim A/mn , is a local ring, with maximal ideal mA.

m, A ^ Hint: show that any
^ not in mA
element of A ^ is invertible.

Proof. Assume otherwise

` that there is some q ∈ Y such that π−1 (q) is not connected,
−1
say π (q) = X1 X2 (where X1 and X2 are nonempty open subsets of the fiber,
see Figure 28.3). Then the nth order formal neighborhood of π−1 (q), having the
same topological space, is also disconnected. We use the useful trick of idempotents
(Remark 3.6.3). Let e1 be the function 1 on X1 and 0 on X2 , and let e2 be the function
1 on X2 and 0 on X1 . These functions make sense for any order formal neighbor-
hood, and they have natural images in the inverse limit. By the isomorphism of
the Theorem on Formal Functions 28.3.2 (applied to H0 ), we get nonzero elements
e1 , e2 ∈ O^Y,q with e1 + e2 = 1, e1 e2 = 0. Now, O^Y,q is a local ring (Exercise 28.4.A).
But e1 e2 = 0 implies that neither e1 nor e2 is invertible (or else the other one would
be 0), so both e1 and e2 are in the maximal ideal, and hence e1 + e2 can’t be 1. We
thus have a contradiction. □

F IGURE 28.3. “Formal neighborhoods” of X1 and X2 in the proof

of Zariski’s Connectedness Lemma 28.4.1

28.4.2. Stein factorization. We next describe the construction known as “Stein fac-
torization”. We could have defined it long before, but now Zariski’s Connectedness
Lemma 28.4.1 will give us some impressive consequences.

28.4.3. Stein Factorization Theorem. — Any proper morphism π : X → Y of locally

Noetherian schemes can be factored into β ◦ α, where α : X → Y ′ is proper and O-connected
(hence has connected fibers, by Zariski’s Connectedness Lemma 28.4.1), and β : Y ′ → Y is
a finite morphism.
O-conn. / Y′
(28.4.3.1) X α
β
π
  finite
Y
We note that by the Cancellation Theorem for projective morphisms (Exer-
cise 17.3.C), if π happens to be projective, then so is α (as δβ , being a closed
embedding, is projective).
Although it is not in the statement of the theorem, the proof produces a specific
factorization, which is called the Stein factorization of π. The picture to have in
780 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

F IGURE 28.4. Stein Factorization

mind is that the Stein factorization (roughly) contracts the continuous parts of the
fibers, in some canonical way; see Figure 28.4.
As usual, the Noetherian hypotheses can be removed (see [Stacks, tag 03H2]).
Proof of the Stein Factorization Theorem 28.4.3. By Grothendieck’s Coherence The-
orem 18.9.1, π∗ OX is a coherent sheaf (of algebras) on Y. Define Y ′ := Spec π∗ OX ,
so (as π∗ OX is finite type) the structure morphism β : Y ′ → Y is finite. We have a
factorization
X / Y′
α
β
π
  finite
Y
By the Cancellation Theorem 11.1.1 for proper morphisms, α is proper. Finally, we
show that α is O-connected, i.e., that OY ′ → α∗ OX is an isomorphism. Because
β is an affine morphism (so β∗ is an exact functor CohY ′ → CohY ), it suffices to
check after applying the pushforward β∗ . But β∗ OY ′ → β∗ α∗ OX is the identity
map π∗ OX → π∗ OX (here we use that β∗ OY ′ = π∗ OX , see §17.1.4, and β∗ α∗ OX =
π∗ OX ). □

28.4.B. E XERCISE . Suppose π : X → Y is a proper birational morphism of irre-

ducible varieties. Suppose further that Y is normal. Show that π is O-connected.
(This applies, for example, to blow-ups of smooth varieties under smooth centers.)
Hence, by Zariski’s Connectedness Lemma 28.4.1, π has connected fibers. This
is called Zariski’s Main Theorem (the version for birational morphisms). Hint:
Exercise 10.7.P.

28.4.C. E XERCISE . Suppose π : X → Y and π ′ : X → Y ′ are two proper birational

morphisms of integral locally Noetherian schemes, each contracting the same
connected set of X to a point (and an isomorphism elsewhere), and suppose Y and
Y ′ are normal. Show that π and π ′ are the same (or more precisely, that there is an
∼
isomorphism i : Y −→ Y ′ such that π ′ = i ◦ π). Informal translation: If X → Y has
connected fibers, then Y is determined by just knowing the locus contracted on X.
Hint: Identify Y and Y ′ as topological spaces, and then identify π and π ′ as maps of
topological spaces. Use Stein factorization to recover the structure sheaf as π∗ OX .
 © 2024 Ravi Vakil. Published by Princeton University Press. 781

28.4.D. E XERCISE . Show that the construction of the Stein factorization of a

morphism π : X → Y of locally Noetherian schemes commutes with any flat base
change Z → Y of locally Noetherian schemes.

28.4.4. Resolution of singularities of curves.

If C is a reduced projective curve over a field k, then we can resolve its singu-
larities by normalization: ν : C̃ → C. (See Remark 22.4.6 for some discussion of
resolution of singularities, and Theorem 13.5.8 for the reason why normal curves
are regular.) But as we have seen from examples, it requires luck and insight to
figure out how to take an integral closure, and I often have neither. We now take the
(inspired) guesswork out of desingularization by explaining how to desingularize
by blowing up. The algorithm is simple: we find a singular point, then blow it
up, then look for more singular points to repeat the process. Clearly if there are
no singular points to be found, then we are done; the problem is to show that
this is guaranteed to terminate. We do this by making use of an integer invari-
ant: the arithmetic genus. We must show (i) that if pa (C) = pa (C̃), then ν is an
isomorphism — C is already nonsingular; (ii) if C ′ → C is a birational morphism,
then pa (C ′ ) ≤ pa (C) (so pa (C̃) will be a lower bound for the arithmetic genus
throughout the process); and (iii) that blowing up a singular point decreases the
arithmetic genus.
To set up these results, we consider the following situation. Suppose π : C ′ → C
is a finite morphism (where C is as described in the previous paragraph, and C ′ is a
reduced curve), that is an isomorphism away from a finite closed subset of C (hence
π is birational). Then the pullback map OC → π∗ OC ′ has no kernel, as C is reduced,
and the map is an isomorphism on the generic points of the components of C. The
cokernel G is supported on a finite set of closed points, and thus Hi (C, G ) = 0 for
i > 0, and if H0 (C, G ) = 0, then G = 0. From the exact sequence

0 / OC / π∗ OC ′ /G / 0,

and the fact that π is affine, χ(C ′ , OC ′ ) = χ(C, OC ) + χ(C, G ), from which pa (C) ≥
pa (C ′ ), with equality if and only if π is O-connected (i.e., G = 0).

28.4.E. E XERCISE . Show that if π (as above) is O-connected, then π is an iso-

morphism. Hint: show that affine O-connected morphisms are isomorphisms.

To complete our strategy, it remains to show that if β : C ′ → C is the blow-up of

C at a point p, and β is an isomorphism, then C is already nonsingular at that point.
But if β is an isomorphism, then p is the exceptional divisor, and hence effective
Cartier, and hence cut out locally by a singular equation. But then mC,p ⊂ OC,p is
principal, so OC,p is a regular local ring of dimension 1.
(You should verify that we have completed our strategy. Can you extend this
argument to the case where C is not projective?)
Resolution of singularities for surfaces will be briefly discussed in §28.6.4.

28.5 Zariski’s Main Theorem

782 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Zariski’s Main Theorem is misnamed in all possible ways (although not horribly
so). It is not a single Theorem, because there are many results that go by this name,
and they often seem quite unrelated. What they have in common is that they
are intellectual descendants of a particular result of Zariski. This result is not the
Main Theorem of Zariski’s prolific career; the name comes because it was the Main
Theorem of a particular paper, [Z]. And finally, it is not Zariski’s in any stronger
sense than this; the modern versions are due to Grothendieck.
We have already seen one version of Zariski’s Main Theorem: the “birational”
form of Exercise 28.4.B. Our goal in this section is to prove a stronger statement
that is close to the optimal version.
Before we start, we note that the fiber of a proper morphism over a point q ∈ Y
is a proper (hence finite type) scheme over the residue field κ(q), and we say that a
point is isolated in its fiber if it forms a component of the fiber of dimension 0. (Here
component means both connected and irreducible component.)

28.5.1. Zariski’s Main Theorem (Grothendieck’s form). — Suppose π : X → Y is a

proper morphism of locally Noetherian schemes.
(a) The set of points of X that are isolated in their fiber forms an open subset X0 ⊂ X.
(b) The morphism π|X0 : X0 → Y factors into an open embedding X0 ,→ Y ′ followed by a
finite morphism Y ′ → Y.
(c) The morphism π : X → Y factors through Y ′ .

(28.5.1.1) ?X
open α

/ 
X0  y π open 3 Y′

 
π|X0 β finite
Y

As with Stein factorization, the proof of the theorem yields a specific factor-
ization. Indeed, X → Y ′ = Spec π∗ OX → Y is the Stein factorization of π : X → Y,
giving (c) and part of (b) basically immediately.
If π is a morphism of varieties, we already know part (a), by upper semicontinu-
ity of fiber dimension Theorem 12.4.3(a). For this reason, the proof in the case of
varieties is easier, and (even if you are interested in the more general case) you are
advised to read the proof restricting to this simpler case, to concentrate on the main
ideas. But even in the case of varieties, we will need some of the ideas arising in
the proof of (a), so you should read it until advised to skip ahead to (b).

Proof. (a) Take the Stein factorization (28.4.3.1) of π:

X
O-conn.
α
π Y′
β

 ~ finite
Y.
 © 2024 Ravi Vakil. Published by Princeton University Press. 783

By the finiteness of β, a point of X is an isolated point in its fiber of π if and only if

it is an isolated point in its fiber of α. Thus we may replace π by α. (Basically, we
are reducing to the case where π is O-connected, but we prefer to carefully call the
morphism under consideration α so we can return to considering π in the proof of
(b).)
Because fibers of proper O-connected morphisms of locally Noetherian schemes
are connected (Zariski’s Connectedness Lemma 28.4.1), a point is isolated in its
fiber of α if and only if it is its fiber.
If you are considering only the variety case, at this point you should jump to
the proof of part (b).
To show that X0 is open, we will identify it with the locus where Ωα = 0 (the
unramified locus of α — not of π!), which is open by Exercise 21.7.H.

28.5.A. E XERCISE . If Ωα |p = 0, show that p is isolated in its fiber of α (and hence

of π).
We now show the other direction. Suppose p is isolated in its fiber, and let
∼
r ∈ Y ′ be α(p). We will show that α induces an isomorphism α♯ : OY ′ ,r −→ OX,p , so
ΩOX,p /OY ′ ,r = 0. As the construction of Ω behaves well with respect to localization
on the source and target (Exercise 21.2.L), this implies that (Ωα )p = 0, so Ωα is 0 at
p.
The stalk OY ′ ,r is obtained by taking the colimit (of sections of OY ′ ) over all
open subsets of Y ′ containing r. For any such open subset V ⊂ Y ′ , the condition
∼
α∗ OX = OY ′ gives an isomorphism Γ (V, OY ′ ) ←→ Γ (α−1 (V), OX ) (compatible with
inclusions). We will show that the system of open subsets α−1 (V) (as V varies
through open neighborhoods of r), which each necessarily contain p, is cofinal with
the system of open neighborhoods of p in X. To do this, we must show that for any
open neighborhood U of p in X, there is an open neighborhood V of r in Y ′ such
that α−1 (V) ⊂ U. To do this, note that X \ U is closed in X, so α(X \ U) is closed in
Y ′ (as α is proper), so its complement V := Y ′ \ α(X \ U) is open. But α−1 (V) ⊂ U,
so we are done.
(b) (set) In our proof of (a), we have established that α gives a bijection between
X0 and an open subset of Y ′ .
(topology) We have basically also showed that α gives a homeomorphism:
any open subset of Y ′ pulls back to an open subset of X0 by the continuity of
α; and if U is an open subset of X0 , then α(X \ U) is a closed subset of Y ′ , so
its complement (which is α(U), using that proper O-connected morphisms are
surjective, Exercise 25.1.H) is open.
(sheaf) Using this isomorphism of topological spaces, the condition of O-
connectedness of α (restricted to α(X0 )) shows that α gives an isomorphism of
∼
ringed spaces (i.e., of schemes) X0 −→ α(X0 ).
We have thus found our desired factorization. (Part (c) is immediate from our
construction.) □

28.5.B. E XERCISE . Prove Zariski’s Main Theorem with slightly different hypothe-
ses: with Y affine, say Spec A, and X a quasiprojective A-scheme. (Hint: by the
Definition 4.5.10 of quasiprojective A-scheme, we can find an open embedding
X ,→ X ′ into a projective A-scheme. Apply Theorem 28.5.1 to X ′ → Spec A.)
784 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

As an application of Zariski’s Main Theorem, we can finally prove a characteri-

zation of finite morphisms that we have mentioned a number of times.

28.5.2. Theorem. — Suppose π : X → Y is a morphism of locally Noetherian schemes.

The following are equivalent.
(a) π is finite.
(b) π is affine and proper.
(c) π is proper and quasifinite.

As usual, Noetherian hypotheses can be removed: [Gr-EGA, IV3 .8.11.1] shows

that proper, quasifinite, locally finitely presented morphisms are finite.

Proof. Finite morphisms are affine by definition, and proper by Proposition 11.4.3,
so (a) implies (b).
To show that (b) implies (c), we need only show that affine proper morphisms
have finite fibers (as the “finite type” part of the definition of quasifiniteness is
taken care of by properness). This was shown in Exercise 18.9.A.
Finally, we assume (c), that π is proper and quasifinite, and show (a), that π is
finite. By Zariski’s Main Theorem 28.5.1, we have a factorization (28.5.1.1).


X0 
open
/ Y′
γ

π finite
 
Y

Now X = X0 . (Do you see why? Hint: Exercise 8.4.D followed by Exercise 8.3.G.)
By the Cancellation Theorem 11.1.1 for proper morphisms (using that Y ′ → Y is
finite hence separated), γ is proper.

28.5.C. E XERCISE . Show that proper quasicompact open embeddings are closed
embeddings. Hint: use Corollary 9.4.5 (for quasicompact morphisms, the closure
of the image is the image of the closure) to show that the image is open and closed
in the target.
Applying this to our situation, we see that X → Y is the composition of two
finite morphisms γ : X ,→ Y ′ and Y ′ → Y, and is thus finite itself. □

28.5.3. Nagata’s compactification theorem. For the final two applications (Exer-
cises 28.5.E and 28.5.F), we will need an annoying hypothesis, which is foreshad-
owed by the previous exercise. We say that a morphism π : X → Y of locally
Noetherian schemes satisfies (†) if for all affines Ui in some open cover of Y, the
restriction π|π−1 (Ui ) : π−1 (Ui ) → Ui factors as an open embedding into a scheme
proper over Ui . A morphism satisfying (†) is necessarily separated and finite type.
It turns out that this is sufficient: Nagata’s Compactification Theorem states that
every separated finite type morphism of Noetherian schemes π : X → Y can be
factored into a dense open embedding into a scheme proper over Y. (See [Lüt]
for a proof. The Noetherian hypotheses can be replaced by the condition that Y is
quasicompact and quasiseparated, see [Co].)
 © 2024 Ravi Vakil. Published by Princeton University Press. 785

28.5.D. E XERCISE . Show that if π : X → Y is a morphism of varieties over k, and

X is an open subset of a proper variety Z, then π satisfies (†). (Hint: X is an open
subscheme of its closure in Z ×k Y.)

28.5.E. E XERCISE . Suppose π : X → Y is a quasifinite birational morphism of

integral locally Noetherian schemes, and that Y is normal. Suppose further π
satisfies (†). Show that π is an open embedding. (Hence if furthermore π is a
bijection, then π must be an isomorphism.) Hint: the condition of being an open
embedding is local on the target.

28.5.F. E XERCISE . Suppose π : X → Y is a separated quasifinite morphism of

Noetherian schemes.
(a) Assuming Nagata’s Compactification Theorem (§28.5.3), show that π factors as
an open embedding into a finite morphism.

X
open
/ Y′

π
  finite
Y

(b) Assuming instead that π satisfies (†), show that π “locally” factors into an open
embedding into a finite morphism, i.e., that there is a open cover Ui of Y such that
over each Ui , there is such a factorization.
This justifies thinking of quasifinite morphisms as “open subsets of finite
morphisms”, as promised in §8.3.12.

28.5.4. ⋆⋆ Other versions of Zariski’s Main Theorem.

The Noetherian conditions in Theorem 28.5.1 can be relaxed (see [Stacks, tag
02LQ], [Gr-EGA, IV3 .8.12.13], or [GW, Thm. 12.73]): if π : X → Y is a separated
morphism of finite type, and Y is quasicompact and quasiseparated, then the set
of points X0 isolated in their fiber is open in X, and for every quasicompact open
subset U of X0 , there is a factorization

U / Y′
γ
β
π|U
  finite
Y

such that β is finite, γ|U is a quasicompact open embedding, and γ−1 (γ(U)) = U.
There are also topological and power series forms of Zariski’s Main Theorem,
see [Mu7, §III.9].

28.6 Castelnuovo’s Criterion for contracting (−1)-curves

We showed in Exercise 22.4.O that if we blow up a regular surface at a (reduced)

point, the exceptional divisor is a (−1)-curve (see Definition 20.2.8: it is isomorphic
to P1 , and has normal bundle OC (−1)). Castelnuovo’s Criterion is the converse: if
we have a quasiprojective surface containing a (−1)-curve, that surface is obtained
786 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

by blowing up another surface at a reduced regular point. We say that we can blow
down the (−1)-curve.

28.6.1. Theorem (Castelnuovo’s Criterion). — Let C ⊂ X be a (−1)-curve on a smooth

projective surface over k. Then there exists a birational morphism π : X → Y such that Y is
a smooth projective surface, π(C) is a k-valued point, π is the blow-up of Y at that point,
and C is the exceptional divisor of the blow-up.
By Exercise 28.4.C, there is only one way to “blow down C”: this contraction is
unique.

Proof. The proof is in three steps. Step 1. We construct π : X → Y that contracts C to

a point q (and is otherwise an isomorphism). At this point we will know that Y is a
projective variety. Step 2. Then we show that Y is smooth; this is the hard step, and
requires the Theorem on Formal Functions 28.3.2. Step 3. Finally, we recognize π as
the blow-up Blq Y → Y.
Step 1. We do this by finding an invertible sheaf L whose complete linear
series will do the job. We start with a very ample invertible sheaf H on X such
that H1 (X, H ) = 0. We can find such an invertible sheaf H by choosing any
very ample invertible sheaf, and then taking a suitably large multiple, invoking
Exercise 16.2.C (very ample ⊗ very ample = very ample) and Serre vanishing
(Theorem 18.1.3(ii)). Let d = H · C := degC H |C . We have d > 0 by the very
ampleness of H (Exercise 20.1.K).

28.6.A. E XERCISE . Show that H1 (X, H (iC)) = 0 for 0 ≤ i ≤ d. Hint: use induction
on i, using

(28.6.1.1) 0 / H ((i − 1)C) / H (iC) / H (iC)|C / 0.

∼ P1 ; which O(n) is it?

Hint: Note that H (iC)|C is a line bundle on C =
Define L := H (dC). We will use the sections of H to obtain sections of L ,
via the map ϕ : H0 (X, H ) → H0 (X, H (dC)).

28.6.B. E XERCISE . Show that L is base-point-free. Hint: To show that L has no

base points away from C, consider the image of ϕ. To show that L has no base
points on C, use that from (28.6.1.1) for i = d, we have that H0 (X, L ) → H0 (C, L |C )
is surjective, and L |C =∼ OC .

Thus the complete linear series |L | yields a morphism π ′ from X to some

projective space PN .

28.6.C. E XERCISE . Show that π ′ precisely contracts C to a point q ′ . More explicitly,

show that π ′ sends C to a point q ′ , and π ′ |X\C : X \ C → PN \ {q ′ } is a closed
embedding. Hint: To show that π ′ gives a closed embedding away from C: note
that |H | separates points and tangent vectors, and argue that |L | = |H (dC)|
separates points and tangent vectors away from C. To show that C is contracted by
π ′ , use the fact that deg L |C = 0.
Let Y ′ be the image of π ′ , so we have a morphism X → Y ′ (which we also
call π ′ ). Let ν : Y → Y ′ be the normalization of Y ′ , so π ′ lifts to π : X → Y by the
 © 2024 Ravi Vakil. Published by Princeton University Press. 787

universal property of normalization (§10.7). (We are normalizing in order to use

Exercise 28.4.B in Step 2.)

28.6.D. E ASY EXERCISE . Show that ν−1 (q ′ ) ⊂ Y is a single point, which we call q;
and that π(C) = q. (Hint: the image of π is closed, and π(C) is connected.)

28.6.E. E XERCISE . Let X be a smooth projective surface, containing a curve C ⊂ X

isomorphic to P1k . If C · C := degC O(C)|C < 0, show that there is a morphism
π : X → X ′ contracting C to a point, and leaving the rest of X unchanged. (This will
not be used, and is included to give practice with the argument of Step 1.)
Step 2. We now show that Y is a smooth surface. As Y is normal, we have
that π∗ OX = ∼ OY (π : X → Y is O-connected), by Exercise 28.4.B. We will show
that Y is smooth at q by showing that O^Y,q = ∼ k[[x, y]] as topological rings, as then
2
dimk (m/m ) = 2. (Do you see why this shows that Y is smooth, not just regular?)
By the Theorem on Formal Functions 28.3.2, O^Y,q = ∼ limn→∞ H0 (C ′ , OC ′ ), where
n n
′
Cn is the closed subscheme of X that is the scheme-theoretic preimage of the nth
order formal neighborhood of q.
We are not precisely sure what the nth order formal neighborhood is, but that
is fine; this inverse system is cofinal with O(−nC) = IC/X n
(Do you see why?), so
we can take this inverse limit instead (by Exercise 28.1.A). It suffices to show that
∼ k[[x, y]]/(x, y)n where Cn is defined by
for all n ≥ 0, H0 (Cn , OCn ) =

0 / O(−nC) / OX / OC / 0.
n

(Be careful: we do not know that Cn = Cn′ or even that C1 = C1′ . For all we know,
the fiber over q might be nonreduced!)
We do this by induction on n. As a k-vector space, this is easy, using

(28.6.1.2) 0 / I n /I n+1 / OC / OC /0
n+1 n

(the closed subscheme exact sequence (9.1.2.1) for Cn in Cn+1 ), using the canonical
∼
isomorphism H0 (P1 , OP1 (n)) ←→ Symn H0 (P1 , OP1 (1)) . The tricky thing is that

we want this isomorphism as rings.

28.6.F. E XERCISE . Prove this. (This may remind you of how we found the ring
of functions on the total space of an invertible sheaf in Exercise 19.11.E. This is no
coincidence.) Hint: interpret I n /I n+1 as Symn (I /I 2 ) (Corollary 22.3.9) and
in turn as OP1 (n), and show that upon taking global sections, (28.6.1.2) remains
exact. At this point you will have shown that some sort of associated graded ring of
H0 (Cn , OCn ) is isomorphic to k[[x, y]]/(x, y)n . Show that this is enough to establish
the desired isomorphism of rings.
Hence we have smoothness, completing Step 2.
Step 3. We now must recognize π as the blow-up of Y at q. As C is an effective
Cartier divisor on X, by the universal property of blowing up means that there is a
788 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

unique morphism α making the following diagram commute

X
α / Blq Y

π β
 }
Y

By the Cancellation Theorem for projective morphisms (Exercise 17.3.C), α is pro-

jective.

28.6.G. E XERCISE . Show that α has finite fibers.

As projective morphisms with finite fibers are finite (Theorem 18.1.6), α is a
finite morphism. But then α is a birational finite morphism to a normal scheme,
and thus an isomorphism by Exercise 10.7.P. □
So far we have only used the Theorem on Formal Functions 28.3.2 for H0 . The
next exercise will give you a chance to see it in action for other cohomology groups.

28.6.H. E XERCISE .
(a) Suppose β : X = Blq Y → Y is the blow-up of a smooth surface over k at a smooth
k-valued point. Show that R1 β∗ OX = R2 β∗ OX = 0.
(b) Show that the natural maps β∗ : H1 (Y, OY ) → H1 (X, OX ) and β∗ : H2 (Y, OY ) →
H2 (X, OX ) are isomorphisms.

28.6.2. Elementary transformations.

Recall the definition of the Hirzebruch surfaces Fn (over a field k) from Exam-
ple 17.2.6. If n > 0, then on Fn , we have a unique curve E of maximally negative
self-intersection (Exercise 20.2.R). It has self-intersection E · E := degE O(E)|E = −n.
It is a section of the projection π : Fn → P1 . (If n = 0, so Fn = ∼ P1 × P1 , E is usually
defined as any “constant” section of π : P × P → P — it is not unique.)
1 1 1

Suppose q is a k-valued point of Fn . Let F be the fiber of π containing q.

28.6.I. E XERCISE . Show that F · F := degF O(F)|F = 0.

Let β : Blq Fn → Fn be the blow-up at q, and let Z be the exceptional divisor
(as the name “E” is taken). Let F ′ be the proper transform of F, and let E ′ be the
proper transform of E.

28.6.J. E XERCISE . Show that F ′ · F ′ := degF ′ O(F ′ )|F ′ = −1.

Thus by Castelnuovo’s Criterion, we can blow down F ′ , to obtain a new surface
Y.

28.6.K. E XERCISE . Suppose n > 0. If q ∈ / E, show that Y = ∼ Fn−1 . If q ∈ E,

show that Y = ∼ Fn+1 . Possible hint: If you knew that Y was a Hirzebruch surface
Fm , you could recognize it by the self-intersection of the unique curve of negative
self-intersection if m > 0.
∼ F1 . Hint: Exercise 22.4.K.
28.6.L. E XERCISE . Suppose n = 0. Show that Y =
This discussion can be generalized to P1 -bundles over more general curves.
The following is just a first step in the story.
 © 2024 Ravi Vakil. Published by Princeton University Press. 789

28.6.M. E XERCISE . Suppose C is a smooth curve over k, π : X → C is a P1 -bundle

over C. If q is a k-valued point of X, then we can blow up q and blow down the
proper transform of the fiber through q. Show that the result is another P1 -bundle
X ′ over C. (This is called an elementary transformation of the ruled surface.)

28.6.N. E XERCISE . Suppose X = Proj C (L ⊕ M ), and q lies on the section corre-

sponding to
0 /L / L ⊕M /M / 0.
′ ∼
Show that X = Proj C (L (π(q)) ⊕ M )).
Can you speculate (or show) what happens if you are just told that X =∼ Proj V ,
C
where V is a locally free sheaf of rank 2, and q lies on a section corresponding to
(20.2.9.1)?

28.6.3. Minimal models of surfaces.

A surface over an algebraically closed field is called minimal if it has no (−1)-
curves. If a surface X is not minimal, then we can choose a (−1)-curve, and blow it
down. If the resulting surface is not minimal, we can again blow down a (−1)-curve,
and so on. By the finiteness of the Picard number ρ (which we have admittedly
not proved, see §18.4.12), as blowing down reduces ρ by 1 (by Exercise 22.4.O),
this process must terminate. By this means we construct a minimal model of X —
a minimal surface birational to X. If minimal models were unique, this would
provide a means of classifying surfaces up to birationality: if X and X ′ were smooth
projective surfaces, to see if they were birational, we would find their minimal
models, and see if they were isomorphic.
Sadly, minimal models are not unique. The example of P1 ×P1 and P2 show that
there can be more than one minimal model in a birational equivalence class. (Why
are these two surfaces minimal?) Furthermore, the isomorphism between the blow-
up of P2 at two points with the blow-up of P1 × P1 at one point (Exercise 22.4.K)
shows that a smooth surface can have more than one minimal model.
Nonetheless, the failure of uniqueness is well-understood: it was shown by
Zariski that the only surfaces with more than one minimal model are those bira-
tional to C × P1 for some curve C. This is part of the Enriques-Kodaira classification
of surfaces. For more on this, see [Be], [Ba], or [BPV, §6].

28.6.4. Here are two more facts worth mentioning but not proving here. First, sin-
gular surfaces can be desingularized by normalizing and then repeatedly blowing
up points; this explicit desingularization is analogous to the version for curves in
§28.4.4, and is generalized by Hironaka’s Theorem on resolution of singularities
(Remark 22.4.6). Second, birational maps X o / Y between smooth surfaces can
be factored into a finite number of blow-ups followed by a finite number of blow-
downs (cf. §22.4.12, on weak factorization). As a consequence, by Exercise 28.6.H,
birational projective surfaces have the same Hi (O) for i = 0, 1, 2. (Using facts stated
in §21.5.10, you can show that over C, this implies that the entire Hodge diamond
except for h1,1 is a birational invariant of surfaces.)

28.7 ⋆⋆ Proof of the Theorem on Formal Functions 28.3.2

790 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

In this section, we will prove the following.

28.7.1. Theorem. — Suppose X → Spec B is a proper morphism, with B Noetherian,

and F is a coherent sheaf on X. Suppose I is an ideal of B, and B ^ is the I-adic completion
lim B/In of B. Let Xn := X ×B (B/In+1 ) (the nth order formal neighborhood of the fiber
X0 ) and let Fn be the pullback of F to Xn . Then for all i, the natural map

(28.7.1.1) Hi (X, F ) ⊗B B (X, F )

^ = Hi\ / lim Hi (Xn , Fn )

is an isomorphism.
This proof is due to Serre, by way of B. Conrad.

28.7.A. E XERCISE . Prove that Theorem 28.7.1 implies Theorem 28.3.2. (This is
essentially immediate if q is a closed point, and might require a little thought if q is
not closed; cf. Exercise 28.3.A.)
On the right side of (28.7.1.1), we have morphisms X1 ,→ X2 ,→ · · · ,→ X, from
which we get restriction maps Hi (X1 , F1 ) ← Hi (X2 , F2 ) ← · · · ← Hi (X, F ). But
notice that the Xi are all supported on the same underlying set. It is helpful to keep
in mind that the Fi ’s are all sheaves on the same topological space.
Our proof will only use properness through the fact that the pushforward of
coherent sheaves under proper morphisms are coherent (Grothendieck’s Coherence
Theorem 18.9.1), which required hard work. You can retreat to the projective case
with little loss.
Write In F for the coherent sheaf on X defined as you might expect: if I is
the pullback of eI from Spec B to X, In F is the image of I n ⊗ F → F . Then
Hi (Xn , Fn ) can be interpreted as Hi (X, F /In F ) (by §18.1 property (v)), so the
terms in the limit on the right side of (28.7.1.1) all live on the single space X. From
now on we work on X.

28.7.2. The idea. The map (28.7.1.1) in the statement of the theorem arises from
the restriction Hi (X, F ) → Hi (X, F /In F ). We are led to consider the long exact
sequence associated to

0 / In F /F / F /In F / 0,

and in particular the portion

(28.7.2.1)
Hi (X, In F )
an
/ Hi (X, F ) bn
/ Hi (X, F /In F ) cn
/ Hi+1 (X, In F ) dn
/ Hi+1 (X, F ).

The map

(28.7.2.2) Hi (X, F )/ im(an )

βn
/ Hi (X, F /In F )

resembles the map (28.7.1.1) in the statement of the Theorem, and we should hope
that the limit (as n → ∞) of (28.7.2.2),

(28.7.2.3) limn→∞ Hi (X, F )/ im(an )

β
/ limn→∞ Hi (X, F /In F )

should be (28.7.1.1). Indeed the right sides are the same: limn→∞ Hi (X, F /In F ) =
limn→∞ Hi (Xn , Fn ).
 © 2024 Ravi Vakil. Published by Princeton University Press. 791

28.7.3. The plan. Thus to prove the result, we wish to show two things.
(i) We have an identification
lim Hi (X, F )/ im(an ) = lim Hi (X, F ) ⊗B B.
^
n→∞ n→∞

(ii) The map β of (28.7.2.3) is an isomorphism.

28.7.4. Step (i). By Theorem 28.1.6(b), we have the isomorphism

∼ / limn→∞ Hi (X, F )/In Hi (X, F ).
limn→∞ Hi (X, F ) ⊗B B
^

So we wish to show that

lim Hi (X, F )/ im(an ) = lim Hi (X, F )/In Hi (X, F ).
n→∞ n→∞

To do this, we show that (im(an )) and (In Hi (X, F )) are cofinal sequences of sub-
groups of Hi (X, F ) (Exercise 28.1.A).
First, we must show that for any m, there is some N(m) such that IN(m) Hi (X, F ) ⊂
im(am ). This is straightforward, as we can take N(m) = m: for any x ∈ Im ,

Hi (X, F )
×x
/ Hi (X, F )
8
am
×x
&
H (X, I F )
i m

commutes.
Second, we must show that for any m, there is some N(m) such that im(aN(m) ) ⊂
Im Hi (X, F ). For this, we use a devious construction. Recall the Rees algebra
B(I)• = B ⊕ I ⊕ I2 ⊕ · · · (§13.9.1). Consider the Cartesian diagram

Xo
affine
(28.7.4.1) Spec X (OX (I)• ) X′
proper proper
 
Spec B o Spec B(I)•
affine

where we define X ′ := Spec X (OX (I)• ) for convenience. The quasicoherent sheaf
F ⊕ IF ⊕ I2 F ⊕ · · · on X is a finitely generated sheaf of modules over OX (I)• , and
thus (by Exercise 17.1.E) corresponds to a coherent sheaf F ′ on X ′ . As X ′ → X is
affine (and property §18.1(v)), we have
(28.7.4.2) Hi (X ′ , F ′ ) = Hi (X, F ⊕IF ⊕I2 F ⊕· · · ) = Hi (X, F )⊕Hi (X, IF )⊕· · · .
By Grothendieck’s Coherence Theorem 18.9.1, Hi (X ′ , F ′ ) is a finitely generated
B(I)• -module. By Proposition 13.9.2, IHi (X, Ik F ) = Hi (X, Ik+1 F ) for k ≥ k0 for
some k0 . Then
Hi (X, Ik0 +m F ) = Im Hi (X, Ik0 F ) ⊂ Im Hi (X, F )
so we may take N(m) = k0 + m, completing Step (i).

28.7.5. Step (ii). From (28.7.2.1) we have a short exact sequence

(28.7.5.1)
0 / Hi (X, F )/ im(an ) / Hi (X, F /In F ) / ker(dn ) / 0.
792 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Apply limn→∞ to both sides. Limits are left-exact (§1.5.14), so the result is
left-exact. Even better, the transition maps of the left term for our exact sequences
Hi (X, F )/ im(an+1 ) / Hi (X, F )/ im(an )

are clearly surjective, so by Exercise 1.5.N we have an exact sequence

(28.7.5.2)
0 / lim Hi (X, F )/ im(an ) β
/ lim Hi (X, F /In F ) / lim ker(dn ) / 0.

To complete step (ii), it remains to show that lim ker(dn ) = 0. We use the
construction of X ′ in (28.7.4.1) again. Note first that we have maps I × ker(dn ) →
ker(dn+1 ), so Q := ⊕n ker(dn ) is a module over B(I)• . Moreover, Q is a submodule
of the B(I)• -module ⊕n Hi (X, In+1 F ), which we showed was finitely generated
(over the Noetherian ring B(I)• ), see (28.7.4.2). Thus Q is also a finitely generated
graded B(I)• -module. Choose a finite set of homogeneous generators of Q, and
let N be the largest degree of a generator. Also, since ker dn = im cn (from the
exactness of (28.7.2.1)), we see that ker dn is annihilated by In , because F /In F is.
Thus the ideal IN ⊕ IN+1 ⊕ · · · ⊂ B(I)• annihilates Q.
Consider the composition
Ir ⊗ ker dn / ker dn+r / ker dn

where the first is from multiplication, and the second is the usual map induced by
Hi (X, In+r F ) → Hi (X, In F ). The first map is surjective for r ≥ 0 and n ≥ N (as
B(I)• is generated in degree 1, and Q a finitely generated B(I)• -module generated
in degrees at most N), and the composition is zero for r ≥ n. Thus map ker dn+r →
ker dn is zero for r, n ≥ N. In particular, ker dN+r → ker dN is zero for r ≥ N, so
lim ker(dn ) = 0 as desired. □
 CHAPTER 29

⋆ Proof of Serre duality

Réfléchissant un peu à ton théorème de dualité, je m’aperçois que sa formulation

générale est à peu près évidente, et d’ailleurs je viens de vérifier qu’elle se trouve implicite-
ment (dans le cas de l’espace projectif) dans ton théorème donnant les T q (M) par des Ext.
(J’ai bien l’impression, salaud, que tes §3 et 4 du Chap. 3 peuvent se faire aussi sans aucun
calcul).
Thinking a bit about your duality theorem, I notice that its general form is almost
obvious, and in fact I just checked that (for a projective space) it is implicitly contained in
your theorem giving the T q (M) in terms of Exts. (I have the impression, you salaud, that
§3 and 4 in your Chap. 3 could be done without any computation).
— A. Grothendieck, letter to J.-P. Serre, December 15, 1955 [GrS, p. 19]
We first met Serre duality in §18.5 (or even Theorem 18.1.2), and we have
repeatedly seen how useful it is. We will prove Theorem 18.5.1 (Corollary 29.3.10,
combined with Exercise 29.4.K, see Remark 29.4.10), as well as stronger versions,
and we will be left with a desire to prove even more. We give three statements
(Serre duality for vector bundles; Serre duality for Hom; and Serre duality for Ext), in two
versions (functorial and trace). (These names are idiosyncratic and nonstandard.)
We give several variants for a number of reasons. First, the easier statements will be
easier to prove, and the hardest statements we won’t be able to prove here. Second,
they may help give you experience in knowing how to know what to hope for, and
what to try to prove.

29.1 Desiderata

Throughout this chapter, we work over a field k; X will be a projective k-scheme,

and n = dim X. We will want a coherent sheaf ω (or with more precision, ωX ,
or even better, ωX/k ) on X, the dualizing sheaf, which will play a role in the
statements of duality. For the best statements, we will want a trace morphism

(29.1.0.1) t : Hn (X, ωX ) → k.

We will finally define ω and t in §29.1.5, but we first discuss what properties we
wish them to have.

This material is published by Princeton University Press as The Rising Sea: Foundations of Algebraic
Geometry by Ravi Vakil. This version is free to view and download for personal use only. Not for
redistribution, resale, or use in derivative works. © 2024 Ravi Vakil.

793
794 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

29.1.1. Desideratum: the determinant of the cotangent bundle is dualizing for

smooth varieties. If X is smooth, we will want ωX = KX in this case (recall KX =
det ΩX ) — the miracle that the canonical bundle is Serre-dualizing (§18.5.2). In
particular, ωX is an invertible sheaf. This will be disturbingly harder to prove than
the basic duality statements we show; we will only get to it later (Exercise 29.4.K).
But we will prove more, for example that ωX is an invertible sheaf if X is a regular
embedding in a smooth variety (Exercise 29.4.J).

29.1.2. Desideratum: Serre duality for vector bundles. The first version of duality,
which (along with Desideratum 29.1.1) gives Theorem 18.5.1, is the following: if F
is locally free of finite rank, then we have a functorial isomorphism
(29.1.2.1) ∼ Hn−i (X, F ∨ ⊗ ωX )∨ .
Hi (X, F ) =
More precisely, we want to construct a particular isomorphism (29.1.2.1), or equiv-
alently, a particular perfect pairing Hi (X, F ) × Hn−i (X, F ∨ ⊗ ωX ) → k. This
isomorphism will be functorial in F , i.e., it gives a natural isomorphism of covariant
functors
∼ / Hn−i (X, ·∨ ⊗ ωX )∨ .
(29.1.2.2) Hi (X, ·)

We call this functorial Serre duality for vector bundles.

Better still, there should be a cup product in cohomology, which can be used
to construct a map Hi (X, F ) × Hn−i (X, F ∨ ⊗ ωX ) → Hn (X, ωX ). This should
be functorial in F (in the sense that we get a natural transformation of func-
tors Hi (X, ·) → Hn−i (X, ·∨ ⊗ ωX )∨ ⊗ Hn (X, ωX )). Combined with the trace map
(29.1.0.1), we get a map
(29.1.2.3)
Hi (X, F ) × Hn−i (X, F ∨ ⊗ ωX ) / Hn (X, F ⊗ F ∨ ⊗ ωX )


Hn (X, ωX )
t /k

which, if it is a perfect pairing, should yield (29.1.2.2). If the composition (29.1.2.3)

is a perfect pairing, we say that X (with the additional data of (ωX , t)) satisfies the
trace version of Serre duality for vector bundles. (This was hinted at after the
statement of Theorem 18.5.1.)
In fact, the trace version of Serre duality for vector bundles (and hence the
functorial version) is true when X is Cohen-Macaulay, and in particular when X is
smooth. We will prove the functorial version (see Corollary 29.3.10). We will give
an indication of how the trace version can be proved in Remark 29.3.15.

29.1.3. Desideratum: duality for more general X and F . A weaker sort of duality
will hold with weaker hypotheses. We will show that (without Cohen-Macaulay
hypotheses) for any coherent sheaf F on a pure n-dimensional projective k-scheme
X, there is a functorial isomorphism (functorial in F )
∼ / Hn (X, F )∨ .
(29.1.3.1) Hom(F , ωX )

We call this functorial Serre duality for Hom.

 © 2024 Ravi Vakil. Published by Princeton University Press. 795

In parallel with Serre duality for vector bundles, we have a natural candidate
for the perfect pairing:

(29.1.3.2) Hom(F , ωX ) × Hn (X, F ) / Hn (X, ωX ) t /k

where the trace map t is some linear functional on Hn (X, ωX ). If this composition
is a perfect pairing, we say that X (with the additional data of (ωX , t)) satisfies the
trace version of Serre duality for Hom. Unlike the trace version of Serre duality
for vector bundles, we already know what the “cup product” map Hom(F , ωX ) ×
Hn (X, F ) → Hn (X, ωX ) is: an element [σ : F → ωX ] of Hom(F , ωX ) induces
— by covariance of Hn (X, ·), see §18.1 — a map Hn (X, F ) → Hn (X, ωX ). The
resulting pairing is clearly functorial in F , so the trace version of Serre duality
for Hom implies the functorial version. Unlike the case of Serre duality for vector
bundles, the functorial version of Serre duality for Hom also implies the trace
version:

29.1.A. E XERCISE . Show that the functorial version of Serre duality for Hom
implies the trace version. In other words, the trace map is already implicit in
∼
the functorial isomorphism Hom(F , ωX ) −→ Hn (X, F )∨ . Hint: consider the
commuting diagram (coming from functoriality)

Hom(F , ωX ) / Hn (X, F )∨
O O

Hom(ωX , ωX ) / Hn (X, ωX )∨ .

29.1.4. Getting down to work. We begin by defining some of our terms.

29.1.5. Definition. Suppose X is a projective k-scheme of dimension n. A coherent

sheaf ω = ωX = ωX/k along with a map t : Hn (X, ωX ) → k is called dualizing if
the natural map (cf. (29.1.3.1))

(29.1.5.1) Hom(F , ωX ) × Hn (X, F ) / Hn (X, ωX ) t /k

is a perfect pairing for all coherent sheaves F on X. We call ωX the dualizing sheaf
and t the trace map. (The earlier discussion of ωX and t was aspirational. This
now is a definition.) If X has such (ωX , t), we say that X satisfies Serre duality (for
Hom). The following proposition justifies the use of the word “the” (as opposed to
“a”) in the phrase “the dualizing sheaf”.

29.1.6. Proposition. — If a dualizing sheaf and trace (ωX , t) exists for X, this data is
unique up to unique isomorphism.

Proof. Suppose we have two such (ωX , t) and (ωX′ , t ′ ). From the two morphisms

(29.1.6.1) Hom(F , ωX ) × Hn (X, F ) / Hn (X, ωX ) t /k

/ Hn (X, ω ′ ) t′ / k,
Hom(F , ωX′ ) × Hn (X, F ) X
796 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

∼ Hom(F , ω ′ ), which is functorial in F .

we get a natural bijection Hom(F , ωX ) = X
By the typical universal property argument (Exercise 1.2.Z), this induces a (unique)
∼
isomorphism ωX ←→ ωX′ . From (29.1.6.1), under this isomorphism, the two trace
′
maps t and t must be the same too. □
We will prove the functorial and trace versions of Serre duality for Hom in
Corollary 29.3.12. The special case of projective space will be a key ingredient. We
prove this case now.

29.1.7. Serre duality (for Hom) for projective space. Define ωPnk (or just ω for
∼
convenience) as OPnk (−n − 1). Let t be any isomorphism Hn (Pn k , ωPk ) −→ k (Theo-
n

rem 18.1.2(b)). As the notation suggests, (ωPnk , t) will be dualizing for projective
space Pnk.

29.1.B. E XERCISE . Suppose F = OPnk (m). Show that the natural map (29.1.5.1) is a
perfect pairing. (Hint: do this by hand. See the discussion after Theorem 18.1.2.)
Hence show that if F is a direct sum of line bundles on Pn k , the natural map
(29.1.5.1) is a perfect pairing.

29.1.8. Proposition. — The functorial version (and hence the trace version by Exer-
cise 29.1.A) of Serre duality for Hom holds for Pn
k.

Proof. We wish to show that the (functorial) natural map

∼ / Hn (Pn , F )∨ ,
Hom(F , ωPnk ) k

coming from the cup pairing and the choice of isomorphism t, is an isomorphism
(cf. (29.1.3.1)). Fix a coherent sheaf F on Pn
k . By Theorem 16.1.1, we can present F
as
(29.1.8.1) 0 /G /E /F /0
where E is a finite direct sum of line bundles, and G is coherent. Applying the
left-exact functor Hom(·, ωPnk ) to (29.1.8.1), we have the exact sequence

(29.1.8.2) 0 / Hom(F , ωPn ) / Hom(E , ωPn ) / Hom(G , ωPn ).

k k k

Taking the long exact sequence in cohomology for (29.1.8.1) and dualizing, we have
the exact sequence

(29.1.8.3) 0 / Hn (Pn , F )∨ / Hn (Pn , E )∨ / Hn (Pn , G )∨

k k k

The (functorial) pairing (29.1.3.2) gives a map from (29.1.8.2) to (29.1.8.3):

(29.1.8.4)
0O /0 / Hn (Pn , F )∨ / Hn (Pn , E )∨ / Hn (Pn , G )∨
O kO kO kO

α β γ δ ϵ

0 /0 / Hom(F , ωPn ) / Hom(E , ωPn ) / Hom(G , ωPn )

k k k

Maps α and β are obviously isomorphisms, and Exercise 29.1.B shows that δ is an
isomorphism. Thus by the subtle version of the Five Lemma (Exercise 1.6.D, as β
and δ are injective and α is surjective), γ is injective. This shows that the natural
′ ∨
map Hom(F ′ , ωPnk ) → Hn (Pn k , F ) is injective for all coherent sheaves F , and
′
 © 2024 Ravi Vakil. Published by Princeton University Press. 797

in particular for F ′ = G . Thus ϵ is injective. Then by the dual of the subtle version
of the Five Lemma (as β and δ are surjective, and ϵ is injective), γ is surjective. □

29.1.9. Mathematical puzzle. Here is a puzzle to force you to confront a potentially

confusing point. We will see that Desideratum 29.1.1 holds for P1 , so ωP1 = ∼ ΩP1 .
What then is the trace map t : H (P , ΩP1 ) → k? The Čech complex for H (P , ΩP1 )
1 1 1 1

(with the usual cover of P1 ) is given by

(29.1.9.1) 0 / ΩP1 (U0 ) × ΩP1 (U1 ) α / ΩP1 (U0 ∩ U1 ) / 0.

If U0 = Spec k[x], and U0 ∩ U1 = Spec k[x, 1/x], then the differentials on U0 ∩ U1

are those of the form f(x) dx where f(x) is a Laurent polynomial (for example:
(x−3 + x−1 + 3 + 17x4 ) dx). To compute H1 (P1 , ΩP1 ), we need to find which such
differentials on U0 ∩ U1 are in the image of α in (29.1.9.1). Clearly any term of the
form xi dx (for i ≥ 0) extends to a differential on U0 , and thus is in the image of
α. A short calculation shows that any term of the form xi dx (i < −1) extends to a
differential on U1 . Thus the cokernel of α can be described as the one-dimensional
k-vector space generated by x−1 dx. We have an obvious isomorphism to k: take
the coefficient of x−1 dx, which can be interpreted as “take the residue at 0”. But
there is another choice, which is equally good: take the residue at ∞ — certainly
there is no reason to privilege 0 over ∞ (or U0 over U1 )! But these two residues are
not the same — they add to 0 (as you can quickly calculate — you may also believe
it because of the Residue Theorem in the theory of Riemann surfaces). So: which
one is the trace?

29.1.10. Desideratum: a stronger version of duality, involving Ext.

Ta formule:
Hn−p (X, F ) ′ = Extp n
O (X, F, Ω )
m’excite beaucoup, car je suis bien convaincu que c’est la bonne façon d’énoncer le théorème
de dualité à la fois dans le cas analytique et dans le cas algébrique (le plus important —
pour moi!).
I find your formula
Hn−p (X, F ) ′ = Extp n
O (X, F, Ω )

very exciting, as I am quite convinced that it is the right way to state the duality theorem in
both the analytic case and the algebraic case (the more important one — for me!).
— J.-P. Serre, letter to A. Grothendieck, December 22, 1955 [GrS, p. 21]
The vector bundle and Hom versions of Serre duality have a common extension.
For this, we need the notion of Ext for O-modules. We will develop what we need
in §29.2, but for now, it suffices to observe that if X is a ringed space, then the
category of OX -modules has enough injectives (Theorem 23.4.1), so we can define
Exti (F , ·) as the derived functor of Hom(F , ·) (like Ext-functors for A-modules,
defined in §23.2.4).
If we have an isomorphism of functors

∼ / Hn−i (X, ·)∨

(29.1.10.1) Exti (·, ωX )
798 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

we say that X satisfies functorial Serre duality for Ext. (The case i = 0 is functorial
Serre duality for Hom, or by Exercise 29.1.A, the trace version.)
In Exercise 29.2.K, we will find that the functorial version of Serre duality for
Ext (resp., the trace version) implies the functorial version of Serre duality for vector
bundles (resp., the trace version). Thus to prove Theorem 18.5.1, it suffices to prove
functorial Serre duality for Ext, and Desideratum 29.1.1 (see Remark 29.4.10). We
will prove the functorial version of Serre duality for Ext when X is Cohen-Macaulay
in Corollary 29.3.14.

29.1.11. Trace version of Serre duality for Ext. As with the previous versions, the
functoriality of functorial Serre duality for Ext should come from somewhere. There
is cup product
Exta (F , G ) × Extb (E , F ) → Exta+b (E , F )
(the “Yoneda cup product”). If the composition

ExtiX (F , ωX ) × Hn−i (X, F ) = ExtiX (F , ωX ) × Extn−i

∪ /
X (OX , F )

Extn n
X (OX , ωx ) = H (X, ωX )
t /k

is a perfect pairing for all i, we say that X satisfies the trace version of Serre duality
for Ext. We will not be able to prove the trace version, as we will not define this
cup product.

29.1.12. Necessity of Cohen-Macaulay hypotheses. We remark that the Cohen-

Macaulay hypotheses are necessary everywhere they are stated. The following
example applies in all cases. Let X be the union of two 2-planes in P4k , meeting at
a point p. If there were a coherent dualizing sheaf ωX on X, then for d ≫ 0, we
would have h1 (X, OX (−d)) = h1 (X, ωX (d)) by Serre duality, which must be 0 by
Serre vanishing (Theorem 18.1.3(ii)). We show that this is not the case.

29.1.C. E XERCISE . Let H be a hyperplane in P4 not passing through p; say H is

hyperplane x0 = 0. Let Z be the intersection of dH (the divisor xd 0 = 0) with X.
Show (cheaply) that for d ≥ 0, h0 (Z, OZ ) > 1. Use the exact sequence

0 / OX (−Z) / OX / OZ /0

to show that h1 (X, OX (−d)) > 0.

29.2 Ext groups and Ext sheaves for O-modules

Recall that for any ringed space X, the category ModOX has enough injectives
(Theorem 23.4.1). Thus for any OX -module F on X, we may define

ExtiX (F , ·) : ModOX / ModΓ (X,O )

X
 © 2024 Ravi Vakil. Published by Princeton University Press. 799

as the ith right derived functor of HomX (F , ·), and we have a corresponding long
exact sequence for ExtiX (F , ·). We similarly define the sheaf version

Ext iX (F , ·) : ModOX / ModO

X

as the ith right derived functor of Hom X (F , ·). In both cases, the subscript X is
often omitted when it is clear from the context, although this can be dangerous
when more than one space is relevant to the discussion. (We saw Ext functors for
A-modules in §23.2.4.)
Warning: it is not clear (and in fact not true, see §23.4.8) that ModOX has enough
projectives, so we cannot define Exti as a derived functor in its left argument.
Nonetheless, we will see that it behaves as though it is a derived functor — it is
“computable by acyclics”, and has a long exact sequence (Remark 29.2.1).
Another warning: with this definition, it is not clear that if F and G are
quasicoherent sheaves on a scheme, then the Ext i (F , G ) are quasicoherent, and
indeed the aside in Exercise 14.3.A(a) points out this need not be true even for i = 0.
But Exercise 29.2.F will reassure you that all is well if F is coherent, and X is locally
Noetherian.
Exercise 23.5.A (an injective OX -module, when restricted to an open subset
U ⊂ X, is injective on U) has a number of useful consequences.

29.2.A. E XERCISE . Suppose I is an injective OX -module. Show that Hom OX (·, I )

is an exact contravariant functor. (A related fact: HomOX (·, I ) is exact, by the
definition of injectivity, Exercise 23.2.C(a).)

29.2.B. E XERCISE . Suppose X is a ringed space, F and G are OX -modules,

∼
and U is an open subset. Describe a canonical isomorphism Ext iX (F , G )|U ←→
Ext U (F |U , G |U ). Hint: the restriction ModOX → ModOY is exact (cf. Exercises 2.7.D
i

and 2.7.E); restriction preserves injectivity of O-modules (Exercise 23.5.A); and the
FHHF Theorem (Exercise 1.5.I).

29.2.C. E XERCISE . Suppose X is a ringed space, and G is an OX -module.

(a) Show that

G if i = 0, and
Ext (OX , G ) =
i
0 otherwise.

∼
(b) Describe a canonical isomorphism ExtiX (OX , G ) ←→ Hi (X, G ).

29.2.D. E XERCISE . Use Exercises 29.2.B and 29.2.C(a) to show that if E is a locally
free sheaf on X, then Ext i (E , G ) = 0 for i > 0.
In the category of modules over a ring, we like projectives more than injectives,
because free modules are easy to work with. It would be wonderful if locally free
sheaves on schemes were always projective, but sadly this is not true. For if OX
were projective, then Exti (OX , ·) would be zero (as a functor) for i > 0, but by
Exercise 29.2.C(b), it is the functor Hi (X, ·), which is often nonzero. Nonetheless,
we can still compute Ext using a locally free resolution, as shown by the following
exercise.
800 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

29.2.E. I MPORTANT E XERCISE . Suppose X is a ringed space, and

(29.2.0.1) ··· / E1 / E0 /F /0
is a resolution of an OX -module F by locally free sheaves. (Of course we are
most interested in the case where X is a scheme, and F is quasicoherent, or even
coherent.) Let E• denote the truncation of (29.2.0.1), where F is removed. Describe
∼
an isomorphism Ext i (F , G ) ←→ Hi (Hom (E• , G )). In other words, Ext • (F , G ))
can be computed by taking a locally free resolution of F , truncating, applying
Hom (·, G ), and taking homology. Hint: choose an injective resolution
0 /G / I0 / I1 / ···
and consider the spectral sequence whose E0 term is
.. ..
.O .O

Hom (EO 0 , I1 ) / Hom (E1 , I1 ) / ···

O

Hom (E0 , I0 ) / Hom (E1 , I0 ) / ··· .

This result is important: to compute Ext , we can compute it using finite rank
locally free resolutions. You can work affine by affine (by Exercise 29.2.B), and
on each affine you can use a free resolution of the left argument. As another
consequence of Exercise 29.2.E:

29.2.F. E XERCISE .
(a) Suppose F is a coherent sheaf and G is a quasicoherent sheaf on a locally
Noetherian scheme X. Show that Ext iX (F , G ) is a quasicoherent sheaf on X. Hint:
by Exercise 29.2.B, because Ext can be computed on an open cover, show that it
suffices to show the result for X affine. But in that case, F has a free resolution. (You
will discover that Ext iX (F , G )(Spec A) = ExtiA (F (Spec A), G (Spec A)).)
(b) If furthermore, G is coherent, show that Ext iX (F , G ) is also coherent.

29.2.1. Remark. The statement “Exti (F , G ) and Ext i (F , G ) behave like a derived
functor in the first argument” is true in a number of ways. For example, we have a
corresponding long exact sequence, as shown in the next exercise.

29.2.G. E XERCISE . Suppose 0 → F ′ → F → F ′′ → 0 is an exact sequence of

OX -modules on a ringed space X. For any OX -module G , describe a long exact
sequence
0 / Hom(F ′′ , G ) / Hom(F , G ) / Hom(F ′ , G )

/ Ext1 (F ′′ , G ) / Ext1 (F , G ) / Ext1 (F ′ , G ) / ··· .

(Thus Exti (·, G ) forms a contravariant δ-functor, as defined in §23.2.7, despite not
being defined as a derived functor.) Prove the analogous results for Ext . Hint: take
 © 2024 Ravi Vakil. Published by Princeton University Press. 801

an injective resolution 0 → G → I 0 → · · · . Use the fact that if I is injective, then

Hom(·, I ) is exact (the definition of injectivity, Exercise 23.2.C(a)). Hence get a
short exact sequence of complexes
0 / Hom(F ′′ , I • ) / Hom(F , I • ) / Hom(F ′ , I • ) /0

and take the long exact sequence in cohomology.

29.2.2. Functorial Serre duality for Ext holds for projective space.
We now prove that functorial Serre duality for Ext holds for projective space.
We will use the machinery of universal δ-functors (§23.2.6), so you may wish to
either quickly skim that section, or else ignore this discussion.

29.2.H. E XERCISE . Show that (ExtiPnk (·, ωPnk )) is a (contravariant) universal δ-

functor. Hint: Ext is not a derived functor in its first argument, so you can’t use
the “projective” version of Corollary 23.2.11. Instead, use Theorem 23.2.9, and the
existence of a surjection O(m)⊕N → F for each F , for some m < 0.
∨
29.2.I. E XERCISE . Show that (Hn−i (Pn
k , ·) ) is a universal δ-functor. (What are the
δ-maps?) Hint: try the same idea as in the previous exercise.
∼
∨
Proposition 29.1.8 gives an isomorphism of functors Ext0Pnk (·, ωPnk ) ←→ Hn (Pn
k , ·) ,
so by the Definition 23.2.8 of universal δ-functor, we have an isomorphism of
∼ ∨
δ-functors (ExtiPnk (·, ωPnk )) ←→ (Hn−i (Pn
k , ·) ), thereby proving functorial Serre
duality for Ext for Pn k.

29.2.3. Two useful exercises, and functorial Serre duality for vector bundles for
projective space.

29.2.J. E XERCISE . Suppose X is a ringed space, F and G are OX -modules, and E is

a locally free sheaf on X. Describe isomorphisms
∼ ∼
Ext i (F ⊗ E ∨ , G ) ←→ Ext i (F , G ⊗ E ) ←→ Ext i (F , G ) ⊗ E
∼
and Exti (F ⊗ E ∨ , G ) ←→ Exti (F , G ⊗ E ).
Hint: show that if I is injective then I ⊗ E is injective.

29.2.K. E XERCISE . If F is a locally free sheaf on a scheme X, and G is any coherent

∼
sheaf on X, describe an isomorphism Exti (F , G ) ←→ Hi (X, F ∨ ⊗ G ) (functorial
in F ). By taking G to be ωX , show that functorial Serre duality for Ext implies
functorial Serre duality for vector bundles. (You may wish to ponder the trace
versions as well.) As a consequence, by §29.2.2, functorial Serre duality for vector
bundles holds for projective space. Hint: Exercises 29.2.J and 29.2.C(b).

29.2.4. The local-to-global spectral sequence for Ext.

“Sheaf” Ext and “global” Ext are related by a spectral sequence. This is a
direct application of the Grothendieck composition-of-functors spectral sequence
(Theorem 23.3.5), once we show that Hom (F , I ) is acyclic for the functor Γ .

29.2.L. E XERCISE . Suppose I is an injective OX -module. Show that Hom (F , I )

is flasque (Definition 23.4.6), and thus Γ -acyclic by Exercise 23.4.I. Hint: suppose
j : U ,→ V is an inclusion of open subsets. We wish to show that Hom (F , I )(V) →
802 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Hom (F , I )(U) is surjective. Note that I |V is injective on V (Exercise 23.5.A).

Apply the exact functor HomV (·, I |V ) to the inclusion j! (F |U ) ,→ F |V of sheaves
on V (§23.4.7).

29.2.M. E XERCISE ( THE LOCAL - TO - GLOBAL SPECTRAL SEQUENCE FOR Ext). Sup-
pose X is a ringed space, and F and G are OX -modules. Describe a spectral
sequence with E2 -term Hi (X, Ext j (F , G )) abutting to Exti+j (F , G ). (Hint: use the
Grothendieck composition-of-functors spectral sequence, Theorem 23.3.5. Recall
that Hom(F , ·) = Γ (Hom (F , ·)), Exercise 2.3.C.)

29.3 Serre duality for projective k-schemes

We now prove various versions of Serre duality for projective k-schemes, by

leveraging Serre duality for projective space.

29.3.1. Recalling the weird functor π!? .

The key construction is a right adjoint π!? to the pushforward π∗ , when π is a
finite morphism, discussed in §17.1.6. You should read this section if you have not
done so already.

29.3.2. Caution. We have defined π!? only for quasicoherent sheaves on Y, not
for all OY -modules. If G is an OX -module (not necessarily quasicoherent), it is in
general not clear how to make sense of this construction to define an OY -module
π!? G . (Try it and see!) However, in the special case where π is a closed embedding,
we can make sense of π!? G , as described in the Exercise 29.4.A.

29.3.3. Remark. If π is finite and flat, which is the case most of interest to us, π!?
agrees (in the only possible sense of the word) with π! , one of Grothendieck’s “six
operations”, and indeed this motivates our notation (“a strange variant of π! ”). But
π! naturally lives in the world of derived categories, so we will not discuss it here.
(For more on π! , see [KS1, Ch. III].)
We now apply the machinery of π!? to Serre duality.

29.3.4. Projective n-dimensional schemes are finite covers of Pn .

29.3.5. Proposition. — Suppose X is a projective k-scheme of dimension at most n.

(a) There exists a finite morphism π : X → Pn .
(b) If furthermore X is Cohen-Macaulay of pure dimension n, then π is flat.

29.3.6. Remark for experts. We carefully allow n to be greater than dim X here,
not because it is useful, but to point out that the proof works, and we are proving
something strange if we don’t have the right hypotheses. Proposition 29.3.5(a) is
still interesting if dim X = n but X is not Cohen-Macaulay. What does Proposi-
tion 29.3.5(a) imply if dim X < n?

Proof. Part (b) follows from part (a) by the Miracle Flatness Theorem 26.2.11, so we
will prove part (a).
 © 2024 Ravi Vakil. Published by Princeton University Press. 803

Choose a closed embedding j : X ,→ PN . For simplicity of exposition, first

assume that k is an infinite field. By Exercise 12.3.D(d), there is a linear space L
of codimension n + 1 (one less than complementary dimension) disjoint from X.
Projection from L yields a morphism π : X → Pn . The morphism π is affine (PN \L →
Pn is affine, and the closed embedding X ,→ PN \ L is, like all closed embeddings,
affine) and projective (by the Cancellation Theorem for projective morphisms,
Exercise 17.3.C; or more directly by Exercise 17.3.D), so π is finite (projective affine
morphisms of locally Noetherian schemes are finite, Corollary 18.1.5).

29.3.A. E XERCISE . Prove Proposition 29.3.5(a), without the assumption that k is

infinite. Hint: using Exercise 12.3.D(c), show that there is some d such that there is
an intersection of n + 1 degree d hypersurfaces missing X. Then apply the above
argument to the dth Veronese embedding of PN (§9.3.6).
□

29.3.7. Remark. More generally, the above argument shows that any projective
k-scheme of dimension n (not necessarily pure dimensional) can be expressed
as a finite cover of Pn . This might be seen as a projective version of Noether
normalization.

29.3.8. Serre duality on X via π!? .

Suppose π : X → Y is a finite morphism of projective k-schemes of pure dimen-
sion n, and we have a coherent sheaf ωY on Y. (We will soon apply this in the case
where Y = Pn k , but we may as well avoid distraction and needless specificity.)

29.3.9. Proposition. — Suppose π is flat. If functorial Serre duality for vector bundles
holds for Y, with dualizing sheaf ωY , then functorial Serre duality for vector bundles holds
for X, with dualizing sheaf π!? (ωY ).

Proof. For each i, and each finite rank locally free sheaf F on X, we have isomor-
phisms (functorial in F ):

∼ Hn−i (Y, π∗ (F ∨ ⊗ π!? ωY )) (affineness of π, §18.1 (v))

Hn−i (X, F ∨ ⊗ π!? ωY ) =
∼ Hn−i (Y, π∗ (Hom X (F , π!? ωY ))) (Exercise 14.1.G)
=
∼ Hn−i (Y, Hom Y (π∗ F , ωY )) (equ. (17.1.8.2))
=
∼ Hn−i (Y, (π∗ F )∨ ⊗ ωY ) (Exercises 24.3.E and 14.1.G)
=
∼ Hi (Y, π∗ F )∨ (Serre duality for vector bundles on Y)
=
∼ Hi (X, F )∨ (affineness of π, §18.1 (v))
=

(The hypothesis of flatness is used to show that π∗ F is locally free.) □

29.3.10. Corollary. — Functorial Serre duality for vector bundles holds for every Cohen-
Macaulay pure dimensional projective k-scheme.

Proof. Combine Proposition 29.3.9 with Proposition 29.3.5(b) and §29.2.2. □

804 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

29.3.11. Proposition. — Suppose π : X → Y is a finite morphism of projective k-schemes.

If Serre duality for Hom holds for Y with dualizing sheaf (ωY , tY ), then Serre duality for
Hom holds for X with dualizing sheaf π!? ωY .
Recall from Exercise 29.1.A that for Serre duality for Hom, the trace version is
equivalent to the functorial version. For this reason we just use the phrase “Serre
duality for Hom”.
Note that we have no flatness hypotheses on π (unlike the corresponding
propositions for Serre duality for vector bundles and, soon, for Ext).

Proof. We have functorial isomorphisms

∼
HomX (F , π!? ωY ) = HomY (π∗ F , ωY ) (adjointness of (π∗ , π!? ))
∼
= Hn (Y, π∗ F )∨ (Serre duality for Hom for Y)
∼
= H (X, F )
n ∨
(affineness of π)
□
You can unwind the argument to obtain the trace map for X in terms of the
trace map for Y, as the composition

Hn (X, π!? ωY ) / Hn (Y, π∗ π!? ωY ) / Hn (Y, ωY ) tY

/ k.

29.3.12. Corollary. — The functorial and trace versions of Serre duality for Hom hold
for all pure dimensional projective k-schemes (with no Cohen-Macaulay hypotheses).

Proof. Combine Proposition 29.3.11 with Proposition 29.3.5(a) and Proposition 29.1.8.
□

29.3.13. Proposition. — If X is a Cohen-Macaulay projective k-scheme of pure dimension

n, and π : X → Pn is a finite flat morphism (as in Proposition 29.3.5(b)), then functorial
Serre duality for Ext holds for X with dualizing sheaf ωX := π!? ωPn .
Unlike Proposition 29.3.9 and 29.3.11, we did not state this for general π : X → Y;
we will use the fact that the target is Pn .

Proof. The argument will parallel that of §29.2.2: we will show an isomorphism of
δ-functors
∼ / Hn−i (X, F )∨ .
(29.3.13.1) ExtiX (F , π!? ωPn )
We already have an isomorphism for i = 0 (Corollary 29.3.12), so it suffices to
show that both sides of (29.3.13.1) are universal δ-functors (§23.2.6). We do this by
verifying the criterion of Theorem 23.2.9. For any coherent sheaf F on X we can
find a surjection O(−m)⊕N → F for m ≫ 0 (and some N), so it suffices to show
that for m ≫ 0,
ExtiX (O(−m), π!? ωPn ) = 0 and Hn−i (X, O(−m))∨ = 0
for i > 0 and m ≫ 0. For the first, we have (for i > 0 and m ≫ 0)
∼ Hi (X, (π!? ωPn )(m))
ExtiX (O(−m), π!? ωPn ) = (Exercise 29.2.K)
=0 (Serre vanishing for X, Thm. 18.1.3(ii))
 © 2024 Ravi Vakil. Published by Princeton University Press. 805

For the second, we have

Hn−i (X, OX (−m)) = Hn−i (X, OX ⊗ π∗ O(−m))
∼ Hn−i (Pn , (π∗ OX ) ⊗ O(−m))
=
by the projection formula (Exercise 18.7.E(b)). Now π∗ OX is locally free by Exer-
cise 24.3.E, so by functorial Serre duality for vector bundles on Pn , we have (for
m ≫ 0)
Hn−i (Pn , (π∗ OX ) ⊗ O(−m)) =∼ Hi (Pn , (π∗ OX )∨ ⊗ O(m) ⊗ ωPn )∨
∼ Hi (Pn , ((π∗ OX )∨ ⊗ ωPn ) ⊗ O(m))∨
=
=0 (Serre vanishing for Pn , Thm. 18.1.3(ii)).
□

29.3.14. Corollary. — Functorial Serre duality for Ext holds for all pure dimensional
Cohen-Macaulay projective k-schemes.

Proof. Combine Proposition 29.3.13 with Proposition 29.3.5(b) and the projective
space case of §29.2.2. □

29.3.15. ⋆⋆ Remark: the trace version of Serre duality for vector bundles. Here is an
outline of how the above proof can be extended to prove the trace version of Serre
duality for vector bundles on all pure dimensional Cohen-Macaulay projective
k-schemes. Continuing the notation of Proposition 29.3.13, the Yoneda cup product
(which we haven’t defined) followed by the trace map gives a pairing

ExtiX (F , ωX ) × Hn−i (X, F )

∪ / Hn (X, ωX ) t /k,

which is functorial in F . This induces a functorial map (29.3.13.1). One shows

that this particular map is a map of δ-functors. Then by the above argument, this
particular map is an isomorphism.

29.3.16. Conclusion. The proofs in this section are perhaps surprisingly easy. Once
we defined π!? , we only used the fact that it was right-adjoint to π∗ . The price we
pay is that we have very little idea of what the dualizing sheaf looks like from the
proof. For example, if X is smooth, it is not clear that ωX is a line bundle! (We
rectify this in the next section.)
You may also have a sense of how these ideas lead to a larger story: to define
and prove duality more generally, we are trying to find a functor π! which is right-
adjoint to π∗ in as much generality as we can manage. This will lead us to work
in a more general setting than coherent sheaves – the derived category of coherent
sheaves (and beyond).

29.4 The adjunction formula for the dualizing sheaf, and ωX = KX

The dualizing sheaf behaves well with respect to slicing by effective Cartier
divisors, and this will be useful to obtaining Desideratum 29.1.1, i.e., the miracle
that the canonical bundle is Serre-dualizing (§18.5.2). In order to show this, we give
806 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

a description for the dualizing sheaf for a subvariety of a variety satisfying Serre
duality for Ext.
As a warm-up, we extend the notion of π!? to O-modules, in the special case
where π is a closed embedding.

29.4.A. E XERCISE . Suppose π : X → Y is a closed embedding of schemes and G

is an OY -module. Explain why Hom Y (π∗ OX , G ) naturally has the structure of an
OX -module. Hint: if I is the ideal sheaf of X, explain how Hom Y (π∗ OX , G ) (over
some open subset U ⊂ Y) is annihilated by “functions vanishing on X” (elements of
I (U)).

29.4.1. Definition. Hence if π : X → Y is a closed embedding, we have defined

a map π!? : ModOY → ModOX extending the map of Exercise 17.1.K(a). In this
case where X is a closed embedding, it is reasonable and correct to write π!? G :=
π−1 Hom (π∗ OX , G ), i.e., to use the symbol π−1 in the definition.

29.4.B. E XERCISE . We remain in the situation of Exercise 29.4.A.

(a) Show that (π∗ , π!? ) is an adjoint pair between ModOX and ModOY .
(b) Show that π!? sends injective OY -modules to injective OX -modules. (Hint: π∗ is
exact; use Exercise 23.5.B.)
We now extend this discussion from Hom to Ext . Suppose π : X → Y is a closed
embedding. Then for any OY -module F , Ext iY (π∗ OX , F ) naturally has the struc-
ture of an OX -module. (Reason: we compute this by taking an injective resolution
of F by OY -modules, then truncating, then applying Hom Y (π∗ OX , ·). But for any
OY -module G , Hom Y (π∗ OX , G ) has the structure of an OX -module, indeed π!? G .)
To emphasize its structure as an OX -module, we write it as π−1 Ext iY (π∗ OX , F ).

29.4.C. E XERCISE . Show that if Y (and hence X) is locally Noetherian, and G is

a coherent sheaf on Y, then π−1 Ext iY (π∗ OX , G ) is a coherent sheaf on X. (Hint:
Ext iY (π∗ OX , G ) is a coherent sheaf on Y, Exercise 29.2.F(b).)
29.4.2. The dualizing sheaf for a subvariety in terms of the dualizing sheaf of the
ambient variety.
For the rest of this section, π : X → Y will be a closed embedding of pure
dimensional projective k-schemes of dimension n and N respectively, where Y
satisfies functorial Serre duality for Ext. Let r = N − n (the codimension of X in Y).
By Corollary 29.3.12, X satisfies functorial Serre duality for Hom. We now identify
ωX in terms of ωY .

29.4.3. Theorem. — We have ωX = π−1 Ext rY (π∗ OX , ωY ).

Before we prove Theorem 29.4.3, we explain what happened to the “earlier”
Ext iY (π∗ OX , ωY )’s.
29.4.4. Proposition. — Suppose that π : X ,→ Y is a closed embedding of pure dimensional
projective k-schemes of dimension n and N respectively, and Y satisfies functorial Serre
duality for Ext. Then for all i < r := N − n, Ext iY (π∗ OX , ωY ) = 0.

Proof. As Ext iY (π∗ OX , ωY ) is coherent (Exercise 29.2.F), it suffices to show that

Ext iY (π∗ OX , ωY ) ⊗ O(m) has no nonzero global sections for m ≫ 0 (as for any
 © 2024 Ravi Vakil. Published by Princeton University Press. 807

coherent sheaf G on Y, G (m) is generated by global sections for m ≫ 0 by Serre’s

Theorem A, Theorem 16.1.5). By Exercise 29.2.J,
Hj (Y, Ext iY (π∗ OX , ωY )(m)) = Hj (Y, Ext iY (π∗ OX , ωY (m))).
If j > 0, then for m ≫ 0, by Serre vanishing, Hj (Y, Ext iY (π∗ OX , ωY )(m)) = 0. Thus
by the local-to-global spectral sequence for Ext (Exercise 29.2.M),
H0 (Y, Ext iY (π∗ OX , ωY (m))) = ExtiY (π∗ OX , ωY (m))).
By Exercise 29.2.J again, then functorial Serre duality for Ext on Y,
ExtiY (π∗ OX , ωY (m))) = ExtiY (π∗ OX (−m), ωY ) = HN−i (Y, π∗ OX (−m))
which is 0 if N − i > n, as the cohomology of a quasicoherent sheaf on a projec-
tive scheme vanishes in degree higher than the dimension of the sheaf’s support
(dimensional cohomology vanishing, Theorem 18.2.6). □
The most difficult step in the proof of Theorem 29.4.3 is the following.

29.4.5. Lemma. — Suppose that π : X → Y is a closed embedding of codimension r of pure

dimensional projective k-schemes, and Y satisfies functorial Serre duality for Ext. Then we
have an isomorphism, functorial in F ∈ CohX :
∼
(29.4.5.1) HomX (F , π−1 Ext rY (π∗ OX , ωY )) o / ExtrY (π∗ F , ωY ).

Proof. Choose an injective resolution

0 / ωY / I0 / I1 / I2 / ···

of ωY . To compute the right side of (29.4.5.1), we drop the ωY from this resolution,
and apply HomY (π∗ F , ·), to obtain

0 / HomY (π∗ F , I 0 ) / HomY (π∗ F , I 1 ) / HomY (π∗ F , I 2 ) / ··· ,

or (by adjointness of π∗ and π!? for O-modules when π is a closed embedding,

Exercise 29.4.B(a)):

0 / HomX (F , π!? I 0 ) / HomX (F , π!? I 1 ) / HomX (F , π!? I 2 ) / ··· .

Motivated by this, consider the complex

(29.4.5.2) 0 / π!? I 0 / π!? I 1 / π!? I 2 / ··· .

Note first that (29.4.5.2) is indeed a complex, and second that π!? I i are injective
OX -modules (by Exercise 29.4.B(b)).

29.4.D. E XERCISE . Show that the cohomology of (29.4.5.2) at the ith step is
π Ext Y (π∗ OX , ωY ).
−1 i

Thus by Proposition 29.4.4, the complex (29.4.5.2) is exact before the rth step.

29.4.E. E XERCISE . Show that there exists a direct sum decomposition π!? I r =
J ⊕ K , so that

0 / π!? I 0 / π!? I 1 / ··· / π!? I r−1 /J /0

808 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

is exact. Hint: work out the case r = 1 first. Another hint: Notice that if
0 / I0 α /F

is exact and I 0 is injective, then there is a map β : F → I 0 “splitting” α, allowing

us to write F as a direct sum I 0 ⊕ J 1 . If F is furthermore injective, then J 1 is
injective too. This is the beginning of an induction.

29.4.F. E XERCISE . (We continue the notation of the previous exercise.) Identify
ker(K → π!? I r+1 ) with π−1 Ext rY (π∗ OX , ωY ).

29.4.G. E XERCISE . Put together the pieces above to complete the proof of Lemma 29.4.5.
□
We are now ready to prove Theorem 29.4.3.

29.4.6. Proof of Theorem 29.4.3. Suppose F is a coherent sheaf on X. We wish to

describe an isomorphism
∼
HomX (F , π−1 Ext rY (π∗ OX , ωY )) o / Hn (X, F )∨ ,

functorial in F . By Lemma 29.4.5, we have a functorial isomorphism

∼
HomX (F , π−1 Ext rY (π∗ OX , ωY )) o / ExtrY (π∗ F , ωY ).

By functorial Serre duality for Ext on Y, we have a functorial isomorphism

∼
ExtrY (π∗ F , ωY ) o / HN−r (Y, π∗ F )∨ .

As F is a coherent sheaf on X, and N − r = n, the right side is precisely Hn (X, F )∨ .

□

29.4.7. Applying Theorem 29.4.3. We first apply Theorem 29.4.3 in the special
case where X is an effective Cartier divisor on Y. We can compute the dualizing
sheaf π−1 Ext 1Y (π∗ OX , ωY ) by computing Ext 1Y (π∗ OX , ωY ) using any locally free
resolution (on Y) of π∗ OX (Exercise 29.2.E). But π∗ OX has a particularly simple
resolution, the closed subscheme exact sequence (9.1.2.1) for X:

(29.4.7.1) 0 / OY (−X) / OY / π∗ OX / 0.

We compute Ext • (π∗ OX , ωY ) by truncating (removing the π∗ OX ), and applying

Hom (·, ωY ): Ext • (π∗ OX , ωY ) is the cohomology of
0 / Hom Y (OY , ωY ) / Hom (OY (−X), ωY ) / 0,

i.e., 0 / ωY / ωY ⊗ OY (X) / 0.

We immediately see that Ext i (π∗ OX , ωY ) = 0 if i ̸= 0, 1. Furthermore, Ext 0 (π∗ OX , ωY ) =

0 by Proposition 29.4.4. (Unimportant aside: you can use this to show that ωY has
no embedded points.)
We now consider ωX = coker(ωY → ωY ⊗ OY (X)). Tensoring (29.4.7.1) with
the invertible sheaf OY (X), and then tensoring with ωY , yields

ωY / ωY ⊗ OY (X) / ωY ⊗ π∗ (OY (X)|X ) / 0.

 © 2024 Ravi Vakil. Published by Princeton University Press. 809

The right term ωY ⊗π∗ (OY (X)|X ) is often (somewhat informally) written as ωY (X)|X .
Thus
ωX = coker(ωY → ωY ⊗ OY (X)) = ωY (X)|X ,
and this identification is canonical.
We have shown the following.

29.4.8. Proposition (the adjunction formula). — Suppose that Y is a Cohen-Macaulay

projective scheme of pure dimension n (which satisfies functorial Serre duality for Ext by
Corollary 29.3.14), and X is an effective Cartier divisor on Y. (Hence X is Cohen-Macaulay
by §26.2.4, and thus satisfies functorial Serre duality for Ext by Corollary 29.3.14.) Then
ωX = ωY (X)|X . In particular, if ωY is an invertible sheaf on Y, then ωX is an invertible
sheaf on X.
As an immediate application, we have the following.

29.4.H. E XERCISE . Suppose X is a complete intersection in Pn , of hypersurfaces

of degrees d1 , . . . , dr . (Note that X is Cohen-Macaulay by Proposition 26.2.6, and
thus satisfies functorial Serre duality for Ext, by Corollary 29.3.14.) Show that
ωX = ∼ OX (−n − 1 + P di ). If furthermore X is smooth, show that ωX = ∼ det ΩX .
(Hint for the last sentence: use the adjunction formula for K , Exercise 21.5.B).
But we can say more.

29.4.I. E XERCISE . Suppose π : X ,→ Y is a codimension r regular embedding with

normal sheaf NX/Y (which is locally free, by Proposition 21.2.16(b)). Suppose L is
an invertible sheaf on Y.
(a) Show that Ext iY (π∗ OX , L ) = 0 if i ̸= r.
∼
(b) Describe a canonical isomorphism π−1 Ext rY (π∗ OX , L ) ←→ (det NX/Y ) ⊗OX L |X .
Hint for both parts: deal with the case L = O first, then notice that the question is
local, if you have solved (b) carefully and correctly.
(Note that because Exercise 29.4.I has nothing explicitly to do with duality, we
have no projectivity assumptions on Y; it is a completely local question.) From
Exercise 29.4.I we deduce the following.

29.4.J. I MPORTANT E XERCISE . Suppose X is a codimension r regular embedding in

Pn
k . (Then X is Cohen-Macaulay by Proposition 26.2.6, and thus satisfies functorial
Serre duality for Ext by Corollary 29.3.14.) Show that
∼ (det NX/Pn ) ⊗ ωPn |X .
ωX = π−1 Ext r n (π∗ OX , ωPn ) =
P
In particular, ωX is an invertible sheaf.

29.4.9. Aside: Gorenstein singularities. If a pure-dimensional projective k-scheme X

has ωX/k an invertible sheaf, we say that X is Gorenstein. Given Proposition 29.4.8,
you can show that this is a condition that can be checked (stalk-)locally at the
closed points of a pure-dimensional projective k-scheme, similar to the definition of
Cohen-Macaulayness. Can you write down such a definition? All “local complete
intersection” k-schemes are Gorenstein, and all Gorenstein k-schemes are Cohen-
Macaulay.

29.4.K. I MPORTANT E XERCISE . Suppose X is a smooth pure codimension r sub-

variety of Pn
k (and hence a regular embedding, by Exercise 13.2.M(b)). Show that
810 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

ωX =∼ KX . Hint: both sides satisfy adjunction (see Exercise 21.5.B for adjunction
∼ (det NX/Y ) ⊗ (KPn |X ).
for Ω): they are isomorphic to (det NX/Y ) ⊗ (ωPn |X ) =

29.4.10. Remark. As long promised, the version of Serre duality given in Theo-
rem 18.5.1 and Further Miracle 18.5.2 now follows by combining Corollary 29.3.10
with Exercise 29.4.K.

Aux yeux de ces amateurs d’inquiétude et de perfection, un ouvrage n’est jamais

achevé, mot qui pour eux n’a aucun sens, mais abandonné; et cet abandon, qui le livre
aux flammes ou au public (et qu’il soit l’effet de la lassitude ou de l’obligation de livrer),
leur est une sorte d’accident, comparable à la rupture d’une réflexion . . .
In the eyes of those lovers of anxiety and perfection, a work is never finished — a word
that for them has no meaning — but abandoned; and this abandonment, whether to the
flames or to the public (and which is the result of weariness or an obligation to deliver) is a
kind of accident to them, like the breaking off of a reflection . . .
— P. Valéry, [Val, p. 1497]

Well, I’ve been talking to you for two years and now I’m going to quit. In some ways
I would like to apologize, and other ways not. I hope –— in fact, I know –— that two or
three dozen of you have been able to follow everything with great excitement, and have had
a good time with it. But I also know that “the powers of instruction are of very little efficacy
except in those happy circumstances in which they are practically superfluous.” So, for
the two or three dozen who have understood everything, may I say I have done nothing
but shown you the things. For the others, if I have made you hate the subject, I’m sorry. I
never taught elementary physics before, and I apologize. I just hope that I haven’t caused a
serious trouble to you, and that you do not leave this exciting business. I hope that someone
else can teach it to you in a way that doesn’t give you indigestion, and that you will find
someday that, after all, it isn’t as horrible as it looks.
— R. Feynman, [FLS, Epilogue]
 Bibliography

[Al] P. Aluffi, Algebra, Chapter 0, Grad. Stud. in Math. 104, Amer. Math. Soc., Providence, RI,
2009.
[Ar1] M. Artin, On the solutions of analytic equations, Invent. Math. 5 (1968), no. 4, 277–291.
[Ar2] , Algebraic approximation of structures over complete local rings, Publ. Math. IHES 36
(1969), 23–58.
[Ar3] , Algebra, Prentice Hall, Inc., Englewood Cliffs, NJ, 1991.
[At1] M. F. Atiyah, On analytic surfaces with double points, Proc. Roy. Soc. London Ser. A 247
(1958), 237—244.
[At2] , Mathematics in the 20th century, Amer. Math. Monthly 108 (2001), no. 7, 654–666.
[AtM] M. F. Atiyah and I. G. Macdonald, Introduction to Commutative Algebra, Addison-Wesley
Publ. Co., Reading, MA, 1969.
[Ba] L. Bădescu, Algebraic Surfaces, V. Maşek trans., Universitext, Springer-Verlag, New York,
2001.
[BCDKT] M. Baker, B. Conrad, S. Dasgupta, K. Kedlaya, and J. Teitelbaum, p-adic geometry, D.
Savitt and D. Thakur ed., Univ. Lect. Series 45, Amer. Math. Soc., Providence, RI, 2008.
[BPV] W. Barth, C. Peters, and A. van de Ven, Compact Complex Surfaces, Ergeb. Math. Grenz. 4,
Springer-Verlag, Berlin, 1984.
[Be] A. Beauville, Complex Algebraic Surfaces, 2nd ed., R. Barlow trans., Cambridge U.P.,
Cambridge, 1996.
[BL] C. Birkenhake and H. Lange, Complex Abelian Varieties, 2nd ed., Grund. Math. Wiss. 302,
Springer-Verlag, Berlin, 2004.
[BLR] S. Bosch, W. Lütkebohmert, and M. Raynaud, Néron Models, Ergeb. Math. Grenz. 21,
Springer-Verlag, Berlin, 1990.
[Bo] N. Bourbaki, Éléments de mathématique, Fasc. XXX: Algèbre Commutative, Actualités Scien-
tifiques et Industrielles, no. 1308, Hermann, Paris, 1964.
[BGS] S. Boissière, O. Gabber, and O. Serman, Sur le produit de variétés localement factorielles ou
Q-factorielles, arxiv:1104.1861v1, preprint 2011.
[BP] R. Brown and T. Porter, Analogy, concepts and methodology, in mathematics, UWB Math
Preprint, May 26, 2006.
[CL] F. Call and G. Lyubeznik, A simple proof of Grothendieck’s theorem on the parafactoriality
of local rings, in Commutative algebra: syzygies, multiplicities, and birational algebra (South
Hadley, MA, 1992), Contemp. Math. 159, Amer. Math. Soc., Providence, RI, 1994, 15–18.
[Carr] L. Carroll, Alice’s Adventures in Wonderland, Macmillan, London, 1865.
[ChJLO] R. Cheng, L. Ji, M. Larson, and N. Olander, Theorem of the base, in Stacks Project Expository
Collection (SPEC), 163–193, London Math. Soc. Lecture Note Ser. 480, Cambridge U.P.,
Cambridge, 2022.
[Co] B. Conrad, Deligne’s notes on Nagata compactifications, J. Ramanujan Math. Soc. 22 (2007),
no. 3, 205–257.
[CP] V. Cossart and O. Piltant, Resolution of singularities of arithmetical threefolds, J. Algebra 529
(2019), no. 7, 268–535.
[Dan] V. I. Danilov, Samuel’s conjecture, Mat. Sb. (N.S.) 81 (123) (1970), 132–144.
[Dav] E. D. Davis, Ideals of the principal class, R-sequences and a certain monoidal transformation,
Pacific J. Math. 20 (1967), 197–205.
[DW] R. Dedekind and H. Weber, Theory of algebraic functions of one variable, J. Stillwell trans.
History of Math. 39, Amer. Math. Soc., Providence, RI; London Math. Soc. London, 2012.

811
812 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

[De] P. Deligne, Courbes elliptiques: formulaire d’après Tate, in Modular Functions of One Variable
IV (Proc. Internat. Summer School, Univ. Antwerp, Antwerp, 1972), 53–73, Lect. Notes in
Math. 476, Springer-Verlag, Berlin, 1975.
[DeI] P. Deligne and L. Illusie, Relèvements modulo p2 et décomposition du complexe de de Rham,
Invent. Math. 89 (1987), no. 2, 247–270.
[Di] J. Dieudonné, On regular sequences, Nagoya Math. J. 27 (1966), 355–356.
[Do] I. Dolgachev, Luigi Cremona and cubic surfaces, in Luigi Cremona (1830–1903), 55–70, Incontr.
Studio, 36, Instituto Lombardo di Scienze e Lettere, Milan, 2005.
[DS] R. Donagi and R. C. Smith, The structure of the Prym map, Acta Math. 146 (1981), no. 1–2,
25–102.
[DZ] H. Duan and X. Zhao, Schubert calculus and the integral cohomology of exceptional Lie groups,
arXiv:0711.2541v10, preprint 2007.
[DF] D. Dummit and R. Foote, Abstract Algebra, 3rd ed., Wiley and Sons, Inc., Hoboken, NJ,
2004.
[E] D. Eisenbud, Commutative Algebra with a View to Algebraic Geometry, Grad. Texts in Math.
150, Springer-Verlag, New York, 1995.
[EGH] D. Eisenbud, M. Green, and J. Harris, Cayley-Bacharach theorems and conjectures, Bull. AMS
33 (1996), no. 3, 295–324.
[Ek] P. Eklof, Lefschetz’s principle and local functors, Proc. Amer. Math. Soc. 37 (1973), no. 2, 1973.
[EE] E. Enochs and S. Estrada, Relative homological algebra in the category of quasi-coherent sheaves,
Adv. Math. 194 (2005), 284–295.
[FGIKNV] B. Fantechi, L. Göttsche, L. Illusie, S. Kleiman, N. Nitsure and A. Vistoli, Fundamental
algebraic geometry: Grothendieck’s FGA explained, Mathematical Surveys and Monographs
123, Amer. Math. Soc., Providence, RI, 2005.
[FK] H. M. Farkas and I. Kra, Riemann Surfaces, Grad. Texts in Math. 71, Springer-Verlag, New
York, 1980.
[FLS] R. Feynman, R. Leighton, and M. Sands, The Feynman Lectures on Physics, Volume III,
Addison Wesley, 1971.
[Fi] F. Scott Fitzgerald, The Crack-Up: A desolately frank document from one for whom the salt of
life has lost its savor, Esquire Magazine, February 1936, Esquire Inc., Chicago, IL.
[FR] G. Frey and H.-G. Rück, The strong Lefschetz principle in algebraic geometry, manuscripta
math. 55 (1986), no. 3–4, 385–401.
[F1] W. Fulton, Young Tableaux: With Applications to Representation Theory and Geometry, Cam-
bridge U.P., Cambridge, 1997.
[F2] , Intersection Theory, 2nd ed., Ergeb. Math. Grenz. 2, Springer-Verlag, New York,
1998.
[Ge] S. Germain, Œuvres Philosophiques de Sophie Germain (nov. éd.), Librarie de Firmin-Didot
et Cie, Paris, 1896.
[GS] P. Gille and T. Szamuely, Central Simple Algebras and Galois Cohomology, Cambridge U.P.,
Cambridge, 2006.
[GW] U. Görtz and T. Wedhorn, Algebraic Geometry I, Vieweg + Teubner, Wiesbaden, 2010.
(Appendices C, D, and E, collecting properties of schemes and morphisms and their
relationships, are alone worth the price.)
[GR] H. Grauert and R. Remmert, Coherent Analytic Sheaves, Grund. Math. Wiss. 265, Springer-
Verlag, Berlin, 1984.
[GH1] P. Griffiths and J. Harris, Principles of Algebraic Geometry, Wiley-Interscience, New York,
1978.
[GH2] , On the Noether-Lefschetz theorem and some remarks on codimension two cycles, Math.
Ann. 271 (1985), 31–51.
[Gr1] A. Grothendieck, Sur quelques points d’algèbre homologique, Tôhoku Math. J. (2) 9 (1957),
119–221.
[Gr2] , Sur la classification des fibrés holomorphes sur la sphère de Riemann, Amer. J. Math.
79 (1957), 121–138.
[Gr3] , The cohomology theory of abstract algebraic varieties, in Proc. Internat. Congress Math.
(Edinburgh, 1958), Cambridge U.P., New York, 1960.
[Gr4] , Fondements de la géométrie algébrique, in Collected Bourbaki Talks, Secrétariat
Mathématique, Paris, 1962.
[Gr5] , Récoltes et Semailles, unpublished.
 © 2024 Ravi Vakil. Published by Princeton University Press. 813

[Gr-EGA] (w. J. Dieudonné), Éléments de Géométrie Algébrique, Publ. Math. IHES, Part I: 4
(1960); Part II: 8 (1961); Part III: 11 (1961), 17 (1963); Part IV: 20 (1964), 24 (1965), 28 (1966),
32 (1967).
[Gr-EGA’] (w. J. Dieudonné), Éléments de Géométrie Algébrique I: Le langage des schémas, Grund.
Math. Wiss. 166 (2nd ed.), Springer-Verlag, Berlin-New York, 1971.
[GrS] A. Grothendieck and J.-P. Serre, Grothendieck-Serre Correspondence, P. Colmez and J.-P.
Serre ed., C. Maclean trans., Amer. Math. Soc. and Soc. Math. de France, Paris, 2004.
[H1] J. Harris, Galois groups of enumerative problems, Duke Math. J. 46 (1979), no. 4, 685–724.
[H2] , Algebraic Geometry: A First Course, Grad. Texts in Math. 133, Springer-Verlag,
New York, 1995.
[Ha1] R. Hartshorne, Algebraic Geometry, Grad. Texts in Math. 52, Springer-Verlag, New York-
Heidelberg, 1977.
[Ha2] , Ample Subvarieties of Algebraic Varieties, Lect. Notes in Math. 156, Springer-Verlag,
Berlin-New York, 1970.
[Hat] A. Hatcher, Algebraic Topology, Cambridge U.P., Cambridge, 2002.
[HM] M. Hazewinkel and C. Martin, A short elementary proof of Grothendieck’s theorem on algebraic
vector bundles over the projective line, J. Pure Appl. Algebra 25 (1982), no. 2, 207–211.
[He] R. C. Heitmann, Examples of noncatenary rings, Trans. Amer. Math. Soc. 247 (1979), 125–136.
[Hil] D. Hilbert, Ueber die Theorie der algebraischen Formen, Math. Ann. 36 (1890), 473–534.
[Hir] H. Hironaka, Resolution of singularities of an algebraic variety over a field of characteristic zero.
I, II, Ann. of Math. (2) 79 (1964), 109-203; ibid. (2) 79 (1964), 205-326.
[Ig] J. Igusa, On some problems in abstract algebraic geometry, Proc. Nat. Acad. Sci. 41 (1955),
964–967.
[Ii] S. Iitaka, Algebraic Geometry: An Introduction to Birational Geometry of Algebraic Varieties,
Grad. Texts in Math. 76, Springer-Verlag, New York-Berlin, 1982.
[Il] L. Illusie, Frobenius and Hodge degeneration, in Introduction to Hodge Theory, SMF/AMS
Texts and Monographs 8, 99–149, 2002.
[dJ] A. J. de Jong, Smoothness, semi-stability and alterations, IHES Publ. Math. 83 (1996), 51–93.
[KN] S. Kaji and M. Nakagawa, The Chow rings of the algebraic groups of type E6 , E7 , and E8 ,
arXiv:0709.3702v3, preprint 2007.
[KS1] M. Kashiwara and P. Schapira, Sheaves on Manifolds, Grund. Math. Wiss. 292, Springer-
Verlag, Berlin, 1994.
[KS2] , Categories and sheaves, Grund. Math. Wiss. 332, Springer-Verlag, Berlin, 2006.
[KPV] K. Kedlaya, B. Poonen, and R. Vakil, The William Lowell Putnam Mathematical Competition
1985–2000: Problems, Solutions, and Commentary, Math. Ass. of Amer., Washington, DC,
2002.
[Kl1] S. Kleiman, Toward a numerical theory of ampleness, Ann. Math. (2) 84 (1966), no. 3, 293–344.
[Kl2] , The transversality of a general translate, Compositio Math. 28 (1974), no. 3, 287–297.
[Kl3] , Cartier divisors versus invertible sheaves, Comm. Alg. 28 (2000), no. 12, 5677–5678.
[Ko1] J. Kollár, Rational Curves on Algebraic Varieties, Ergeb. Math. Grenz. 32, Springer-Verlag,
Berlin, 1996.
[Ko2] , Lectures on Resolutions of Singularities, Annals of Math. Stud. 166, Princeton U.P.,
Princeton, NJ, 2007.
[Lan] S. Lang, Algebra (rev. 3rd ed.), Grad. Texts in Math. 211, Springer-Verlag, New York, 2002.
[Laz] R. Lazarsfeld, Positivity in Algebraic Geometry I: Classical setting: line bundles and linear
series, Ergeb. Math. Grenz. 48, Springer-Verlag, Berlin, 2004.
[Lef] S. Lefschetz, L’analysis situs et la géométrie algébrique, Gauthier-Villars, Paris, 1924.
[Lew] R. Lewontin, The Triple Helix: Gene, Organism, and Environment, Harvard U.P., Cambridge
MA, 2000.
[Lit] J. E. Littlewood, Littlewood’s Miscellany, Cambridge U.P., Cambridge, 1986.
[Liu] Q. Liu, Algebraic Geometry and Arithmetic Curves, R. Erné trans., Oxford Grad. Texts in
Math. 6, Oxford U.P., Oxford, 2002.
[Lur] J. Lurie, Higher Algebra, preprint, September 18, 2017.
[Lüt] W. Lütkebohmert, On compactification of schemes, Manuscripta Math. 80 (1), 95–111.
[Mac] S. Mac Lane, Categories for the Working Mathematician, 2nd ed., Grad. Texts in Math. 5,
Springer-Verlag, New York, 1998.
[Mat1] H. Matsumura, Commutative Algebra, 2nd ed., Math. Lecture Note Series 56, Ben-
jamin/Cummings Publ. Co., Inc., Reading, MA, 1980.
814 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

[Mat2] , Commutative Ring Theory, M. Reid trans., 2nd ed, Cambridge Stud. in Adv. Math.
8, Cambridge U.P., Cambridge, 1989.
[MO33489] http://mathoverflow.net/questions/33489/
[MO68421] Answers by G. Elencwajg and M. Kim to http://mathoverflow.net/questions/68421/
[MO70920] Answers by T. Ekedahl to http://mathoverflow.net/questions/70920/
[MO90551] Answer by M. Brandenburg to http://mathoverflow.net/questions/90551/
[MO129242] Answer by L. Moret-Bailly to http://mathoverflow.net/questions/129242/
[MO152008] Answer by J. Stillwell to http://mathoverflow.net/questions/152008/
[MO286390] Answers to https://mathoverflow.net/questions/286390/
[Mc] C. McLarty, The Rising Sea: Grothendieck on simplicity and generality, in J. J. Gray and K.
H. Parshall eds., Episodes in the History of Modern Algebra (1800–1950), Amer. Math. Soc.,
Providence, RI, 2007.
[Mu1] D. Mumford, Pathologies of modular algebraic surfaces, Amer. J. Math. 83 (no. 2), Apr. 1961,
339–342.
[Mu2] , Lectures on Curves on an Algebraic Surface, Princeton U.P., Princeton NJ, 1966.
[Mu3] , Abelian Varieties, TIFR Studies in Math. 5, Oxford U.P., London, 1970.
[Mu4] , Curves and Their Jacobians, Univ. of Michigan Press, Ann Arbor, MI, 1975.
[Mu5] , Hilbert’s fourteenth problem — the finite generation of subrings such as rings of
invariants, in Mathematical Developments Arising from Hilbert Problems, Part 2, F. Browder
ed., Amer. Math. Soc., Providence RI, 1976.
[Mu6] , Algebraic Geometry I: Complex Projective Varieties, Classics in Math., Springer-
Verlag, Berlin, 1995.
[Mu7] , The Red Book of Varieties and Schemes (2nd expanded ed.), Lect. Notes in Math.
1358, Springer-Verlag, Berlin, 1999. (Caution: in order to keep readers on their toes, a
number of typos were introduced between the first and second editions.)
[Mur] D. Murfet, http://therisingsea.org/notes/ModulesOverAScheme.pdf
[N] F. Nietzsche, Twilight of the Idols, or, How to Philosophize with a Hammer, 1889.
[OB] A. Ogus and G. Bergman, Nakayama’s lemma for half-exact functors, Proc. Amer. Math. Soc.
31 (1972), 67–74.
[OSS] C. Okonek, M. Schneider, and H. Spindler, Vector Bundles on Complex Projective Spaces,
corrected reprint of the 1988 edition, Birkhäuser/Springer Basel AG, Basel, 2011.
[Pi] P. Picasso, Statement to Marius de Zayas, 1923, in Picasso Speaks, The Arts, New York, May
1923, pp. 315-26.
[Po1] H. Poincaré, Analysis situs, J. de l’École Polytechnique (2) 1 (1895), 1–123.
[Po2] , Papers on topology: Analysis situs and its five supplements, J. Stillwell trans., Amer.
Math. Soc., Providence, RI, 2010.
[P1] B. Poonen, Varieties without extra automorphisms, I: Curves, Math. Res. Lett. 7 (2000), no. 1,
67–76.
[P2] , Rational Points on Varieties, Grad. Stud. Math., 186, Amer. Math. Soc., Providence,
RI, 2017.
[Ra] M. Raynaud, Contre-exemple au vanishing theorem en caractéristique p > 0, in C. P. Ra-
manujam — A Tribute, 273–278, TIFR Studies in Math. 8, Springer, Berlin-New York,
1978.
[RS] M. Reed and B. Simon, Methods of Modern Mathematical Physics I: Functional Analysis,
Academic Press, Inc., New York, 1980.
[Rei1] M. Reid, Undergraduate Algebraic Geometry, London Math. Soc. Student Texts 12, Cam-
bridge U.P., Cambridge, 1988.
[Rei2] M. Reid, Undergraduate Commutative Algebra, London Math. Soc. Student Texts 29, Cam-
bridge U.P., Cambridge, 1995.
[Rem] R. Remmert, Local theory of complex spaces, in Several complex variables VII: Sheaf-theoretical
methods in complex analysis, Encl. of Math. Sci. 74, Springer-Verlag, Berlin, 1994.
[Roh] F. Rohrer, Quasicoherent sheaves on toric schemes, Expo. Math. 32 (2014), 33–78.
[RotV] M. Roth and R. Vakil, The affine stratification number and the moduli space of curves, CRM
Proceedings and Lecture Notes 38, Univ. de Montréal, 2004, 213–227.
[Sa] P. Salmon, Sulla fattorialità delle algebre graduate e degli anelli locali, Rend. Sem. Mat. Univ.
Padova 41 (1968), 119–138.
[Schm] H. L. Schmid, Über die Automorphismen eines algebraischen Funktionenkörpers von
Primzahlcharacteristik, J. Reine Angew. Math. 179 (1938), 5–14.
 © 2024 Ravi Vakil. Published by Princeton University Press. 815

[Se1] J.-P. Serre, Faisceaux algébriques cohérents, Ann. of Math. (2) 61 (1955), 197–278.
[Se2] , Sur la dimension cohomologique des anneaux et des modules noethériens, in Proc. Intern.
Symp. on Alg. Number Theory, 176–189, Tokyo-Nikko, 1955.
[Se3] , Géométrie algébrique et géométrie analytique, Ann. de l’institut Fourier 6 (1956),
1–42.
[Se4] , Sur la topologie des variétés algébriques en caractéristique p, in Symposium Interna-
cional de Topologı́a Algebraica, International Symposium on Algebraic Topology, 24–53, Univ.
Nacional Autónoma de México and UNESCO, Mexico City, 1958.
[Se5] , A Course in Arithmetic, Grad. Texts in Math. 7, Springer-Verlag, New York, 1973.
[SGA2] A. Grothendieck, Séminaire de Géométrie Algébrique du Bois Marie 1962: Cohomologie locale
des faisceaux cohérents et Théorèmes de Lefschetz locaux et globaux (SGA2), North-Holland
Publ. Co., Amsterdam, 1968.
[SGA6] P. Berthelot, A. Grothendieck, and L. Illusie eds., Séminaire de Géométrie Algébrique du Bois
Marie 1966–67: Théorie des intersections et théorème de Riemann Roch (SGA 6), Lect. Notes in
Math. 225, Springer-Verlag, Berlin-New York, 1971.
[Sh] I. Shafarevich, Basic Algebraic Geometry 1: Varieties in Projective Space, 2nd ed., M. Reid
trans., Springer-Verlag, Berlin, 1994.
[Si1] J. Silverman, The Arithmetic of Elliptic Curves, 2nd ed., Grad. Texts in Math. 206, Springer-
Verlag, New York-Berlin, 2009.
[Si2] , Advanced Topics in the Arithmetic of Elliptic Curves, Grad. Texts in Math. 151,
Springer-Verlag, New York-Berlin, 1994.
[Stacks] The Stacks Project Authors, Stacks Project, http://stacks.math.columbia.edu.
(To look up tags: http://stacks.math.columbia.edu/tags. To search:
http://stacks.math.columbia.edu/search.)
[Stic] H. Stichtenoth, Über die Automorphismengruppe eines algebraischen Funktionenkörpers von
Primzahlcharakteristik. I. Eine Abschätzung der Ordnung der Automorphismengruppe, Arch.
Math. (Basel) 24 (1973), 527–544.
[Ta] A. Tarski, A decision method for elementary algebra and geometry, RAND Corp., Santa Monica,
CA, 1948.
[TWi] R. Taylor and A. Wiles, Ring-theoretic properties of certain Hecke algebras, Ann. of Math. (2)
141 (3) (1995), 553–572.
[To] L. Tolstoy, Anna Karenina, http://www.gutenberg.org/files/1399/1399-h/1399-h.htm.
[Tu] L. Tu, An Introduction to Manifolds, Universitext, Springer-Verlag, New York, 2011.
[Vak1] R. Vakil, The Rising Sea: Foundations of Algebraic Geometry, preprint.
[Vak2] , Puzzling through exact sequences: A bedtime story with pictures,
https://www.3blue1brown.com/content/blog/exact-sequence-picturebook/PuzzlingThroughExactSeq
[Val] P. Valéry, Au sujet du Cimetière Marin, in Œuvres, édition Jean Hytier, tome I, Gallimard,
La Pléiade, Paris, 1957.
[van] B. L. van der Waerden, Algebra vol. II, J. R. Schulenberger trans., Springer-Verlag, New
York, 1991.
[Var] V. S. Varadarajan, Lie Groups, Lie Algebras, and Their Representations, Grad. Texts in Math.
102, Springer-Verlag, New York-Berlin, 1984.
[Vo] C. Voisin, Hodge Theory and Complex Algebraic Geometry I and II, L. Schneps trans., Cam-
bridge Stud. in Adv. Math. 76 and 77, Cambridge U.P., Cambridge, 2007.
[Weib] C. Weibel, An Introduction to Homological Algebra, Cambridge Stud. in Adv. Math. 38,
Cambridge U.P., Cambridge, 1994.
[Weil] A. Weil, Scientific works — Collected papers Vol. II (1951–1964), Springer-Verlag, New
York-Heidelberg, 1979.
[Wey] H. Weyl, Philosophy of Mathematics and Natural Science, Princeton U.P., Princeton, NJ, 2009.
[Wi] A. Wiles, Modular elliptic curves and Fermat’s last theorem, Ann. of Math. (2) 141 (3) (1995),
443–551.
[Wł] J. Włodarczyk, Toroidal varieties and the weak factorization theorem, Invent. Math. 154 (2003),
no. 2, 223–331.
[Z] O. Zariski, Foundations of a general theory of birational correspondences, Trans. Amer. Math.
Soc. 53 (1943), 490–542.
 Index

(−1)-curve, 582, 653, 759, 765, 767, 769, 778, Y red (reduction of a scheme), 269, 308
785, 786, 789, see also Castelnuovo’s ∆ (diagonal locally closed subscheme of
Criterion; blowing down X ×Y X), 314
A(I)• (Rees algebra), 395 Γ (U, F ) (section of sheaf F over U), 74
A^ (completion), 769 Γ adjoint to Spec, 209
A-bilinear, 38 Γ• (graded analog of Γ ), 454, 454
Creg , 564 ΩA/B (module of Kähler differentials), 596,
F ⊗ G for quasicoherent sheaves, 176 597, 604
F (U) (section of sheaf F over U), 74 ΩX/Y (relative cotangent sheaf), 606
F (m) (“twist” of F , = F ⊗ O(m)), 428 ΩiX/Y (sheaf of relative i-forms), 622
F |U (restriction of sheaf to open subset), 76 χ(X, F ) (Euler characteristic), 427, 507
F sh (sheafification of presheaf F ), 84 `
(coproduct), 41, 46, 47, 140
F ∨ (dual of OX -module F ), 80 γπ (graph morphism), 316
Fp (stalk of (pre)sheaf F ), 74 ,→, 231, 253, 257, see (open, closed, locally
Flf (torsion-free quotient of F ), 412 closed) embedding
Ftors (torsion subsheaf of F ), 412 λ-invariant, 557
f |U = resVU (f) (restriction of f to U), 74 µn (group scheme), 227
G |X (pullback of G to X), 416 νd (Veronese embedding), 262
√
I (radical of ideal I), 116 ω (dualizing sheaf), 514, 538, 539, 542, 548, 550,
ip (inclusion of point p), 77 551, 622, 793, 794, 795, 795, 796, 796–810
ip,∗ (pushforward along ip ), 76 ⊕ (direct sum, in an additive category), 53
(L1 · L2 · · · Ln · F ) (intersection of L1 , . . . , ⊗ (tensor product), 37
Ln with F ), 575 ⊗OX (tensor product of OX -modules), 92
| L | (complete linear series), 430 ⊠ (box-times), 433, 433, 460, 461, 583, 734
(M : x), 693 ϕp p
q , ϕZ (cohomology and base change map),
M(I)• (A(I)• -module), 395 729, 730, see also push-pull map
M(m)• (“shift” or “twist” of M• by m), 428 π-ample, see ampleness, with respect to π
M• (Z-graded module), 422 (relatively ample)
Mtors (torsion submodule of M), 173 π-very ample, see very ampleness, with respect
M • (quasicoherent sheaf corresponding to to π (relatively very ample)
g
graded module M• ), 422 π∗ s (pullback of section s), 419
in terms of “compatible germs”, 423 π∗ (pullback of O-module / quasicoherent
is an exact functor, 423 sheaf), 205, 416, 418, 419, see also pullback;
may not be an isomorphism, 423 projection formula; push-pull map
Me (quasicoherent sheaf corresponding to a π!? (occasional right adjoint to π∗ ), 480, 480,
module M), 134 481, 528, 802–804
[p] (point of Spec A corresponding to p), 102 computable affine-locally, 480, 481
pa (X) (arithmetic genus), 510 for O-modules and closed embeddings, 806
Pg (X) (geometric genus), 623 π−1 , 92, 93, see inverse image sheaf
S−1 A (localization of ring), 35 π∗ (pushforward sheaf), 77, see sheaf,
S−1 M (localization of module), 37 pushforward
S (constant sheaf), 76 π! , 96, 675, see extension by zero
s ⊗ t (tensor of sections), 177, 401 ρ(X) (Picard number), 513, 513, 585, 586, 626

817
818 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

finiteness of, 789 Affine Communication Lemma, 134, 161, 162,

∧• M, ∧• F (exterior algebra), 408, 408, 409, 164, 171, 172, 181, 182, 206, 229, 231, 237,
409, 473, 474, 622, 654, 704 241, 243, 244, 449
affine cone, 149, 155, 263, 263, 264, 339, 341,
A1 (affine line), 102, 104, 105 348, 349, 369, 424, 449, 644, 752
An (affine n-space), 106 relative version, 485
AnX , 282 affine cover cohomology vanishing, 494, 495,
A-scheme, 163 496, 502, 527, 530
Ad′ (X) (variant of Chow group), 588 affine line, 102, 103, 104, 105, 255
map to homology, 589 functorial characterization, 223
An -singularities, 648, 649, 774 affine morphisms, 237
Ab (category of abelian groups), 30 Ri>0 π∗ F = 0 for all quasicoherent F , 525
is an abelian category, 52 π∗ exact (for quasicoherent sheaves), 415,
AbX (category of sheaves of abelian groups on 780
X), 79, 80, 81, 91, 92 π∗ preserves cohomology of quasicoherent
has enough injectives, 673 sheaves, 494
is an abelian category, 90 affine-local on target, 237, 238
abelian category, 29, 31, 42, 51–53, 54, 54–60, 62, are quasicompact and quasiseparated, 237
81, 89, 171, 178, 179, 182, 183, 199, 399, 402, are separated, 316
663–665, 671, 674, 711, see also Ab; AbX ; as Spec , 479
CohX ; ComC ; complex; f.d.Veck ; ModA ; e.g., closed embeddings, 253
ModOX ; objects, in an abelian category; e.g., finite morphisms, 238
QCohX ; category; Veck ; enough injectives;
preserved by base change, 291
enough projectives; FHHF Theorem;
preserved by composition, 237
Jordan-Hölder package; spectral sequence
reasonable class, 292
e.g., coherent A-modules, 180
affine n-space, 106, 108, 112, 121
e.g., presheaves of abelian groups on X, 80
(over a field) is regular, 374
abelian cone associated to coherent sheaf, 480
(over any base) is flat, 687
abelian group scheme, 225
coordinate functions, 107
abelian semigroup, see semigroup (abelian)
functorial characterization, 224
abelian varieties, 328, 328, 330, 563
are abelian, 330 minus origin, or with doubled origin, see
recurring (counter)examples
absolute Frobenius, see Frobenius
abstract nonsense, 27 over a scheme X (An X ), 282
action of group variety/scheme, 227, 448, 475 affine open subscheme/subset, 139, 140
transitive, 630 affine plane, 105, 109, 111, 114, 115, 138, 142,
acyclic object, 667, 669, 670, 674, 676, 799, 801, 145, 324, 377, 635, 648
see also resolution (complexes) affine schemes/varieties, see schemes; varieties
additive category, 52, 53, 53, 54, 80, 81, 89 affine-diagonal morphisms, 320
additive functor, 53, 53, 58, 61, 665–667, 669, affine-local properties, 161, 162
677, 711 implied by stalk-local, 162, 230
ADE singularities, 649, 774 on the source, 231
adj(M) (adjugate matrix), 233, 235, 308 on the target, 230, 231
adjoints, 49 Alexandrov space, 676
(co)unit, 49 algebraic equivalence, 719
left, 49 implies numerical equivalence, 719
are right-exact, 60 algebraic field extension, see field extension
commute with colimits, 51 algebraic group, 328, 328
of forgetful functors, 51 Algebraic Hartogs’s Lemma, 146, 148, 165, 167,
right, 49 383, 383–385, 405, 440, 442, 450, 542, 622,
are left-exact, 60 765
commute with limits, 51, 61 algebraic space, 435
preserves injectives, 677 Alteration Theorem, 650
table of, 52 ample cone, 512, 585, 590, 592
adjunction formula, 759, 810 = interior of nef cone, 592
for a curve on a surface, 581 is open, 590, 592
for the canonical bundle, 623, 623, 759 ampleness, 457, 459, 585, see also Kleiman’s
for the dualizing sheaf, 805, 809, 809 criterion; Kodaira Embedding Theorem;
 © 2024 Ravi Vakil. Published by Princeton University Press. 819

Nakai-Moishezon criterion; Serre’s of locally Noetherian scheme, 195

cohomological criterion zerodivisors must vanish at one, 188, 191
⊠ ample is ample, 461 associated prime ideals, 187, 188, 189, 192, 194,
⊗ ample is ample, 461 347, 384, 690, 691, 748–750, see also
⊗ base-point-free is ample, 461 associated points; primary ideals
⊗ nef is ample, 592 and depth, 750
= positive degree for smooth projective annihilator of some element of M, 188, 191
integral curves, 540 e.g., minimal prime ideals, 193
better-behaved than very ampleness, 461 localization at, 190
cohomological criterion, 461 of a ring, 190, 190
implies nef, 513, 592 zerodivisors = element of associated prime,
in the absolute setting, 464 191
properties of, 464, 465 Associated-to-Separated Theorem, 324
many characterizations of, 460 Asymptotic Riemann-Roch, 578, 579, 581, 590
of vector bundle, 490 Atiyah flop, 653
open condition in families, 461 Auslander-Buchsbaum Theorem, 394, 394, 453,
over A, see ampleness, with respect to π 580, 747, 755
preserved by finite pullback, 461, 488, 511, automorphism group
528 infinitesimal, 626
with respect to π (relatively ample), 460, 465, of P1k , 447, 549, 550, 556, 557, 559, 567, 626
488 of Pnk , 432, 447, 551, 626, 701
affine-local on the target, 489 of a curve, 549, 550, 552, 554, 555, 567, 621,
analytification, 213, 213, 214, 319, 328, 387, 404, 622
453, 469, 509, 510, 513, 521, 570, 580, 610, of a scheme, 208
620, 625, see also varieties, complex of an object in a category, 30
of a morphism, 213 Axiom of Choice, 19, 661, 663, see also Zorn’s
André-Quillen homology, 601, 602 Lemma
AnnA m (annihilator ideal of m), 188
annihilator ideal, 188, 188, 191, 192, 633, 697
anticanonical bundle, 766, see also canonical, BlX Y (blow-up of Y along X), 637, 638
bundle base (of a topology), 86, see also presheaf, on a
arithmetic genus, see genus, arithmetic base; sheaf, on a base
arrow (category), see morphism (category) base (target of a morphism), 285
Artin-Rees Lemma, 359, 379, 394, 395, 395, 396, base change, 285
773 base ring, 151
Artin-Schreier cover, 619 base scheme, 208, 285
Artinian, see also finite length; Jordan-Hölder base change diagram/square, 285, see also
package; length Cartesian diagram/square; fibered
local ring, 186, 337, 387 diagram/square; pullback,
ring, 186 diagram/square
scheme, 186, 387 base field extension, 284
ascending chain condition, 125, 126, 127 base identity axiom / base gluability axiom,
AssA M (associated points/primes of M), 188, 87, 132–134, see also gluability axiom;
189 identity axiom
assassin, 189 base locus, 430, 651, 653, see also
associated graded, 362, 588, 787 scheme-theoretic, base locus
associated points, 138, 187, 188, 189, 195–197, base point, 430, 539, 591, 786
268, 270, 300, 324, 448, 452, 453, 469, 471, base-point-freeness, 430, 430, 457, see also
511, 518, 520, 576–578, 580, 581, 588, 610, global generation
611, 639, 690, 699, 700, 708, 747, see also ⊗ ample is ample, 461
Associated-to-Separated Theorem ⊗ very ample is very ample, 460
and depth, 747 and maps to projective space, 430
and flatness, 690 base-point-free locus, 430
behaviour in exact sequences, 191, 193 implied by very ampleness, 460
commutes with localization, 188, 194 independent of base field, 503
finitely many, 188 preserved by ⊗, see global generation
of coherent sheaf, 188 with respect to π (relatively base-point-free),
of integral schemes, 197 465, 488
820 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Bertini’s Theorem, 359, 372, 375, 375, 521, 523, to resolve singularities, 648, 650
551, 631, 652, 758, 759, see also Blow-Up Closure Lemma, 638, 639, 641, 641,
Kleiman-Bertini Theorem 648, 652
improved characteristic 0 version, 631 blowing down, 653, see also Weak Factorization
with weaker hypotheses, 377 Theorem
Betti numbers, 625, 730 (−1)-curve, 786, see also Castelnuovo’s
Bézout’s Theorem, 260, 263, 305, 373, 433, 520, Criterion
520, 521, 521, 553, 554, 563, 576, 752, 764 Borel-Moore homology, 589
for plane curves, 520, 543, 545, 581 boundary (homological algebra), 56
bilinear form (index of), 585 box-times, see ⊠
bilinear map, 38 branch
birational, 626 divisor, 617, 619
(rational) map, 217, 217, 218, 220, 221, 304, locus, 617, 631
469, 650, 651, 653, 765, 789 points, 545, 546, 547, 558, 618, 621
factorization of, 653 branched cover, 239, 546–548, 551, 556, 569,
invariant, 622, 789 612, 616, 619, 620, 685, see also double
e.g., geometric genus, 623 cover
e.g., h0 (X, , ΩiX/k ) = hi,0 622
e.g., jth plurigenus, 623 c1 (L)∪, 588, see also Chern class
e.g., Kodaira dimension, 624 Calabi-Yau variety, 353, 623, 624, see also
model, 789 trichotomy
model of integral curves, 465 Cancellation Theorem, 230, 312, 312, 313, 316
morphism, 217, 217, 304, 307, 466, 531, 636, for affine morphisms, 694
764, 765, 780, 781, 785, 786, 788 for closed embeddings, 460
Zariski’s Main Theorem version, 780, 782 for projective morphisms (“unreasonable”
varieties/schemes, 217, 217, 218, 466, 468, class), 487, 779, 788, 803
582, 623, 628, 635, 653, 789 for proper morphisms, 326, 328, 780, 784
may not be isomorphic, 449 for quasicompact morphisms, 237, 491, 494
often means isomorphic dense open for reasonable classes of morphisms, 312
subschemes, 217 for separated morphisms, 494
birational transform, see proper transform in Sets, 312
blow-up, 288, 362, 635, 638, see also BlX Y; EX Y; canonical
exceptional divisor; Hironaka’s Theorem; bundle KX , 514, 622, 623, 654, 759, 766
proper transform; resolution is Serre-dualizing, 514, 555, 612, 620, 622,
(singularities); total transform; universal 623, 625, 626, 794, 805
property; Weak Factorization Theorem curve, 551, 553, 554, 624
and line bundles, 653 embedding | K |, 550, 551, 551–554, 562
and the canonical bundle, 654 map | K |, 549, 550–552
commutes with flat base change, 692, 774 sheaf, see canonical, bundle KX
computable locally, 638, 641, 643 Caroll, L., 115
does not commute with general base change, Cartan-Eilenberg resolution, 669, 671, 672, 672,
692 673
exists, 642 Cartesian diagram/square, 40, see also base
is projective, 642 change diagram/square; fibered
locally principal closed subscheme, 639 diagram/square; pullback,
not flat, 691 diagram/square
of a nonreduced subscheme, 650 Cartier divisors, 18, 453, see also effective
of affine space, 480, 637 Cartier divisors
of an effective Cartier divisor, 638 Castelnuovo’s Criterion, 644, 765, 769, 778, 785,
of regular embeddings, 645 786, 788
of smooth in smooth is smooth, 646 category, 29, see also Ab; AbX ; abelian category;
of the plane, 288, 449, 582, 583, 635, 647, 758, additive category; CohX ; ComC ; f.d.Veck ;
763, 764, 766, 767 equivalence; Groups; ModA ; ModOX ;
as graph of rational map, 324 opposite; QCohX ; Rings; Sets; SetsX ; Sch;
preserves irreducibility and reducedness, SchS ; subcategory; Top; Veck
639 concrete, 31
Proj description, 642 filtered index, 47
to resolve base locus, 651 index, 44
 © 2024 Ravi Vakil. Published by Princeton University Press. 821

locally small, 29 intersections of, 282

of k-schemes/A-schemes/S-schemes, see preserved by base change, 282, 291
Schk /SchA /SchS preserved by composition, 254
of complexes, see ComC reasonable class, 291
of open subsets, 31 closed immersions, see closed embeddings
of schemes, see Sch closed map, 239, 250, 325, 533, see also
of subsets of a set, 31 universally closed morphisms
of vector bundles, 406 not preserved by base change, 325
small, 34, 44, 44, 48, 57 closed points, 107, 121, 157, 611, see also degree,
catenary rings, 342, 754 of closed point; Nullstellensatz
Cayley-Bacharach Theorem, 543 dense for finitely generated k-algebras, 121
Čech cohomology, 18, 311, 493, 499, 500, 501, dense for varieties, 158
666, 677–682, see also cohomology may not be dense, 122
= derived functor cohomology, 676 of Pn , 145, 147, 154, 261
k
= derived functor cohomology, 659, 674, 676, of Spec Q(t) ⊗Q C, 284
677 of Spec A correspond to maximal ideals, 121
and change of base field, 503 of affine n-space, 121
behaves well in flat families, 717 of locally finite type k-schemes, 164
commutes with filtered colimits, 494 quasicompact schemes have one, 157
commutes with flat base change, 691, 730 schemes need not have one, 158, 436, 437
dimensional vanishing, 494, 504, 654, 655, closed subscheme exact sequence, 254, 406,
703, 807 421, 422, 424, 451, 452, 508, 519, 529, 692,
more generally, 504 744, 787, 808
relative, 527 closed subscheme-theoretic image, see
of line bundles on projective space, 504 scheme-theoretic, image
pullback, 494 closed subschemes, 136, 140, 150, 154, 231, 246,
Čech complex, 499, 499–501, 503, 505, 506, 524, 253, 253, see also effective Cartier divisors;
547, 679–681, 717, 736, 737, 797 closed embeddings
Čech cover, 677, 680 = quasicoherent sheaves of ideals, 254
Čech resolution, 501, 718 arbitrary intersections of, 255
change of base, see base, change cut out by function s, 256
Chasles’ Theorem, 543 cut out by section s of locally free sheaf, 405
Chern class, 509, 510, 579, 587–589 finite union of, 255
Chevalley’s Theorem, 245, 247, 247–250, 265, intersections of, 282
289, 291, 352, 579, 629, 708 locally principal, 269, 270, 640, 642, 651, 715,
for finitely presented morphisms, 247, 290 see also divisor, locally principal
for locally finitely presented morphisms, 290 codimension ≤ 1, 347
Chinese Remainder Theorem, 136, 186, 283 pull back, 291
as geometric fact, 147 pull back, 282
Chow group, 589 closure, see scheme-theoretic, closure; Zariski,
Chow’s Lemma, 530, 531, 531, 532, 576 closure
other versions, 534 cocycle condition, 89, 144, 400
Cl(An k ) = 0, 445 codim (codimension), 335
Cl(X) (class group), 445 codimension
class (in set theory), 29 at most difference of dimensions, 336
class group, 170, 445, 445, 446, 448–450, see also inequality for fibers, 353
excision exact sequence of a point, 335
and unique factorization, 450, 451 of a prime ideal, 335
in number theory, 404, 446, 450 of an irreducible subset, 335
closed embeddings, 239, 253, 253 of regular embedding, 273
affine-local on target, 254, 302 pathologies, 336, 351
are affine, 253 sometimes difference of dimensions, 342
are finite, 253 cofiltered, 60
are monomorphisms, 256, 284 cofinal, 769, 770, 773, 783, 787, 791
are proper, 325, 326 Cohen Structure Theorem, 770
are separated, 320 Cohen-Macaulayness, 187, 342, 394, 515, 581,
base change by, 282 747, 751, 751–756, 794, 798, 802–805, 809,
in projective space, 259 see also Miracle Flatness; Serre duality
822 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

and adding a variable, 754 presheaf, 81

e.g., all dimension 0 Noetherian schemes, sheaf, 90
752 colim (colimit), 46
e.g., all dimension 0 Noetherian schemes, colimits, 46, see also universal property
751 are right-exact, 60
e.g., all dimension 1 Noetherian schemes commute with colimits, 51, 60
without embedded points, 751 commute with left-adjoints, 51, 60
e.g., all dimension 2 normal varieties, 755 e.g., cokernels, 60
e.g., Gorenstein, 809 e.g., localization, 48
e.g., regular embeddings in smooth e.g., stalks, 74
k-schemes, 752 filtered, 47
e.g., regular local rings, 747 commute with homology in ModA , 61,
fancy properties, 753 697
implies catenary, 754 in ModA are exact, 60, 61, 94
implies no embedded points, 751, 752 not always exact, 60
open condition, 754 in ModA (non-filtered case), 48
preserved by localization, 751, 754 may not commute with right-exact functors,
slicing criterion, 751 62
stalk-local, 751 collapse (of spectral sequence), 66
Cohen-Maculayness
comathematician, 46
e.g., regular local rings, 751
ComC (category of complexes in C ), 56
Coherence Theorem, see Grothendieck’s
is abelian category, 57
Coherence Theorem
commutative monoid, 53
coherent modules, 180, 181, 197–199, see also
compatible germs, 54, 82, 83, 83, 84, 86, 87, 94,
coherent sheaves
154, 175, 254, 423, 435
form an abelian category, 199
complete k-scheme, 325, see also proper
Homs are coherent, 199
morphisms
over non-Noetherian rings, 197
complete intersections, 274, 523, 554, 563, 623,
coherent sheaves, 155, 182, 399, see also
624, see also local complete intersections
coherent modules; finite type,
in affine space
quasicoherent sheaves; finitely presented,
are Cohen-Macaulay, 752
sheaves; Grothendieck’s Coherence
Theorem implies no embedded points, 752
abelian cone of, 480 in projective space, 523, 552–554, 562, 563,
affine-local nature, 182 624, 809
dual, 410 positive-dimensional ones are connected,
Euler characteristic, 507 523
flat = locally free, 698 complete linear series, see linear series
higher pushforward, see sheaf, higher complete with respect to an ideal I, 770
pushforward completion, 46, 198, 362, 701, 727, 769, 769–774,
non-Noetherian context, 198 777, 778, 790
not good notion in smooth geometry, 182 and Cohen-Macaulayness, 754
on X form an abelian category CohX , 182, commutes with finite direct sums, 773
199, 402 is flat, 772, 774
pushforward, see sheaf, pushforward not always exact, 772
cohomology, 18, 33, 56, see also Čech of A along I, 769
cohomology; cup product; derived preserves exact sequences of finitely
functor cohomology; homology; long generated modules, 772, 773
exact sequence complex (in abelian category), 55
Čech = derived functor, 659, 674, 676, 677 double complex, 63, 63
of a double complex, 64 (hyper)cohomology of, 64
Cohomology and Base Change Theorem, 415, cohomology, 64
524, 526, 731, 731–735, 737–740, 743, see first quadrant, 65
also push-pull map total complex of, 64
CohX (category of coherent sheaves on X), 182 exactness, 55, 55
coker (cokernel), 54 (for sheaves) is stalk-local, 179
cokerpre (cokernel presheaf), 81 of sequence of sheaves is stalk-local, 91
cokernel, 54, see also universal property factorization into short exact sequences, 56
 © 2024 Ravi Vakil. Published by Princeton University Press. 823

homotopic maps give same map on constant rank implies locally free on reduced
homology, 57, 661 (for finite type quasicoherent sheaves),
complex geometry, see analytification; 413, 471, 699, 718
manifolds, complex; varieties, complex constructible subsets, 246, see also Chevalley’s
complex manifolds, see manifolds, complex Theorem
complex varieties, see varieties, complex are finite disjoint unions of locally closed
component subsets, 246
connected, 123, 124, 125, 157, 163, 297, 402, in more general situations, 290
see also idempotent-connectedness contravariant functor, see functor
package coordinates, 102
are closed, 124 on affine space, e.g., xi , 107
may not be open, 124 projective xi , 146, 147, 154, 615
irreducible, 123, 123, 125, 129, 157, 300 coproduct, 41, 46
= minimal primes for Spec A, 128 of schemes (is disjoint union), 211, 275
are closed, 123 of sets, 41
Noetherian spaces have finite #, 124 corank, 364, see also Jacobian corank function;
composition series, 183, see also Jordan-Hölder semicontinuity
package; modules cotangent, see also tangent
composition-of-functors spectral sequence, see bundle, 514
Grothendieck composition-of-functors complex, 601, 728
spectral sequence exact sequence, 601, 608, 725
compositum of subfields, 338 affine version, 600, 601
cone, see also abelian cone; affine cone; ample left-exact in good circumstances, 600, 609,
cone; conormal cone; effective cone; 617, 727
mapping cone; nef cone; normal cone; map, 363
projective cone; tangent cone sheaf, 595, 606, see also tangent sheaf
over smooth quadric surface, 135, 148, 166, relative, 606
168, 169, 221, 303, 339, 362, 369, 372, 383, space, 73, 359
450, 451, 652, 653, see also recurring computed by Jacobian at rational points,
(counter)examples; quadric surface 364
not factorial, 362 computed by Jacobian at separable closed
over twisted cubic / rational normal curve, points, 599, 611
120, 265, 339 is fiber of Ω, 599
traditional, i.e. x2 + y2 = z2 , 152, 263–265, vector, 360, 595, 595
305, 361, 362, 382, 444, 449, 451, 553, 569, = differential, 360
649, 650 relative, 596
conic (curve in P2 ), 259 counterexamples, see recurring
∼ P1 if regular with a rational point, 220,
= (counter)examples
542 covariant functor, see functor
classification of, 261 cover
universal, 511 by open subfunctors, 281
coniveau filtration on K(X), 588 open, 72
connected, 119, see also component; fibers; covering space, 386, 722
idempotent-connectedness package Cremona transformation, 221, 324, 439, 652,
schemes, 157 767
connecting homomorphism, 58 cross-ratio, 557, 557
conormal and the j-invariant, 557
bundle, 601, 635 classifies 4 points on P1 , 557
cone, 635 cubic, 305, 545, 553, 559–567, 571, 572, 652, 701,
exact sequence, 601, 608 703
affine version, 601 curves, 652, see also cuspidal cubic curve;
for smooth varieties, 610 nodal cubic curve; twisted cubic curve
left-exact in good circumstances, 602, 609, hypersurface, 259
610, 727 plane curves, 371, 551
module, 602 surface, 16, 18, 346, 552, 553, 652, 757–761,
sheaf, 602, 603, 635 763–766, see also del Pezzo surfaces;
of regular embedding is locally free, 604 Fermat cubic surface
constant (pre)sheaf, 76 is blown-up plane, 764
824 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

singular, 759 morphism of, 666

cup product, 794, 795, 798, 805 universal, 665, 666, 801, 804
Curve-to-Projective Extension Theorem, δ-invariant, 775
219–221, 437, 437–439, 466–469, 558, 561 d-uple embedding νd , see Veronese
necessity of hypotheses, 438 embedding, 262
curves, 259, 334, see also (−1)-curve; canonical, Dedekind domain, 169, 170, 306, 342, 382, 404
curve; elliptic curves; genus; Fermat unique factorization of ideals, 183, 382
curve; hyperelliptic curves; Klein quartic deformation theory, 363, 626, 686, 697, 757, 760
curve; Mg ; rational normal curve; Tate deformation to the normal cone, 655, 656, 699,
curve; twisted cubic curve see also normal cone
curve singularities, 774 deg F = degC F (degree of coherent sheaf on
exists regular projective birational model, curve), 510
465 when F is line bundle, 508
genus, 535 degenerate linear series, 430, see also linear
arithmetic = topological, 510 series
geometric = topological, 620 degree
genus 0, 522, 541–543, 545, 548–550, 554, 555, d map, 432
620, 626, 734, 775 algebraic = analytic for maps of curves, 469
are conics, 541 in projective space, 520, 685
genus 1, 522, 539, 545, 546, 548, 549, 554, 555, additive for unions, 522
566, 567, 569, 571, 703, see also elliptic constant in flat families, 718
curves of curve, 433
classified by j-invariant, 567 topological definition, 521
not classified by j-invariant, 567 of closed point of finite type k-scheme, 164
genus 2, 548–550, 619 of coherent sheaf on curve, 510
hyperelliptic in precisely one way, 548 additive in exact sequences, 510
genus 3, 510, 522, 548, 550–554, 619, 776 of discriminant of degree d polynomials, 620
not all hyperelliptic, 548 of divisor on regular curve over k, 508
genus 4, 523, 552, 553, 562 of finite flat morphism is locally constant,
genus 5, 552–554 288, 469, 579, 620, 654, 699, 700, 710, 761
genus 6, 522, 554, 776 of finite map from curve to regular curve is
moduli space Mg , 686 constant, 469, 470, 509, 545, 558, 718
non-rational, 221 of finite morphism at point, 414, 685
nonhyperelliptic, 548, 551–554, 562 of generically finite morphism, 710
of every genus, 548 of generically finite rational map, 339
plane, 372, 520, 543, 551, 552, 568, 581, 598, multiplicative under composition, 339
641, 648, 703, 757, 774, 775, see also of homogeneous element, 151, 151
Bézout’s Theorem of hypersurface, 259
proper implies projective, 540 of line bundle on curve, 470, 508, 576, 685
trigonal, 551, 554 additive for ⊗, 511
various categories are equivalent, 468 is homomorphism, 511
cusp, 305, 372, 535, 536, 632, 651, 774, 775, see of line bundle on projective space, 446
also normalization, of cusp of projective morphism of curves, 468, 469,
cuspidal cubic curve, 305, 372, 632, 648 700
normalization, 708 of projective variety, 377
cycle (homological algebra), 56 of torsion sheaf on curve, 510, 511
del Pezzo surfaces, 554, 624, 767, see also cubic,
d : OX → ΩX/Y , 606 surface; Fano variety (classification
d : A → ΩA/B , 596 theory); quadric surface; surfaces,
D+ (f) (projective distinguished open set), 119, Enriques-Kodaira classification
153, 428 delta-functor, see δ-functor
D(f) (distinguished open set), see open set, depth, 187, 747, 748, 748, 752, 754
distinguished as cohomological property, 748
Dn -singularities, 649, 774 bounded by dimension of associated primes,
δπ (diagonal morphism), 40, 312 750
δ-functors, 665, 800, 804, 805 bounded by dimension of support, 748
covariant/contravariant, 666 of regular local ring = dimension, 750
e.g., derived functors, 666 derivation, 604, 604, 606–608, 614
 © 2024 Ravi Vakil. Published by Princeton University Press. 825

in differential geometry, 360 of fiber at a point (upper semicontinuous),

universal d : A → ΩA/B , 604 356
derived category, 500, 738, 802, 805 of fiber of closed map (upper
derived functors, 18, 19, 62, 493, 504, 659 semicontinuous), 356
= Čech cohomology, 659, 674, 676, 677 of fibers, 353
and spectral sequences, 667 never lower than expected, 760
computed with acyclic resolutions, 667 dimensional cohomology vanishing, 494, 504,
left, 524, 663, 664 703, 807
right, 524, 664 for quasiprojective schemes, 654, 655
derived pushforward, see higher pushforward more generally, 504
descending chain condition, 124, 124, 348 relative, 527
descent, 173, 284, 566, 686, see gluing dimensional filtration on K(X), 588
desingularization, 636, see resolution Diophantine equations, 212, 219–222, 542, 549,
det F (determinant bundle), 408 757
determinant (line) bundles, 408, 514, 794 direct image sheaf, see sheaf, pushforward
behave well in exact sequences, 409 direct limit, see colimits
determinant map GLn → Gm , 226 discrete topology, 76
dévissage, 19, 576 discrete valuation, 380, 390, 392, see also
discrete valuation rings; valuations
diagonal ∆ (locally closed subscheme of
X ×Y X), 314 discrete valuation rings, 110, 305, 359, 367, 380,
390, 392, see also finitely generated
diagonal morphism δπ , 40, 41, 42, 311, 312, 312,
modules, over discrete valuation rings
314, 317, 318, 321, 533, 607, 632
(classification)
is a locally closed embedding, 314
differentials over a field, 613, 618
Diagonal-Base-Change diagram, 41, 42, 45, 312,
flatness criterion, 697
314, 320
discriminant ideal, 633, see also relative
diagonalizing quadratic forms, 168, 261, 263,
different and discriminant ideals
303, 361, 585
disjoint union of schemes, 135, 140, 725, see also
diagram indexed by, see category, index
recurring (counter)examples, infinite
Dieudonné (completely functorised), 51
disjoint union of schemes
different ideal, 633, see also relative different
is coproduct, 275
and discriminant ideals
distinguished affine base, 174
differentiable manifolds, see manifolds,
distinguished open set, see open set,
differentiable
distinguished
differentials, 312, 360, 596, see also cotangent
div(s) (Weil divisor of zeros and poles), 441
exact sequence; universal property
divisible abelian group, 665
= cotangent vectors, 360
divisor, see also branch divisor; effective Cartier
behave well with respect to base change, divisors; exceptional divisor; Q-Cartier
605, 609 divisor; Weil divisor
explicit description, 597 Cartier, 18, 453
fiber at a rational point, 599 Doesn’t-vanish set, see open set, distinguished
module of (for ring maps), 596 domain of definition, 197, 323, see also
on hyperelliptic curves, 612 indeterminacy locus
on projective space, see Euler exact sequence dominant morphism, 217
pulling back, 605, 609 dominant rational map, 217
dimension, 106, see also pure dimension double complex, see complex
= transcendence degree, 338 double cover, 111, 239, 545, 546–551, 556, 558,
additive for products of varieties, 341 561, 569, 571, 612, 616, 685, see also
at a point, 344 branched cover; hyperelliptic map
constant in flat families, 718 dual
Krull, 334 of coherent sheaf, 410
of a ring, 334 of locally free sheaf, 403
of a scheme, 16, 165 of O-module, 80
topological in nature, 159 projective space, 375
of a topological space, 333 variety, 377
of a variety at a point (upper dual numbers, 104, 112, 113, 136, 140, 160, 213,
semicontinuous), 344, 345 253, 266, 322, 334, 363, 413, 414, 455, 687,
of a variety preserved by field extension, 345 697, 722
826 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

flatness criterion, 690, 697 embedded points/primes, 188, 188, 195, 195,
dualizing sheaf, 495, 514, 553, 793, 795 384, 441, 469, 471, 520, 581, 611, 750–752,
= canonical sheaf, 805 754, 808
DVR, see discrete valuation rings of locally Noetherian scheme, 195
Dynkin diagrams, 649, 767 embedding (unwise terminology), 257
enough injectives, 664, 665, 667, 669
e.g., Ab, 665
E6 , 758, 764, 774
e.g., AbX , 673
E8 , 589, 774
e.g., ModA , 19, 664, 665
En , 774
e.g., ModOX , 673, 797, 798
En -singularities, 649, 774
e.g., QCohX , 682
Ep,q (entry in double complex), 63
enough projectives, 663, 664, 667, 676
Ep,q
r (entry (p, q) in rth page of spectral
not ModOX , 676, 799
sequence), 65
epimorphisms, 42, 42
EX Y (exceptional divisor), 637, 638
may not be surjective, 42
effaceable functor, see functor, effaceable
equalizer exact sequence, 75, 132, 280
effective Cartier divisors, 18, 187, 270, 270, 271,
equidimensional, see pure dimension
274, 289, 324, 366, 438, 448–450, 452, 453,
equivalence of categories, 34, 34, 88, 175, 208,
504, 518, 576–578, 580, 581, 636–646, 650,
219, 423, 468
651, 653–655, 692, 712, 715, 747, 751, 764,
espace étalé, see sheaf, space of sections
781, 787, 808, 809, see also regular
essentially surjective functor, 34
sequence; slicing criterion
étale cohomology, 178, 659
adding, 453
étale cover, 387, 619, 725
and Cohen-Macaulayness, 752
étale morphisms, 359, 368, 385, 386, 387, 632,
and invertible sheaves, 425, 452
721, 722, 723, 725, 727, 728, 761, 763, 771,
contain no associated point, 270
see also formally étale
e.g., exceptional divisor EX Y on blow-up,
= locally finitely presented + flat +
637
unramified, 725
how to restrict, 448 = smooth + unramified, 725
is pure codimension 1, 347 differential geometric motivation, 385
normal (line) bundle to, 453, 603 local on source and target, 387
not easily affine-local condition, 270 not same as “local isomorphism”, 387
principal, 643 open condition, 386
relative, 715, 715 preserved by base change, 387
slicing by, 269, 269, 270, 521, 523, 725, 747, preserved by composition, 387
805 reasonable class, 387
effective cone, 584, 585 étale topology, 178, 619, 722, 771
Eisenstein’s Criterion, 371 Euclidean algorithm / Euclidean domain, 105,
elementary transformation (ruled surface), 788, 106, 167
789 Euler characteristic χ(X, F ), 427, 507, 521, 522,
elimination of quantifiers, 249, 250 581, 685, 686, 704
elimination theory, 245, 250, see also additive in exact sequences, 508, 519, 588
Fundamental Theorem of Elimination constant in projective flat families, 687, 704,
Theory 717, 737
ellipse, 544, 568 constant in proper flat families, 524, 737
elliptic curves, 167, 213, 221, 328, 405, 547, 549, without Noetherian conditions, 740
554, 555, 555–558, 561, 563, 566–573, 583, is finite for proper X, 588
730 polynomiality, 517, 554
are group varieties, 563 Euler exact sequence, 614, 614, 616, 625, 626
counterexamples and pathologies, 569 generalizations, 616
degenerate, 567 exact functor, 58, see also π−1 , π! , FHHF
degree 1 points = degree 0 line bundles, 555 Theorem, flat morphisms, localization,
group law, 556, 562, 563, 571, 764 stalkification
geometrically, 562 exact modules / exact morphisms, 685
level n structure, 557 exact sequence, see also Euler exact sequence;
non-torsion point, 570, 571 Grassmannian, universal exact sequence;
not same as genus 1 curves, 555 excision exact sequence; equalizer exact
elliptic fibration, 652 sequence; exponential exact sequence;
 © 2024 Ravi Vakil. Published by Princeton University Press. 827

long exact sequence; normal exact Faltings’s Theorem (Mordell’s Conjecture), 549
sequence Fano variety (classification theory), 353, 623,
factored into short exact sequences, 56 624, 764, see also del Pezzo surfaces;
grafting, 68 trichotomy
split, 673 Fano variety (of lines), 764
exceptional divisor EX Y, 289, 582, 636, 638, f.d.Veck , 34
643–645, 648, 649, 651, 653, 654, 656, Fermat cubic surface, 759, 761, 763, 764
765–767, 774, 781, 785, 786, 788 Fermat curve, 221, 372
in minimal model program, 638 Fermat’s Last Theorem, 221, 549
normal bundle to, 644 FHHF Theorem, 59, 61, 503, 523, 524, 526, 693,
excision exact sequence for class groups, 445 695, 799
exponential exact sequence, 84, 86, 88, 91, 510 fibered coproduct, 41, see also universal
Ext functors, 659 property
first version, 664 fibered diagram/square, 40, see also base
for O-modules, 798 change diagram/square; Cartesian
computable by acyclics, 799 diagram/square; pullback,
long exact sequence, 799 diagram/square
second version, 664 fibered product, 39, see also base change;
two definitions are same, 667 product; pullback; universal property
Ext functors, 799 “commutes with localization”, 283
Ext iX (coherent, coherent) = coherent, 800 in category of open sets, 40
Ext iX (coherent, quasicoherent) = of functors, 279
quasicoherent, 800 of schemes, 275
behaves like derived functor in first of sets, 40
argument, 800 with open embedding, 231
computable with locally free resolutions, 800 fibers of a morphism of schemes, 229, 285, 286,
long exact sequence in first argument, 800 see also generic fiber; scheme-theoretic
may not preserve quasicoherence, 799 connected, 297, 779, 780
extension by zero π! , 96, 480, 675, 675–677, 682 from Zariski’s Connectedness Lemma, 778
is exact, 675
dimension of, 353
left adjoint to inverse image π−1 , 96, 675
well-behaved for flat morphisms, 691, 706,
extension of an ideal, 282
708
extension of fields, see field extension
geometric, 293, 294, 686, 724, 728, 734, 735
extension of scalars, 52
geometrically connected (etc.), 293
exterior algebra, 408
properties preserved by base change, 294
Fn (Hirzebruch surface), 484, 584 fibers of an OX -module, 142
F |p (stalk of F at p), 142 field
factorial, 167 imperfect, 328
̸= normal, 169, 363 perfect, 300, 301, 393, 394, 627, 755
but no affine open is UFD (example), 569 separably closed, 294, 297–299, 319
implied by regular (Auslander-Buchsbaum field extension, see also transcendence theory
Theorem), 394 algebraic, 232, 233, 306, 335, 338, 345, 599
implies normal, 167 and differentials, 599, 605
open condition, 167 finite, 107, 122, 164, 169, 239, 240, 247, 283,
ring, 167 288, 294, 301, 306, 307, 335, 339, 344, 370,
stalk-local, 167 465, 546, 570, 599, 632, 633, 722, 725
weaker than regular (example), 394 finitely generated, 219, 219, 300, 338, 354,
faithful functor, 32, 33, see also fully faithful 522, see also transcendence degree
functor Galois, 283, 344, 620
faithful module, 236 normal, 344
faithfully flat algebra, 707 purely inseparable, 293, 296, 296, 297, 297,
faithfully flat morphisms, 173, 706, 707 308, 628
local on the target, 707 purely transcendental, 219, 345
preserved by base change, 707 separable, 301, 307, 599, 616, 617, 632, 633,
preserved by composition, 707 722, 725
reasonable class, 707 separably generated, 300, 300, 605
faithfully flat sheaves, 707 filtered, 47, see also colimits, filtered
828 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

final object, see object, in a category, support is closed, 411

initial/final/zero finite type morphisms, 242
finite B-algebra, 233, 238 affine-local on the target, 242
finite extension, see also field extension e.g., proper morphisms, 325
of integrally closed domains, 342, 343, 707, preserved by base change, 291
see also Going-Down Theorem preserved by composition, 242
of rings, 240, 343 reasonable class, 292
finite fibers not preserved by base change, 292 finitely generated domain, 338
finite flat morphisms, 698, 741, 758, 802, 804 finitely generated field extension, see field
have locally constant degree, 288, 469, 579, extension
620, 654, 699, 700, 710, 761 finitely generated modules, 180, see also finite
finite global generation, see global generation type, quasicoherent sheaves
finite length, 183, see also Artinian; over discrete valuation rings (classification),
Jordan-Hölder package; length 180
scheme, 186 over discrete valuation rings (classification),
finite morphisms, 238 180, 405, 412, 428, 471, 556, 614
= integral + finite type, 242 over principal ideal domains (classification),
= projective + affine, 497 20, 180, 183, 405, 412, 428, 513, 556, 614,
= projective + finite fibers, 468, 486, 497, 498, 688
511, 537, 761, 788 over regular local rings have finite free
= projective + quasifinite, 497 resolutions, 705
= proper + affine, 327, 497, 531, 784 finitely globally generated, see global
= proper + affine (Noetherian setting), 784 generation
= proper + finite fibers, 497 finitely presented, see also local finite
= proper + finite fibers (Noetherian setting), presentation
784 = finite type under Noetherian hypotheses,
= proper + quasifinite, 327 243, 245, 290
= proper + quasifinite (Noetherian setting), algebras (ring map), 243, 243, 244, 244, 597
784 affine-local property, 243
= proper + quasifinite + locally finitely implies always finitely presented, 243
presented, 784 sheaves, 182, 740, see also modules, finitely
affine-local on target, 238 presented
are affine, 238 affine-local, 182
are closed, 241, 261 flat = locally free, 698
are integral, 233, 241 local freeness is stalk-local property, 411
are projective, 240, 486 finitely presented modules, 59, 180, 198, 199,
are proper, 325 597, 698, 699, see also finitely presented,
are quasifinite, 240, 243 sheaves
are separated, 316 = coherent for Noetherian rings, 199
e.g., closed embeddings, 253 “finitely presented implies always finitely
have finite fibers, 240, 286, 288 presented”, 181, 411, 698
not same as finite fibers, 241 graded, 424
preserved by base change, 291 localization and Hom commute, 37
preserved by composition, 240 finitely presented morphisms, 236, 242, 243,
reasonable class, 292 244, 461, 686, 700, 710, 726, 728, 733,
separable, 616 740–744
to Spec k, 239 = “finite in all ways”, 245
finite presentation, see finitely presented affine-local on the target, 245
(algebras, sheaves, modules, morphisms) Chevalley’s Theorem, 247, 290
finite type locally pullbacks of nice morphisms, 290
A-scheme, 163, 163 preserved by base change, 291
implied by quasiprojective, 163 preserved by composition, 244
over a ring, 163 reasonable class, 314
quasicoherent sheaves, 182, see also modules, finiteness of integral closure, see integral
finitely generated closure
affine-local, 182 finiteness of normalization, see normalization
constant rank implies locally free on Five Lemma, 55, 69, 70
reduced, 413, 471, 699, 718 subtler, 70, 796
 © 2024 Ravi Vakil. Published by Princeton University Press. 829

flabby sheaf, see sheaf, flasque/flabby over integral domains, 687, 688
flag variety, 228, 616 over local rings, 687
bundles, 616 over principal ideal domains, 687, 688, 697
partial, 228, 475 over regular curves, 699
flasque sheaf, see sheaf, flasque/flabby over the dual numbers, 687, 690, 697
flatness, 245, 327, 469, 503, 513, 524, 526, 575, preserved by base change, 688, 690, 712
589, 656, 659, 660, 662, 663, 689, 720, see preserved by composition, 688, 690
also Euler characteristic, constant in reasonable class, 690
projective flat families; faithful flatness; regular embeddings pull back under flat
finite flat morphisms; generic flatness; morphisms, 692
Going-Down Theorem; Miracle Flatness sends associated points to associated points,
Theorem; resolution (complexes); slicing 690
criterion stalk/prime-local property, 688, 689
= free = projective for finitely presented topological aspects, 687, 706
modules over local rings, 687, 698 Tor1 -criterion, 662
= locally free for finitely presented sheaves, valuative criterion, 699
698 variation of cohomology, 701
= projective for finitely presented modules, flex, 560, 560, 561
699 forgetful functor, see functor, forgetful
= torsion-free for PID, 697 form of degree d, 151
and blowing up, 692 formal neighborhood, 771, 777, 777, 779, 787,
and regular sequences, 271 790
at a point, 689 formal power series, 45, see power series
constancy of fiber degree / dimension / formal schemes, 16, 19, 777
arithmetic genus, 718 formally étale, 727, 728, 728, 771
criterion over regular curve, 699 formally smooth, 727, 728, 728
e.g., affine space, 687 formally unramified, 631, 727, 728, 728
e.g., completion, 772, 774 e.g., localization, 631
e.g., free modules, 687 Four Squares Theorem, 221
e.g., localization, 687 fraction field, 36, see also total fraction ring
e.g., open embeddings, 689 fractional ideal, 399, 404, 450
e.g., projective modules, 687 fractional linear transformations, 447, 447, 551
equational criterion, 696 free resolution, see resolution (complexes)
fiber dimension well-behaved, 691, 708, 718 free sheaf, 179, 399, see also trivial bundle
fibral, 715, 716 Freyd-Mitchell Embedding Theorem, 55
flat limit, 700, 700, 702–704 Frobenius, 328, 339, 368, 535, 536, 617, 619
explicit example, 701 absolute, 242
ideal-theoretic criteria, 687, 693, 696, 696, full functor (unimportant), 32
697, 711, 773 full rank quadratic form, see maximal rank
finitely generated version, 697 full subcategory, see subcategory, full
stronger form, 773 fully faithful functor, 33, 34, 43, 51, 55, 322
implies torsion-free, 688, 693, 697, 712, 753 function field, 160, see also rational function
local criterion, 687, 711, 711, 731, 744 and dominant rational maps, 217
no higher Tor, 662, 663, 686, 693, 694, 696, determines irreducible regular projective
697 curve, 468
of modules, 60, 687 functional field
of quasicoherent sheaf over a base, 689 and birational maps, 217
of relative dimension n, 708, 709, 723, 724, functions, 78, see also differentiable functions;
753 rational function
includes locally finite type hypotheses, = maps to A1 , 223
708 determined by values on reduced scheme,
preserved by base change, 709 159
reasonable class, 709 global, 78
of ring morphism, 687 multiplicity at regular point, 650, 654
open condition, 711 not determined by their values, 112
open map in good situations, 687, 708 on a ringed space, 78
over a field, 687 on a scheme, 102, 139
over discrete valuation rings, 687, 697 on an open subset, 78
830 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

on projective space, 146 genus, 517, see also curves, genus 0, 1, . . .

value at a point, 102, 140, 141 arithmetic, 510, 553, 623, 685, 718, 741, 776
vanishes at a point, 141 constant in projective flat families, 718
functor, see also additive; derived; essentially arithmetic = topological (for curves), 510
surjective; exact; faithful; full; fully geometric, 611, 623
faithful; half-exact; left-exact; natural is a birational invariant, 622, 623
transformation; representable; right-exact geometric = topological (for curves), 620
composition of, 32 topological, 510
contravariant, 33 geometric fiber, see fibers, geometric
covariant, 32 geometric genus, see genus, geometric
effaceable, 666 Geometric Nakayama’s Lemma, see
forgetful, 32, 33, 50–52, 84, 91, 226, 454, 456 Nakayama’s Lemma
left adjoints to, 51, see also groupification; geometric Noether normalization, see Noether
saturation; sheafification normalization
functor category, 43, 279 geometric point, 293
identity, 32 geometrically connected / integral /
natural isomorphism, 34 irreducible / reduced, 294
of points, 33, 211, 212 germs, 72, 74, 75, see also compatible germs
fundamental groupoid, 30 determine section of sheaf, 82
fundamental point (of rational map), 323 GLn (group scheme), 226
Fundamental Theorem of Elimination Theory, global functions, 78
245, 250, 250, 251, 261, 327, 353, 438, 497, global generation (of quasicoherent sheaf), 457,
498 458, 527, see also base-point-freeness
at a point, 458
G(k, n) (Grassmannian), 227 e.g., any quasicoherent sheaf on any affine
G(k − 1, n − 1) (Grassmanian), 228 scheme, 458
GAGA Theorem, 625, 685 finite, 458
Galois cohomology, 178 e.g., every finite type quasicoherent sheaf
Galois theory, 105, 108, 114, 283, 308, 343, 344, on every affine scheme, 458
371, 546, 558, 620, 686, 764 open condition, 459
Gauss’s Lemma, 167, 168 preserved by ⊕, 458
Gaussian integers Z[i], 167, 288, 305, 373, 374, preserved by ⊗, 433, 459
382 with respect to π (relatively globally
general fiber, 289, see also generic fiber generated), 465, 487
general linear group scheme, 226 affine-local on target, 488
general point, 123, 339, 378, 652, 764 for locally ringed spaces, 488
general type (variety), 353, 624, 624, see also global section, 74
trichotomy gluability axiom, 75, see also base gluability
generated by (a finite number of) global axiom
sections, see global generation gluing morphisms
generated in degree 1, 151 of locally ringed spaces, 207
generic fiber, 286, 289, 289, 355, 700 of ringed spaces, 204
generic flatness, 123, 710, 710 of schemes, 209, 277, 279
generic point, 103, 105, 110, 113–115, 120, 122, of topological spaces, 77
157 gluing ringed spaces along open sets, 143
and function field, 160 gluing schemes
generic smoothness, 123, 551, 627, 630 along closed subschemes, 434
of varieties, 627 along open subschemes, 143, 144, 254, 277,
on the source, 627, 628 434
on the target, 629 Gm (multiplicative group scheme), 224
generically finite, 289, 289, 650 Going-Down Theorem, 235
degree, 339, 710 for finite extensions of integrally closed
implies generally finite, 357, 710 domains, 342, 343, 343, 707
usually means generally finite, 290 for flat morphisms, 343, 706, 707, 707, 708
generically separable morphisms, 616, 617, see for integral extensions of integrally closed
also separable finite morphisms domains, 343
generization, 122, 157 Going-Up Theorem, 233, 234, 335, 343, 344,
stable under, 247, 708 707, 708
 © 2024 Ravi Vakil. Published by Princeton University Press. 831

Gorenstein, 809 left adjoint to forgetful functor, 51

graded ideal, see homogeneous ideal groupoid, 30
graded modules, 51, 422, 423, 447, 454, 455, see Groups (the category of groups), 225
also saturated graded module; saturation
functor H ((co)homology), 55
quasicoherent sheaf corresponding to, 422, H0 (U, F ) (section of sheaf F over U), 74
454 hF (m) (Hilbert function), 517
graded rings, 51, 148, 151 hA (functor of maps from A), 32
= Z≥0 -graded ring, 151 hA , see functor, of points
finitely generated over A, 151 Hairy Ball Theorem, 612
generated in degree 1, 151
half-exact functor, 711
maps of, 214
Hartogs’s Lemma, 143, 383, see also Algebraic
over A, 151
Hartogs’s Lemma
standing assumptions on, 151
Hausdorff, 311, see also separated morphisms
Z-, 151
in terms of separatedness, 315
graph (of a) morphism, 316, 533, 534, 583
height, see codimension
is a locally closed embedding, 316, 534
Hensel’s Lemma, 771
to a separated scheme is a closed
Hermitian metric, 461
embedding, 316, 534
higher direct image sheaf, see sheaf, higher
graph of a rational map, 316, 323, 324, 652
pushforward
and blow-ups, 652
Hilbert Basis Theorem, 125, 126
Grassmannian, 156, 227, 228, 228, 448, 472, 473,
Hilbert function, 126, 517, 522, 554, 718
475, 616, 741, 742, 744, 745, 760, 761, see
of coherent sheaf, 517
also Plücker
of projective scheme, 517
as moduli space, 471
Hilbert functor, 741, 742
bundle, 475, 616
functor, 471, 472 Hilbert polynomial, 495, 517, 518, 518–520, 522,
553, 576, 579, 685, 742
G(2, 4) = G(1, 3), 346, 474, 475, 760
G(2, 5), 554, 563 locally constant in flat families, 718
tautological bundles, 473, 616, 741 Hilbert scheme, 686, 741, 742, 745, 764
universal exact sequence, 473 universal family, 741
Grauert’s Theorem, 731, 731, 734, 735, 737–740 Hilbert Syzygy Theorem, 126, 457, 458, 518,
Grothendieck composition-of-functors spectral 704, 706
sequence, 70, 669, 669, 674, 801, 802 proof, 705
proof, 672 Hilbert’s (Weak) Nullstellensatz, see
Grothendieck functor, 224 Nullstellensatz
Grothendieck group of coherent sheaves on X Hilbert’s fourteenth problem, 573
(K(X)), 588 Hironaka’s example, 327, 720
Grothendieck topology, 177, 178, 686, 722 Hironaka’s Theorem, 650, 653, 789, see also
Grothendieck’s “six operations”, 802 resolution (singularities)
Grothendieck’s Coherence Theorem, 327, 415, Hirzebruch surfaces Fn , 484, 484, 584, 648, 776,
457, 488, 495, 496, 507, 530, 531, 531, 532, 788
588, 737, 776, 780, 790, 791 Hirzebruch surfaces Fn , see also elementary
for projective morphisms, 497, 525 transformation; ruled surface
Grothendieck’s Theorem (vector bundles on Hirzebruch-Riemann-Roch, 579
P1 ), 515 Hodge diamond, 625, 625, 626, 789
Grothendieck-Riemann-Roch, 579 Hodge Index Theorem, 513, 585, 585, 586
group (perverse definition), 30 Hodge numbers, 409, 625, 625, 731
group law on elliptic curve, see elliptic curves, birational invariance of some, 625
group law not all are birational invariants, 626
group object (in a category), 224 Hodge theory, 622, 625, 730
group variety/scheme, 223, 224, 225, 225–227, Hom
328, 328, 330, 448, 475, 563, 567, 630, 734, is left-exact, 58
see also action of group variety/scheme may not commute with localization, 59
abelian, 225 right-adjoint to tensor product, 49
morphism of, 226 Hom , 80, 410
structure on elliptic curves, 566 is left-exact, 92
groupification (of abelian semigroup), 50, 52 may not commute with stalkification, 80
832 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

of quasicoherent sheaves not necessarily identity axiom, 75, see also base identity axiom
quasicoherent, 410 identity functor, 32
homogeneous element, 151 identity morphism idA , 30
homogeneous ideal, 151 image (of morphism in a category), 54
homogeneous space, 630 image (of morphism of schemes)
homology, 55, see also cohomology set-theoretic, see Chevalley’s Theorem;
commutes with exact functors, 59, see also closed map; Fundamental Theorem of
FHHF Theorem Elimination Theory; open maps;
commutes with filtered colimits in ModA , 61 surjective; universally closed; universally
homomorphism (in additive category), 53 open morphisms
homotopic maps of complexes give same map image (of morphism of schemes)
on homology, 57, 57, 661 scheme-theoretic, see scheme-theoretic,
Hopf algebra, 227 image
horseshoe construction, 662, 672 image presheaf, 81
Hurwitz’s Automorphisms Theorem, 621 immersion
hypercohomology, see cohomology, of a double of manifolds, 101, 721
complex of schemes (unwise terminology), 257
hyperelliptic imperfect field, see field, imperfect
curves, 535, 545, 545, 547–552, 561, 612, 613, incidence correspondence/variety, 346, 375,
see also curves, nonhyperelliptic; 631, 759, 760
Riemann-Hurwitz formula, hyperelliptic inclusion-exclusion via sheaf cohomology, 680
differentials on, 612 indeterminacy locus, 197, 323
moduli spaces, 549 index category, 44
most not if g > 2, 548 index of symmetric bilinear form, 585
involution, 563 induced reduced subscheme, see reduced
map, 545, 548, 549 subscheme structure
unique for g ≥ 2, 548 inductive limit, see colimits
hyperplane, 150, 259, 260 infinitesimal neighborhood, see formal
universal, 375, 631, 741 neighborhood
hyperplane class, 521 initial object, see object, in a category,
hyperplane class, 444 initial/final/zero
hypersurface, 107, 148, 150, 259, 336, see also injection of vector bundles, 406, 407, 725, see
quadric also maps of vector bundles
associated points of, 190 injective limit, see colimits
degree, 259 injective maps of rings induce dominant maps
discriminant, 759 of schemes, 119
in Ank have no embedded points, 195
injective morphisms of schemes not preserved
by base change, 292, see also universally
id (identity morphism), 30 injective morphisms
I(S) (ideal of functions vanishing on S), 127 injective objects, 664, see also enough injectives;
I-adic completion, 769 projective objects; resolution (complexes)
I-adic filtration, 395 closed under product, 674
I-adically separated, 770 have no Čech cohomology, 679, 680
I-depth, see depth preserved by right adjoints, 677
I-filtration of a module, 395 integral closure, 236, see also normalization
I-stable filtration, 395 finiteness of, 307, 308
ideal, see also different ideal; discriminant of Noetherian need not be Noetherian, 307
ideal; fractional ideal; homogeneous ideal; integral element of B-algebra, 232
maximal ideals; minimal prime ideals; integral extensions of rings, 232, see also
prime ideals; principal ideal Going-Down Theorem
extension of, 282 preserved by localization of source, 232
of denominators, 166, 166, 167, 384 integral morphisms of rings, 232
product of two, 116 does not imply finite, 233
ideal sheaves, 176, 254 e.g., finite, 233
product of, 452, 603, 642, 692 preserved by composition, 233
idempotent, 119, 211, 297, 298, 779 preserved by localization of source, quotient
idempotent-connectedness package, 119, 297, of source and target, 232
298 integral morphisms of schemes, 241
 © 2024 Ravi Vakil. Published by Princeton University Press. 833

affine-local on target, 241 bounds dimension of variety, 365

are affine, 241 for projective k-schemes, 365
are closed, 241 independent of presentation, 364
e.g., finite morphisms, 241 preserved by field extension, 365
fibers of, 241, 334 Jacobson radical, 235
may not be finite, 241 JC (Jacobian corank function), 364
preserved by base change, 241, 291 Jordan-Hölder package, 183
preserved by composition, 241 Jordan-Hölder Theorem, 183, 183
reasonable class, 292
integral scheme, 160 K• (the Koszul complex), 704
= irreducible and reduced, 160 k[ϵ]/(ϵ2 ), see dual numbers
almost a stalk-local property, 161, 163 KX (canonical divisor), 581
integrally closed, 165, see also integral closure KX (canonical sheaf/bundle), 514, 622
intersection, see scheme-theoretic, intersection k[[x1 , . . . , xn ]], 769
intersection number/product of L1 , . . . , Ln K(A) (fraction field), 36, 160
with F , 575, 575–580, 586–588 κ(p) (residue field), 141
deformation invariance, 575 κ(X) (Kodaira dimension), 623
is symmetric multilinear, 576 K(X) (function field of X), 160
locally constant in flat families, 719 K(X) (Grothendieck group of coherent sheaves
on a surface, 580 on X), 587
intersection theory, 433, 575, 655, 656, 685, 757 K(X)≤d (dimensional filtration), 588
on a surface, 580 k-point / k-rational point, see k-valued point
inverse image (scheme), see scheme-theoretic k-scheme, see A-scheme
preimage k-smooth, see smoothness, over a field
inverse image sheaf, 92, 92, 93, 416 k-valued point, 211
construction, 93 k-variety, see variety
exact (for sheaves of abelian groups), 94 K3 surfaces, 623, 624, 624
left adjoint to pushforward, 92, 93 Kähler differentials, see differentials
preserves stalks, 94 ker (kernel), 54
right adjoint to extension by zero, 96, 675 kerpre (presheaf kernel), 81
inverse limit, see limit is a sheaf, 82
invertible sheaves, see line bundles kernels, 53, see also universal property
irreducible, 120, 120, see also component are limits, 60
criterion to be, 356 commute with limits, 60
schemes, 157 presheaf, 81
Weil divisor, 440 sheaf, 82
irreducible object in abelian category, 183, see Kleiman’s criterion for ampleness, 514, 585,
also Jordan-Hölder package 589, 591, 592, see also Nakai-Moishezon
irregularity (surface), 626 criterion
irrelevant ideal, 151, 152, 153, 260, 264 Kleiman’s Theorem, 591, 591, 592
isomorphism Kleiman-Bertini Theorem, 375, 629, 630, 630,
(category), 30 631, see also Bertini’s Theorem
of ringed spaces, 139 poor man’s, 631
of schemes, 139, 208 Klein quartic curve, 552
Kodaira dimension κ(X), 623, 624
j-invariant, 547, 554, 558, 558, 561, 567 is a birational invariant, 624
and the cross-ratio, 557 Kodaira Embedding Theorem, 461
Jacobian Kodaira Vanishing Theorem, 495, 625, 625
computes cotangent space at rational points, Koszul complex, 272, 614, 704, 704, 705, 715,
364 747
computes cotangent space at separable Koszul homology, 705
closed points, 599, 611 Koszul resolution, 705, 705
criterion for regularity, 366, 366–368, 370, Krull dimension, see dimension
373, 546, 598, 610 Krull Intersection Theorem, 396, 770
for projective hypersurfaces, 371 Krull’s Height Theorem, 273, 338, 350, 350, 351,
description of Ω, 599, 608 374, 388, 523
matrix, 364 Krull’s Principal Ideal Theorem, 273, 339, 347,
Jacobian corank function JC, 364 347–352, 360, 373, 378, 519, 752
834 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

for tangent spaces, 361, 366 ∼ invertible sheaves, 401

=
proof, 347 additive notation, 577
Künneth formula for quasicoherent sheaves, and maps to projective schemes, 429
507, 694, 696 nef, 512, 513, 592, see also nef cone
= limit of amples, 591
Li F (left derived functors), 663, 664 numerical property, 513
ℓ(·) (length of module or coherent sheaf), 183, numerically effective, see line bundles, nef
186 numerically trivial, 512, 512, 513, 513, 585,
Lefschetz principle, 625, 731 586, 719
left derived functors, see derived functors, left on Pn k , 428
left-adjoints, see adjoint, left cohomology, 504
left-exact functors, 58 on projective schemes, 428
and derived functors, 493 semiample, 464
and right derived functors, 664 twisting by, 428
and the FHHF Theorem, 59 twisting by divisors, 446
and universal δ-functors, 666 linear independence of characters, 283, 308
e.g., global section functor, 91 linear series, 430, 431–433, 464, 513, 519, 539,
e.g., Hom, 58 543, 545, 561, 566, 572, 580, 583, 585, 631,
e.g., Hom , 92 651, 759
e.g., limits, 60 (non)degenerate, 430
e.g., pushforward, 91
anticanonical, 767
e.g., right adjoints, 60, 61
base points, base-point-free, base locus, ...,
e.g., sections over U, 91
430
not pullback of quasicoherent sheaves, 420
canonical, 549
Leibniz rule, 360
complete, 430, 432, 433, 537, 548, 562, 566,
length, 183, see also Jordan-Hölder package
786
additive in exact sequences, 185
is always nondegenerate, 430
at a point, 186
linear space, 260
of a coherent sheaf, 186
linear system, 430, see linear series
of a scheme, 186
local, see also affine-local; stalk-local
of dimension 0 scheme, 581
on the source, 231
of regular sequence, 271
on the target, 230
Leray spectral sequence, 494, 526, 531, 532, 659,
local finite presentation, see also finite
669, 674, 676
presentation
Lie groups, 328, 330, 757, 758
lim (colimit), 46 morphisms, 236, 242, 243, 244, 386, 631, 632,
−→ 708, 711, 722–725, 728, 784
lim (limit), 44
←− ”limit-preserving”, 245
limits, 44, see also universal property
affine-local on target, 244
are left-exact, 60, 792
Chevalley’s Theorem, 290
commute with kernels, 60
commute with limits, 51, 60 preserved by base change, 291
commute with right adjoints, 51, 60, 61 preserved by composition, 244
e.g., kernels, 60 reasonable class, 292
lim is exact if transition maps of source are over B, 244
surjective, 62 local freeness of finitely presented sheaf is
line, see also affine line; projective line; ruling stalk-local, 411
of quadric; tangent local isomorphism, 721
in Pn local rings, 19, see also regular local rings
k , 259, 260
on cone, 361 morphisms of, 207
with doubled origin, 135, 144, 145, 164, 311, are faithfully flat, 707
317, 320, 322, 323, 390, see also recurring localization, 35, see also universal property
(counter)examples as colimit, 48
line bundles, 18, 155, 212, 383, 399, 400, 401, commutes with direct sums, 37
401, see also algebraic equivalence; give monomorphisms of schemes, 284
ampleness; base-point-freeness; degree of is exact, 58, 60, 61, 172, 413, 617, 687
line bundle on curve; global generation; is flat, 687
numerically trivial; Picard group; Q-line may not commute with Hom, 37, 59
bundle; total space; very ampleness may not commute with infinite products, 37
 © 2024 Ravi Vakil. Published by Princeton University Press. 835

of regular local rings are regular local rings, lower semicontinuity, see semicontinuity
393 Lüroth’s Theorem, 558, 620, 620
”preserved by base change”, 283 Lutz-Nagell Theorem, 571
locally (of) finite presentation, see local finite Lying Over Theorem, 233, 234, 235, 241, 304,
presentation 340, 343, 356, 391
locally closed embeddings, 231, 253, 256, 257 geometric translation, 234
are locally of finite type, 257
are monomorphisms, 257, 284 mp (maximal ideal of stalk of functions), 72
are separated, 316 Mg (moduli space of curves), 554
finite intersections of, 257, 282, 282
manifolds, 15, 20, 29, 71, 76, 78, 99–102, 141,
preserved by base change, 282
142, 203, 206, 213, 223, 227–229, 311, 333,
reasonable class, 291 386, 399, 402, 405, 572, 596, 603, 606, 721,
locally closed immersions, see locally closed see also immersion; submersion
embeddings
analytic, 73, 99
locally closed subschemes, 257, see also locally
complex, 20, 101, 111, 142, 143, 182, 198,
closed embeddings
213–215, 221, 325, 383, 445, 461, 555, 559,
locally closed subset, 246
570, 610, 618, 636, 776, see also varieties,
locally constructible subsets, see constructible
complex
subsets
differentiable, 73, 99–101, 142, 203, 204, 360,
locally finite type
363, 386
A-scheme, 163
morphisms of, 100
over a ring, 163
smooth, 142
locally finite type morphisms, 241
vector bundle on, 399
affine-local on the target, 242
map, see also morphism; rational map
preserved by base change, 291
(category), 29, see morphism (category)
preserved by composition, 242
of graded rings, 214
reasonable class, 292
of vector bundles, 407, 407, 408, 414
locally finitely presented, see local finite
mapping cone, 65, 70, 70, 502, 737, 738
presentation
Max Noether’s AF + BG Theorem, 752
locally free resolution, see resolution
maximal ideals of A
locally free sheaf, 399–401, 402, see also vector
bundle = closed points of Spec A, 121
= finitely presented sheaf with free stalks, exist if A ̸= 0, 19
411 maximal rank (quadratic form), 168, 263, 303
in short exact sequences, 405 = smooth, 372
is quasicoherent, 405 maximum principle, 327
rational/regular section, 405 minimal prime ideals, 128
∼ vector bundle, 479 = irreducible components of Spec A, 128
locally of finite type, see locally finite type are associated primes, 193
locally principal closed subscheme, see closed Noetherian ring has finitely many, 128
subscheme, locally principal Miracle Flatness Theorem, 747, 752, 753, 753,
locally principal divisor, see Weil divisor, 802
locally principal algebraic version, 753
locally ringed spaces, 102, 140, see also gluing Möbius strip, 399
morphisms ModA (category of A-modules), 31
functions have values; values pull back; has enough injectives, 19
zeros of functions pull back, 207 is an abelian category, 52
morphisms of, 207, 488 ModOX (category of OX -modules), 79, 178
LocPrin (group of locally principal divisors), has enough injectives, 673
444 is an abelian category, 92
locus where two morphisms agree, 321, 322, not enough projectives, 676
329, 330 modular curve, 557
long exact sequence, 57 modular form, 557
for Ext, 664, 748–750, 799 modules, see also coherent; depth; finitely
for Ext-functors of O-modules, 799 generated; finitely presented; flatness;
for Tor, 660, 661, 694 graded; length; Noetherian; Kähler
for higher pushforward sheaves, 525, 743, differentials; projective
744 ModA has enough injectives, 664, 665
836 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Cohen-Macaulay, 751, see also multiplicity of zero or pole, see order of

Cohen-Macaulayness pole/zero
free implies projective implies flat, 687 multp D, 654
topological, 770 Mumford complex, 735
moduli space, 228, 390, 392, 431, 549, 551, 553,
555, 557, 563, 686, 735, 740, 741, 758, 761, N (nilradical), 113
764, see also parameter space N1 (X) = Pic X/ Picτ X, 513, 513
of curves Mg , 554, 686, 732 N1Q (x) = N1 (X) ⊗Z Q, 513, 513, 514, 583, 584,
universal families, 741 586, 590, 653
monoid, 30 ND/X (normal bundle to effective Cartier
monomorphisms, 41 divisor), 453
are separated, 316 NS(X) (Néron-Severi group), 513
are universally injective, 296 NX Y (normal cone), 644
∨ (conormal sheaf), 603
NX/Y
diagonal characterization, 42
e.g., closed embeddings, 256 NX/Y (normal sheaf), 603
e.g., locally closed embeddings, 257 n-fold, 334
e.g., open embeddings, 205 n-plane, 260
e.g., open embeddings of schemes, 231 Nagata’s Compactification Theorem, 784, 785
local on the target, 314 Nagata’s Lemma, 167, 450, 450, 451
preserved by base change, 314 Nagell-Lutz Theorem, 571
preserved by composition, 42, 284 Nakai-Moishezon criterion for ampleness, 540,
575, 580, 589, 589, 590, 592, see also
reasonable class, 314
Kleiman’s criterion
Mor, 29
Nakayama’s Lemma, 183, 232, 235, 235, 236,
Mordell’s conjecture (Faltings’s Theorem), 549
236, 273, 348, 351, 370, 379, 380, 396, 413,
Mordell-Weil Theorem, group, and rank, 556,
537, 613, 632, 698, 711, 727, 733, 750
570
for half-exact functors, 711, 712
morphisms, 29, see also diagonal morphism;
geometric, 411, 413, 459, 654
gluing morphisms; morphisms of
graded version, 706
schemes
natural isomorphism of functors, 34
in a category, 29
natural transformation of functors, 33, 34, 43,
more fundamental than objects, 229
49, 60, 79, 279, 280, 473, 666, 794
of (pre)sheaves, 79
nef, see line bundles, nef
induce morphisms of stalks, 79 nef cone, 512, 513, 513, 514, 583, 585, 592
of graded rings, 214 = closure of ample cone, 592
of local rings, 207 Néron-Severi group NS(X), 512, 513, 719
of locally ringed spaces, 207 Néron-Severi Theorem (Theorem of the Base),
of manifolds, 100 513, 513, see also Picard number ρ(X)
of ringed spaces, 204 nilpotent, 15, 112, 112
of sheaves determined by stalks, 83 =vanishes at every associated point, 188
morphisms of schemes, 16, 206, 208, see also nilradical, 113
affine morphisms; closed embeddings; is intersection of all primes, 113
gluing morphisms; locally closed nodal cubic curve, 240, 304, 645, 648, 686, 699,
embeddings; open embeddings; 701, 703
projective morphisms; proper morphisms; normalization of, 304, 701
quasiaffine morphisms; quasiseparated node, 305, 372, 372, 651, 769, 773, 775
morphisms; reasonable class; separated normalization, 536
morphisms; smooth morphisms Noether normalization, 308, 338, 339, 340,
and tangent spaces, 363 341–343, 354–356, 465
criterion to be a closed embedding, 535, 536 geometric, 340
dominant, 217 projective, 803
from Spec of local ring, 211 Noether-Lefschetz Theorem, 346
glue, 207 Noetherian, see also scheme, locally
of group schemes, 226 Noetherian; scheme, Noetherian
to affine schemes, 209 hypotheses, 16, 19, 127, 740
multiplicative group Gm , 224 induction, 124
multiplicative subset, 35 local ring has finite dimension, 338, 351
multiplicity of function at regular point, 650 module, 126
 © 2024 Ravi Vakil. Published by Princeton University Press. 837

rings, 121, 124, 125 proofs, 247, 339

e.g., principal ideal domains, 125 weak, 107, 114, 122, 213, 292
finitely generated module over, 190, 192 number field, 306, see also ring, of integers in
finitely many minimal primes, 128 number field
important facts about, 125, 126 numerical criterion
infinite-dimensional, 334, 337, 351 for ampleness, see Kleiman’s criterion
preserved by quotients and localization, for line bundle on curve to be
125 base-point-free, 539, 540
topological space, 124, 124–126, 157, 163, for line bundle on curve to be very ample,
246, 504 540
finitely many irreducible components, 124 numerical equivalence, 513, 578, 589
open subsets are quasicompact, 126 implied by algebraic equivalence, 719
non-affine scheme (example), 140, 143–146, 241 numerically effective line bundles, see line
non-archimedean, 178, 182, 198, 380 bundles, nef
non-zerodivisor, 36, 36, 270–273, 453, 604, 640, numerically trivial line bundles, see line
643, 646, 647, 651, 688, 690, 691, 712–714, bundle, numerically trivial
749, 751–753, see also zerodivisor
nondegenerate linear series, 430 O (structure sheaf), 78, 132
nonprojective proper variety, see proper O-modules, 78, 171
nonprojective variety (examples) Ext-functors for, see Ext functors
nonsingular, see regular may not have enough projectives, 676
normal (ring/scheme), 143, 165, 165, see also not always quasicoherent, 173
normalization; Serre’s criterion for on X form an abelian category, 92, 402
normality on a base, 88
= R1 + S2 , see Serre’s criterion for normality tensor product, 92
implied by factorial, 167 O(1) on relative Proj, 483
may not be integral, 166 OP1 (m), OPn (m), 426, 427
may not be UFD, 169 O ⊕I = O I , free sheaf of “rank I”, 399
stalk-local property, 165, 166 OK (ring of integers in number field K), 306
normal bundle, 603, see also normal exact OSpec A , 131
sequence; normal sheaf OX (−D) for D an effective Cartier divisor, 452
projectivized, 643, 645 OX (D), 442
to effective Cartier divisor, 453, 581–583, 603 for D an effective Cartier divisor, 452
to exceptional divisors, 644, 645, 649, 785 is line bundle if X Noetherian factorial, 445
to regular embedding, 604 is quasicoherent, 442
normal cone, 644, see also deformation to the OX,p , 78
normal cone O-connected morphisms, 732, 732–734, 780,
projective completion, 656 781, 783
normal exact sequence for smooth varieties, + proper implies connected fibers, 778
606, 610, see also conormal, exact sequence + proper implies surjective, 733, 783
normal form for quadrics, see diagonalizing + proper not reasonable class, 733
quadratic forms local on the target, 732
normal sheaf, 603, 809, see also normal bundle not preserved by base change, 733
to regular embedding is locally free, 604 preserved by composition, 732
normalization, 239, 303, 304, 304–307, 335, 383, preserved by flat base change, 733
391, 466, 468, 469, 477, 478, 481, 483, 528, object
541, 546, 551, 623, 632, 699, 701, 775, 781, in a category, 29, see also automorphism;
786, see also integral closure; universal group object
property initial / final / zero, 35
finiteness of, 307, 465, 469 in an abelian category, see acyclic; injective;
in field extension, 306, 306, 307, 373, 465 irreducible; projective; quotient; simple;
is a birational morphism, 304, 307 subobject
is integral and surjective, 304 Oka’s Theorem, 182, 198
not flat, 699 open embeddings, 230, 785
of cusp, 535, 536, 632, 648, 708 are flat, 689
of node, 469, 536 are locally finitely presented, 244
Nullstellensatz, 107, 107, 108, 114, 115, 121, are locally of finite type, 242
122, 128, 158, 249, 250, 276, 370, 555 are monomorphisms, 205, 284
838 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

are separated, 320 Pfaffian, 563

base change by, 281 Pic(Ank ) = 0, 445
Pic(Pn ∼
local on the target, 231 k ) = Z, 446
not local on the source, 231 Pic0 (X), 719
of ringed spaces, 78, 204 Pic0 (E), 556, 563, 572
of schemes are monomorphisms, 231 =∼ E, 562
preserved by base change, 231, 291 Picd (C) (the degree d line bundles on C), 508
preserved by composition, 231 Picτ (X) (group of numerically trivial line
reasonable class, 231 bundles), 513
open immersions, see open embeddings Pic(X) (Picard group), 403
open maps of topological spaces, 297, 687, see Picard group, 403, 404, 426, 442, 444–446, 448,
also universally open morphisms 449, 453, 509, 512, 513, 556, 566, 567, 569,
e.g., flat morphisms (in reasonable 570, 572, 580–584, 588, 653, 719, 734, see
situations), 708 also line bundles
open subfunctors, 280 of Spec of UFD is 0, 445
open subschemes, 140, 231, see also affine open of cone, 449
subscheme/subset of projective space, 444
open (sub)sets, see also affine open torsion example, 448
subscheme/subset Picard number ρ(X), 513, 626, 789
distinguished D(f), 109, 114, 118 finiteness of, see also Néron-Severi Theorem
form base for Zariski topology, 118 (Theorem of the Base)
projective distinguished D+ (f), 119, 153, 153 complex case, 513
form base for Zariski topology, 153 PID, see principal ideal domains
opp (opposite category), 33 pinched plane, 383, 429, 755
opposite category, 33, 208, 219, 468 normalization of, 306, 383
Rings and affine schemes, 208 not Cohen-Macaulay, 755
orbifolds, 16, 555 plane, 260, see also affine space; curves, plane
order of pole/zero, 381, 382 with doubled origin, see recurring
ordinary double point, 774 (counter)examples
ordinary multiple (m-fold) point, 774 Plücker
orientation (of spectral sequence), 64, 67 embedding, 228, 473–475, 554, 563
equations, 474
P1k (projective line), 145 plurigenus (birational invariant), 623
P(d1 , . . . , dn ) (weighted projective space), 264 point, see also closed point; general point;
PF (projectivization of finite type generic point; geometric point
quasicoherent sheaf, with caution over field-valued/ring-valued/scheme-valued,
notation), 484 211
Pn∨ (dual projective space), 375 pole, 165, 381, 382
Pn (projective space), 154 of order n for discrete valuation rings, 381
Pn -bundle (projective bundle), 484 of order n for Noetherian scheme, 382
PnA (projective space over A), 146 Poncelet’s Porism, 567, 568, 569
is quasicompact, 157 poset, see partially ordered set
Pnk (proective space over a field k), 146 power series (formal, convergent, etc.), 45, 45,
PV (projectivization of vector space), 155 46, 648, 769, 770, 785
p-adics, 44, 45, 60, 380, 393, 698, 769 preimage, 286, see also scheme-theoretic
page of spectral sequence, 64 preimage
Pappus’s Theorem, 543, 543, 545 preserved by base change
parabola, 106, 109–111, 123, 138, 150, 287, 288, e.g., any reasonable class of morphisms, 229
294, 470, 521, 722 e.g., projective morphisms, 486, 487
parameter space, 346, 352, 512, 735, 742, 745, not “has finite fibers”, 292
759, see also moduli space not closed maps, 325
partial flag variety, see flag variety not injective morphisms of schemes, 292
partially ordered set (poset), 31, see also filtered not O-connected morphisms, 733
partition of unity, 132, 165, 232, 501, 695 preserved by composition
Pascal’s “Mystical Hexagon” Theorem, 543, e.g., any reasonable class of morphisms, 229
543–545, 757 e.g., O-connected morphisms, 732
perfect field, see field, perfect projective morphisms, usually, 489
perfect pairing of vector bundles, 409 preserved by products
 © 2024 Ravi Vakil. Published by Princeton University Press. 839

e.g., any reasonable class of morphisms, 230, commutes with base change, 483
312 finite generation in degree 1, 483
e.g., projective morphisms (despite no great universal property, 481
unreasonableness), 487 O(1), 483
presheaf, 73 projection formula, 519, 526, 528, 579, 580, 589,
̸= sheaf (examples), 75 805
as contravariant functor, 74 for quasicoherent sheaves, 421
cokernel, 81 more general, 579
espace étalé, see sheaf, space of sections projective
germ, 74, see germ A-scheme, 150
image, 81 S-scheme, 438
kernel, 81 Y-scheme, 485
is a sheaf, 82 over Y, 485
on a base, 87 projective morphisms, 487
stalk, 87 projective bundle, 213, 346, 376, 484, 734
quotient, 81 projective change of coordinates, 447, see also
section of, 74 coordinates, projective; projective
separated, 75 transformation
stalk, 74 projective completion, 264, 703
prevarieties (complex analytic), 213, see also of normal cone, 656
varieties, complex analytic projective cone, 149, 155, 264, 264, 341
primary decomposition, 187, 188, 197 relative version, 485
primary ideal, 188, 197, 197, 347 projective coordinates, see coordinates
prime avoidance, 343, 344, 349, 351, 690 projective distinguished open set, see open set
prime ideals, 129, see also maximal ideals; projective limit, see limit
minimal prime ideals; primary ideals projective line, 145, 611
are the points of Spec, 102 projective module
correspond to irreducible closed subsets, 128 = direct summand of free module, 663
Prin (group of principal Weil divisors), 444 = free if finitely generated over local ring,
principal divisor, see effective Cartier divisors, 698
principal; Weil divisor, principal e.g., free module, 663
principal fractional ideal, 404, see fractional is flat, 663
ideal projective morphisms, 438, 468, 477, 485,
principal ideal, 105, 142, 190, 269, 361, 362, 378, 485–487, 489, 496, 498, 524, 527, 532, 533,
381, 404, see also Krull’s Principal Ideal 536, 537, 700, 717, 718, 720, 761, 779, 788,
Theorem 803
principal ideal domains, 105, 373, 378, 380, 412, are proper, 486
428, 556, 687, 688, 697, see also finitely not a reasonable class, 487
generated modules, over principal ideal notion not local on target, 487, 720
domains (classification) preserved by base change, 486, 487
are Noetherian, 125 preserved by product (despite
Principal Ideal Theorem, see Krull’s Principal unreasonableness), 487
Ideal Theorem usually preserved by composition, 487, 489,
principal Weil divisor, see Weil divisor, 533, 642, 720
principal projective objects, 663, see also injective objects
principality not affine-local, 270, 572 projective resolution, see resolution
product, 27, 28, 45, see also fibered product; projective schemes, see scheme, projective
ideal sheaves, product of; universal projective space, 146, 154, see also coordinates,
property projective xi ; projective, change of
of irreducible varieties /k is irreducible, 292 coordinates; xi/j (coordinates on patches
of schemes, 275 for projective space)
of schemes isn’t product of sets, 275 automorphisms, 447, 551
of varieties, 319 cohomology of line bundles on, 504
profinite topology, 283 functions on, 146
Proj S• , 148, 152, 153 functorial definition, 431
ProjA S• , 481 is finite type, 242
Proj (relative Proj), 477, 481, 482 is separated, 317
and affine base change, 481 line bundles on, 425
840 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Serre duality for, 796, 801 of schemes, 285, see also base change;
tautological bundle, 480 scheme-theoretic, pullback
projective transformation, 447, see also of closed subschemes by flat morphisms,
projective change of coordinates 692
projective variety, see variety, projective pure dimension, 334
projectively normal, 756 purely inseparable, see field extension
projectivization push-pull map
of finite type quasicoherent sheaf, 484, see for O-modules, 205
also normal bundle, projectivized; tangent for cohomology, 525, 526, 729, 730, see also
cone, projectivized Cohomology and Base Change Theorem
of vector space, 155 isomorphism for flat base change, 691
projectivized tangent cone, see tangent cone, for quasicoherent sheaves, 421
projectivized for sheaves, 94
proper (=strict) transform (of blow-up), 582, pushforward
636, 638, 640, 640, 641, 648, 649, 654, 720, occasionally left-adjoint to π!? , 806, 807
767, 788, 789 of O-modules, 205
of nodal curve, 648 of quasicoherent sheaves, 414
proper (=strict=birational) transform (of is exact for affine morphisms, 415
blow-up), 638 right adjoint to inverse image, 92, 93
proper morphisms, 120, 148, 311, 325, see also sheaf, 77, see sheaf, pushforward
Chow’s Lemma; Grothendieck’s Pythagorean triples, 219, 220, 222, 542
Coherence Theorem; valuative criteria
e.g., projective morphisms, 486 Q-Cartier Weil divisor, 450
local on the target, 326 Q-line bundles, 513
of projective schemes, 326 ampleness, 514
over A, 325 nef, 513
preserved by base change, 326, 329 QCohX (category of quasicoherent sheaves on
preserved by composition, 326 X), 171, 178
preserved by product, 326 is an abelian category, 178
reasonable class, 326 qcqs (quasicompact quasiseparated), 19, 158,
proper nonprojective variety (examples), 327, 158, 159, 177, 236, 236–238, 245, 392, 415,
434, 583, 720 420, 421, 447, 463, 487, 490, 491, 530, 784,
property Rk , see Rk 785
property Sk , see Sk down-to-earth interpretation, 158
pseudo-morphisms, 216 Qcqs Lemma, 177, 238
Puiseux series, 380 generalization, 446
pullback, see also π∗ ; universal property quadratic forms, see diagonalizing quadratic
diagram/square, 40, 285, see also base forms; maximal rank; rank
change diagram/square; Cartesian quadric hypersurface, 150, 259, 263, 475, 553
diagram/square; fibered diagram/square quadric surface, 135, 148, 166, 168, 169, 221,
in cohomology, 494 262, 263, 303, 339, 362, 369, 383, 451, 523,
of O-modules, 205, see also π∗ 552, 553, 569, 652, 653, see also del Pezzo
of a section, 419 surfaces; cone over quadric surface;
of quasicoherent sheaves, 414, 428, see also recurring (counter)examples
π∗ quartic, 623
construction via affines, 417 hypersurface, 259
explicit construction, 419 plane curve, 295, 551–553, 561, 757, see also
left-adjoint to pushforward for qcqs Klein quartic curve
morphisms, 420 surface, 346, 623
not left-exact, 420 most contain no lines, 346
of ideal sheaves (subtleties), 421, 422 quasiaffine morphisms, 461, 490, 491
preserves ⊗, 420 affine-local on target, 491
preserves finite type and finite e.g., quasicompact locally closed
presentation, 420 embeddings, 491
pullback of vector bundle is vector preserved by base change, 491
bundle, 419 preserved by composition, 491
right-exact functor, 420 properties of, 491
stalks and fibers, 420 reasonable class, 491
 © 2024 Ravi Vakil. Published by Princeton University Press. 841

quasiaffine schemes, 490 preserved by base change, 291

= OX ample, 491 preserved by composition, 236
quasicoherent sheaves, 16, 134, 171, 171, 399, reasonable class, 314
402, see also coherent; finite type; finitely quasiseparated schemes, 19, 158, 158, 177, 237,
presented; global generation; Künneth 238, 447, 463, 465, 491, 784, 785
formula e.g., affine schemes, 158
characterization using distinguished affine e.g., locally Noetherian schemes, 163, 237
base, 174, 175 e.g., projective A-schemes, 159
corresponding to graded modules, 422, 422, not A∞ with doubled origin, see also
454 recurring (counter)examples, 159
corresponding to modules, 171 quasiseparated topological space, 158
higher pushforward, see sheaf, higher Quillen-Suslin Theorem, 445
pushforward quotient
local nature of, 422 object, 54
morphisms of, 178 of monomorphism in abelian category, 54
not all locally free, 405 presheaf, 81
not necessarily preserved by Hom , 410 sheaf, 85
on X form an abelian category QCohX , 178,
254, 399 Ri π∗ (higher pushforward sheaf), 524
on affines have no derived functor Ri F (right derived functor), 664
cohomology, 680 Rk (regular in codimension k), 754
on projective schemes, 422 R0 = generically reduced, 754
on ringed spaces, 179 R1 (regular in codimension 1), 382, 440
pullback of, 419 radical ideal, 116
pushforward, see sheaf, pushforward radical of ideal, 116
pushforward usually quasicoherent, 415 = intersection of primes containing ideal,
restriction of, 416 116
quasicompact morphisms, 19, 236 commutes with finite intersections, 116
affine-local on the target, 237 radicial morphisms, see universally injective
preserved by base change, 291 (radicial) morphisms
preserved by composition, 236 ramification
reasonable class, 292 divisor, 617
quasicompact schemes, 157 and number of preimages, 619
e.g., affine schemes, 121 locus, 617, 631
have closed points, 157 order, 618
quasicompact topological space, 120 point, 545, 618
quasifinite morphisms, 240, 243, 784, 785 tame/wild, 618
from proper to separated is finite, 328 rank
how to picture, 243, 785 of (locally) free sheaf, 399
may not be finite, 243 of coherent sheaf is additive in exact
most are open subsets of finite morphisms, sequences, 413
785 of finite type quasicoherent sheaf at point,
preserved by base change, 292 412
reasonable class, 292 of OX -module, 142
to fields are finite, 243, 247, 289, 292, 784 of quadratic form, 168, see also maximal rank
quasiisomorphism, 500, 736, 737 (quadratic form)
quasiprojective, see also scheme; variety of quasicoherent sheaf, 413
implies finite type, 163 of vector bundle on manifold, 400
implies separated, 320 rankp (rank of O-module at p), 142
morphisms, 490 rational function, 78, 102, 104, 142, 160, 165,
quasiseparated morphisms, 229, 236, 236, 237, 196
245, 314, 314, 316, 317, 391, 392, 415, 420, rational map, 216, see also birational; degree;
421, 487, 682, 691 domain of definition; dominant; graph;
affine-local on target, 237 Reduced-to-Separated Theorem
e.g., morphisms from locally Noetherian rational normal curve, 120, 262, 265, 432, 523,
schemes, 237 548, 549, 585, see also twisted cubic curve;
e.g., morphisms from quasiseparated Veronese embedding
schemes, 237 not complete intersection for d > 2, 523
842 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

rational point of k-scheme, see k-valued point reduced subscheme structure, 269, 391, 629,
rational section 655, 816
of line bundle, 441 Reduced-to-Separated Theorem, 216, 321, 322,
of locally free sheaf, 405 323, 324, 389, 390, 438, 765
rational variety, 217, 221, 765 reducedness, 159
reasonable classes of morphisms, 229 is stalk-local, 159, 163, 165, 301
not projective morphisms, 487 not open condition in general, 160
by definition include all isomorphisms, are of ring, 113
preserved by composition, are preserved often open condition, 269
by base change, are local on the target, 229 reduced rings have no embedded primes,
e.g., closed embeddings, 291 195
e.g., étale morphisms, 387 reducible, see irreducible
e.g., faithfully flat morphisms, 707 reduction
e.g., finitely presented, 314 of a morphism, 309
e.g., flat morphisms, 690 of a scheme, 269, 308, 483, 511, 528, 703, see
e.g., flat of relative dimension n, 709 also universal property
e.g., isomorphisms, 230 Rees algebra, 395, 643, 791
e.g., locally closed embeddings, 291 reflexive sheaf, 410, 603
e.g., monomorphisms, 314 regular
e.g., open embeddings, 231 An k is, 374, 375
e.g., proper morphisms, 326 does not imply smooth over field, 368
e.g., quasiaffine morphisms, 491 implies Cohen-Macaulay, 751
e.g., quasicompact; affine; finite; integral; in codimension 1, 382, see R1
locally finite type; finite type; locally in codimension k, 754, see Rk
finitely presented, 292 ring, 365, see also regular local ring
e.g., quasifinite morphisms, 292 scheme, 365
e.g., quasiseparated morphisms, 314 stalk-local property, 365
e.g., separated morphisms, 315 regular embeddings, 269, 273, 273, 274, 349,
e.g., smooth morphisms, 387 363, 369, 603, 604, 610, 643, 645, 646, 692,
e.g., surjective morphisms, 292 747, 752, 755, 794, 809
e.g., universally closed morphisms, 325 blowing up, 645
e.g., universally injective (radicial) e.g., regular in regular, 369
morphisms, 297, 321 in smooth k-schemes are Cohen-Macaulay,
e.g., unramified morphisms, 632 752
P reasonable implies Pδ reasonable, 313 normal bundle is locally free, 604
preserved by products, 230 open condition, 273
satisfy the Cancellation Theorem, 312 pull back under flat morphisms, 692
recurring (counter)examples, 16, 135 regular function, 78, 197, see function
Q∞
F2 , 121, 124, 135, 699 regular local rings, 365, 365, 366, 368–370, 374,
k(x) ⊗k k(y), 135, 284, 345, 709 375, 378–382, 614, 725, 753, 770, 781
x2 = 0 in the projective plane, 135, 160, 327, are Cohen-Macaulay, 394, 747, 751, 755
523 are integral domains, 368
(cone over) quadric surface, 135, 148, 166, are integrally closed, 394, 747, 755
168, 169, 221, 259, 262, 303, 339, 362, 369, are Noetherian by definition, 365
372, 383, 449–451, 552, 569, 652, 653 are unique factorization domains, see
affine space minus the origin, 126 Auslander-Buchsbaum Theorem
affine space minus the origin, 121, 135, 142, finitely generated modules have finite free
143, 241, 243, 266, 504 resolutions, 705
affine space with doubled origin, 135, 144, preserved by localization, 366, 393, 394, 598,
145, 159, 241, 311, 317, 320, 322, 323, 390 627, 754, 755
embedded point on line, 135, 138, 159, 187, regular in regular is regular embedding, 369
195, 196, 323, 691 regular section of locally free sheaf, 405
infinite disjoint union of schemes, 135, 140, regular sequence, 187, 269, 270, 271, 271, 273,
143, 160, 177, 266, 269, 722, 772 274, 604, 645, 647, 715, 748, 749, 751
Spec Q → Spec Q, 135, 233, 241, 243, 276, and the Koszul complex, 705
283, 292, 294 cohomological criterion for existence, 748
two planes meeting at a point, 134, 135, 515, length, 271
523, 699, 714, 750, 751, 753, 755, 798 maximal, 748
 © 2024 Ravi Vakil. Published by Princeton University Press. 843

order doesn’t matter (over Noetherian local Riemann-Hurwitz formula, 535, 559, 612, 613,
rings), 271 616, 619, 619–621, 623
order matters, 271 applications, 619
preserved by flat base change, 271 hyperelliptic, 546, 547, 548, 550, 556, 613
relative cotangent sheaf, 606 Riemann-Roch, 507, 508, 508, 514, 520, 538, 552,
relative different and discriminant ideals, 633 555, 560, 579, 582, see also Asymptotic
relative dimension, see flat morphisms, relative Riemann Roch;
dimension n; smooth morphisms, of Hirzebruch-Riemann-Roch;
relative dimension n Grothendieck-Riemann-Roch
relative Proj and relative Spec, see Proj and common formulation, 510
Spec for coherent sheaves or vector bundles, 510
relative tangent sheaf, 606 for nonreduced curves, 511
relatively ample, see ampleness, with respect for surfaces, 579, 582, 582, 586
to π restatement using Serre duality, 514
relatively base-point-free, see right adjoints, see adjoints, right
base-point-freeness, with respect to π right derived functors, see derived functors,
relatively globally generated, see global right
generation, with respect to π right-exact functors, 38, 58
relatively very ample, see very ample, with and derived functors, 493
respect to π and left derived functors, 664
Remainder Theorem, 646 and the FHHF Theorem, 59
representable functor, 42, 43, 223, 223, 224, e.g., left adjoints, 60, 61
279–281, 448, 471–473, 475, 478, 741–745, e.g., tensor product, 38, 58
see also fibered product; Grassmannian; may not commute with colimits, 62
Hilbert scheme; Yoneda’s Lemma; Zariski Rigidity Lemma, 328, 329, 329, 330, 497
sheaf ring, see also Noetherian; graded ring; local
residue field κ(p) of a point p, 141 rings; regular local rings
Residue Theorem, 470, 797 catenary, 342, 754
resolution (complexes), 458, see also of integers in number field, 306, 306, 382,
Cartan-Eilenberg resolution; Čech 399, 404, 450, 633
resolution; Ext functors topological, 770
acyclic, 667, 669 ring scheme, 225
flat, 669 ringed spaces, 15, 78, see also locally ringed
free, 660, 667, 693, 704, 705, 800 spaces; gluing morphisms; structure sheaf
graded, 706 (locally) free sheaf on, 399, 401
injective, 667, 670–673, 677, 680, 749, (quasi)coherent sheaves on, 179, 182
799–801, 807 functions on, 78, 102, 141
locally free, 799, 800, 808 gluing along open sets, 143
projective, 663 higher pushforwards for, 674
resolution (singularities), 635, 636, 648–650, isomorphism of, 139, 139, 140, 204, 783
775, 789, see also Alteration Theorem; Leray spectral sequence, 674
Hironaka’s Theorem; Weak Factorization morphisms of, 203, 204, 204–208, 418, 419,
Theorem 478, 674–676
for curves, 781 open embedding of, 78, 204, 208, 230, 418
for surfaces, 781, 789 Picard functor for, 404
of indeterminacy of rational maps, 635 pullback for, 416
small, 618, 652, 653, 653 pushforward of O-modules, 205
non-isomorphic, 653 Rings (category of rings), 31
restriction, see also pullback; quasicoherent Rk (regular in codimension k), 754
sheaves ruled surface, 484, 583, 584, 789, see also
map (sections of sheaves), 74 elementary transformation; Hirzebruch
of scalars, 52 surface
of sheaf to open subset G |U , 76 self-intersection of section of, 584
of Weil divisor, 440 ruling of quadric, 148, 262, 262, 263, 303, 346,
resultant, 250 369, 451, 569
Riemann surface, 20, 101, 468, 549, 620, 797 of higher-dimensional quadrics, 263, 263
genus, 510 of quadric hypersurfaces in P5 , 475
844 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

S• (graded ring), 151 are qcqs, 159

S+ (irrelevant ideal), 151 line bundles on, 428
Sk , 754, 754, 755 morphisms to, 431
and Cohen-Macaulayness, 754 quasiprojective, 148, 155
S1 = no embedded points, 754 visualizing, 106, 110, 113, 135, 136, 213, 243,
S2 (elementary interpretation), 754 264, 414
saturated graded module, 456 Schubert cell, 227, 228
saturation functor, 423, 454, 455, 455, 456 Second Riemann Extension Theorem, 383
left-adjoint to forgetful functor, 454, 456 section
may not be injective or surjective, 455 of a morphism, 316
Sch (category of schemes), 208 of sheaf
SchS (category of schemes over S), 210 means global section, 74
scheme-theoretic, 196, 256 over open set, 74
base locus, 430 Segre embedding, 301, 302, 318, 434, 460, 490,
closure, 196, 265, 268, 323, 532, 533, 639–641, 537
700 as complete linear series, 433
fiber, 196, 229, 286, 287, 288, 297, 725, 761 coordinate-free description, 303
image, 196, 265, 265–268, 324, 326, 327, 430 Segre variety, 302
and set-theoretic image, 267 self-intersection of curve on a surface, 581
of proper is proper, 326 semiample line bundle, 464
usually computable affine-locally, 267 semicontinuity, 345
intersection, 196 (lower) rank of matrix, 345, 736–738
of closed subschemes, 138, 255, 274, 361, (upper) Jacobian corank function, 365
377, 430, 521, 581, 638, 641 (upper) degree of finite morphism, 345, 414,
of locally closed subschemes, 291 537
inverse image, see scheme-theoretic, (upper) dimension of a variety at a point,
preimage 345
preimage, 196, 285, 286, 414, 469, 582, 637, (upper) dimension of tangent space, 599
638, 641, 643, 761, 787 (upper) dimension of tangent space at
pullback, see scheme-theoretic preimage closed points, 345
support (of a module or finite type (upper) fiber dimension, 345, 353, 498, 537,
quasicoherent sheaf), 196, 256, 511, 532 760, 782
union (finite) of closed subschemes, 255, 361, (upper) fiber dimension (for closed maps),
656 356
scheme-theoretically dense, 196 (upper) fiber dimension at a point, 356
schemes, 139, see also disjoint union; finite type; (upper) rank of cohomology groups, 345, see
functions; gluing; isomorphism; locally of Semicontinuity Theorem
finite type; morphisms; gluing (upper) rank of sheaf, 345, 413, 414, 627, 730
morphisms; normal scheme; quasiaffine; list of examples, 345
quasiseparatedness; regular; . . . Semicontinuity Theorem, 345, 701, 704, 717,
affine, 132, 139, see also Serre’s 730, 730, 731, 735, 737, 740
cohomological criterion for affineness semigroup (abelian), 50, see also groupification
is quasicompact, 121 graded, 214
is quasiseparated, 158 separable field extension, see field extension
morphisms to, 209 separable finite morphisms, 616
locally Noetherian, 158, 162 separably closed, see field
implies quasiseparated, 163 separably generated, see field extension
Noetherian, 19, 157, 162 separated morphisms, 144, 148, 158, 164, 216,
fibered product need not be Noetherian, 236, 311, 312, 315, see also
276 Associated-to-Separated Theorem;
finitely many connected and irreducible Reduced-to-Separated Theorem; valuative
components, 163 criteria
finiteness of zeros and poles, 382 e.g., affine morphisms, 316
with non-Noetherian ring of functions, e.g., proper morphisms, 325
573 implies quasiseparated, 316
non-affine, 140 local on the target, 315
over a ring, 163 over A, 315
projective, 51, 148, 149, 155 preserved by base change, 315
 © 2024 Ravi Vakil. Published by Princeton University Press. 845

preserved by composition, 320 higher pushforward, 457, 523–527, 532, 589,

reasonable class, 315 677, 682, 691, 692, 717, 729, 731, 735, 744
topological characterization, 315 and base change, 525
separated presheaf, 75 as derived functor, 674
separating transcendence basis, 300 commutes with affine flat base change,
proof of existence, 300 526
Serre duality, 479, 495, 514–516, 535, 538, 552, commutes with flat base change, 691
555, 581, 586, 587, 612, 613, 622, 625, 664, quasicoherent definition, 524
665, 747, 752, 793, 797 identity axiom, 75
for Cohen-Macaulay projective schemes, 515 injective, see also injective objects
for Ext, 793, 798 have no Čech cohomology, 679
functorial, 798, 801, 803, 805–809 implies flasque, 675
functorial (for projective space), 801 morphisms determined by stalks, 83
trace version, 798, 798 of (relative) i-forms ΩiX/Y , 622
for Hom, 793, 795, 796, 797, 804 of differentiable functions, 100
for projective space, 796 of ideals, see ideal sheaves
functorial, 794, 804, 806 of nilpotents, 176
functorial = trace version, 795 of relative differentials, 595
trace version, 804 of sections
for projective schemes, 802 of a map, 77
for smooth projective varieties, 514 of a vector bundle, 78, 400
for vector bundles, 793, 794, 794, 797 of smooth functions, 71
functorial, 798, 801, 803, 805 on a base, 86, 87, 174
functorial (for projective space), 801 determines sheaf, 87
trace version, 798, 805 pushforward, 77, 77, 79, 91, 92, 100, 205, 414,
functorial vs. trace version, 793 415, 418, 419, 421, 480, 496, 526, 589, 674,
Serre vanishing, 495, 507, 516, 518, 529, 591, 731, 733, 790, 802
717, 786, 798, 805, 807 induces maps of stalks, 77
relative, 527 preserves coherence for projective
Serre’s cohomological criterion for affineness, morphisms, 496
494, 504, 527, 530, 530 preserves coherence for proper
Serre’s cohomological criterion for ampleness, morphisms, see Grothendieck’s
461, 527, 528, 529, 541 Coherence Theorem
Serre’s criterion for normality, 385, 747, 754, right adjoint to π−1 (for sheaves of sets),
755, 755 93
Serre’s property Sk , see Sk right-adjoint to π∗ (for O-modules), 205
Serre’s Theorem A, 459, 461, 516, 807 usually right-adjoint to π∗ (for
Sets (category of sets), 30 quasicoherent sheaves), 420
SetsX (category of sheaves of sets on X), 79 restriction to open subset G |U , 416
sextic, 259, 552, 558, 559 sections determined by germs, 82
sheaf, 75, see also coherent; dualizing; space of sections (espace étalé), 71, 74, 77, 85,
extension by zero; finite presentation; 94
finite type; germs; inverse image; kernels; sheaf Hom, sheaf Proj, sheaf Spec, see Hom ;
line bundle; locally free; pullback; Proj ; Spec
quasicoherent; quotient; section; sheafification, 52, 81, 84, see also universal
skyscraper; structure sheaf; subsheaf; property
support; Zariski sheaf; . . . construction, 84
determined by sheaf on distinguished affine induces isomorphism of stalks, 85
base, 87, 175 is a functor, 84
direct image, see sheaf, pushforward left-adjoint to forgetful functor, 84
espace étalé, see sheaf, space of sections short exact sequence, 55
flasque (flabby), 674 signature of symmetric bilinear form, 585
have no Čech cohomology, 680 simple group (of order 168), 552
implies acyclic, 676 simple object in abelian category, 183, see
preserved by pushforward, 675 Jordan-Hölder package
gluability axiom, 75 simply connected (étale), 619
gluing, 89 singular (point), see regular
846 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

singularities, 365, 769, 773, see also ADE Spec k, 104

singularities; An -singularities; curves; Spec Z, 104
Dn -singularities; En -singularities; specialization, 122, 122, 123, 157, 246, 247, 352
resolution (singularities) spectral functor, 70, 682
site, see Grothendieck topology spectral sequence, 58, 63, 64, 64–70, 272, 273,
skyscraper sheaf, 76 531, 667, 669, 670, 673, 677–679, 681, 682,
slicing criterion, 713, see also effective Cartier 696, 800, see also Grothendieck
divisors, slicing by composition-of-functors spectral
for Cohen-Macaulayness, 751, 752 sequence; Leray spectral sequence;
geometric interpretation, 751 orientation of spectral sequence
for flatness, 712, 751 abut, 66
on the source, 714, 723 and derived functors, 667
on the target, 713, 713–715, 753 collapse of, 66
for regularity, 366, 564, 646, 725, 751 convergence, 66
SLn (group scheme), 227, 328 local-to-global for Ext, 801, 802, 807
small resolution, see resolution (singularities) page of, 64
smooth functions (sheaf), 71, 72, 76, 78, 79, split exact sequence, 673
99–101, 203 split node, 774
not coherent, 183 stacks, 16, 89, 178, 320, 555, 557, 732
smooth morphisms, 101, 359, 385, 386, 387, 721, stalk-local properties, 82, 159
725, 727, 728, see also formally smooth; Cohen-Macaulayness, 751
generic smoothness; smooth varieties exactness for sheaves, 91, 179
differential geometric motivation, 385 factoriality, 167
implies flat, 723 flatness, 689
local on source and target, 387 implies affine-local, 162, 230
of relative dimension n, 386, 386, 388, 608, inverse image sheaf, 94
630, 632, 723, 725 local freeness of finitely presented sheaves,
open condition, 386 411
preserved by base change, 387 morphisms of sheaves, 83
open condition, 551 being epimorphisms, 85
preserved by base change, 387 being injective, 85
preserved by composition, 387 being isomorphisms, 83
preserved by product, 387 cokernels, 90
“reasonable” class, 387 kernels, 89
smooth varieties, 327, 328, 359, 366, 367, 609, normality, 165, 166
630, 636, 725 reducedness, 159, 163, 165, 301
are Cohen-Macaulay, 751 regularity of a locally Noetherian scheme,
implies reduced, 369, 611 365
of dimension d, 367, 388 sections of a sheaf, 82
smoothness preserved by base field support, 95
extension, 367 tensor product of sheaves, 92
smoothness over a ring, 388 torsion-freeness, 173
Smoothness-Regularity Comparison Theorem, stalkification, 79
368, 646 doesn’t commute with Hom , 80
proof, 725 is exact, 61, 91
Sn• ⊂ S• (Veronese subring), 215 stalks
Snake Lemma, 55, 58, 66, 68, 68, 198 as colimit, 74
space of sections of sheaf, see sheaf, space of determine isomorphisms, 83
sections determine morphisms, 83
Spec (relative Spec), 477, 478, see also universal of (pre)sheaf on base, 87
property of a presheaf, 74
and affine morphisms, 479 of a sheaf, 75
and base change, 479 preserved by inverse image, 94
computable affine-locally, 478 Stein factorization, 778, 779, 779, 780, 782
universal property, 478 commutes with flat base change, 781
Spec 0 = ∅, 104 Stein Factorization Theorem, 769, 779, 780
Spec A, 102 Stein space, 530, 570
Spec adjoint to Γ , 209 stereographic projection, 219
 © 2024 Ravi Vakil. Published by Princeton University Press. 847

strict transform, see proper transform tacnode, 305, 372, 650, 774, 775
structure morphism, 210 tame ramification, 618
structure sheaf, 132 tangent
of ringed space, 78 bundle, 399
subcategory, 32 cone, 645, 645
full, 33, 33, 51, 55, 79, 83, 135, 178, 185, 279 = tangent space for regular point, 645, 646
submersion, 101, 386, 596, 606, 721, 722, see also projectivized, 645
smooth morphisms line, 360, 361, 372, 373, 373, 377, 560, 561,
subobject, 54 564, 568, 569
subpresheaf, 81 map, 535, 536
subschemes = locally closed subschemes, 257, plane, 372, 373
see also closed / locally closed / open sheaf, 606, see also cotangent, sheaf
subschemes relative, 600, 606
subsheaf, 85 space, 101, 345, 359, 360, 360–363, 365, 373,
subtler Five Lemma, see Five Lemma 382, 388, 444, 536, 595, 596, 599, 600, 602,
subvariety (e.g., open or closed), 319 611, 645, 646, 721, 722
Supp D (support of divisor), 440 and morphisms, 363
Supp G (support of a sheaf), 95 vector, 360, 595
Supp m (support of element of module), 135 relative, 596
Supp M (support of module), 135 Taniyama-Shimura Conjecture, 549
Supp s (support of a section), 95 Tate curve, 559
support tautological bundle
commutes with localization, 194 on projective space, 480
of m ∈ M, 135 on the Grassmannian, 473, 616, 741
of a module, 135 tensor algebra, 408
= closure of associated points, 188 tensor product, 37, see also universal property
of a section s, 95, 411 by locally free sheaf is exact, 403
is closed, 95 commutes with direct sum, 39, 62, 660, 773
of a sheaf, 95 is not left-exact, 38
of a Weil divisor, 440 is right-exact, 38, 58, 688, 773
of finite type quasicoherent sheaf is closed, left-adjoint to Hom, 49
411 of OX -modules, 92
scheme-theoretic, 256 of quasicoherent sheaves, 176
stalk-local notion, 95 of rings, 39
surfaces, 259, 334, 623, see also cubic; Fermat is fibered coproduct, 41, 276
cubic surface; intersection theory; tensor-finiteness trick, 298, 298–301
irregularity; quadric surface; quartic; Theorem on Formal Functions, 769, 776, 777,
resolution (singularities); Riemann-Roch; 778, 778, 779, 786–788
ruled surface proof, 789
Enriques-Kodaira classification, see also del thickening, see formal neighborhood
Pezzo surfaces; Hirzebruch surfaces; K3 Top (category of topological spaces), 31
surfaces; minimal models, 624, 789 topological spaces, see gluing morphisms;
proper but no nontrivial line bundles, 572 Noetherian
proper nonprojective, 572, 583 topology, see base of a topology; discrete
surjective morphisms, 234 topology; étale topology; Grothendieck
checking on closed points, 247 topology; profinite topology; Zariski
preserved by base change, 292 topology
reasonable class, 292 topos, 27, 178
Sylvester’s law of intertia, 585 Tor functors, 659, 663, 686, 687, 693
Sym• M, Sym• F (symmetric algebra), 408, computable with flat resolution, 669
409 criterion for flatness, 662
symbolic power of an ideal q(n) , 347 long exact sequence, 661
symmetric algebra, 151, 408 symmetry of, 667, 693
system of parameters, 351, 353, 366 torsion
(sub)module, 173
t : Hn (X, ωX ) → k (trace map), 793 quasicoherent sheaf on reduced scheme, 174,
T • M, T • F (tensor algebra), 408, 409 412
TX/Y (relative tangent sheaf), 606 torsion-freeness, 173
848 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

= flat for PID, 697 universal family over moduli spaces, 741, 741,
implied by flatness, 688, 693, 697, 712, 753 742, 744
is a stalk-local notion, 173 universal hyperplane, see hyperplane,
module, 173 universal
sheaf, 174, 412 universal property, 27, 35
torsor, 283 and Yoneda’s Lemma, 42
total complex (of a double complex), 64 of blow-up, 635, 637, 637–639, 641–643, 692,
total fraction ring (total quotient ring), 196, 787
197, 646, see also fraction field of cokernel, 54, 81, 90
total space of colimit, 46, 90
of line bundle, 570, 573, 584, 787 of coproduct, 41
of vector bundle / locally free sheaf, 402, of differentials, 597, 604, 604, 605, 607
479, 479, 573, 645 of fibered coproduct, 41, 41
total transform (of blow-up), 638, 640, 641, 648, of fibered product, 39, 40, 276–279, 291
651 of inverse image sheaf, 92, 93
trace map, 410, 793, 794, 795, 795–798, 804, 805 of kernel, 54, 81, 82
transcendence basis, 338, see also separating of limit, 44, 61
transcendence basis of localization, 35, 36, 37, 176
transcendence degree, 338, 354 of normalization, 304, 304, 466, 787
= dimension, 338 of product, 28, 28, 35
transcendence theory, 233, 299, 338 of Proj (not great), 481
transition functions/matrices, 400, 400–403, of pullback of O-modules, 418, 419
406, 409, 425–428, 431, 441, 479, 480, 482, of pullback of quasicoherent sheaves, 416,
509, 517, 519, 584 418, 418, 420
trdeg (transcendence degree), 338 of reduction of a scheme, 308
trichotomy (classification of varieties), 549, 624 of sheafification, 83, 84, 84, 90
triple point, 774, 774 of Spec , 477, 477, 478
trivial bundle/sheaf of rank n, 400 of stalks, 74, 78
trivialization of vector bundle / locally free
of Sym and ∧, 408
sheaf, 400
of tensor product, 37, 38, 39, 49, 607
tropical geometry, 127
universally, 293
Tsen’s Theorem, 735
universally closed morphisms, 293, 325, 392,
tubular neighborhood, 603, 645
436, 486, 532, see also valuative criteria
twisted cubic curve, 120, 259, 259, 260, 262, 339,
e.g., proper morphisms, 325
372, 373, 423, 519, 520, 701, 703, see also
reasonable class, 325
rational normal curve; Veronese
universally injective (radicial) morphisms, 293,
embedding
295, 296, 321
two planes meeting at a point, see recurring
= surjective diagonal, 321
(counter)examples
are separated, 321
not S2 or Cohen-Macaulay, 755
e.g., monomorphisms, 296
ultrafilter, 121 local on the target, 297
uniformizer, 379, 380, 410, 439, 469, 471, 613, of varieties over k, 321
618, 697 preserved by composition, 297
union (scheme-theoretic), see scheme-theoretic, reasonable class, 297, 321
union universally open morphisms, 297
unique factorization domain, 102, 104, 166–170, e.g., maps to Spec k, 297–299, 709
183, 190, 222, 305, 349, 362, 378, 381, 383, unramified morphisms, 101, 253, 257, 536, 622,
394, 445, 446, 449–451, 570, 627, 653, 755, 631, 721, 722, 725–728, 783
see also factorial characterizations by fibers, 632
= all codimension 1 primes principal, 167, diagonal characterization, 632
337, 349, 362, 445, 450 e.g., étale morphisms, 725
= integrally closed and class group 0, 450 examples, 631
is integrally closed, 167 formal, see formally unramified
list of useful criteria, 167 in number theory, 633
not affine-local property, 169, 449 open condition, 633
uniruled variety, 352 reasonable class, 632
universal δ-functors, see δ-functors, universal upper semicontinuity, see semicontinuity
 © 2024 Ravi Vakil. Published by Princeton University Press. 849

V(S) (vanishing set), 115 as complete linear series, 432, 537

projective case, 153 Veronese subring Sn• , 215, 215, 261
val = valp (valuation), 380 Veronese surface, 262, 262
valuation rings, 127, 198, 380, 380, 384, 389, very ample, 459, 459
391, 392, see also discrete valuation rings ⊠ very ample is very ample, 460
rank 2 example, 437 ⊗ base-point-free is very ample, 460
Serre makes the case, 392 ⊗ very ample is very ample, 460
valuations, 380, 441, 466, see also discrete implies base-point-free, 460
valuations with respect to π (relatively very ample),
valuative criteria, 19, 359, 389 459, 483, 487
for flatness, 699 affine-local on the target, 489
for properness, 392, 392, 438, 728 worse-behaved than ample, 461
for separatedness, 390, 390, 391, 391, 392, 728 very general, 346
for universal closedness, 392, 392
value of a function at a point, see functions, Weak Factorization Theorem, 653, 789
value at a point Weierstrass normal form, 561, 562, 563, 566
vanishing scheme, 150, 154, 154, 256, 259, 617 and double cover of P1 , 561
vanishing set, 115, 115, 127, 140, 150, 153 weighted projective space, 264
properties of, 116 Weil divisor, 425, 440, 440
vanishing theorems, 507, see also affine cover effective, 440
cohomology vanishing; dimensional irreducible, 440
cohomology vanishing; Kodaira locally principal, 444, 444, 446, 653, see also
vanishing; Serre vanishing closed subscheme, locally principal
varieties, 164, 318 not necessarily locally principal, 444, 450
affine, 164 of zeros and poles, 441
classically, 122 principal, 444, 445
complex algebraic, 14, 99, 108, 110, 114, 148, Q-Cartier, 450
210, 211, 213, 214, 316, 319, 328, 387, 404, restriction to open set, 440
453, 469, 509, 510, 513, 521, 570, 573, 580, support of, 440
610, 620, 625, see also analytification twisting line bundles by, 446
complex analytic, 14–16, 101, 179, 182, 213, Weil X (group of Weil divisors on X), 440
214, 319, 328, 387, 404, 453, 469, 509, 510, Weyl group, 758, 764, 767
513, 521, 530, 556, 559, 570, 580, 610, 620, wild ramification, 618
625, 776
Xan (analytification of X), 213
product of, 319
Xf (f ∈ Γ (X, OX )), 119, 177
projective, 164
xi/j (coordinates on patches for projective
quasiprojective, 164
space), 146, 146, 150, 152, 228, 259, 302,
with non-finitely generated ring of global
317, 318, 425–427, 432, 474, 612, 615
sections, 572
Xs (s ∈ Γ (X, L )), 447
Veck (category of vector spaces over k), 30
vector bundles, 16, 77, 78, 155, 182, 399–402, Yoda embedding, 43
405, 407, 410, 414, 457, 479, 484, 510, 517, Yoneda’s Lemma, 15, 42, 42, 43, 43, 212, 223,
583, 603, 615, 616, 643, 686, 693, 731, 735, 276, 279, 389, 431, 473, 638, 741
793–795, 797, 798, 801, 803–805, see also
Serre duality, for vector bundles; total Z-graded rings, see graded rings, Z-
space Z≥0 -graded rings =: graded rings, 151, see
ample, 490 graded rings
classification on P1 , 515 Zp (p-adics), 45
∼ locally free sheaves, 479 Zariski closure, 116
not always direct sum of line bundles, 615 Zariski cotangent space, see cotangent, space
on manifold, 399 Zariski sheaf (sheaf on the big Zariski site),
perfect pairing, 409 279, 280, 280, 281, 472
pulling back, 416 Zariski tangent space, see tangent, space
trivialization, 400 Zariski topology, 114–118, 120, 136, 139
n
Veronese embedding νd , 120, 215, 260, 261, on k , 249, 345
262, 302, 318, 375, 432, 433, 519–521, 538, on Proj S• , 153
548, 803, see also rational normal curve, on Spec A, 116
twisted cubic curve on a scheme, 139
850 Pre-publication version (2024-07-27) of The Rising Sea: Foundations of Algebraic Geometry

Zariski’s Connectedness Lemma, 733, 769, 778,

778–780, 783
Zariski’s Lemma, 107, see Nullstellensatz
Zariski’s Main Theorem, 243, 327, 765, 778, 784
for birational morphisms, 780, 782
Grothendieck’s form, 769, 782, 784
misnamed, 781
other versions, 783, 785
zero object, see object, in a category,
initial/final/zero
zero/pole of order n
for discrete valuation rings, 381
for Noetherian scheme, 382
zerodivisor, 36, 187, see also non-zerodivisor
= element of associated prime, 191
= vanish at some associated point, 188, 191
Zorn’s Lemma, 19, 113, 123, 292, 664, 665, see
also Axiom of Choice