Graduate Texts in Mathematics 52

Editorial Board
S. Axler F.W. Gehring K.A. Ribet
Graduate Texts in Mathematics
most recent titles in the GTM series

129 FULTON/HARRIS. Representation Theory: 159 CONWAY. Functions of One

A First Course. Complex Variable II.
Readings in Mathematics 160 LANG. Differential and Riemannian
130 DODSON/PoSTON. Tensor Geometry. Manifolds.
131 LAM. A First Course in Noncommutative 161 BORWEINIERDEL Yl. Polynomials and
Rings. Polynomial Inequalities.
132 BEARDON. Iteration of Rational Functions. 162 ALPERIN/BELL. Groups and
133 HARRIS. Algebraic Geometry: A First Representations.
Course. 163 DIXON/MORTIMER. Permutation
134 RoMAN. Coding and Information Theory. Groups.
135 ROMAN. Advanced Linear Algebra. 164 NATHANSON. Additive Number Theory:
136 ADKINS/WEINTRAUB. Algebra: An The Classical Bases.
Approach via Module Theory. 165 NATHANSON. Additive Number Theory:
137 AxLERIBouRDoN/RAMEY. Harmonic Inverse Problems and the Geometry of
Function Theory. Sumsets.
138 CoHEN. A Course in Computational 166 SHARPE. Differential Geometry: Cartan's
Algebraic Number Theory. Generalization of Klein's Erlangen
139 BREDON. Topology and Geometry. Program.
140 AUBIN. Optima and Equilibria. An 167 MORANDI. Field and Galois Theory.
Introduction to Nonlinear Analysis. 168 EWALD. Combinatorial Convexity and
141 BECKERIWEISPFENNING/KREDEL. Grtibner Algebraic Geometry.
Ba~es. A Computational Approach to 169 BHATIA. Matrix Analysis.
Commutative Algebra. 170 BREDON. Sheaf Theory. 2nd ed.
142 LANG. Real and Functional Analysis. 171 PETERSEN. Riemannian Geometry.
3rd ed. 172 REMMERT. Classical Topics in Complex
143 DOOB. Measure Theory. Function Theory.
144 DENNISIFARB. Noncommutative 173 DIESTEL. Graph Theory.
Algebra. 174 BRIDGES. Foundations of Real and
145 VICK. Homology Theory. An Abstract Analysis.
Introduction to Algebraic Topology. 175 LICKORISH. An Introduction to Knot
2nd ed. Theory.
146 BRIDGES. Computability: A 176 LEE. Riemannian Manifolds.
Mathematical Sketchbook. 177 NEWMAN. Analytic Number Theory.
147 ROSENBERG. Algebraic K-Theory 178 CLARKE/LEDYAEV/STERN/W OLENSKI.
and Its Applications. Nonsmooth Analysis and Control
148 ROTMAN. An Introduction to the Theory.
Theory of Groups. 4th ed. 179 DOUGLAS. Banach Algebra Techniques in
149 RATCLIFFE. Foundations of Operator Theory. 2nd ed.
Hyperbolic Manifolds. 180 SRIVASTAVA. A Course on Borel Sets.
150 EISENBUD. Commutative Algebra 181 KRESS. Numerical Analysis.
with a View Toward Algebraic 182 WALTER. Ordinary Differential
Geometry. Equations.
151 SILVERMAN. Advanced Topics in 183 MEGGINSON. An Introduction to Banach
the Arithmetic of Elliptic Curves. Space Theory.
152 ZIEGLER. Lectures on Polytopes. 184 BOLLOBAS. Modem Graph Theory.
153 FULTON. Algebraic Topology: A 185 COXILITILEIO'SHEA. Using Algebraic
First Course. Geometry.
154 BROWN/PEARCY. An Introduction to 186 RAMAKRISHNAN/V ALENZA. Fourier
Analysis. Analysis on Number Fields.
155 KASSEL. Quantum Groups. 187 HARRIS/MORRISON. Moduli of Curves.
156 KECHRIS. Cla~sical Descriptive Set 188 GOLDBLATT. Lectures on the Hyperreals:
Theory. An Introduction to Nonstandard Analysis.
157 MALLIAVIN. Integration and 189 LAM. Lectures on Modules and Rings.
Probability. 190 ESMONDEIMURTY. Problems in Algebraic
158 ROMAN. Field Theory. Number Theory.
Robin Hartshorne

Algebraic Geometry

~Springer
Robin Hartshorne
Department of Mathematics
University of California
Berkeley, California 94720
USA

Editorial Board
S. Axler K.A. Ribet
Mathematics Department Department of Mathematics
San Francisco State Universi~y University of California at Berkeley
San Francisco, CA 94132 Berkeley, CA 94720-3840
USA USA
ribet@ math. berkeley .edu

Mathematics Subject Classification (2000): 13-xx, 14Al0, 14A15, 14Fxx, 14Hxx, 14Jxx

Library of Congress Cataloging-in-Publication Data

Hartshorne, Robin.
Algebraic geometry.
(Graduate texts in mathematics: 52)
Bibliography: p.
Includes index.
1. Geometry, Algebraic. I. Title II. Series.
QA564.H25 516'.35 77-1177

ISBN 978-1-4419-2807-8 ISBN 978-1-4757-3849-0 (eBook) Printed on acid-free paper.

DOI 10.1007/978-1-4757-3849-0

© 1977 Springer Science+Business Media, Inc.

Softcover reprint of the hardcover I st edition 1977
All rights reserved. This work may not be translated or copied in whole or in part without the
written permission of the publisher (Springer Science+Business Media, Inc., 233 Spring Street, New
York, NY 10013, USA), except for brief excerpts in connection with reviews or scholarly analysis.
Use in connection with any form of information storage and retrieval, electronic adaptation, com-
puter software, or by similar or dissimilar methodology now known or hereafter developed is for-
bidden.
The use in this publication of trade names, trademarks, service marks, and similar terms, even if
they are not identified as such, is not to be taken as an expression of opinion as to whether or not
they are subject to proprietary rights.

(ASC/SBA)

15 14

springeronline.com
For Edie, Jonathan, and Berifamin
Preface

This book provides an introduction to abstract algebraic geometry using

the methods of schemes and cohomology. The main objects of study are
algebraic varieties in an affine or projective space over an algebraically
closed field; these are introduced in Chapter I, to establish a number of
basic concepts and examples. Then the methods of schemes and
cohomology are developed in Chapters II and III, with emphasis on appli-
cations rather than excessive generality. The last two chapters of the book
(IV and V) use these methods to study topics in the classical theory of
algebraic curves and surfaces.
The prerequisites for this approach to algebraic geometry are results
from commutative algebra, which are stated as needed, and some elemen-
tary topology. No complex analysis or differential geometry is necessary.
There are more than four hundred exercises throughout the book, offering
specific examples as well as more specialized topics not treated in the
main text. Three appendices present brief accounts of some areas of
current research.
This book can be used as a textbook for an introductory course in
algebraic geometry, following a basic graduate course in algebra. I re-
cently taught this material in a five-quarter sequence at Berkeley, with
roughly one chapter per quarter. Or one can use Chapter I alone for a
short course. A third possibility worth considering is to study Chapter I,
and then proceed directly to Chapter IV, picking up only a few definitions
from Chapters II and Ill, and assuming the statement of the Riemann-
Roch theorem for curves. This leads to interesting material quickly, and
may provide better motivation for tackling Chapters II and III later.
The material covered in this book should provide adequate preparation
for reading more advanced works such as Grothendieck [EGA], [SGA],
Hartshorne [5], Mumford [2], [5], or Shafarevich [1].

Vll
Preface

Acknowledgements
In writing this book, I have attempted to present what is essential for a
basic course in algebraic geometry. I wanted to make accessible to the
nonspecialist an area of mathematics whose results up to now have been
widely scattered, and linked only by unpublished "folklore." While I
have reorganized the material and rewritten proofs, the book is mostly a
synthesis of what I have learned from my teachers, my colleagues, and
my students. They have helped in ways too numerous to recount. I owe
especial thanks to Oscar Zariski, J.-P. Serre, David Mumford, and Arthur
Ogus for their support and encouragement.
Aside from the "classical" material, whose origins need a historian to
trace, my greatest intellectual debt is to A. Grothendieck, whose treatise
[EGA] is the authoritative reference for schemes and cohomology. His
results appear without specific attribution throughout Chapters II and III.
Otherwise I have tried to acknowledge sources whenever I was aware of
them.
In the course of writing this book, I have circulated preliminary ver-
sions of the manuscript to many people, and have received valuable
comments from them. To all of these people my thanks, and in particular
to J.-P. Serre, H. Matsumura, and Joe Lipman for their careful reading
and detailed suggestions.
I have taught courses at Harvard and Berkeley based on this material,
and I thank my students for their attention and their stimulating questions.
I thank Richard Bassein, who combined his talents as mathematician
and artist to produce the illustrations for this book.
A few words cannot adequately express the thanks I owe to my wife,
Edie Churchill Hartshorne. While I was engrossed in writing, she created
a warm home for me and our sons Jonathan and Benjamin, and through
her constant support and friendship provided an enriched human context
for my life.
For financial support during the preparation of this book, I thank the
Research Institute for Mathematical Sciences of Kyoto University, the
National Science Foundation, and the University of California at
Berkeley.

August 29, 1977

Berkeley, California ROBIN HARTSHORNE

viii
Contents

Introduction Xlll

CHAPTER I
Varieties 1
I Affine Varieties
2 Projective Varieties 8
3 Morphisms 14
4 Rational Maps 24
5 Nonsingular Varieties 31
6 Nonsingular Curves 39
7 Intersections in Projective Space 47
8 ·what Is Algebraic Geometry? 55

CHAPTER II
Schemes 60
Sheaves 60
2 Schemes 69
3 First Properties of Schemes 82
4 Separated and Proper Morphisms 95
5 Sheaves of Modules 108
6 Divisors 129
7 Projective Morphisms 149
8 Differentials 172
9 Formal Schemes 190

CHAPTER III
Cohomology 201
Derived Functors 202
2 Cohomology of Sheaves 206
3 Cohomology of a Noetherian Affine Scheme 213

ix
Contents

4 Cech Cohomology 218

5 The Cohomology of Projective Space 225
6 Ext Groups and Sheaves 233
7 The Serre Duality Theorem 239
8 Higher Direct Images of Sheaves 250
9 Flat Morphisms 253
10 Smooth Morphisms 268
11 The Theorem on Formal Functions 276
12 The Semicontinuity Theorem 281

CHAPTER IV
Curves 293
Riemann-Roch Theorem 294
2 Hurwitz's Theorem 299
3 Embeddings in Projective Space 307
4 Elliptic Curves 316
5 The Canonical Embedding 340
6 Classification of Curves in P" 349

CHAPTER V
Surfaces 356
1 Geometry on a Surface 357
2 Ruled Surfaces 369
3 Monoidal Transformations 386
4 The Cubic Surface in P" 395
5 Birational Transformations 409
6 Classification of Surfaces 421

APPENDIX A
Intersection Theory 424
I Intersection Theory 425
2 Properties of the Chow Ring 428
3 Chern Classes 429
4 The Riemann-Roch Theorem 431
5 Complements and Generalizations 434

APPENDIX B
Transcendental Methods 438
1 The Associated Complex Analytic Space 438
2 Comparison of the Algebraic and Analytic Categories 440
3 When is a Compact Complex Manifold Algebraic? 441
4 Kahler Manifolds 445
5 The Exponential Sequence 446

X
 Contents

APPENDIX C
The Weil Conjectures 449
I The Zeta Function and the Weil Conjectures 449
2 History of Work on the Weil Conjectures 451
3 The /-adic Cohomology 453
4 Cohomological Interpretation of the Weil Conjectures 454

Bibliography 459

Results from Algebra 470

Glossary of Notations 472

Index 478

XI
Introduction

The author of an introductory book on algebraic geometry has the difficult

task of providing geometrical insight and examples, while at the same
time developing the modem technical language of the subject. For in
algebraic geometry, a great gap appears to separate the intuitive ideas
which form the point of departure from the technical methods used in
current research.
The first question is that of language. Algebraic geometry has
developed in waves, each with its own language and point of view. The
late nineteenth century saw the function-theoretic approach of Riemann,
the more geometric approach of Brill and Noether, and the purely alge-
braic approach of Kronecker, Dedekind, and Weber. The Italian school
followed with Castelnuovo, Enriques, and Severi, culminating in the clas-
sification of algebraic surfaces. Then came the twentieth-century "'Ameri-
can" school of Chow, Wei!, and Zariski, which gave firm algebraic foun-
dations to the Italian intuition. Most recently, Serre and Grothendieck
initiated the French school, which has rewritten the foundations of alge-
braic geometry in terms of schemes and cohomology, and which has an
impressive record of solving old problems with new techniques. Each of
these schools has introduced new concepts and methods. In writing an
introductory book, is it better to use the older language which is closer to
the geometric intuition, or to start at once with the technical language of
current research?
The second question is a conceptual one. Modern mathematics tends to
obliterate history: each new school rewrites the foundations of its subject
in its own language, which makes for fine logic but poor pedagogy. Of
what use is it to know the definition of a scheme if one does not realize
that a ring of integers in an algebraic number field, an algebraic curve, and
a compact Riemann surface are all examples of a ''regular scheme of

xiii
Introduction

dimension one"? How then can the author of an introductory book indi-
cate the inputs to algebraic geometry coming from number theory, com-
mutative algebra, and complex analysis, and also introduce the reader to
the main objects of study, which are algebraic varieties in affine or pro-
jective space, while at the same time developing the modem language of
schemes and cohomology? What choice of topics will convey the meaning
of algebraic geometry, and still serve as a firm foundation for further study
and research?

My own bias is somewhat on the side of classical geometry. I believe

that the most important problems in algebraic geometry are those arising
from old-fashioned varieties in affine or projective spaces. They provide
the geometric intuition which motivates all further developments. In this
book, I begin with a chapter on varieties, to establish many examples and
basic ideas in their simplest form, uncluttered with technical details. Only
after that do I develop systematically the language of schemes, coherent
sheaves, and cohomology, in Chapters II and III. These chapters form the
technical heart of the book. In them I attempt to set forth the most
important results, but without striving for the utmost generality. Thus, for
example, the cohomology theory is developed only for quasi-coherent
sheaves on noetherian schemes, since this is simpler and sufficient for
most applications; the theorem of "coherence of direct image sheaves" is
proved only for projective morphisms, and not for arbitrary proper
morphisms. For the same reasons I do not include the more abstract
notions of representable functors, algebraic spaces, etale cohomology'
sites, and topoi.
The fourth and fifth chapters treat classical material, namely nonsingu-
lar projective curves and surfaces, but they use techniques of schemes
and cohomology. I hope these applications will justify the effort needed to
absorb all the technical apparatus in the two previous chapters.
As the basic language and logical foundation of algebraic geometry, I
have chosen to use commutative algebra. It has the advantage of being
precise. Also, by working over a base field of arbitrary characteristic,
which is necessary in any case for applications to number theory, one
gains new insight into the classical case of base field C. Some years ago,
when Zariski began to prepare a volume on algebraic geometry, he had to
clevelop the necessary algebra as he went. The task grew to such pro-
portions that he produced a book on commutative algebra only. Now we
are fortunate in having a number of excellent books on commutative
algebra: Atiyah-Macdonald [1], Bourbaki [1], Matsumura [2], Nagata [7],
and Zariski-Samuel [1]. My policy is to quote purely algebraic results as
needed, with references to the literature for proof. A list of the results
used appears at the end of the book.
Originally I had planned a whole series of appendices-short expos-
itory accounts of some current research topics, to form a bridge between
the main text of this book and the research literature. Because of limited

xiv
 Introduction

time and space only three survive. I can only express my regret at not
including the others, and refer the reader instead to the Arcata volume
(Hartshorne, ed. [1]) for a series of articles by experts in their fields,
intended for the nonspecialist. Also, for the historical development of
algebraic geometry let me refer to Dieudonne [1]. Since there was not
space to explore the relation of algebraic geometry to neighboring fields as
much as I would have liked, let me refer to the survey article of Cassels [1]
for connections with number theory, and to Shafarevich [2, Part III] for
connections with complex manifolds and topology.
Because I believe strongly in active learning, there are a great many
exercises in this book. Some contain important results not treated in the
main text. Others contain specific examples to illustrate general
phenomena. I believe that the study of particular examples is inseparable
from the development of general theories. The serious student should
attempt as many as possible of these exercises, but should not expect to
solve them immediately. Many will require a real creative effort to under-
stand. An asterisk denotes a more difficult exercise. Two asterisks denote
an unsolved problem.
See (1, §8) for a further introduction to algebraic geometry and this
book.
Terminology
For the most part, the terminology of this book agrees with generally
accepted usage, but there are a few exceptions worth noting. A variety is
always irreducible and is always over an algebraically closed field. In
Chapter I all varieties are quasi-projective. In (Ch. II, §4) the definition is
expanded to include abstract varieties, which are integral separated
schemes of finite type over an algebraically closed field. The words curve,
surface, and 3-fold are used to mean varieties of dimension 1, 2, and 3
respectively. But in Chapter IV, the word curve is used only for a nonsin-
gular projective curve; whereas in Chapter V a curve is any effective
divisor on a nonsingular projective surface. A surface in Chapter V is
always a nonsingular projective surface.
A scheme is what used to be called a prescheme in the first edition of
[EGA], but is called scheme in the new edition of [EGA, Ch. I].
The definitions of a projective morphism and a very ample invertible sheaf
in this book are not equivalent to those in [EGA]-see (II, §4, 5). They are
technically simpler, but have the disadvantage of not being local on the
base.
The word nonsingular applies only to varieties; for more general
schemes, the words regular and smooth are used.

Results from algebra

I assume the reader is familiar with basic results about rin~s. ideals,
modules, noetherian rings, and integral dependence, and is willing to ac-
cept or look up other results, belonging properly to commutative algebra

XV
Introduction

or homological algebra, which will be stated as needed, with references to

the literature. These results will be marked with an A: e.g., Theorem
3.9A, to distinguish them from results proved in the text.
The basic conventions are these: All rings are commutative with iden-
tity element I. All homomorphisms of rings take 1 to 1. In an integral
domain or a field, 0 -=I 1. A prime ideal (respectively, maximal ideal) is an
ideal p in a ring A such that the quotient ring A/p is an integral domain
(respectively, a field). Thus the ring itself is not cc'1sidered to be a prime
ideal or a maximal ideal.
A multiplicative system in a ring A is a subsetS, containing I, and closed
under multiplication. The localizationS ~ 1A is defined to be the ring formed
by equivalence classes of fractions a/s, a EA, s E S, wherea/s and a 'Is' are
said to be equivalent if there is an s" E S such that s"(s 'a -sa') = 0 (see
e.g. Atiyah-Macdonald [I, Ch. 3]). Two special cases which are used
constantly are the following. If p is a prime ideal in A, then S = A - p is a
multiplicative system, and the corresponding localization is denoted by
A,. Iff is an element of A, then S = {I} U {f" In~ I} is a multiplicative
system, and the corresponding localization is denoted by A 1 • (Note for
example that ifjis nilpotent, thenA 1 is the zero ring.)
References
Bibliographical references are given by author, with a number in square
brackets to indicate which work, e.g. Serre, [3, p. 75]. Cross references to
theorems, propositions, lemmas within the same chapter are given by
number in parentheses. e.g. (3.5). Reference to an exercise is given by
(Ex. 3.5). References to results in another chapter are preceded by the
chapter number, e.g. (II, 3.5), or (II, Ex. 3.5).

XVI
CHAPTER I

Varieties

Our purpose in this chapter is to give an introduction to algebraic geometry

with as little machinery as possible. We work over a fixed algebraically
closed field k. We define the main objects of study, which are algebraic
varieties in affine or projective space. We introduce some of the most
important concepts, such as dimension, regular functions, rational maps,
nonsingular varieties, and the degree of a projective variety. And most im-
portant, we give lots of specific examples, in the form of exercises at the end
of each section. The examples have been selected to illustrate many inter-
esting and important phenomena, beyond those mentioned in the text. The
person who studies these examples carefully will not only have a good under-
standing of the basic concepts of algebraic geometry, but he will also have
the background to appreciate some of the more abstract developments of
modern algebraic geometry, and he will have a resource against which to
check his intuition. We will continually refer back to this library of examples
in the rest of the book.
The last section of this chapter is a kind of second introduction to the book.
It contains a discussion of the "classification problem," which has motivated
much of the development of algebraic geometry. It also contains a discussion
of the degree of generality in which one should develop the foundations of
algebraic geometry, and as such provides motivation for the theory of
schemes.

1 Affine Varieties

Let k be a fixed algebraically closed field. We define affine n-space over k,

denoted Ai: or simply An, to be the set of all n-tuples of elements of k. An
element P E An will be called a point, and if P = (ab . .. ,an) with ai E k, then
the ai will be called the coordinates of P.
I Varieties

Let A = k[x 1 , . . . ,xn] be the polynomial ring in n variables over k.

We will interpret the elements of A as functions from the affine n-space
to k, by defining f(P) = f(a 1 , .•• ,an), where f E A and P E An. Thus if
f E A is a polynomial, we can talk about the set of zeros of f, namely
Z(f) = {P E Anlf(P) = 0}. More generally, if T is any subset of A, we
define the zero set of T to be the common zeros of all the elements of T,
namely
Z(T) = {P E Anlf(P) = 0 for all f E T}.

Clearly if a is the ideal of A generated by T, then Z(T) = Z(a). Further-

more, since A is a noetherian ring, any ideal a has a finite set of generators
f 1, . . . ,fr. Thus Z(T) can be expressed as the common zeros of the finite
set of polynomials f1> ... ,fr.

Definition. A subset Y of An is an algebraic set if there exists a subset T ~ A

such that Y = Z(T).

Proposition 1.1. The union of two algebraic sets is an algebraic set. The
intersection of any family of algebraic sets is an algebraic set. The empty
set and the whole space are algebraic sets.

PROOF. If Y1 = Z(T 1 ) and Y2 = Z(T 2 ), then Y1 u Y2 = Z(T 1 T 2 ), where

T 1 T 2 denotes the set of all products of an element of T 1 by an element of
T 2 . Indeed, if P E Y1 u Y2 , then either P E Y1 or P E Y2 , so P is a zero of
every polynomial in T 1 T 2 . Conversely, if P E Z(T 1 T 2 ), and P ¢; Y1 say,
then there is an f E T 1 such that f(P) # 0. Now for any g E T 2 , (fg)(P) = 0
implies that g(P) = 0, so that P E Y2 .
= Z(Ta.) is any family of algebraic sets, nZ(l),=andZ(UTa.),
n If~ then ~ so
also an algebraic set. Finally, the empty set 0
~is = the whole
space An = Z(O).

Definition. We define the Zariski topology on An by taking the open subsets

to be the complements of the algebraic sets. This is a topology, because
according to the proposition, the intersection of two open sets is open,
and the union of any family of open sets is open. Furthermore, the empty
set and the whole space are both open.

Example 1.1.1. Let us consider the Zariski topology on the affine line A1 .
Every ideal in A = k[ x] is principal, so every algebraic set is the set of zeros
of a single polynomial. Since k is algebraically closed, every nonzero poly-
nomial f(x) can be written f(x) = c(x - a 1) · · · (x - an) with c,a 1 , . . . ,an E
k. Then Z(f) = {a 1, . . . ,an}· Thus the algebraic sets in A 1 are just the finite
subsets (including the empty set) and the whole space (corresponding to
f = 0). Thus the open sets are the empty set and the complements of finite
subsets. Notice in particular that this topology is not Hausdorff.

2
 1 Affine Varieties

Definition. A nonempty subset Y of a topological space X is irreducible if

it cannot be expressed as the union Y = Y1 u Y2 of two proper subsets,
each one of which is closed in Y. The empty set is not considered to be
irreducible.

Example 1.1.2. A 1 is irreducible, because its only proper closed subsets are
finite, yet it is infinite (because k is algebraically closed, hence infinite).

Example 1.1.3. Any nonempty open subset of an irreducible space is irre-

ducible and dense.

Example 1.1.4. If Y is an irreducible subset of X, then its closure Y in X is

also irreducible.

Definition. An affine algebraic variety (or simply affine variety) is an irre-

ducible closed subset of An (with the induced topology). An open subset
of an affine variety is a quasi-affine variety.

These affine and quasi-affine varieties are our first objects of study. But
before we can go further, in fact before we can even give any interesting
examples, we need to explore the relationship between subsets of A" and
ideals in A more deeply. So for any subset Y c:; A", let us define the ideal of
Yin A by
I(Y) = {f E Alf(P)_ = 0 for all P E Y}.
Now we have a function Z which maps subsets of A to algebraic sets, and a
function I which maps subsets of A" to ideals. Their properties are sum-
marized in the following proposition.

Proposition 1.2.
(a) If T 1 c:; T 2 are subsets of A, then Z(T 1 ) ::::2 Z(T 2 ).
(b) If ¥1 c:; ¥2 are subsets of An, then I(Yd ::::2 I(¥2 ).
(c) For any two subsets ¥1 , ¥2 of A", we have I(¥1 u Y2 ) = I(Y1 ) n I(¥2 ).
(d) For any ideal a c:; A, I(Z(a)) = JO., the radical of a.
(e) For any subset Y c:; A", Z(J(Y)) = Y, the closure of Y.

PROOF. (a), (b) and (c) are obvious. (d) is a direct consequence of Hilbert's
Nullstellensatz, stated below, since the radical of a is defined as

JO. = {f E Alf' E a for some r > 0}.

To prove (e), we note that Y c:; Z(J(Y) ), which is a closed set, so clearly
Y c:; Z(I(Y) ). On the other hand, let W be any closed set containing Y.
Then W = Z(a) for some ideal a. So Z(a) ::::2 Y, and by (b), IZ(a) c:; I(Y).
But certainly a c:; IZ(a), so by (a) we have W = Z(a) ::::2 ZI(Y). Thus
ZI(Y) = Y.

3
I Vanetles

Theorem 1.3A (Hilbert's Nullstellensatz). Let k be an algebraically closed

field, let a be an ideal in A = k[ x b . . . ,x,], and let f E A be a polynomial
which vanishes at all points of Z(a). Then rEa for some integer r > 0.
PROOF. Llng [2, p. 256] or Atiyah-Macdonald [1, p. 85] or Zariski-Samuel
[1. vol. 2, p. 164].

Corollary 1.4. There is a one-to-one inclusion-reversing correspondence

between algebraic sets in A" and radical ideals (i.e., ideals which are equal
to their own radical) in A, given by Y f-> /(Y) and a f-> Z(a). Furthermore,
an algebraic set is irreducible if and only if its ideal is a prime ideal.
PROOF. Only the last part is new. If Y is irreducible, we show that J(Y) is
prime. Indeed, if fg E l(Y), then Y s Z(fg) = Z(f) u Z(g). Thus Y =
( Y n Z(.j')) u ( Y n Z(g) ), both being closed subsets of Y. Since Y is irre-
ducible, we have either Y = Y n Z(f), in which case Y s Z(.f), or Y s
Z(y). Hence either f E l(Y) or g E l(Y).
Conversely, let p be a prime ideal, and suppose that Z(p) = Y1 u Y2 .
Then p = /(Y1 ) n /(Y2 ), so either p = l(YJ or p = /(Y2 ). Thus Z(p) = Y1
or Y2 , hence it is irreducible.

Example 1.4.1. A" is irreducible, since it corresponds to the zero ideal in A,

which is prime.

Example 1.4.2. Let f be an irreducible polynomial in A = k[x,y]. Then f

generates a prime ideal in A, since A is a unique factorization domain, so
the zero set Y = Z(f) is irreducible. We call it the affine curve defined by
the equationf(x,y) = 0. Iff has degree d, we say that Y is a curve of degree d.

Example 1.4.3. More generally, iff is an irreducible polynomial in A =

k[x 1 , . . . ,x,], we obtain an affine variety Y = Z(f), which is called a surface
if n = 3, or a hyperswface if n > 3.

Example 1.4.4. A maximal ideal m of A = k[ x 1 , . . . ,x,] corresponds to

a minimal irreducible closed subset of A", which must be a point, say
P = (ab ... ,a,). This shows that every maximal ideal of A is of the form
m = (x 1 - a 1 , . . . ,x, - an), for some a 1 , . . . ,a, E k.

Example 1.4.5. If k is not algebraically closed, these results do not hold. For
example, if k = R, the curve x 2 + y 2 + 1 = 0 in Ai has no points. So (1.2d)
is false. See also (Ex. 1.12).

Definition. If Y s A" is an affine algebraic set, we define the affine coordinate

riny A(Y) of Y, to be A/l(Y).

Remark 1.4.6. If Y is an affine variety, then A(Y) is an integral domain.

Furthermore, A( Y) is a finitely generated k-algebra. Conversely, any

4
 1 Affine Varieties

finitely generated k-algebra B which is a domain is the affine coordinate

ring of some affine variety. Indeed, write Bas the quotient of a polynomial
ring A = k[x 1 , . . . ,xn] by an ideal a, and let Y = Z(a).

Next we will study the topology of our varieties. To do so we introduce

an important class of topological spaces which includes all varieties.

Definition. A topological space X is called noetherian if it satisfies the de-

scending chain condition for closed subsets: for any sequence Y1 ::::::> Y2 ::::::> •••
of closed subsets, there is an integer r such that Y,. = Y,.+ 1 = ...

Example 1.4.7. An is a noetherian topological space. Indeed, if Y1 ::::::> Y2 ::::::> •••

is a descending chain of closed subsets, then I(Y1 ) s;; I(Y2 ) s;; ... is an as-
cending chain of ideals in A = k[x 1 , . . . ,xnJ. Since A is a noetherian ring,
this chain of ideals is eventually stationary. But for each i, Y; = Z(J( Y;) ),
so the chain Y; is also stationary.

Proposition 1.5. In a noetherian topological space X, every nonempty closed

subset Y can be expressed as a finite union Y = Y1 u ... u Y,. of irreducible
closed subsets Y;. If we require that Y; ~ lj for i # j, then the Y; are
uniquely determined. They are called the irreducible components of Y.
PROOF. First we show the existence of such a representation of Y. Let 6
be the set of nonempty closed subsets of X which cannot be written as a
finite union of irreducible closed subsets. If 6 is nonempty, then since X
is noetherian, it must contain a minimal element, say Y. Then Y is not
irreducible, by construction of 6. Thus we can write Y = Y' u Y", where
Y' and Y" are proper closed subsets of Y. By minimality of Y, each of Y'
and Y" can be expressed as a finite union of closed irreducible subsets, hence
Y also, which is a contradiction. We conclude that every closed set Y can
be written as a union Y = Y1 u ... u Y,. of irreducible subsets. By throwing
away a few if necessary, we may assume Y; ~ 1j for i # j.
Now suppose Y = Y~ u ... u Y~ is another such representation. Then
Y~ s;; Y = Y1 u ... u Y,., so Y~ = U(Y~ n Y;). But Y~ is irreducible, so
Y~ s;; Y; for some i, say i = 1. Similarly, Y1 s;; Yj for some j. Then Y~ s;; Yj,
so j = 1, and we find that Y1 = Y~. Now let Z = (Y - ¥1 )-. Then Z =
Y2 u ... u Y,. and also Z = Y2 u ... u Y~. So proceeding by induction on
r, we obtain the uniqueness of the r;.

Corollary 1.6. Every algebraic set in An can be expressed uniquely as a union

of varieties, no one containing another.

Definition. If X is a topological space, we define the dimension of X (denoted

dim X) to be the supremum of all integers n such that there exists a chain
Z 0 c Z 1 c ... c Zn of distinct irreducible closed subsets of X. We
define the dimension of an affine or quasi-affine variety to be its dimen-
sion as a topological space.

5
I Varieties

Example 1.6.1. The dimension of A 1 is 1. Indeed, the only irreducible closed

subsets of A 1 are the whole space and single points.

Definition. In a ring A, the height of a prime ideal p is the supremum of all

integers n such that there exists a chain p 0 c p 1 c ... c Pn = p of
distinct prime ideals. We define the dimension (or Krull dimension) of A
to be the supremum of the heights of all prime ideals.

Proposition 1.7. If Y is an affine algebraic set, then the dimension of Y is

equal to the dimension of its affine coordinate ring A( Y).
PROOF. If Y is an affine algebraic set in An, then the closed irreducible subsets
of Y correspond to prime ideals of A = k[x~o ... ,xn] containing I(Y).
These in turn correspond to prime ideals of A( Y). Hence dim Y is the length
of the longest chain of prime ideals in A( Y), which is its dimension.

This proposition allows us to apply results from the dimension theory of

noetherian rings to algebraic geometry.

Theorem l.SA. Let k be a field, and let B be an integral domain which is a

finitely generated k-algebra. Then:
(a) the dimension of B is equal to the transcendence degree of the
quotient field K(B) of B over k;
(b) For any prime ideal pin B, we have
height p + dim B/p = dim B.
PROOF. Matsumura [2, Ch. 5, §14] or, in the case k is algebraically closed,
Atiyah-Macdonald [ 1, Ch. 11 J

Proposition 1.9. The dimension of An is n.

PROOF. According to (1.7) this says that the dimension of the polynomial
ring k[ x ~o ... ,xn] is n, which follows from part (a) of the theorem.

Proposition 1.10. If Y is a quasi-affine variety, then dim Y = dim Y.

PROOF. If Zo c z1 c ... c zn is a sequence of distinct closed irreducible
subsets of Y, then Z 0 c Z 1 c ... c Zn is a sequence of distinct closed
irreducible subsets of Y (1.1.4), so we have dim Y ~ dim Y. In particular,
dim Y is finite, so we can choose a maximal such chain Z 0 c ... c Zn,
with n = dim Y. In that case Z 0 must be a point P, and the chain P =
Z 0 c ... c Zn will also be maximal (1.1.3). Now P corresponds to a
maximal ideal m of the affine coordinate ring A( f) of Y. The Z; correspond
to prime ideals contained in m, so height m = n. On the other hand, since
Pis a point in affine space, A(f)/m ~ k (1.4.4). Hence by (1.8Ab) we find
that n = dim A( f) = dim Y. Thus dim Y = dim Y.

6
 1 Affine Varieties

Theorem l.llA (Krull's Hauptidealsatz). Let A be a noetherian ring, and let

f E A be an element which is neither a zero divisor nor a unit. Then every
minimal prime ideal p containing f has height 1.

PROOF. Atiyah-Macdonald [1, p. 122].

Proposition 1.12A. A noetherian integral domain A is a unique factorization

domain if and only if every prime ideal of height 1 is principal.
PROOF. Matsumura [2, p. 141], or Bourbaki [1, Ch. 7, §3].

Proposition 1.13. A variety Yin A" has dimension n - 1 if and only if it is

the zero set Z(f) of a single nonconstant irreducible polynomial in A =
k[ x 1 , . . . ,xnJ.
PROOF. Iff is an irreducible polynomial, we have already seen that Z(f) is
a variety. Its ideal is the prime ideal p = (f). By (1.11A), p has height 1,
so by (1.8A), Z(f) has dimension n - 1. Conversely, a variety of dimension
n - 1 corresponds to a prime ideal of height 1. Now the polynomial ring A
is a unique factorization domain, so by (1.12A), p is principal, necessarily
generated by an irreducible polynomial f Hence Y = Z(f).

Remark 1.13.1. A prime ideal of height 2 in a polynomial ring cannot

necessarily be generated by two elements (Ex. 1.11 ).

EXERCISES

1.1. (a) Let Y be the plane curve y = x 2 (i.e., Y is the zero set of the polynomial f =
y - x 2 ). Show that A(Y) is isomorphic to a polynomial ring in one variable
over k.
(b) Let Z be the plane curve xy = 1. Show that A(Z) is not isomorphic to a poly-
nomial ring in one variable over k.
*(c) Let f be any irreducible quadratic polynomial in k[ x,y], and let W be the
conic defined by f Show that A(W) is isomorphic to A(Y) or A(Z). Which one
is it when?
1.2. The Twisted Cubic Curve. Let Y <::::: A 3 be the set Y = { (t,t 2 ,t 3 Jlt E k}. Show that Y
is an affine variety of dimension 1. Find generators for the ideal J(Y). Show that
A(Y) is isomorphic to a polynomial ring in one variable over k. We say that Y
is given by the parametric representation x = t, y = t 2 , z = t 3 •
1.3. Let Y be the algebraic set in A3 defined by the two polynomials x 2 - yz and
xz - x. Show that Y is a union of three irreducible components. Describe them
and find their prime ideals.
1.4. If we identify A2 with A 1 x A1 in the natural way, show that the Zariski topology
on A2 is not the product topology ofthe Zariski topologies on the two copies of A 1 .

7
I Varieties

1.5. Show that a k-algebra B is isomorphic to the affine coordinate ring of some alge-
braic set in A", for some n, if and only if B is a finitely generated k-algebra with no
nilpotent elements.
1.6. Any nonempty open subset of an irreducible topological space is dense and
irreducible. If Y is a subset of a topological space X, which is irreducible in its
induced topology, then the closure Y is also irreducible.
1.7. (a) Show that the following conditions are equivalent for a topological space X:
(i) X is noetherian; (ii) every nonempty family of closed subsets has a minimal
element; (iii) X satisfies the ascending chain condition for open subsets;
(iv) every nonempty family of open subsets has a maximal element.
(b) A noetherian topological space is quasi-compact, i.e., every open cover has a
finite subcover.
(c) Any subset of a noetherian topological space is noetherian in its induced
topology.
(d) A noetherian space which is also Hausdorff must be a finite set with the discrete
topology.
1.8. Let Y be an affine variety of dimension r in A". Let H be a hypersurface in A",
and assume that Y <;/;. H. Then every irreducible component of Y n H has
dimension r - 1. (See (7.1) for a generalization.)
1.9. Let a£; A = k[x1o ... ,xnJ be an ideal which can be generated by r elements.
Then every irreducible component of Z(a) has dimension ;::, n - r.
1.10. (a) IfYisanysubsetofatopologicalsp aceX,thendim Y ~dim X.
(b) If X is a topological space which is covered by a family of open subsets {U;},
then dim X = sup dim U;.
(c) Give an example of a topological space X and a dense open subset U with
dim U <dim X.
(d) If Y is a closed subset of an irreducible finite-dimensional topological space X,
and if dim Y = dim X, then Y = X.
(e) Give an example of a noetherian topological space of infinite dimension.
*1.11. Let Y £; A 3 be the curve given parametrically by x = t 3 , y = t 4 , z = t 5 . Show
that J(Y) is a prime ideal of height 2 in k[x,y,z] which cannot be generated by
2 elements. We say Y is not a local complete intersection-d. (Ex. 2.17).
1.12. Give an example of an irreducible polynomial fER[ x, y], whose zero set
Z(f) in A~ is not irreducible (cf. 1.4.2).

2 Projective Varieties

To define projective varieties, we proceed in a manner analogous to the

definition of affine varieties, except that we work in projective space.
Let k be our fixed algebraically closed field. We defined projective n-space
over k, denoted Pi:, or simply pn, to be the set of equivalence classes of
(n + 1)-tuples (a 0 , .. . ,an) of elements of k, not all zero, under the equiva-
lence relation given by (a 0 , . . . ,an) ~ (.lca 0 , . . . ,.lean) for all A E k, A =f. 0.
Another way of saying this is that pn as a set is the quotient of the set
8
 2 Projective Varieties

An+ 1 - {(0, ... ,0)} under the equivalence relation which identifies points
lying on the same line through the origin.
An element of pn is called a_point. If P is a point, then any (n + 1)-
tuple (a 0 , . . . ,an) in the equivalence class P is called a set of homogeneous
coordinates for P.
LetS be the polynomial ring k[x 0 , . . . ,xnJ. We want to regardS as a
graded ring, so we recall briefly the notion of a graded ring.
A graded ring is a ring S, together with a decomposition S = EBd;.o Sd
of S into a direct sum of abelian groups Sd, such that for any d,e ): 0,
Sd · Se c;:: Sd+e· An element of Sd is called a homogeneous element of degree
d. Thus any element of S can be written uniquely as a (finite) sum of
homogeneous elements. An ideal a c;:: S is a homogeneous ideal if a =
EBd;.o (an Sd). We will need a few basic facts about homogeneous ideals
(see, for example, Matsumura [2, §10] or Zariski-Samuel [1, vol. 2, Ch. VII,
§2]). An ideal is homogeneous if and only if it can be generated by homo-
geneous elements. The sum, product, intersection, and radical of homo-
geneous ideals are homogeneous. To test whether a homogeneous ideal is
prime, it is sufficient to show for any two homogeneous elements f,g, that
fg E a implies f E a or g E a.
We make the polynomial ring S = k[x 0 , . . . ,xn] into a graded ring by
taking Sd to be the set of all linear combinations of monomials of total
weight d in x 0 , . . . ,xn- Iff E S is a polynomial, we cannot use it to define
a function on pn, because of the nonuniqueness of the homogeneous co-
ordinates. However, if f is a homogeneous polynomial of degree d, then
f(Aa 0 , . .. ,Aan) = Adf(a 0 , . .. ,an), so that the property off being zero or
not depends only on the equivalence class of (a 0 , . . . ,an). Thus f gives a
function from pn to {0,1} by f(P) = 0 if f(a 0 , . .. ,an) = 0, and f(P) = 1
if f(a 0 , . . . ,an) =f. 0.
Thus we can talk about the zeros of a homogeneous polynomial, namely
Z(f) = {P E pnlf(P) = 0}. If Tis any set of homogeneous elements of S,
we define the zero set of T to be
Z(T) = {P E pnlf(P) = 0 for all f E T}.
If a is a homogeneous ideal of S, we define Z(a) = Z(T), where Tis the set
of all homogeneous elements in a. Since S is a noetherian ring, any set of
homogeneous elements T has a finite subset f 1 , . •• ,f,. such that Z(T) =
Z(f1 , . . . ,f,.).

Definition. A subset Y of pn is an algebraic set if there exists a set T of ho-

mogeneous elements of S such that Y = Z(T).

Proposition 2.1. The union of two algebraic sets is an algebraic set. The
intersection of any family of algebraic sets is an algebraic set. The empty
set and the whole space are algebraic sets.
PROOF. Left to reader (it is similar to the proof of (1.1) above).
9
1 Varieties

Definition. We define the Zariski topology on pn by taking the open sets

to be the complements of algebraic sets.

Once we have a topological space, the notions of irreducible subset and

the dimension of a subset, which were defined in §1, will apply.

Definition. A projective algebraic variety (or simply projective variety) is an

irreducible algebraic set in pn, with the induced topology. An open
subset of a projective variety is a quasi-projective variety. The dimension
of a projective or quasi-projective variety is its dimension as a topo-
logical space.
If Y is any subset of pn, we define the homogeneous ideal of Y in S,
denoted J(Y), to be the ideal generated by {f E Slf is homogeneous and
f(P) = 0 for all P E Y}. If Y is an algebraic set, we define the homo-
geneous coordinate ring of Y to be S(Y) = Sjl(Y). We refer to (Ex. 2.1 ~
2.7) below for various properties of algebraic sets in projective space
and their homogeneous ideals.

Our next objective is to show that projective n-space has an open covering
by affine n-spaces, and hence that every projective (respectively, quasi-
projective) variety has an open covering by affine (respectively, quasi-affine)
varieties. First we introduce some notation.
If f E S is a linear homogeneous polynomial, then the zero set of f is
called a hyperplane. In particular we denote the zero set of X; by H;, for
i ;= 0, ... ,n. Let U; be the open set pn- H;. Then pn is covered by the
open sets U;, because if P = (a 0 , • •• ,an) is a point, then at least one a; -:f= 0,
hence PE U;. We define a mapping <p;: U;----*An as follows: if P=(a 0 , . . . ,an) E
U;, then <p;(P) = Q, where Q is the point with affine coordinates

with a;/a; omitted. Note that <p; is well-defined since the ratios a)a; are
independent of the choice of homogeneous coordinates.

Proposition 2.2. The map <p; is a homeomorphism of U; with its induced

topology to An with its Zariski topology.

PROOF. <p; is clearly bijective, so it will be sufficient to show that the closed
sets of U; are identified with the closed sets of An by <fJ;· We may assume
i = 0, and we write simply U for U 0 and <p: U----* An for <p 0 .
Let A = k[Yl, . .. ,ynJ. We define a map a from the set Sh of homo-
geneous elements of S to A, and a map f3 from A to Sh. Given f E S\ we
set a(f) = f(l,Yl, . .. ,yn). On the other hand, given g E A of degree e, then

10
 2 Projective Varieties

•.• ,x"jx 0 ) is a homogeneous polynomial of degree e in the x;,

x 0g(xdx 0 ,
which we call f3(g).
Now let Y <;; U be a closed subset. Let Y be its closure in P". This is
an algebraic set, so Y = Z(T) for some subset T <;; Sh. Let T' = cx(T).
Then straightforward checking shows that cp(Y) = Z(T'). Conversely, let
W be a closed subset of A". Then W = Z(T') for some subset T' of A, and
one checks easily that cp - 1(W) = Z(f3(T')) n U. Thus cp and cp - l are both
closed maps, so cp is a homeomorphism.
Corollary 2.3. If Y is a projective (respectively, quasi-projective) variety, then
Y is covered by the open sets Y n U;, i = 0, ... ,n, which are homeomorphic
to affine (respectively, quasi-affine) varieties via the mapping qJ; defined
above.

EXERCISES

2.1. Prove the "homogeneous Nullstellensatz," which says if a <;; S is a homogeneous

ideal, and iff E Sis a homogeneous polynomial with deg f > 0, such that f(P) = 0
for all P E Z( a) in P", then fq E a for some q > 0. [Hint: Interpret the problem in
terms of the affine (n + 1)-space whose affine coordinate ring is S, and use the
usual Nullstellensatz, (1.3A).]
2.2. For a homogeneous ideal a <;; S, show that the following conditions are equi-
valent:
(i) Z(a) = 0 (the empty set);
(ii) .jO. = either s or the ideals+ = ffid>O Sd;
(iii) a ;::> Sd for some d > 0.

2.3. (a) If T 1 <;; T 2 are subsets of Sh, then Z(Td ;::> Z(T2 ).
(b) If Y1 <;; Y2 are subsets ofP", then J(Y1 ) ;::> J(Y2 ).
(c) For any two subsets Y1 ,Y2 ofP", J(Y1 u Y2 ) = J(Ytl n J(Y2 ).
(d) If a <;; Sis a homogeneous ideal with Z(a) #- 0, then J(Z(a)) = .jO..
(e) For any subset Y <;; P", Z(J(Y)) = Y.

2.4. (a) There is a 1-1 inclusion-reversing correspondence between algebraic sets in

P", and homogeneous radical ideals of S not equal to S+, given by Y H J(Y)
and a H Z( a). Nate: Since S + does not occur in this correspondence, it is
sometimes called the irrelevant maximal ideal of S.
(b) An algebraic set Y <;; P" is irreducible if and only if J(Y) is a prime ideal.
(c) Show that P" itself is irreducible.

2.5. (a) P" is a noetherian topological space.

(b) Every algebraic set in P" can be written uniquely as a finite union of irreducible
algebraic sets, no one containing another. These are called its irreducible
components.

2.6. If Y is a projective variety with homogeneous coordinate ring S(Y), show that
dim S(Y) = dim Y + 1. [Hint: Let cp;: U;--> A" be the homeomorphism of(2.2),
let Y; be the affine variety cp;(Y n U;), and let A(Y;) be its affine coordinate ring.

11
I Varieties

Show that A( Y;) can be identified with the sub ring of elements of degree 0 of the
localized ring S(YJx,· Then show that S(Y)x, ~ A(Y;)[x;,X;- 1 ]. Now use (1.7),
(1.8A), and (Ex 1.10), and look at transcendence degrees. Conclude also that
dim Y = dim Y; whenever Y; is nonempty.J
2.7. (a) dim P" = n.
(b) If Y s; P" is a quasi-projective variety, then dim Y = dim f.
[Hint: Use (Ex. 2.6) to reduce to (1.10).]

2.8. A projective variety Y s; P" has dimension n - 1 if and only if it is the zero set of
a single irreducible homogeneous polynomial f of positive degree. Y is called a
hypersurface in P".
2.9. Projective Closure of an Affine Variety. If Y s; A" is an affine variety, we identify
A" with an open set U 0 s; P" by the homeomorphism <p 0 . Then we can speak of
Y, the closure of Yin P", which is called the projective closure of Y.
(a) Show that J(Y) is the ideal generated by f3(I(Y)), using the notation of the
proof of (2.2).
(b) Let Y s; A 3 be the twisted cubic of (Ex. 1.2). Its projective closure Y s; P 3
is called the twisted cubic curve in P 3 . Find generators for J(Y) and J(Y), and
use this example to show that if f 1 , ..• ,f.. generate J(Y), then f3(fd, ... ,{3(!,.)
do not necessarily generate J(Y).
2.10. The Cone Over a Projective Variety (Fig. 1). Let Y s; P" be a nonempty algebraic
set, and let e:A"+ 1 - {(0, ... ,0)}--+ P" be the map which sends the point with
affine coordinates (a 0 , .•• ,an) to the point with homogeneous coordinates
(a 0 , . .. ,anl· We define the affine cone over Y to be

C(Y) = e- 1 (Y) u {(0, ... ,0)}.

(a) Show that C(Y) is an algebraic set in A"+l, whose ideal is equal to J(Y),
considered as an ordinary ideal in k[ x 0 , . . . ,xnJ.
(b) C(Y) is irreducible if and only if Y is.
(c) dim C(Y) = dim Y + 1.
Sometimes we consider the projective closure C( Y) of C( Y) in P" + 1 . This is called
the projective cone over Y.

Figure 1. The cone over a curve in P 2 .

12
 2 Projective Varieties

2.11. Linear Varieties in P". A hypersurface defined by a linear polynomial is called a

hyperplane.
(a) Show that the following two conditions are equivalent for a variety Yin P":
(i) J(Y) can be generated by linear polynomials.
(ii) Y can be written as an intersection of hyperplanes.
In this case we say that Y is a linear variety in P".
(b) If Y is a linear variety of dimension r in P", show that J(Y) is minimally gen-
erated by n - r linear polynomials.
(c) Let Y,Z be linear varieties in P", with dim Y = r, dimZ = s. Ifr + s - n )' 0,
then Y n Z =f. 0. Furthermore, if Y n Z =f. 0, then Y n Z is a linear
variety of dimension )' r + s - n. (Think of A"+ 1 as a vector space over k,
and work with its subspaces.)

2.12. The d-Uple Embedding. For given n,d > 0, let M 0 ,M 1 , . . . ,MN be all the mono-
mials of degree d in the n + 1 variables x 0 , . •. ,x", where N = ("!d) - 1. We
define a mapping Pd: P" --+ pN by sending the point P = (a 0, ... ,a") to the point
pAP) = (M 0 (a), ... ,MN(a)) obtained by substituting the a, in the monomials Mi.
This is called the d-uple embedding ofP" in pN_ For example, ifn = 1, d = 2, then
N = 2, and the image Y of the 2-uple embedding ofP 1 in P 2 is a conic.
(a) Let 8: k[y 0 , . . . ,yNJ --+ k[ x 0 , . . . ,x"J be the homomorphism defined by
sending y, to M,, and let a be the kernel of 8. Then a is a homogeneous prime
ideal, and so Z( a) is a projective variety in pN_
(b) Show that the image of Pd is exactly Z(a). (One inclusion is easy. The other will
require some calculation.)
(c) Now show that Pd is a homeomorphism ofP" onto the projective variety Z(a).
(d) Show that the twisted cubic curve in P 3 (Ex. 2.9) is equal to the 3-uple embed-
ding of P 1 in P 3 , for suitable choice of coordinates.
2.13. Let Y be the image of the 2-uple embedding of P 2 in P 5 . This is the Veronese
surface. If Z £; Y is a closed curve (a curve is a variety of dimension 1), show that
there exists a hypersurface V £; P 5 such that V n Y = Z.
2.14. The Segre Embedding. Let 1/J:P' x P'--+ pN be the map defined by sending the
ordered pair (a 0 , . .. ,a,) x (b 0 , . .. ,b,) to (... ,a,bi, . .. ) in lexicographic order,
where N = rs + r + s. Note that 1/J is well-defined and injective. It is called the
Segre embedding. Show that the image ofljJ is a subvariety ofPN. [Hint: Let the
homogeneous coordinates of pN be {ziili = 0, ... ,r, j = 0, ... ,s}, and let a be
the kernel of the homomorphism k[{zii}] --+ k[x 0 , . . . ,x,y0 , . . . ,y,] which sends
zii to x,yi. Then show that Im 1/1 = Z(a).]
2.15. The Quadric Surface in P 3 (Fig. 2). Consider the surface Q (a swface is a variety of
dimension 2) in P 3 defined by the equation xy - zw = 0.
(a) Show that Q is equal to the Segre embedding of P 1 x P 1 in P 3 , for suitable
choice of coordinates.
(b) Show that Q contains two families of lines (a line is a linear variety of dimen-
sion 1) {L,},{M,}, each parametrized by t E P 1, with the properties that if
L, =f. Lu, then L, n Lu = 0; if M, =f. Mu, M, n Mu = 0. and for all t,u,
L, n Mu = one point.
(c) Show that Q contains other curves besides these lines, and deduce that the
Zariski topology on Q is not homeomorphic via 1/J to the product topology on
P 1 x P 1 (where each P 1 has its Zariski topology).

13
I Varieties

Figure 2. The quadric surface in P 3 .

2.16. (a) The intersection of two varieties need not be a variety. For example, let Q1
and Q2 be the quadric surfaces in P 3 given by the equations x 2 - yw = 0
and x y - zw = 0, respectively. Show that Q1 n Q2 is the union of a twisted
cubic curve and a line.
(b) Even if the intersection of two varieties is a variety, the ideal of the inter-
section may not be the sum of the ideals. For example, let C be the conic in
P 2 given by the equation x 2 - yz = 0. Let L be the line given by y = 0.
Show that C n L consists of one point P, but that J( C) + J(L) f= J(P).
2.17. Complete intersections. A variety Y of dimension r in P" is a (strict) complete
intersection if J(Y) can be generated by n - r elements. Y is a set-theoretic com-
plete intersection if Y can be written as the intersection of n - r hypersurfaces.
(a) Let Y be a variety in P", let Y = Z(a); and suppose that a can be generated
by q elements. Then show that dim Y ;, n - q.
(b) Show that a strict complete intersection is a set-theoretic complete inter-
section.
*(c) The converse of (b) is false. For example let Y be the twisted cubic curve in
P 3 (Ex. 2.9). Show that J(Y) cannot be generated by two elements. On the
other hand, find hypersurfaces H l>H 2 of degrees 2,3 respectively, such that
Y = H 1 n H2 .
**(d) It is an unsolved problem whether every closed irreducible curve in P 3 is
a set-theoretic intersection of two surfaces. See Hartshorne [1 J and Hart-
shorne [5, III, §5] for commentary.

3 Morphisms

So far we have defined affine and projective varieties, but we have not dis-
cussed what mappings are allowed between them. We have not even said
when two are isomorphic. In this section we will discuss the regular func-
tions on a variety, and then define a morphism of varieties. Thus we will
have a good category in which to work.

14
 3 Morphisms

Let Y be a quasi-affine variety in An. We will consider functions f from

Yto k.

Definition. A function f: Y --+ k is regular at a point P E Y if there is an open

neighborhood UwithPE Us; Y,andpolynomialsg,hEA = k[x 1 , . . . ,xn],
such that h is nowhere zero on U, and f = gjh on U. (Here of course we
interpret the polynomials as functions on An, hence on Y.) We say that
f is regular on Y if it is regular at every point of Y.

Lemma 3.1. A regular function is continuous, when k is identified with Al

in its Zariski topology.

PROOF. It is enough to show that f- 1 of a closed set is closed. A closed set

of Al is a finite set of points, so it is sufficient to show that f- 1 (a) =
{P E Ylf(P) = a} is closed for any a E k. This can be checked locally: a
subset Z of a topological space Y is closed if and only if Y can be covered
by open subsets U such that Z n U is closed in U for each U. So let U be
an open set on which f can be represented as gjh, with g,h E A, and h no-
where 0 on U. Thenf- 1(a) n U = {P E Ulg(P)/h(P) =a}. But g(P)/h(P) =
a if and only if (g - ah)(P) = 0. So f- 1(a) n U = Z(g - ah) n U which
is closed. Hence f- 1 (a) is closed in Y.

Now let us consider a quasi-projective variety Y s; pn_

Definition. A function f: Y --+ k is regular at a point P E Y if there is an open

neighborhood U with P E U s; Y, and homogeneous polynomials
g,h E S = k[ x 0 , . . . ,xn], of the same degree, such that h is nowhere zero
on U, and f = gjh on U. (Note that in this case, even though g and h
are not functions on pn, their quotient is a well-defined function whenever
h i= 0, since they are homogeneous of the same degree.) We say that
f is regular on Y if it is regular at every point.

Remark 3.1.1. As in the quasi-affine case, a regular function is necessarily

continuous (proof left to reader). An important consequence of this is the
fact that iff and g are regular functions on a variety X, and iff = g on
some nonempty open subset U s; X, then f = g everywhere. Indeed, the
set of points where f - g = 0 is closed and dense, hence equal to X.

Now we can define the category of varieties.

Definition. Let k be a fixed algebraically closed field. A variety over k (or

simply variety) is any affine, quasi-affine, projective, or quasi-projective
variety as defined above. If X, Yare two varieties, a morphism cp: X --+ Y

15
I Varieties

is a continuous map such that for every open set V ~ Y, and for every
regular functionf: V--+ k, the function! o cp:cp- 1 (V)--+ k is regular.

Clearly the composition of two morphisms is a morphism, so we have a

category. In particular, we have the notion of isomorphism: an isomorphism
cp: X --+ Y of two varieties is a morphism which admits an inverse morphism
ljJ: Y --+ X with ljJ o cp = idx and cp o 1jJ = idy. Note that an isomorphism is
necessarily bijective and bicontinuous, but a bijective bicontinuous mor-
phism need not be an isomorphism (Ex. 3.2).

Now we introduce some rings of functions associated with any variety.

Definition. Let Y be a variety. We denote by @(Y) the ring of all regular

functions on Y. If P is a point of Y, we define the local ring of P on Y,
@P.Y (or simply @p) to be the ring of germs of regular functions on Y
near P. In other words, an element of @p is a pair <U,f) where U is an
open subset of Y containing P, and f is a regular function on U, and
where we identify two such pairs (U,f) and (V,g) iff = g on U n V.
(Use (3.1.1) to verify that this is an equivalence relation!)

Note that @P is indeed a local ring: its maximal ideal m is the set of germs
of regular functions which vanish at P. For if f(P) =1= 0, then 1/f is regular
in some neighborhood of P. The residue field @pjm is isomorphic to k.

Definition. If Yis a variety, we define the function field K(Y) of Y as follows:

an element of K(Y) is an equivalence class of pairs (U,f) where U is a
nonempty open subset of Y, f is a regular function on U, and where
we identify two pairs (U,f) and (V,g) iff= g on U n V. The elements
of K ( Y) are called rational functions on Y.

Note that K(Y) is in fact a field. Since Y is irreducible, any two non-
empty open sets have a nonempty intersection. Hence we can define addition
and multiplication in K~ Y), making it a ring. Then if <U,.f) E K( Y) with
f =I= 0, we can restrict f to the open set V = U - U n Z(f) where it never
vanishes, so that 1/f is regular on V, hence (V,1/f) is an inverse for (U,f).
Now we have defined, for any variety Y, the ring of global functions @( Y),
the local ring @P at a point of Y, and the function field K( Y). By restricting
functions we obtain natural maps @(Y)--+ @p --+ K(Y) which in fact are
injective by (3.1.1). Hence we will usually treat @(Y) and @pas subrings of
K(Y).
If we replace Y by an isomorphic variety, then the corresponding rings are
isomorphic. Thus we can say that @(Y), @p, and K(Y) are invariants of the
variety Y (and the point P) up to isomorphism.
Our next task is to relate @( Y), (l] p, and K ( Y) to the affine coordinate
ring A(Y) of an affine variety, and the homogeneous coordinate ring S(Y)

16
 3 Morphisms

of a projective variety, which were introduced earlier. We will find that for
an affine variety Y, A( Y) = (()( Y), so it is an invariant up to isomorphism.
However, for a projective variety Y, S( Y) is not an invariant: it depends on
the embedding of Y in projective space (Ex. 3.9).

Theorem 3.2. Let Y c:; An be an affine variety with affine coordinate ring
A(Y). Then:
(a) @(Y) ~ A(Y);
(b) for each point P E Y, let mp c:; A(Y) be the ideal of functions
vanishing at P. Then P 1--> mp gives a 1-1 correspondence between the
points of Y and the maximal ideals of A( Y);
(c) for each P, @p ~ A(Y)"'P' and dim @p = dim Y;
(d) K(Y) is isomorphic to the quotient field of A(Y), and hence K(Y)
is a.finitely generated extension field of k, of transcendence degree = dim Y.

PROOF. We will proceed in several steps. First we define a map a:A(Y)-->

(()( Y). Every polynomial f E A = k[ x 1, ... ,xn] defines a regular function
on An and hence on Y. Thus we have a homomorphism A --> @(Y). Its
kernel is just l(Y), so we obtain an injective homomorphism a: A( Y) --> @(Y).
From (1.4) we know there is a 1-1 correspondence between points of Y
(which are the minimal algebraic subsets of Y) and maximal ideals of A
containing J(Y). Passing to the quotient by l(Y), these correspond to the
maximal ideals of A(Y). Furthermore, using a to identify elements of A(Y)
with regular functions on Y, the maximal ideal corresponding to P is just
mp = {f E A(Y)if(P) = 0}. This proves (b).
For each P there is a natural map A( Y)mp --> @p. It is injective because a
is injective, and it is surjective by definition of a regular function! This
shows that (OP ~ A(Y)mp· Now dim (()P =height mp. Since A(Y)/mp ~ k,
we conclude from (1.7) and (l.SA) that dim @p = dim Y.
From (c) it follows that the quotient field of A(Y) is isomorphic to the
quotient field of (OP for every P, and this is equal to K(Y), because every
rational function is actually in some @p. Now A(Y) is a finitely generated
k-algebra, so K(Y) is a finitely generated field extension of k. Furthermore,
the transcendence degree of K(Y)/k is equal to dim Y by (1.7) and (l.SA).
This proves (d).
To prove (a) we note that (O(Y) c:; nPEY @p, where all our rings are re-
garded as subrings of K ( Y).
Using (b) and (c) we have

A(Y) c:; @(Y) c:; n A(Y)"''

"'
where m runs over all the maximal ideals of A(Y). The equality now follows
from the simple algebraic fact that if B is an integral domain, then B is
equal to the intersection (inside its quotient field) of its localizations at all
maximal ideals.

17
I Varieties

Propositi' , 3.3. Let Ui s P" be the open set defined by the equation xi #- 0.
Then tne mapping cpi: Ui --+ A" of (2.2) above is an isomorphism of varieties.

PROOF. We have already shown that it is a homeomorphism, so we need

only check that the regular functions are the same on any open set. On Ui
the regular functions are locally quotients of homogeneous polynomials in
x 0 , . .. ,x" of the same degree. On A" the regular functions are locally
quotients of polynomials in y 1 , . . . ,Yn· One can check easily that these two
concepts are identified by the maps ex and f3 of the proof of (2.2).

Before stating the next result, we introduce some notation. If S is a

graded ring, and p a homogeneous prime ideal in S, then we denote by S<Pl
the subring of elements of degree 0 in the localization of S with respect to
the multiplicative subset T consisting of the homogeneous elements of S
not in p. Note that T- 1 S has a natural grading given by deg(f/g) = deg f -
deg g for f homogeneous in S and g E T. S<Pl is a local ring, with maximal
ideal (p · T- 1S) n S<vJ· In particular, if Sis a·domain, then for p = (0) we
obtain a field s((O))• Similarly, iff E s is a homogeneous element, we denote
by s(f) the sub ring of elements of degree 0 in the localized ring sf•

Theorem 3.4. Let Y s P" be a projective variety with homogeneous co-

ordinate ring S( Y). Then:
(a) (O(Y) = k;
(b) for any point P E Y, let mp S S(Y) be the ideal generated by the
set of homogeneous f E S(Y) such that f(P) = 0. Then @p = S(Y)(mp);
(c) K(Y) ~ S(Y)<<OJJ·
PROOF. To begin with, let Ui s P" be the open set xi #- 0, and let Y; =
Y n Ui. Then Ui is isomorphic to A" by the isomorphism cpi of (3.3), so we
can consider Y; as an affine variety. There is a natural isomorphism cpf
of the affine coordinate ring A( Y;) with the localization S( Y)<xil of the homo-
geneous coordinate ring of Y. We first make an isomorphism of k[ y 1 , . . . , Yn]
with k[x 0 , ••• ,xnJx;) by sending f(y 1 , •.• ,yn) to f(x 0 jxi> ... ,xn/xJ, leaving
out xdxi> as in the proof of (2.2). This isomorphism sends /( }/) to /( Y)S<x.J
(cf. Ex. 2.6), so passing to the quotient, we obtain the desired isomorphism
cpf:A(Y;) ~ S(Y)<x.J·
Now to prove (b), let P E Y be any point, and choose i so that P E Y;.
Then by (3.2), @p ~ A(}/)111 " ' where m~ is the maximal ideal of A(Y;) corre-
sponding toP. One checks easily that cpf(m~) = mp · S(Y)<x.J· Now xi f! mp,
and localization is transitive, so we find that A( Y;)mJ. ~ S( Y)(mp)' which
proves (b).
To prove (c), we use (3.2) again to see that K(Y), which is equal to K(Y;),
is the quotient field of A(}/). But by cpf, this is isomorphic to S( Y)<<on·
To prove (a), Jet f E (0( Y) be a global regular function. Then for each i,
f is regular on Y;, so by (3.2), f E A( Y;). But we have just seen that A( Y;) ~
S(Y)<x,J• so we conclude that f can be written as gdxf' where gi E S(Y) is

18
 3 Morphisms

homogeneous of degree N;. Thinking of (O(Y), K(Y) and S(Y) all as sub-
rings of the quotient field L of S(Y), this means that xf'f E S(Y)N,, for each i.
Now choose N ~ L,N;. Then S(Y)N is spanned as a k-vector space by
monomials of degree N in x 0 , . . . ,xn, and in any such monomial, at least
one X; occurs to a power ~N;. Thus we have S(Y)N · f s:;; S(Y)N. Iterating,
we have S(Y)N ° r s:;; S(Y)N for all q > 0. In particular, x~r E S(Y) for
all q > 0. This shows that the sub ring S( Y)[f] of Lis contained in x 0 N S( Y),
which is a finitely generated S(Y)-module. Since S(Y) is a noetherian ring,
S(Y)[f] is a finitely generated S(Y)-module, and therefore f is integral
over S(Y) (see, e.g., Atiyah-Macdonald [1, p. 59]). This means that there
are elements ab ... ,am E S(Y) such that
fm + alfm-1 + 0 0 0 +am= 0.

Since f has degree 0, we can replace the a; by their homogeneous components

of degree 0, and still have a valid equation. But S(Y) 0 = k, so the a; E k,
and f is algebraic over k. But k is algebraically closed, so f E k, which
completes the proof.

Our next result shows that if X and Y are affine varieties, then X is iso-
morphic to Y if and only if A( X) is isomorphic to A( Y) as a k-algebra.
Actually the proof gives more, so we state the stronger result.

Proposition 3.5. Let X be any variety and let Y be an affine variety. Then
there is a natural bijective mapping of sets

o::Hom(X,Y) ~ Hom(A(Y),(O(X))

where the left Hom means morphisms of varieties, and the right Hom
means homomorphisms of k-algebras.
PROOF. Given a morphism q>: X ----> Y, q> carries regular functions on Y to
regular functions on X. Hence q> induces a map (O(Y) to (O(X), which is
clearly a homomorphism of k-algebras. But we have seen (3.2) that (O(Y) ~
A(Y), so we get a homomorphism A(Y)----> (O(X). This defines o:.
Conversely, suppose given a homomorphism h: A( Y)----> (O(X) of k-algebras.
Suppose that Y is a closed subset of A", so that A(Y) = k[x 1 , . . . ,xn]/I(Y).
Let :X; be the image of X; in A(Y), and consider the elements~; = h(xJ E (O(X).
These are global functions on X, so we can use them to define a mapping
1/J:X----> A" by 1/J(P) = (~ 1 (P), ... '~"(P)) for P EX.
We show next that the image of 1jJ is contained in Y. Since Y = Z(J(Y) ),
it is sufficient to show that for any P EX and any f E I(Y),f(lji(P)) = 0. But

Now f is a polynomial, and h is a homomorphism of k-algebras, so we have

19
I Varieties

since f E J(Y). So 1/1 defines a map from X to Y, which induces the given
homomorphism h.
To complete the proof, we must show that 1jJ is a morphism. This is a
consequence of the following lemma.

Lemma 3.6. Let X be any variety, and let Y £ An be an affine variety. A

map of sets 1/1: X --+ Y is a morphism if and only if X; o 1/J is a regular
function on X for each i, where x 1, . . . ,xn are the coordinate functions
on An.

PROOF. If 1jJ is a morphism, the X; o 1jJ must be regular functions, by definition

of a morphism. Conversely, suppose the x; o 1jJ are regular. Then for any
polynomial f = f(x 1 , . . . ,xn), f o 1/J is also regular on X. Since the closed
sets of Y are defined by the vanishing of polynomial functions, and since
regular functions are continuous, we see that 1/J- 1 takes closed sets to closed
sets, so 1/J is continuous. Finally, since regular functions on open subsets of
Y are locally quotients of polynomials, g o 1jJ is regular for any regular
function g on any open subset of Y. Hence 1jJ is a morphism.

Corollary 3.7. If X, Y are two affine varieties, then X and Y are isomorphic
if and only if A(X) and A(Y) are isomorphic ask-algebras.
PROOF. Immediate from the proposition.

In the language of categories, we can express the above result as follows:

Corollary 3.8. The functor X 1---> A( X) induces an arrow-reversing equivalence
of categories between the category of affine varieties over k and the category
of finitely generated integral domains over k.

We include here an algebraic result which will be used in the exercises.

Theorem 3.9A (Finiteness of Integral Closure). Let A be an integral domain

which is a finitely generated algebra over a field k. Let K be the quotient
field of A, and let L be a finite algebraic extension of K. Then the integral
closure A' of A in L is a finitely generated A -module, and is also a finitely
generated k-algebra.
PROOF. Zariski-Samuel [1, vol. 1, Ch. V., Thm. 9, p. 267.]

EXERCISES

3.1. (a) Show that any conic in A2 is isomorphic either to A 1 or A 1 - { 0} (cf. Ex. 1.1 ).
(b) Show that A 1 is not isomorphic to any proper open subset of itself. (This result
is generalized by (Ex. 6.7) below.)
(c) Any conic in P 2 is isomorphic to P 1 .
(d) We will see later (Ex. 4.8) that any two curves are homeomorphic. But show
now that A2 is not even homeomorphic to P 2 •

20
 3 Morphisms

(e) If an affine variety is isomorphic to a projective variety, then it consists of only

one point.
3.2. A morphism whose underlying map on the topological spaces is a homeomor-
phism need not be an isomorphism.
(a) For example, let <p:A 1 --> A 2 be defined by t f--> (t 2 ,t 3 ). Show that <p defines a
bijective bicontinuous morphism of A 1 onto the curve y 2 = x 3 , but that <p is
not an isomorphism.
(b) For another example, let the characteristic of the base field k be p > 0, and
define a map <p :A 1 --> A 1 by t f--> tP. Show that <p is bijective and bicontinuous
but not an isomorphism. This is called the Frobenius morphism.
3.3. (a) Let <p:X--> Y be a morphism. Then for each P EX, <p induces a homomor-
phism of local rings <pt:(!J<PiP).Y--> (!JP.x·
(b) Show that a morphism <p is an isomorphism if and only if <p is a homeomor-
phism, and the induced map <pp on local rings is an isomorphism, for all P EX.
(c) Show that if <p(X) is dense in Y, then the map <pp is injective for all P EX.
3.4. Show that the d-uple embedding of P" (Ex. 2.12) is an isomorphism onto its
image.
3.5. By abuse of language, we will say that a variety "is affine" if it is isomorphic to
an affine variety. If H <;::; P" is any hypersurface, show that P" - H is affine.
[Hint: Let H have degree d. Then consider the d-uple embedding of P" in pN
and use the fact that pN minus a hyperplane is affine.
3.6. There are quasi-affine varieties which are not affine. For example, show that
X = A 2 - {(0,0)} is not affine. [Hint: Show that (!)(X) ~ k[x,y] and use (3.5).
See (Ill, Ex. 4.3) for another proof.]
3.7. (a) Show that any two curves in P 2 have a nonempty intersection.
(b) More generally, show that if Y <;::; P" is a projective variety of dimension ;, 1,
and if H is a hypersurface, then Y n H =f. 0. [Hint: Use (Ex. 3.5) and (Ex.
3.1e). See (7.2) for a generalization.]
3.8. Let H; and Hi be the hyperplanes in P" defined by X; = 0 and x1 = 0, with i =f. j.
Show that any regular function on P" - (H; n Hi) is constant. (This gives an
alternate proof of (3.4a) in the case Y = P".)
3.9. The homogeneous coordinate ring of a projective variety is not invariant under
isomorphism. For example, let X = P 1 , and let Y be the 2-uple embedding of
P 1 in P 2 . Then X ~ Y (Ex. 3.4). But show that S(X) "1. S( Y).

3.10. Subvarieties. A subset of a topological space is locally closed if it is an open

subset of its closure, or, equivalently, if it is the intersection of an open set with
a closed set.
If X is a quasi-affine or quasi-projective variety and Y is an irreducible locally
closed subset, then Y is also a quasi-affine (respectively, quasi-projective) variety,
by virtue of being a locally closed subset of the same affine or projective space.
We call this the induced structure on Y, and we call Y a subvariety of X.
Now let <p:X--> Y be a morphism, let X' s::; X and Y' s::; Y be irreducible
locally closed subsets such that <p(X') <;::; Y'. Show that <fJix·: X' --> Y' is a mor-
phism.

21
I Varieties

3.11. Let X be any variety and let P E X. Show there is a 1-1 correspondence between
the prime ideals of the local ring {!)P and the closed subvarieties of X containing P.
3.12. If P is a point on a variety X, then dim {!)P = dim X. [Hint:Reduce to the
affine case and use (3.2c).]
3.13. The Local Ring of a Subvariety. Let Y <:; X be a subvariety. Let {!)Y.x be the set
<
of equivalence classes U,f) where U <:; X is open, U n Y =/= 0, and f is a
regular function on U. We say <UJ) is equivalent to <V,g), iff= g on U n V.
Show that {!)Y.x is a local ring, with residue field K(Y) and dimension = dim X --
dim Y. It is the local ring of Yon X. Note if Y = Pis a point we get {!)p, and if
Y = X we get K(X). Note also that if Y is not a point, then K(Y) is not alge-
braically closed, so in this way we get local rings whose residue fields are not
algebraically closed.
3.14. Projection from a Point. Let P" be a hyperplane in pn+ 1 and let P E pn+ 1 - P".
Define a mapping cp :P"+ 1 - { P}--> P" by cp(Q) =the intersection of the unique
line containing P and Q with P"-
(a) Show that cp is a morphism.
(b) Let Y <:; P 3 be the twisted cubic curve which is the image of the 3-uple em-
bedding of P 1 (Ex. 2.12). If t,u are the homogeneous coordinates on PI, we
say that Y is the curve given parametrically by (x,y,z,w) = (t 3 ,t 2 u,tu 2 ,u 3 ). Let
P = (0,0,1,0), and let P 2 be the hyperplane z = 0. Show that the projection of
Y from P is a cuspidal cubic curve in the plane, and find its equation.

3.15. Products of Affine Varieties. Let X <:; A" and Y <:; Am be affine varieties.
(a) Show that X x Y <:; An+m with its induced topology is irreducible. [Hint:
Suppose that X x Y is a union of two closed subsets Z 1 u Z 2 . Let X; =
{xEXIx X y <:; Z;}, i = 1,2. Show that X= XI u Xz and xl,x2 are
closed. Then X= X 1 or X 2 so X x Y = Z 1 or Z 2 .] The affine variety
X x Y is called the product of X and Y. Note that its topology is in general
not equal to the product topology (Ex. 1.4).
(b) Show that A(X x Y) ~ A(X) @k A(Y).
(c) Show that X x Y is a product in the category of varieties, i.e., show (i) the
projections X x Y --> X and X x Y --> Y are morphisms, and (ii) given a
variety Z, and the morphisms Z --> X, Z --> Y, there is a unique morphism
Z --> X x Y making a commutative diagram
z ---------+ X X y

V><~
~Y. X
(d) Show that dim X x Y = dim X + dim Y.
3.16. Products of Quasi-Projective Varieties. Use the Segre embedding (Ex. 2.14) to
identify P" x pm with its image and hence give it a structure of projective variety.
Now for any two quasi-projective varieties X <:; P" and Y <:; pm, consider
X X Y <:; P" X pm_
(a) Show that X x Y is a quasi-projective variety.
(b) If X, Y are both projective, show that X x Y is projective.
*(c) Show that X x Y is a product in the category of varieties.

22
 3 Morphisms

3.17. Normal Varieties. A variety Y is normal at a point P E Y if @p is an integrally

closed ring. Y is normal if it is normal at every point.
(a) Show that every conic in P 2 is normal.
(b) Show that the quadric surfaces Q 1,Q 2 in P 3 given by equations Q 1 :xy = zw;
Q2 :xy = z2 are normal (cf. (II. Ex. 6.4) for the latter.)
(c) Show that the cuspidal cubic y 2 = x 3 in A2 is not normal.
(d) If Y is affine, then Y is normal= A(Y) is integrally closed.
(e) Let Y be an affine variety. Show that there is a normal affine variety Y, and a
morphism n: Y--+ Y, with the property that whenever Z is a normal variety,
and cp:Z--+ Y is a dominant morphism (i.e., cp(Z) is dense in Y), then there is
a unique morphism e:z--+ Y such that cp = no e. Y is called the normaliza-
tion of Y. You will need (3.9A) above.
3.18. Projectively Normal Varieties. A projective variety Y s; P" is projectively normal
(with respect to the given embedding) if its homogeneous coordinate ring S(Y)
is integrally closed.
(a) If Y is projectively normal, then Y is normal.
(b) There are normal varieties in projective space which are not projectively
normal. For example, let Y be the twisted quartic curve in P 3 given para-
metrically by (x,y,z,w) = (t 4 ,t 3 u,tu 3 ,u4 ). Then Yis normal but not projectively
normal. See (III, Ex. 5.6) for more examples.
(c) Show that the twisted quartic curve Y above is isomorphic to P 1 , which is
projectively normal. Thus projective normality depends on the embedding.
3.19. Automorphisms of A". Let cp:A"--+ A" be a morphism of A" to A" given by n
polynomials f 1 , . . . ,fn of n variables x 1 , . . . ,x"" Let J = detliJ/;/iJxjl be the
Jacobian polynomial of cp.
(a) If cp is an isomorphism (in which case we call cp an automorphism of A") show
that J is a nonzero constant polynomial.
**(b) The converse of (a) is an unsolved problem, even for n = 2. See, for example.
Vitushkin [1].
3.20. Let Y be a variety of dimension ~ 2, and let P E Y be a normal point. Let f be
a regular function on Y - P.
(a) Show that f extends to a regular function on Y.
(b) Show this would be false for dim Y = 1.
See (III, Ex. 3.5) for generalization.
3.21. Group Varieties. A group variety consists of a variety Y together with a morphism
J1: Y x Y -+ Y, such that the set of points of Y with the operation given by J1 is a
group, and such that the inverse map y --+ y- 1 is also a morphism of Y--+ Y.
(a) The additive group Ga is given by the variety A 1 and the morphism J1: A2 --+ A 1
defined by Jl(a,b) = a + b. Show it is a group variety.
(b) The multiplicative qroup Gm is given by the variety A 1 - {(0)} and the mor-
phism !liu.h) = ah. Show it is a group variety.
(c) If G is a group variety, and X is any variety, show that the set Hom(X,G) has a
natural group structure.
(d) For any variety X, show that Hom(X,Ga) is isomorphic to (1J(X) as a group
under addition.
(e) For any variety X, show that Hom(X,Gml is isomorphic to the group of units
in @(X), under multiplication.

23
I Varieties

4 Rational Maps

In this section we introduce the notions of rational map and birational

equivalence, which are important for the classification of varieties. A rational
map is a morphism which is only defined on some open subset. Since an
open subset of a variety is dense, this already carries a lot of information.
In this respect algebraic geometry is more "rigid" than differential geometry
or topology. In particular, the concept of birational equivalence is unique
to algebraic geometry.

Lemma 4.1. Let X and Y be varieties, let cp and ljJ be two morphisms fi'om
X to Y, and suppose there is a nonempty open subset U s; X such that
({Jiu = 1/Jiu· Then cp = 1/J.
PROOF. We may assume that Y s; P" for some n. Then by composing with
the inclusion morphism Y --> P", we reduce to the case Y = P". We consider
the product P" x P", which has a structure of projective variety given by its
Segre embedding (Ex. 3.16). The morphisms cp and ljJ determine a map
cp x 1/f:X __. P" x P", which in fact is a morphism (Ex. 3.16c). Let .d =
{P x PIP E P"} be the diagonal subset of P" x P". It is defined by the
equations {xd'j = xjy;li,j = 0,1, ... ,n} and so is a closed subset ofP" x P".
By hypothesis cp x 1/J(U) s; .d. But U is dense in X, and .d is closed, so
cp x 1/f(X) s; .d. This says that cp = 1/J.

Definition. Let X, Y be varieties. A rational map cp: X --> Y is an equivalence

class of pairs <U,cpu) where U is a nonempty open subset of X, ({Ju is a
morphism of U to Y, and where <U,cpu) and <V,<pv) are equivalent if
({Ju and ({Jv agree on U n V. The rational map cp is dominant if for some
<
(and hence every) pair U ,({Ju), the image of ({Ju is dense in Y.

Note that the lemma implies that the relation on pairs <U,cpu) just
described is an equivalence relation. Note also that a rational map cp: X --> Y
is not in general a map of the set X to Y. Clearly one can compose dvminant
rational maps, so we can consider the category of varieties and dominant
rational maps. An "isomorphism" in this category is called a birational map:

Definition. A birational map cp: X --> Y is a rational map which admits an

inverse, namely a rational map ljJ: Y --> X such that ljJ a cp = idx and
cp o ljJ = idy as rational maps. If there is a birational map from X to Y,
we say that X and Yare birationally equivalent, or simply birational.

The main result of this section is that the category of varieties and domi-
nant rational maps is equivalent to the category of finitely generated field

24
 4 Rational Maps

extensions of k, with the arrows reversed. Before giving this result, we need
a couple of lemmas which show that on any variety, the open affine subsets
form a base for the topology. We say loosely that a variety is affine if it is
isomorphic to an affine variety.

Lemma 4.2. Let Y be a hypersurface in An given by the equationf(x 1 , •.. ,xn) =

0. Then An - Y is isomorphic to the hypersurface H in An+ 1 given by
xn + d = 1. In particular, An - Y is affine, and its affine ring is
k[xl, ... ,xn]f·
PROOF. For P = (a~o ... ,an+ 1 ) E H, let cp(P) = (a 1 , . . • ,an). Then clearly cp
is a morphism from H to An, corresponding to the homomorphism of rings
A --> A 1 , where A = k[ x 1 , . . . ,xn]. It is also clear that cp gives a bijective
mapping of H onto its image, which is An - Y To show that cp is an isomor-
phism, it is sufficient to show that cp- 1 is a morphism. But cp- 1 (a 1 , ••• ,an) =
(a 1 , . • . ,an,l/f(a 1 , . . . ,an)), so the fact that cp- 1 is a morphism on An - Y
follows from (3.6).

Proposition 4.3. On any variety Y, there is a base for the topology consisting
of open affine subsets.
PROOF. We must show for any point P E Y and any open set U containing P,
that there exists an open affine set V with P E V <:; U. First, since U is also
a variety, we may assume U = Y Secondly, since any variety is covered by
quasi-affine varieties (2.3), we may assume that Y is quasi-affine in An.
Let Z = Y - Y, which is a closed set in An, and let a <:; A = k[x 1 , . . . ,xn]
be the ideal of Z. Then, since Z is closed, and P ¢ Z, we can find a polynomial
f E a such that f(P) =f. 0. Let H be the hypersurface f = 0 in An. Then
Z <:; H but P ¢ H. Thus P E Y - Y n H, which is an open subset of
Y Furthermore, Y - Y n H is a closed subset of An - H, which is affine
by (4.2), hence Y - Y n H is affine. This is the required affine neighbor-
hood of P.

Now we come to the main result of this section. Let cp: X --> Y be a
<
dominant rational map, represented by U,ffJu ). Let f E K( Y) be a rational
function, represented by <V,f) where Vis an open set in Y, and f is a regular
function on V. Since ffJu(U) is dense in Y, cp{) 1(V) is a nonempty open subset
of X, sofa ffJu is a regular function on cp{) 1 (V). This gives us a rational
function on X, and in this manner we have defined a homomorphism of
k-algebras from K(Y) to K(X).

Theorem 4.4. For any two varieties X and Y, the above construction gives a
bijection between
(i) the set of dominant rational maps from X to Y, and
(ii) the set of k-algebra homomorphisms from K( Y) to K(X).

25
I Varieties

Furthermore, this correspondence gives an arrow-reversing equivalence of

categories of the category of varieties and dominant rational maps with the
category of finitely generated field extensions of k.
PROOF. We will construct an inverse to the mapping given by the construction
.tbove. Let B:K(Y)--+ K(X) be a homomorphism of k-algebras. We wish
to define a rational map from X to Y. By (4.3), Y is covered by affine varieties,
so we may assume Y is affine. Let A(Y) be its affine coordinate ring, and let
y 1 , . . . ,y" be generators for A(Y) as a k-algebra. Then B(y 1 ), . . . ,B(y") are
rational functions on X. We can find an open set U s; X such that the func-
tions B(y;) are all regular on U. Then e defines an injective homomorphism
of k-algebras A( Y) --+ (?)( U). By (3.5) this corresponds to a morphism
q>: U --+ Y, which gives us a dominant rational map from X to Y. It is easy
to see that this gives a map of sets (ii) --+ (i) which is inverse to the one defined
above.
To see that we have an equivalence of categories as stated, we need only
check that for any variety Y, K(Y) is finitely generated over k, and conversely,
if K/k is a finitely generated field extension, then K = K(Y) for some Y.
If Y is a variety, then K( Y) = K(U) for any open affine subset, so we may
assume Y affine. Then by (3.2d), K(Y) is a finitely generated field extension
of k. On the other hand, let K be a finitely generated field extension of k.
Let y 1 , . • . ,y" E K be a set of generators, and let B be the sub-k-algebra of K
generated by y 1 , . . . ,y.. Then B is a quotient of the polynomial ring
A = k[x 1 , . . . ,x"]' soB ~ A(Y) for some variety Yin A". Then K ~ K(Y)
so we are done.

Corollary 4.5. For any two varieties X,Y the following conditions are equiv-
ulent:
(i) X and Yare birationally equivalent;
(ii) there are open subsets U s; X and V s; Y with U isomorphic to V,
(iii) K(X) ~ K( Y) us k-algebrus.

PROOF.
(i) = (ii). Let q>: X --+ Y and lj;: Y --+ X be rational maps which are inverse
to each other. Let q> be represented by <U,q>) and let lj; be represented by
<V,l/J). Then lj; a q> is represented by <<r>- 1 (V),lj; o q>), and since lj; o q> = idx
as a rational map, lj; o q> is the identity on q> - 1 ( V). Similarly q> o lj; is the
identity on lj;- 1(U). We now take q>- 1(lj;- 1(U)) as our open set in X, and
lj;- 1 ( q>- 1 ( V)) as our open set in Y. It follows from the construction that these
two open sets are isomorphic via q> and lj;.
(ii) =(iii) follows from the definition of function field.
(iii) = (i) follows from the theorem.

As an illustration of the notion of birational correspondence, we will use

some algebraic results on field extensions to show that every variety is bi-

26
 4 Rational Maps

rational to a hypersurface. We assume familiarity with the notion of sepa-

rable algebraic field extensions, and the notions of transcendence base and
transcendence degree for infinite field extensions (see, e.g., Zariski-Samuel
[1, Ch. II]).

Theorem 4.6A (Theorem of the Primitive Element). Let L be a finite separable

extension field of afield K. Then there is an element r:t.. E L which generates
L as an extension field of K. Furthermore, if {3 1 , . . . ,f3n is any set of
generators of L over K, and if K is infinite, then r:t.. can be taken to be a
linear combination r:t.. = c 1 {3 1 + ... + cnf3n of the f3i with coefficients
C;EK.

PROOF. Zariski-Samuel [1, Ch. II, Theorem 19, p. 84]. The second statement
follows from the proof given there.

Definition. A field extension Kjk is separably generated if there is a tran-

scendence base {x;} for Kjk such that K is a separable algebraic extension
of k( {x;} ). Such a transcendence base is called a separating transcendence
base.

Theorem 4.7A. If a field extension Kjk is finitely generated and separably

generated, then any set of generators contains a subset which is a separating
transcendence base.

PROOF. Zariski-Samuel [1, Ch. II, Theorem 30, p. 104].

Theorem 4.8A. If k is a perfect field (hence in particular if k is algebraically

closed), any .finitely generated field extension Kjk is separably generated.

PROOF. Zariski-Samuel [1, Ch. II, Theorem 31, p. 105], or Matsumura

[2, Ch. 10, Corollary, p. 194].

Proposition 4.9. Any variety X of dimension r is birational to a hypersurface

Yin pr+ I.

PROOF. The function field K of X is a finitely generated extension field of k.

By (4.8A), K is separably generated over k. Hence we can find a transcendence
base x 1, ... ,xr E K such that K is a finite separable extension of k(x 1 , . •• ,xrl·
Then by (4.6A) we can find one further element y E K such that
K = k(x~> . .. ,x.,y). Now y is algebraic over k(x~> . .. ,xr), so it satisfies a
polynomial equation with coefficients which are rational functions in
x~> ... ,xr. Clearing denominators, we get an irreducible polynomial
f(x~> . .. ,x.,y) = 0. This defines a hypersurface in Ar+ 1 with function
field K, which, according to (4.5}, is birational to X. Its projective closure
(Ex. 2.9) is the required hypersurface Y c:; pr+ 1 .

27
I Varieties

Blowing Up
As another example of a birational map, we will now construct the blowing-
up of a variety at a point. This important construction is the main tool in
the resolution of singularities of an algebraic variety.
First we will construct the blowing-up of A" at the point 0 = (0, ... ,0).
Consider the product A" x P"- 1, which is a quasi-projective variety
(Ex. 3.16). If x 1 , . . . ,x" are the affine coordinates of A", and if y 1 , . . . ,y"
are the homogeneous coordinates of P"- 1 (observe the unusual notation!),
then the closed subsets of A" x P"- 1 are defined by polynomials in the
X;,yi, which are homogeneous with respect to the Yi·
We now define the blowing-up of A" at the point 0 to be the closed subset
X of A" x P"- 1 defined by the equations {x;Yi = XiY;Ii,j = 1, ... ,n}.

X A" X pn-1

We have a natural morphism cp: X --+ A" obtained by restricting the pro-
jection map of A" x P"- 1 onto the first factor. We will now study the
properties of X.
(1) If PEA", P # 0, then cp- 1(P) consists of a single point. In fact,
cp gives an isomorphism of X- cp- 1 (0) onto A"- 0. Indeed, let P =
(ab ... ,an), with some a; # 0. Now if P x (Yb ... ,y") E cp- 1 (P), then for
each j, Yi = (aia;)y;, so (y 1, . . . ,yn) is uniquely determined as a point in
P"- 1 . In fact, setting Y; = a;, we can take (yb ... ,yn) = (a 1 , . . . ,an). Thus
cp - 1 (P) consists of a single point. Furthermore, for PEA" - 0, setting
t/J(P) = (ab . .. ,an) x (a 1 , . . . ,an) defines an inverse morphism to cp,
showingX- cp- 1 (0)isisomorphictoA"- 0.
(2) cp- 1 ( 0) ~ P"- 1 . Indeed, cp- 1 ( 0) consists of all points 0 x Q, with
Q = (y 1 , . . . ,yn) E P"-1, subject to no restriction.
(3) The points of cp - 1 ( 0) are in 1-1 correspondence with the set of lines
through 0 in A". Indeed, a line L through 0 in A" can be given by para-
metric equations X; = a;t, i = 1, ... ,n, where a; E k are not all zero, and
tEA 1. Now consider the line L' = cp - 1(L - 0) in X - cp - 1( 0). It is given
parametrically by X; = a;t, Y; = a;t, with t E A 1 - 0. But the Y; are homo-
geneous coordinates in P"- 1 , so we can equally well describe L' be the
equations X; = a;t,y; = a;, for tEA 1 - 0. These equations make sense
also for t = 0, and give the closure L' of L' in X. Now L' meets cp - 1 ( 0) in
the point Q = (a 1 , . . . ,an) E pn- \ so we see that sending L to Q gives a
1-1 correspondence between lines through 0 in A" and points of cp- 1 ( 0).
(4) X is irreducible. Indeed, X is the union of X - cp - 1( 0) and cp - 1( 0).
The first piece is isomorphic to A" - 0, hence irreducible. On the other
hand, we have just seen that every point of cp- 1 ( 0) is in the closure of some
28
 4 Rational Maps

uo
Figure 3. Blowing up.

subset (the line L') of X - <p- 1( 0). Hence X - <p- 1( 0) is dense in X, and
X is irreducible.
Definition. If Y is a closed subvariety of A" passing through 0, we define
the blowing-up of Y at the point 0 to be Y = (<p - 1 ( Y - 0) )-, where
<p:X--> A" is the blowing-up of A" at the point 0 described above. We
denote also by <p: Y --> Y the morphism obtained by restricting <p :X --> A"
to Y. To blow up at any other point P of A", make a linear change of
coordinates sending P to 0.
Note that <p induces an isomorphism of Y- <p- 1 (0) to Y- 0, so that
<pis a birational morphism of Y to Y. Note also that this definition apparently
depends on the embedding of Y in A", but in fact, we will see later that
blowing-up is intrinsic (II, 7.15.1).
The effect of blowing up a point of Y is to "pull apart" Y near 0 according
to the different directions of lines through 0. We will illustrate this with
an example.
Example 4.9.1. Let Y be the plane cubic curve given by the equation y 2 =
x 2 (x + 1). We will blow up Y at 0 (Fig. 3). Let t,u be homogeneous co-
ordinates for P 1 . Then X, the blowing-up of A2 at 0, is defined by the
equation xu= ty inside A 2 x P 1 . It looks like A2 , except that the point 0
has been replaced by a P 1 corresponding to the slopes of lines through 0.
We will call this P 1 the exceptional curve, and denote it by E.
We obtain the total inverse image of Yin X by considering the equations
y 2 = x 2 (x + 1) and xu = ty in A2 x P 1 . Now P 1 is covered by the open
sets t i= 0 and u i= 0, which we consider separately. If t i= 0, we can set
t = 1, and use u as an affine parameter. Then we have the equations

y2 = x 2 (x + 1)
y = xu

29
I Varieties

in A 3 with coordinates x, y,u. Substituting, we get x 2 u 2 - x 2 (x + 1) = 0,

which factors. Thus we obtain two irreducible components, one defined by
x = 0, y = 0, u arbitrary, which is E, and the other defined by u 2 = x + 1,
y = xu. This is Y. Note that Y meets Eat the points u = ±1. These points
correspond to the slopes of the two branches of Y at 0.
Similarly one can check that the total inverse image of the x-axis consists
of E and one other irreducible curve, which we call the strict transform of the
x-axis (it is the curve L' described earlier corresponding to the line L = x-axis).
This strict transform meets E at the point u = 0. By considering the other
open set u =1= 0 in A 2 x P 1 , one sees that the strict transform of the y-axis
meets E at the point t = 0, u = 1.
These conclusions are summarized in Figure 3. The effect of blowing up
is thus to separate out branches of curves passing through 0 according to
their slopes. If the slopes are different, their strict transforms no longer meet
in X. Instead, they meet E at points corresponding to the different slopes.

EXERCISES

4.1. Iff and g are regular functions on open subsets U and V of a variety X, and if
f = g on U n V, show that the function which is f on U and g on V is a regular
function on U u V. Conclude that iff is a rational function on X, then there is
a largest open subset U of X on which f is represented by a regular function.
We say that f is defined at the points of U.
4.2. Same problem for rational maps. If 1.fJ is a rational map of X to Y, show there
is a largest open set on which l.fJ is represented by a morphism. We say the ra-
tional map is defined at the points of that open set.
4.3. (a) Letfbe the rational function on P 2 given by f = xdx 0 . Find the set of points
where f is defined and describe the corresponding regular function.
(b) Now think of this function as a rational map from P 2 to A 1 • Embed A 1 in P 1 ,
and let I.{J:P 2 --> P 1 be the resulting rational map. Find the set of points where
1.fJ is defined, and describe the corresponding morphism.

4.4. A variety Y is rational if it is birationally equivalent toP" for some n (or, equiva-
lently by (4.5), if K(Y) is a pure transcendental extension of k).
(a) Any conic in P 2 is a rational curve.
(b) The cuspidal cubic y 2 = x 3 is a rational curve.
(c) Let Y be the nodal cubic curve y 2 z = x 2 (x + z) in P 2 . Show that the pro-
jection 1.fJ from the point P = (0,0,1) to the line z = 0 (Ex. 3.14) induces a
birational map from Y to P 1 . Thus Y is a rational curve.

4.5. Show that the quadric surface Q:xy = zw in P 3 is birational to P 2 , but not
isomorphic to P 2 (cf. Ex. 2.15).

4.6. Plane Cremona Tramiformations. A birational map of P 2 into itself is called a

plane Cremona transformation. We give an example, called a quadratic transfor-
mation. It is the rational map I.{J:P 2 --> P 2 given by (a 0 ,a 1 ,a 2 )--> (a 1a 2 ,a 0 a 2 ,a 0 a 1 )
when no two of a 0 ,a 1 ,a 2 are 0.

30
 5 Nonsingular Varieties

(a) Show that <p is birational, and is its own inverse.

(b) Find open sets U,V s; P 2 such that <p: U ---> Vis an isomorphism.
(c) Find the open sets where <p and <p- 1 are defined, and describe the correspond-
ing morphisms. See also (V, 4.2.3).
4.7. Let X and Y be two varieties. Suppose there are points P EX and Q E Y such
that the local rings ~P.x and (r;;Q.Y are isomorphic as k-algebras. Then show
that there are open sets P E U s; X and Q E V s; Y and an isomorphism of
U to V which sends P to Q.

4.8. (a) Show that any variety of positive dimension over k has the same cardinality as
k. [Hints: Do A" and P" first. Then for any X, use induction on the dimension
11. Use (4.9) to make X birational to a hypersurface H <;: pn+ 1 . Use (Ex. 3.7)

to show that the projection of H to P" from a point not on H is finite-to-one

and surjective.]
(b) Deduce that any two cwTes over k are homeomorphic (cf. Ex. 3.1).
4.9. Let X be a projective variety of dimension r in P". with 11 ? r + 2. Show that
for suitable choice of P ¢X, and a linear P"- 1 s; P". the projection from P to
pn-t (Ex. 3.14) induces a hirotionol morphism of)( onto its image x· s; P"- 1 •
You will need to use (4.6Al. (4.7A). and (4.8A). This shows in particular that the
birational map of (4.9) can be obtained by a finite number of su~.:h proje~.:tions.
4.10. Let r be the cusp ida! cubic curve _r 2 = x 3 in A2 Blow up the point 0 = (0,0).
let E be the exceptional curve, and let Y be the strict transform of Y. Show that
E meets Y in one point, and that Y ::; A 1 • In this case the morphism <p: Y ---> Y
is bijective and bicontinuous, but it is not an isomorphism.

5 Nonsingular Varieties

The notion of nonsingular variety in algebraic geometry corresponds to the

notion of manifold in topology. Over the complex numbers, for example,
the nonsingular varieties are those which in the "usual" topology are complex
manifolds. Accordingly, the most natural (and historically first) definition
of nonsingularity uses the derivatives of the functions defining the variety:

Definition. Let Y ~ A" be an affine variety, and let j 1 , ••• ,J; E A =

k [ x 1 , . . . ,x"] be a set of generators for the ideal of Y. Y is nonsingular at a
point P E Y if the rank of the matrix IIU)Rx)(Plll is 11 - r, where r is
the dimension of Y. Y is nonsingular if it is nonsingular at every point.

A few comments are in order. In the first place, the notion of partial
derivative of a polynomial with respect to one of its variables makes sense
over any field. One just applies the usual rules for differentiation. Thus no
limiting process is needed. But funny things can happen in characteristic
p > 0. For example, if f(x) = xP, then df/dx = pxp-J = 0, since p = 0 ink.
In any case, iff E A is a polynomial, then for each i, of/ex; is a polynomial.

31
I Varieties

The matrix ll(o/;/oxi)(P)II is called the Jacobian matrix at P. One can show
easily that this definition of nonsingularity is independent of the set of
generators of the ideal of Y chosen.
One drawback of our definition is that it apparently depends on the
embedding of Y in affine space. However, it was shown in a fundamental
paper, Zariski [1 ], that nonsingularity could be described intrinsically in
terms of the local rings. In our case the result is this.

Definition. Let A be a noetherian local ring with maximal ideal m and residue
field k = A/m. A is a regular local ring if dimk m/m 2 = dim A.

Theorem 5.1. Let Y <;; An be an affine variety. Let P E Y be a point. Then Y

is nonsingular m P if and only if the local ring (!) P,Y is a regular local ring.
PROOF. Let p be the point (ab ... ,an) in An, and let Op = (xl - ({1, ... ,Xn- an)
be the corresponding maximal ideal in A = k[x 1 , . . • ,xnJ. We define a
linear map 0: A --+ kn by

of (P), ... , -0cf (P) )

O(f) = \ -~
cx 1 Xn
for any f E A. Now it is clear that O(x; - a;) for i = 1, ... ,n form a basis of
kn, and that O(a~) = 0. Thus 0 induces an isomorphism O':ap/a~--+ kn.
Now let b be the ideal of Yin A, and let f 1 , ••. ,fr be a set of generators of b.
Then the rank of the Jacobian matrix J = li(o/;/oxi)(P)II is just the dimension
of O(b) as a subspace of kn. Using the isomorphism 0', this is the same as the
dimension of the subspace (b + a~)/a~ of ap/a~. On the other hand, the
local ring(!) P of P on Y is obtained from A by dividing by b and localizing at
the maximal ideal ap. Thus if m is the maximal ideal of (!)p, we have
m/m 2 ~ ap/(b + a~).
Counting dimensions of vector spaces, we have dim m/m 2 + rank J = n.
Now let dim Y = r. Then (!)Pis a local ring of dimension r (3.2), so (!)Pis
regular if and only if dimk mjm 2 = r. But this is equivalent to rank J = n - r,
which says that P is a nonsingular point of Y.
Note. Later we will give another characterization of nons in gular points in
terms of the sheaf of differential forms on Y (II, 8.15).
Now that we know the concept of nonsingularity is intrinsic, we can extend
the definition to arbitrary varieties.

Definition. Let Y be any variety. Y is nonsingular at a point P E Y if the local

ring (! P.r is a regular local ring. Y is nonsingulur if it is nonsingular at
every point. Y is sinyular if it is not nonsingular.
Our next objective is to show that most points of a variety are nonsingular.
We need an algebraic preliminary.

32
 5 Nonsingular Varieties

Proposition 5.2A. If A is a noetherian local ring with maximal ideal m and

residue field k, then dimk m/m 2 ~ dim A.
PROOF. Atiyah-Macdonald [1, Cor. 11.15, p. 121] or Matsumura [2,
p. 78].

Theorem 5.3. Let Y be a variety. Then the set Sing Y of singular points of Y
is a proper closed subset of Y.

PROOF. (See also II, 8.16.) First we show Sing Y is a closed subset. It is
sufficient to show for some open covering Y = U Y; of Y, that Sing Y; is
closed for each i. Hence by (4.3) we may assume that Y is affine. By (5.2) and
the proof of (5.1) we know that the rank of the Jacobian matrix is always
~ n - r. Hence the set of singular points is the set of points where the rank is
< n - r. Thus Sing Y is the algebraic set defined by the ideal generated by
I( Y) together with all determinants of (n - r) x (n - r) submatrices of the
matrix II8N8xill· Hence Sing Y is closed.
To show that Sing Y is a proper subset of Y, we first apply (4.9) to get Y
birational to a hypersurface in Pn. Since birational varieties have isomorphic
open subsets, we reduce to the case of a hypersurface. It is enough to consider
any open affine subset of Y, so we may assume that Y is a hypersurface in An,
defined by a single irreducible polynomial f(xb . .. ,xn) = 0.
Now Sing Y is the set of points P E Y such that (8f/8x;)(P) = 0 for i =
1, ... ,n. If Sing Y = Y, then the functions 8fj8x; are zero on Y, and hence
8fj8x; E I(Y) for each i. But I(Y) is the principal ideal generated by f, and
deg(8f/8x;) ~ deg f - 1 for each i, so we must have 8f/8x; = 0 for each i.
In characteristic 0 this is already impossible, because if X; occurs in f,
then 8f/8x; i= 0. So we must have char k = p > 0, and then the fact that
8f/8x; = 0 implies that f is actually a polynomial in xf. This is true for each
i, so by taking pth roots of the coefficients (possible since k is algebraically
closed), we get a polynomial g(x 1 , . . . ,xn) such that f = gP. But th:s
contradicts the hypothesis that f was irreducible, so we conclude that
Sing Y < Y.

Completion
For the local analysis of singularities we will now describe the technique of
completion. Let A be a local ring with maximal ideal m. The powers of m
define a topology on A, called the m-adic topology. By completing with
respect to this topology, one defines the completion of A, denoted A. Alter-
natively, one can define A as the inverse limit !!!!! A/mn. See Atiyah-
Macdonald [1, Ch. 10], Matsumura [2, Ch. 9], or Zariski-Samuel [1, vol. 2,
Ch. VIII] for general information on completions.
The significance of completion in algebraic geometry is that by passing
to the completion &P of the local ring of a point P on a variety X, one can
study the very local behavior of X near P. We have seen (Ex. 4.7) that if

33
I Varieties

points P EX and Q E Y have isomorphic local rings, then already P and Q

have isomorphic neighborhoods, so in particular X and Y are birational.
Thus the ordinary local ring (np carries information about almost all of X.
However, the completion r!p, as we will see, carries much more local in-
formation, closer to our intuition of what "local" means in topology or
differential geometry.
We will recall some of the algebraic properties of completion and then
give some examples.

Theorem 5.4A. Let A be a noetherian local ring with maximal ideal m, and
let A be its completion.
(a) A is a local ring, with maximal ideal m = mA, and there is a natural
injective homomorphism A --> A.
(b) If M is a finitely generated A-module, its completion M with respect
to its m-adic topology is isomorphic toM ®A A.
(c) dim A = dim A.
(d) A is regular if and only if A is regular.
PROOF. See Atiyah~Macdonald [1, Ch. 10, 11] or Zariski~Samuel [1, vol. 2,
Ch. Vlll].

Theorem 5.5A (Cohen Structure Theorem). If A is a complete regular local

ring of dimension n contain in?] some field, then A ~ k[[x 1 , •.• ,xn]J, the
ring of formal power series over the residue field k of A.
PROOF. Matsumura [2, Cor. 2, p. 206] or Zariski~Samuel [1, vol. 2, Cor.,
p. 307].

Definition. We say twu points P EX and Q E Y are analytically isomorphic

if there is an isomorphism &P ~ &Q as k-algebras.

Example 5.6.1. If P EX and Q E Y are analytically isomorphic, then

dim X = dim Y. This follows from (5.4A) and the fact that any local ring
of a point on a variety has the same dimension as the variety (Ex. 3.12).

Example 5.6.2. If P E X and Q E Y are nonsingular points on varieties of the

same dimension, then P and Q are analytically isomorphic. This follows
from (5.4A) and (5.5A). This example is the algebraic analogue of the fact
that any two manifolds (topological, differentiable, or complex) of the same
dimension are locally isomorphic.

Example 5.6.3. Let X be the plane nodal cubic curve given by the equation
y 2 = x 2 (x + 1). Let Y be the algebraic set in A 2 defined by the equation
x.r = 0. We will show that the point 0 = (0,0) on X is analytically iso-
morphic to the point 0 on Y. (Since we haven't yet developed the general
theory of local rings of points on reducible algebraic sets, we use an ad hoc

34
 5 Nonsingular Varieties

definition @0 ,Y = (k[x,y]/(xy))(x,y)· Thus &o,Y ~ k[[x,y]]/(xy).) This ex-

ample corresponds to the geometric fact that near 0, X looks like two lines
crossing.
To prove this result, we consider the completion &o,x which is isomorphic
to k[[x,y]]/(y 2 - x 2 - x 3 ). The key point is that the leading form of the
equation, namely y 2 - x 2 , factors into two distinct factors y + x andy - x
(we assume char k # 2). I claim there are formal power series
g = Y+ x + gz + g3 + · · ·
h = y - x + h2 + h 3 + ...
in k[[ x, y]], where g;,h; are homogeneous of degree i, such that y 2 - x 2 -
x 3 = gh. We construct g and h step by step. To determine g 2 and h2 , we
need to have
(y - x)gz + (y + x)h 2 = -x 3 .
This is possible, because y - x and y + x generate the maximal ideal of
k[[x,y]]. To determine g 3 and h 3, we need
(y - x)g 3 + (y + x)h 3 = -g 2 h 2
which is again possible, and so on.
Thus &o,x = k[[ x,y ]]/(gh). Since g and h begin with linearly independent
linear terms, there is an automorphism of k[[ x, y]] sending g and h to x
andy, respectively. This shows that &o,x ~ k[[ x,y]]/(xy) as required.
Note in this example that @0 ,x is an integral domain, but its completion
is not.

We state here an algebraic result which will be used in (Ex. 5.15) below.

Theorem 5.7A (Elimination Theory). Let f 1 , . . . ,f,. be homogeneous polyno-

mials in x 0 , . . . ,x", having indeterminate coefficients aii. Then there is a
set g 1 , . . . ,g 1 of polynomials in the a;i, with integer coefficients, which are
homogeneous in the coefficients of each J; separately, with the following
property: for any field k, and for any set of special values of the aii E k,
a necessary and sufficient condition for the J; to have a common zero different
from (0,0, ... ,0) is that the aii are a common zero of the polynomials gi.
PROOF. Vander Waerden [1, vol. II, §80, p. 8].

EXERCISES

5.1. Locate the singular points and sketch the following curves in A2 (assume char
k =1- 2). Which is which in Figure 4?
(a) x2 = + .l:
x4
(b) xy = x 6 + y 6;
(c) x3 = )'2 + x4 + )'4;
(d) x2y + xi = x4 + y4.

35
I Varieties

Node Triple point Cusp Tacnode

Figure 4. Singularities of plane curves.

5.2. Locate the singular points and describe the singularities of the following sur-
faces in A3 (assume char k # 2). Which is which in Figure 5?
(a) xy 2 = z 2 ;
(b) xz + yz = zz;
(c) xy + x 3 + y 3 = 0.

Conical double point Double line Pinch point

Figure 5. Surface singularities.

5.3. Multiplicities. Let Y ~ A 2 be a curve defined by the equation f(x,y) = 0. Let

P = (a,b) be a point of A2 . Make a linear change of coordinates so that P be-
comes the point (0,0). Then write f as a sum f = fo + f 1 + ... + f:J, where
j; is a homogeneous polynomial of degree i in x and y. Then we define the multi-
plicity of P on Y, denoted ,Up(Y), to be the least r such that fr # 0. (Note that
P E Y = ,Up(Y) > 0.) The linear factors of fr are called the tangent directions
at P.
(a) Show that ,Up(Y) = 1 =Pis a nonsingular point of Y.
(b) Find the multiplicity of each of the singular points in (Ex. 5.1) above.
5.4. Intersection Multiplicity. If Y,Z ~ A2 are two distinct curves, given by equations
f = 0, g = 0, and if P E Y n Z, we define the intersection multiplicity ( Y · Z)p
of Y and Z at P to be the length of the C9p-module C9p j(f,g).
(a) Show that (Y · Z)p is finite, and (Y · Z)p ~ flp(Y) · ,Up(Z).
(b) If P E Y, show that for almost all lines L through P (i.e., all but a finite number),
(L . Y)p = ,Up(Y).
(c) If Y is a curve of degree din P 2 , and if Lis a line in P 2 , L # Y, show that
(L · Y) = d. Here we define (L · Y) = L)L · Y)p taken over all points P E
L n Y, where (L · Y)p is defined using a suitable affine cover ofP 2 .

36
 5 N onsingular Varieties

5.5. For every degree d > 0, and every p = 0 or a prime number, give the equation
of a nonsingular curve of degree d in P 2 over a field k of characteristic p.

5.6. Blowing Up Curve Singularities.

(a) Let Y be the cusp or node of (Ex. 5.1). Show that the curve Y. obtained by
blowing up Yat 0 = (0,0) is nonsingular (cf. (4.9.1) and (Ex. 4.10)).
(b) We define a node (also called ordinary double point) to be a double point
(i.e., a point of multiplicity 2) of a plane curve with distinct tangent directions
(Ex. 5.3). If P is a node on a plane curve Y, show that rp - 1 (P) consists of two
distinct nonsingular points on the blown-up curve Y. We say that "blowing
up P resolves the singularity at P".
(c) Let P E Y be the tacnode of(Ex. 5.1). If rp: Y--> Y is the blowing-up at P, show
that rp - 1(P) is a node. Using (b) we see that the tacnode can be resolved by
two successive blowings-up.
(d) Let Y be the plane curve y 3 = x 5 , which has a "higher order cusp" at 0. Show
that 0 is a triple point; that blowing up 0 gives rise to a double point (what
kind?) and that one further blowing up resolves the singularity.
Note: We will see later (V, 3.8) that any singular point of a plane curve can be
resolved by a finite sequence of successive blowings-up.

5.7. Let Y <::::: P 2 be a nonsingular plane curve of degree > 1, defined by the equation
f(x,y,z) = 0. Let X <::::: A 3 be the affine variety defined by f (this is the cone
over Y; see (Ex. 2.1 0) ). Let P be the point (0,0,0), which is the vertex of the cone.
Let rp:X--> X be the blowing-up of X at P.
(a) Show that X has just one singular point, namely P.
(b) Show that X is nonsingular (cover it with open affines).
(c) Show that rp - 1(P) is isomorphic to Y.
5.8. Let Y <::::: P" be a projective variety of dimension r. Let f 1, ... ,j, E S =
k[ x 0 , . •• ,x"] be homogeneous polynomials which generate the ideal of Y. Let
P E Y be a point, with homogeneous coordinates P = (a 0 , . .• ,a"). Show that
Pis nonsingular on Y if and only if the rank of the matrix jj(8};/8xj)(a 0 , . . . ,anlll
is n - r. [Hint: (a) Show that this rank is independent of the homogeneous
coordinates chosen for P; (b) pass to an open affine U, <::::: P" containing P and
use the affine Jacobian matrix; (c) you will need Euler's lemma, which says that
iff is a homogeneous polynomial of degree d, then I,x;(3fj3x;) = d · f]
5.9. Let f E k[x,y,z] be a homogeneous polynomial, let Y = Z(f).::::: P 2 be the
algebraic set defined by f, and suppose that for every P E Y, at least one of
w;ax)(P), W/3y)(P), w;az)(P) is nonzero. Show that f is irreducible (and hence
that Y is a nonsingular variety). [Hint: Use (Ex. 3.7).]
5.10. For a point P on a variety X, let m be the maximal ideal of the local ring {!)P·
We define the Zariski tangent space Tp(X) of X at P to be the dual k-vector space
ofm/m 2 .
(a) For any point P EX, dim Tp(X) ;::, dim X, with equality if and only if Pis
nonsingular.
(b) For any morphism rp:X--> Y, there is a natural induced k-linear map Tp(rp):
T p(X) --> T cp<Pl( Y).
(c) If rp is the vertical projection of the parabola x = y 2 onto the x-axis, show that
the induced map T 0 (rp) of tangent spaces at the origin is the zero map.

37
I Varieties

5.11. The Elliptic Quartic Curve in P 3 . Let Y be the algebraic set in P 3 defined by the
equations x 2 - xz - yw = 0 and yz - xw - zw = 0. Let P be the point
(x,y,z,w) = (0,0,0,1), and let cp denote the projection from P to the plane w = 0.
Show that cp induces an isomorphism of Y - P with the plane cubic curve
y 2 z - x 3 + xz 2 = 0 minus the point (1,0,-1). Then show that Y is an irre-
ducible nonsingular curve. It is called the elliptic quartic curve in P 3 • Since it
is defined by two equations it is another example of a complete intersection
(Ex. 2.17).

5.12. Quadric Hypersurfaces. Assume char k i= 2, and let f be a homogeneous poly-

nomial of degree 2 in x 0 , •.. ,x•.
(a) Show that after a suitable linear change of variables,! can be brought into the
"form f = x6 + ... + x; for some 0 ~ r ~ n.
(b) Show that f is irreducible if and only if r ~ 2.
(c) Assume r ~ 2, and let Q be the quadric hypersurface in P" defined by f Show
that the singular locus Z = Sing Q of Q is a linear variety (Ex. 2.11) of dimen-
sion n - r - 1. In particular, Q is nonsingular if and only if r = n.
(d) In case r < n, show that Q is a cone with axis Z over a nonsingular quadric
hypersurface Q' s:::: P'. (This notion of cone generalizes the one defined in
(Ex. 2.10). If Y is a closed subset of P', and if Z is a linear subspace of dimen-
sion n - r - 1 in P", we embed P' in P" so that P' n Z = 0, and define
the cone over Y with axis Z to be the union of all lines joining a point of Y
to a point of Z.)
5.13. It is a fact that any regular local ring is an integrally closed domain (Matsumura
[2, Th. 36, p. 121 ]). Thus we see from (5.3) that any variety has a nonempty
open subset of normal points (Ex. 3.17). In this exercise, show directly (without
using (5.3)) that the set of nonnormal points of a variety is a proper closed sub-
set (you will need the finiteness of integral closure: see (3.9A) ).
5.14. Analytically Isomorphic Singularities.
(a) If P E Y and Q E Z are analytically isomorphic plane C'lrve singularities, show
that the multiplicities Jlp(Y) and Jlq(Z) are the same (Ex. 5.3).
(b) Generalize the example in the text (5.6.3) to show that iff = f.. + f..+ 1 + ... E
k[[ x, y]], and if the leading form f.. off factors as f.. = gsh,, where g.,h, are
homogeneous of degrees s and t respectively, and have no common linear
factor, then there are formal power series

g = Os + 9s+ 1 + · · ·
h=h,+ht+1+ ...
ink[[ x,y ]] such that f = gh.
(c) Let Ybedefinedbytheequationf(x,y) = OinA 2 ,andletP = (O,O)beapoint
of multiplicity r on Y, so that when f is expanded as a polynomial in x and y,
we have f = f.. + higher terms. We say that Pis an ordinary r-fold point if
f.. is a product of r distinct linear factors. Show that any two ordinary double
points are analytically isomorphic. Ditto for ordinary triple points. But show
that there is a one-parameter family of mutually nonisomorphic ordinary
4-fold points.
*(d) Assume char k i= 2. Show that any double point of a plane curve is analy-
tically isomorphic to the singularity at (0,0) of the curve i = x', for a uniquely

38
 6 Nonsingular Curves

determined r ~ 2. If r = 2 it is a node (Ex. 5.6). If r = 3 we call it a cusp;

if r = 4 a tacnode. See (V, 3.9.5) for further discussion.
5.15. Families of Plane Curves. A homogeneous polynomial f of degree d in three
variables x,y,z has (di 2 ) coefficients. Let these coefficients represent a point in
PN, where N = (di 2) - 1 = ±d(d + 3).
(a) Show that this gives a correspondence between points of pN and algebraic
sets in P 2 which can be defined by an equation of degree d. The correspondence
is 1-1 except in some cases where f has a multiple factor.
(b) Show under this correspondence that the (irreducible) nonsingular curves of
degree d correspond 1-1 to the points of a nonempty Zariski-open subset of
PN. [Hints: (1) Use elimination theory (5.7A) applied to the homogeneous
polynomials 8f/8x 0 , . • . ,8f/8x"; (2) use the previous (Ex. 5.5, 5.8, 5.9) above.]

6 Nonsingular Curves

In considering the problem of classification of algebraic varieties, we can

formulate several subproblems, based on the idea that a nonsingular pro-
jective variety is the best kind: (a) classify varieties up to birational equiva-
lence; (b) within each birational equivalence class, find a nonsingular
projective variety; (c) classify the nonsingular projective varieties in a given
birational equivalence class.
In general, all three problems are very difficult. However, in the case of
curves, the situation is much simpler. In this section we will answer problems
(b) and (c) by showing that in each birational equivalence class, there is a
unique nonsingular projective curve. We will also give an example to show
that not all curves are birationally equivalent to each other (Ex. 6.2). Thus
for a given finitely generated extension field K of k of transcendence degree 1
(which we will call a function field of dimension 1) we can talk about the
nonsingular projective curve CK with function field equal to K. We will see
also that if K 1 ,K 2 are two function fields of dimension 1, then any k-homo-
morphism K 2 ~ K 1 is represented by a morphism of CK, to CKz·
We will begin our study in an oblique manner by defining the notion of an
"abstract nonsingular curve" associated with a given function field. It will
not be clear a priori that this is a variety. However, we will see in retrospect
that we have defined nothing new.
First we have to recall some basic facts about valuation rings and Dede-
kind domains.

Definition. Let K be a field and let G be a totally ordered abelian group. A

valuation of K with values in G is a map v:K - {0} ~ G such that for all
x,y E K, x,y i= 0, we have:

(1) v(xy) = v(x) + v(y);

(2) v(x + y) ;;:;, min(v(x),v(y) ).

39
I Varieties

Ifv is a valuation, then the set R = {xEKJv(x) ~ 0} u {0} is a subring of K,

which we call the valuation ring ofv. The subset m = {xEKJv(x) > 0} u
{0} is an ideal in R, and R,m is a local ring. A valuation ring is an integral
domain which is the valuation ring of some valuation of its quotient field.
If R is a valuation ring with quotient field K, we say that R is a valuation
ring of K. If k is a subfield of K such that v(x) = 0 for all x E k - {0},
then we say vis a valuation of Kjk, and R is a valuation ring of Kjk. (Note
that valuation rings are not in general noetherian!)

Definition. If A,B are local rings contained in a .field K, we say that B dominates
A if A S:::: Band m 8 n A = rnA-

Theorem 6.1A. Let K be afield. A local ring R contained inK is a valuation

ring of K if and only if it is a maximal element of the set of local rings con-
tained in K, with respect to the relation of domination. Every local ring
contained in K is dominated by some valuation ring of K.
PROOF. Bourbaki [2, Ch. VI, §1, 3] or Atiyah-Macdonald [1, Ch. 5, p. 65,
and exercises, p. 72].

Definition. A valuation v is discrete if its value group G is the integers. The

corresponding valuation ring is called a discrete valuation ring.

Theorem 6.2A. Let A be a noetherian local domain of dimension one, with

maximal ideal m. Then the following conditions are equivalent:
(i) A is a discrete valuation ring;
(ii) A is integrally closed;
(iii) A is a regular local ring;
(iv) m is a principal ideal.
PROOF. Atiyah-Macdonald [1, Prop. 9.2, p. 94].

Definition. A Dedekind domain is an integrally closed noetherian domain of

dimension one.

Because integral closure is a local property (Atiyah-Macdonald [1, Prop.

5.13, p. 63]), every localization of a Dedekind domain at a nonzero prime
ideal is a discrete valuation ring.

Theorem 6.3A. The integral closure of a De de kind domain in a finite extension

field of its quotient field is again a Dedekind domain.
PROOF. Zariski-Samuel [1, vol. 1, Th. 19, p. 281].

We now turn to the case of a function field K of dimension 1 over k, where

k is our fixed algebraically closed base field. We wish to establish a connection

40
 6 Nonsingular Curves

between non-singular curves with function field K and the set of discrete
valuation rings of Kjk. If Pis a point on a nonsingular curve Y, then by (5.1)
the local ring (!JP is a regular local ring of dimension one, and so by (6.2A) it
is a discrete valuation ring. Its quotient field is the function field K of Y,
and since k ~ (!Jp, it is a valuation ring of Kjk. Thus the local rings of Y
define a subset of the set CK of all discrete valuation rings of Kjk. This
motivates the definition of an abstract nonsingular curve below. But first
we need a few more preliminaries.

Lemma 6.4. Let Y be a quasi-projective variety, let P,Q E Y, and suppose that
(!JQ ~ (!JP as subrings of K(Y). Then P = Q.
PROOF. Embed Y in pn for some n. Replacing Y by its closure, we may
assume Y is projective. After a suitable linear change of coordinates in pn,
we may assume that neither P nor Q is in the hyperplane H 0 defined by
x 0 = 0. Thus P,Q E Y n (Pn - H 0 ) which is affine, so we may assume that
Y is an affine variety.
Let A be the affine ring of Y. Then there are maximal ideals m,n ~ A
such that (!JP = A"' and (!JQ = Alt. If (!JQ ~ (!Jp, we must have m ~ n. But
m is a maximal ideal, so m = n, hence P = Q, by (3.2b).

Lemma 6.5. Let K be a function field of dimension one over k, and let x E K.
Then {R E CKix ¢ R} is a finite set.
PROOF. If R is a valuation ring, then x ¢ R if and only if 1/x E mR. So letting
y = 1/x, we have to show that if y E K, y # 0, then {R E CKIY E mR} is a
finite set. If y E k, there are no such R, so let us assume y ¢ k.
We consider the sub ring k[ yJof K generated by y. Since k is algebraically
closed, y is transcendental over k, hence k[y] is a polynomial ring. Further-
more, since K is finitely generated and of transcendence degree 1 over k,
K is a finite field extension of k(y). Now let B be the integral closure of
k[y] inK. Then by (6.3A), B is a Dedekind domain, and it is also a finitely
generated k-algebra (3.9A).
Now if y is contained in a discrete valuation ring R of Kjk, then k[y J ~ R,
and since R is integrally closed in K, we have B ~ R. Let n = mR n B.
Then n is a maximal ideal of B, and B is dominated by R. But Bit is also a
discrete valuation ring of Kjk, hence Bit = R by the maximality of valuation
rings (6.1A).
If furthermore y E mR, then yEn. Now B is the affine coordinate ring
of some affine variety Y (1.4.6). Since B is a Dedekind domain, Y has di-
mension one and is nonsingular. To say that yEn says that y, as a regular
function on Y, vanishes at the point of Y corresponding to n. But y # 0,
so it vanishes only at a finite set of points; these are in 1-1 correspondence
with the maximal ideals of B by (3.2), and R = Bit is determined by the
maximal ideal n. Hence we conclude that y E mR for only finitely many
RECK, as required.

41
I Varieties

Corollary 6.6. Any discrete valuation ring of Kjk is isomorphic to the local
ring of a point on some nonsingular affine curve.
PROOF. Given R, let y E R - k. Then the construction used in the proof
of (6.5) gives such a curve.

We now come to the definition of an abstract nonsingular curve. Let

K be a function field of dimension 1 over k (i.e., a finitely generated exten-
sion field of transcendence degree 1). Let CK be the set of all discrete valua-
tion rings of Kjk. We will sometimes call the elements of CK points, and
write P E CK, where P stands for the valuation ring Rp. Note that the set
CK is infinite, because it contains all the local rings of any nonsingular
curve with function field K; those local rings are all distinct (6.4), and there
are infinitely many of them (Ex. 4.8). We make CK into a topological space
by taking the closed sets to be the finite subsets and the whole space. If
U c:; C K is an open subset of C K• we define the ring of regular functions
on u to be (!J(U) = nPeU Rp. An element/ E (!J(U) defines a function from
U to k by taking f(P) to be the residue off modulo the maximal ideal of
Rp. (Note by (6.6) that for any RECK, the residue field of R is k.) If two
elements f,g E @(U) define the same function, then f - g E mp for infi-
nitely many PECK, so by (6.5) and its proof, f = g. Thus we can identify
the elements of @(U) with functions from U to k. Note also by (6.5) that
any f E K is a regular function on some open set U. Thus the function
field of CK, defined as in §3, is just K.

Definition. An abstract nonsingular curve is an open subset U c:; CK, where

K is a function field of dimension 1 over k, with the induced topology,
and the induced notion of regular functions on its open subsets.

Note that it is not clear a priori that such an abstract curve is a variety.
So we will enlarge the category of varieties by adjoining the abstract curves:

Definition. A morphism q>: X ~ Y between abstract nonsingular curves or

varieties is a continuous mapping such that for every open set V c:; Y,
and every regular function f: V ~ k, f o q> is a regular function on
q> -l(V).

Now that we have apparently enlarged our category, our task will be
to show that every nonsingular quasi-projective curve is isomorphic to an
abstract nonsingular curve, and conversely. In particular, we will show
that CK itself is isomorphic to a nonsingular projective curve.

Proposition 6.7. Every nonsingular quasi-projective curve Y is isomorphic

to an abstract nonsingular curve.

42
 6 Nonsingular Curves

PROOF. Let K be the function field of Y. Then each local ring (!JP of a point
P E Y is a discrete valuation ring of Kjk, by (5.1) and (6.2A). Furthermore,
by (6.4), distinct points give rise to distinct sub rings of K. So let U s; CK
be the set of local rings of Y, and let <p: Y ---> U be the bijective map defined
by <p(P) = @p.
First, we need to show that U is an open subset of CK. Because open
sets are complements of finite sets, it is sufficient to show that U contains a
nonempty open set. Thus, by (4.3), we may assume Y is affine, with affine
ring A. Then A is a finitely generated k-algebra, and by (3.2), K is the
quotient field of A, and U is the set of localizations of A at its maximal
ideals. Since these local rings are all discrete valuation rings, U consists in
fact of all discrete valuation rings of Kjk containing A. Now let xt> ... ,x"

n
be a set of generators of A over k. Then A s; Rp if and only if x 1 , . . . ,
Xn E Rp. Thus u = U;, where U; = {P E CKixi E Rp }. But by (6.5),
{P E CKix; ¢ Rp} is a finite set. Therefore· each U; and hence also U is open.
So we have shown that the U defined above is an abstract nonsingular
curve. To show that <p is an isomorphism, we need only check that the
regular functions on any open set are the same. But this follows from the
definition of the regular functions on U and the fact that for any open set
v s; Y, G(V) = nPeV GP,Y·

Now we need a result about extensions of morphisms from curves to

projective varieties, which is interesting in its own right.

Proposition 6.8. Let X be an abstract nonsingular curve, let P E X, let Y be

a projective variety, and let <p:X - P---> Y be a morphism. Then there
exists a unique morphism q5: X ---> Y extending <p.

PROOF. Embed Y as a closed subset of P" for some n. Then it will be suffi-
cient to show that <p extends to a morphism of X into P", because if it does,
the image is necessarily contained in Y. Thus we reduce to the case Y = P".
Let pn have homogeneous coordinates x 0 , . . . ,x", and let U be the open
set where x 0 , . . . ,xn are all nonzero. By using induction on n, we may
assume that <p(X - P) n U =/= 0. Because if <p(X - P) n U = 0. then
<p(X - P) s; P" - U. But pn - U is the union of the hyperplanes H;
defined by X; = 0. Since <p(X - P) is irreducible, it must be contained in
H; for some i. Now H; ~ pn-1, so the result would follow by induction.
So we will assume that <p(X - P) n U =1- 0.
For each i,j, x;/xi is a regular function on U. Pulling it back by <p, we
obtain a regular function fii on an open subset of X, which we view as a
rational function on X, i.e., fii E K, where K is the function field of X.
Let v be the valuation of K associated with the valuation ring Rp. Let
r; = v(f;o), i = 0,1, ... ,n, r; E Z. Then since x;/xi = (x;/x 0 )/(x)x 0 ), we have

i,j = 0, ... ,n.

43
I Varieties

Choose k such that rk is minimal among r 0 , . .. ,rn- Then v(};d ~ 0 for all i,
hence j~b ... Jnk E Rp. Now define ip(P) = (f0k(P), ... ,j,k(P) ), and ip(Q) =
cp(Q) for Q -=f. P. I claim that ip is a morphism of X to pn which extends cp,
and that ip is unique. The uniqueness is clear by construction (it also follows
from (4.1) ). To show that ip is a morphism, it will be sufficient to show that
regular functions in a neighborhood of ip(P) pull back to regular functions
on X. Let uk <:; pn be the open set where xk -=f. 0. Then ip(P) E ub since
hk(P) = 1. Now Uk is affine, with affine coordinate ring equal to

k[ x 0/xb ... ,xn/xk].

These functions pull back to fob ... ,j,k which are regular at P by con-
struction. It follows immediately that for any smaller neighborhood ip(P) E
V <:; Ub regular functions on V pull back to regular functions on X. Hence
ip is a morphism, which completes the proof.

Now we come to our main result.

Theorem 6.9. Let K be a function field of dimension 1 over k. Then the

abstract nonsingular curve CK defined above is isomorphic to a nonsingular
projective curve.

PROOF. The idea of the proof is this: we first cover C = CK with open
subsets Ui which are isomorphic to nonsingular affine curves. Let Y; be
the projective closure of this affine curve. Then we use (6.8) to define a
morphism <pi: C --> Y;. Next, we consider the product mapping cp: C --> flY;,
and let Y be the closure of the image of C. Then Y is a projective curve,
and we show that cp is an isomorphism of C onto Y.
To begin with, let P E C be any point. Then by (6.6) there is a nonsingular
affine curve V and a point Q E V with Rp ~ (IJQ· It follows that the function
field of Vis K, and then by (6.7}, Vis isomorphic to an open subset of C.
Thus we have shown that every point P E C has an open neighborhood
which is isomorphic to an affine variety.
Since C is quasi-compact, we can cover it with a finite number of open
subsets Ui, each of which is isomorphic to an affine variety V;. Embed
V; <:; A"', think of An' as an open subset of pn,, and let Y; be the closure of
V; in P"'. Then Y; is a projective variety, and we have a morphism tp;: U; --> Y;
which is an isomorphism of Ui onto its image.
By (6.8) applied to the finite set of points C - U;, we can find a morphism
i[5;: C --> Y; extending <pi. Let f1 Y; be the product of the projective varieties
Y; (Ex. 3.16). Then f1Y; is also a projective variety. Let cp:C--> f1Y; be the
"diagonal" map cp(P) = f1ip;(P), and let Y be the closure of the image of
<p. Then Y is a projective variety, and cp: C --> Y is a morphism whose
image is dense in Y. (It follows that Y is a curve.)
Now we must show that cp is an isomorphism. For any point P E C, we
have P E U; for some i. There is a commutative diagram

44
 6 Nonsingular Cunt

c y

J
of dominant morphisms, where n is the projection map onto the ith factor.
Thus we have inclusions of local rings
(!) rp;(P),Y; ~ (!) rp(P),Y ~ (!) P,C

by (Ex. 3.3). The two outside ones are isomorphic, so the middle one is
also. Thus we see that for any P E C, the map cpp: (!) rp(P),Y --+ (!) P,c is an
isomorphism.
Next, let Q be any point of Y. Then (!)Q is dominated by some discrete
valuation ring R of Kjk (take for example a localization of the integral
closure of (!)Qat a maximal ideal). But R = Rp for some P E C, and (l)rp(Pl ~
R, so by (6.4) we must have Q = cp(P). This shows that cp is surjective.
But cp is clearly injective, because distinct points of C correspond to distinct
subrings of K.
Thus cp is a bijective morphism of C to Y, and for every P E C, cpp is an
isomorphism, so by (Ex. 3.3b), cp is an isomorphism.

Corollary 6.10. Every abstract nonsingular curve is isomorphic to a quasi-

projective curve. Every nonsingular quasi-projective curve is isomorphic
to an open subset of a nonsingular projective curve.

Corollary 6.11. Every curve is birationally equivalent to a nonsingular pro-

jective curve.
PROOF. Indeed, if Y is any curve, with function field K, then Y is birationally
equivalent to CK which is nonsingular and projective.

Corollary 6.12. The following three categories are equivalent:

(i) nonsingular projective curves, and dominant morphisms;
(ii) quasi-projective curves, and dominant rational maps;
(iii) function fields of dimension 1 over k, and k-homomorphisms.
PROOF. We have an obvious functor from (i) to (ii). We have the functor
Y --+ K( Y) from (ii) to (iii), which induces an equivalence of categories by
(4.4). To complete the cycle, we need a functor from (iii) to (i).
To a function field K, associate the curve CK, which by the theorem is a
projective nonsingular curve. If K 2 --+ K 1 is a homomorphism, then by
(ii) ~ (iii), it induces a rational map of the corresponding curves. This can
be represented by a morphism cp: U --+ CK 2 , where U ~ CK, is an open
subset. By (6.8) cp extends to a morphism cp:CK,--+ CK 2 • If K 3 --+ K 2 --+ K 1

45
I Varieties

are two homomorphisms, it follows from the uniqueness part of (6.8) that
the corresponding morphisms C 1 -+ C 2 -+ C 3 and C 1 -+ C 3 are compatible.
Hence K ~ CK is a functor from (iii) -+ (i). It is clearly inverse to the given
functor (i) -+ (ii) -+ (iii), so we have an equivalence of categories.

EXERCISES

6.1. Recall that a curve is rational if it is birationally equivalent to P 1 (Ex. 4.4). Let Y
be a nonsingular rational curve which is not isomorphic to P 1 .
(a) Show that Y is isomorphic to an open subset of A 1 .
(b) Show that Y is affine.
(c) Show that A(Y) is a unique factorization domain.
6.2. An Elliptic Curve. Let Y be the curve y 2 = x 3 - x in A 2 , and assume that the
characteristic of the base field k is =1= 2. In this exercise we will show that Y is not a
rational curve, and hence K(Y) is not a pure transcendental extension of k.
(a) Show that Yisnonsingular,anddeducethatA = A(Y):::::: k[x,y]j(y 2 - x 3 + x)
is an integrally closed domain.
(b) Let k[x] be the subring of K = K(Y) generated by the image of x in A. Show
that k[ x] is a polynomial ring, and that A is the integral closure of k[ x] in K.
(c) Show that there is an automorphism u: A ->A which sends y to - y and leaves
x fixed. For any a E A, define the norm of a to be N(a) = a· u(a). Show that
N(a) E k[x], N(l) = 1, and N(ab) = N(a) · N(b) for any a,b EA.
(d) Using the norm, show that the units in A are precisely the nonzero elements of
k. Show that x and y are irreducible elements of A. Show that A is not a
unique factorization domain.
(e) Prove that Y is not a rational curve (Ex. 6.1). See (II, 8.20.3) and (III, Ex. 5.3)
for other proofs of this important result.

6.3. Show by example that the result of (6.8) is false if either (a) dim X :;:, 2, or (b) Y is
not projective.
6.4. Let Y be a nonsingular projective curve. Show that every nonconstant rational
function f on Y defines a surjective morphism tp: Y -> P 1, and that for every P E P 1,
tp - 1 (P) is a finite set of points.

6.5. Let X be a nonsingular projective curve. Suppose that X is a (locally closed)

subvariety of a variety Y (Ex. 3.10). Show that X is in fact a closed subset of Y.
See (II, Ex. 4.4) for generalization.
6.6. Automorphisms ofP 1 • Think of P 1 as A 1 u { oo }. Then we define a fractional
linear transformation of P 1 by sending x f-> (ax + b)j(cx + d), for a,b,c,d E k,
ad - be =1= 0.
(a) Show that a fractional linear transformation induces an automorphism of P 1
(i.e., an isomorphism of P 1 with itself). We denote the group of all these
fractional linear transformations by PGL(l).
(b) Let Aut P 1 denote the group of all automorphisms of P 1 . Show that Aut P 1 ::::::
Aut k(x), the group of k-automorphisms of the field k(x).
(c) Now show that every automorphism of k(x) is a fractional linear transforma-
tion, and deduce that PGL(l)-> Aut P 1 is an isomorphism.

46
 7 Intersections in Projective Space

Note: We will see later (II, 7.1.1) that a similar result holds for P": every automor-
phism is given by a linear transformation of the homogeneous coordinates.
6.7. LetP 1 , . . . ,P, Q1 , . . . ,Qs be distinct points of A 1 . IfA 1 - {P 1 , . . . ,P,} is isomor-
phic to A 1 - {Q 1 , . . . ,Qs}, show that r = s. Is the converse true? Cf. (Ex. 3.1).

7 Intersections in Projective Space

The purpose of this section is to study the intersection of varieties in a

projective space. If Y, Z are varieties in pn, what can one say about Y n Z?
We have already seen (Ex. 2.16) that Y n Z need not be a variety. But it
is an algebraic set, and we can ask first about the dimensions of its irreducible
components. We take our cue from the theory of vector spaces: if U,V are
subspaces of dimensions r,s of a vector space W of dimension n, then
U n V is a subspace of dimension ~ r + s - n. Furthermore, if U and V
are in sufficiently general position, the dimension of U n V is equal to
r + s - n (provided r + s - n ~ 0). This result on vector spaces imme-
diately implies the analogous result for linear subspaces of pn (Ex. 2.11).
Our first result in this section will be to prove that if Y,Z are subvarieties of
dimensions r,s ofPn, then every irreducible component of Y n Z has dimen-
sion ~ r + s - n. Furthermore, if r + s - n ~ 0, then Y n Z is nonempty.
Knowing something about the dimension of Y n Z, we can ask for more
precise information. Suppose for example that r + s = n, and that Y n Z
is a finite set of points. Then we can ask, how many points are there? Let
us look at a special case. If Y is a curve of degree din P 2 , and if Z is a line
in P 2 , then Y n Z consists of at most d points, and the number comes to d
exactly if we count them with appropriate multiplicities (Ex. 5.4). This
result generalizes to the well-known theorem of Bezout, which says that if
Y,Z are plane curves of degrees d,e, with Y -:f. Z, then Y n Z consists of
de points, counted with multiplicities. We will prove Bezout's theorem
later in this section (7.8).
The ideal generalization of Bezout's theorem to pn would be this. First,
define the degree of any projective variety. Let Y,Z be varieties of dimen-
sions r,s, and of degrees d,e in pn. Assume that Y and Z are in a sufficiently
general position so that all irreducible components of Y n Z have di-
mension = r + s - n, and assume that r + s - n ~ 0. For each ir-
reducible component W of Y n Z, define the intersection multiplicity
i(Y,Z; W) of Y and Z along W. Then we should have

l)(Y,Z; W) · deg W = de,

where the sum is taken over all irreducible components of Y n Z.

The hardest part of this generalization is the correct definition of the
intersection multiplicity. (And, by the way, historically it took many at-
tempts before a satisfactory treatment was given by Severi [3] geometrically

47
I Varieties

and by Chevalley [1] and Wei! [1] algebraically). We will define the inter-
section multiplicity only in the case where Z is a hypersurface. See Appendix
A for the general case.
Our main task in this section will be the definition of the degree of a
variety Y of dimension r in P". Classically, the degree of Y is defined as the
number of points of intersection of Y with a sufficiently general linear space
L of dimension 11 - r. However, this definition is difficult to use. Cutting
Y successively with r sufficiently general hyperplanes, one can find a
linear space L of dimension 11 - r which meets Y in a finite number of
points (Ex. 1.8). But the number of intersection points may depend on L,
and it is hard to make precise the notion "sufficiently general."
Therefore we will give a purely algebraic definition of degree, using the
Hilbert polynomial of a projective variety. This definition is less geo-
metrically motivated, but it has the advantage of being precise. In an
exercise we show that it agrees with the classical definition in a special case
(Ex. 7.4).

Proposition 7.1 (Affine Dimension Theorem). Let Y,Z be varieties of dimen-

sions r,s in A". Then every irreducible component W of Y n Z has
dimension ;?! r + s - n.
PROOF. We proceed in several steps. First, suppose that Z is a hypersurface,
defined by an equation f = 0. If Y c::::: Z, there is nothing to prove. If
Y ¢. Z, we must show that each irreducible component W of Y n Z has
dimension r - 1. Let A( Y) be the affine coordinate ring of Y Then the
irreducible components of Y n Z correspond to the minimal prime ideals
p of the principal ideal (f) in A(Y). Now by Krull's Hauptidealsatz (1.11A),
each such p has height one, so by the dimension theorem (1.8A), A( Y)/p
has dimension r - 1. By (1.7) this shows that each irreducible component
W has dimension r - 1.
Now for the general case. We consider the product Y x Z c::::: A 2 ",
which is a variety of dimension r + s (Ex. 3.15). Let L1 be the diagonal
{P x PIPE A"} c::::: A 2 ". ThenA"isisomorphicto.dbythemapP ~ P x P,
and under this isomorphism, Y n Z corresponds to ( Y x Z) n .d. Since
~ has dimension n, and since r + s - n = (r + s) + n - 2n, we reduce
to proving the result for the two varieties Y x Z and L1 in A2 ". Now L1 is
an intersection of exactly n hypersurfaces, namely, x 1 - .h = 0, ... ,x" -
Yn = 0, where xb . . . ,xn, y 1 , . . . ,yn are the coordinates of A 2 ". Now ap-
plying the special case above n times, we have the result.

Theorem 7.2 (Projective Dimension Theorem). Let Y,Z be varieties of dimen-

sions r,s in P". Then every irreducible component of Y n Z has dimension
;?! r + s - n. Furthermore, if r + s - n ;?! 0, then Y n Z is nonempty.

PROOF. The first statement follows from the previous result, since P" is
covered by affine n-spaces. For the second result, let C(Y) and C(Z) be the

48
 7 Intersections in Projective Space

cones over Y,Z in A"+ 1 (Ex. 2.10). Then C(Y), C(Z) have dimensions r + 1,
s + 1, respectively. Furthermore, C(Y) n C(Z) # 0, because both contain
the origin P = (0, ... ,0). By the affine dimension theorem, C(Y) n C(Z) has
dimension ~(r + 1) + (s + 1) - (n + 1) = r + s - n + 1 > 0. Hence
C(Y) n C(Z) contains some point Q # P, and soY n Z # 0.

Next, we come to the definition of the Hilbert polynomial of a projective

variety. The idea is to associate to each projective variety Y s;: Pi: a poly-
nomial Py E Q[ z] from which we can obtain various numerical invariants
of Y. We will define Pr starting from the homogeneous coordinate ring S(Y).
In fact, more generally, we will define a Hilbert polynomial for any graded
S-module, where S = k[x 0 , . . . ,xnJ. Although the next few results are almost
pure algebra, we include their proofs, for lack of a suitable reference.

Definition. A numerical polynomial is a polynomial P(z) E Q[ z] such that

P(n) E Z for all n » 0, n E Z.

Proposition 7.3.
(a) If P E Q[ z] is a numerical polynomial, then there are integers
c0 ,ct. ... ,c, such that

where

(z)
r
= J_ z(z
r!
- 1) · · · (z - r + 1)

is the binomial coefficient function. In particular P(n) E Z for all n E Z.

(b) If f:Z --+ Z is any function, and if there exists a numerical poly-
nomial Q(z) such that the difference function LJf = f(n + 1) - f(n) is equal
to Q(n) for all n » 0, then there exists a numerical polynomial P(z) such
that f(n) = P(n) for all n » 0.

PROOF.
(a) By induction on the degree of P, the case of degree 0 being obvious.
Since (;) = z'/r! + ... , we can express any polynomial P E Q[ z] of degree r
in the above form, with c0 , ... ,c, E Q. For any polynomial P we define the
difference polynomialLJP by LJP(z) = P(z + 1) - P(z). Since L1 (~) = {r..: d,

L1P = c (r ~ 1) + (r ~ 2) + ... +
0 C1 cr-1·

By induction, c0 , ... ,c,_ 1 E Z. But then c, E Z since P(n) E Z for n » 0.

(b) Write

Q = c0 (:) + ... + c,

49
I Varieties

with c0 , ... ,c, E Z. Let

Then L1P = Q, so L1(f - P)(n) = 0 for all n » 0, so (f - P)(n) =

constant c,+ 1 for all n » 0, so
f(n) = P(n) + c,+ 1
for all n » 0, as required.

Next, we need some preparations about graded modules. Let S be a

graded ring (cf. §2). A graded S-module is an S-module M, together with a
decomposition M = (£ldEZ Ma, such that Sa· Me ~ Md+e· For any graded
S-module M, and for any IE Z, we define the twisted module M(l) by shifting
l places to the left, i.e., M(/)d = Md+l· If M is a graded S-module, we define
the annihilator of M, Ann M = {s E Sis· M = 0} .. This is a homogeneous
ideal inS.
The next result is the analogue for graded modules of a well-known result
for modules of finite type over a noetherian ring (Bourbaki [1, Ch. IV,
§1, no. 4] or Matsumura [2, p. 51]). Again, we include the proof for lack
of an adequate reference.

Proposition 7.4. Let M be a.finitely generated graded module over a noetherian

graded ringS. Then there exists a .filtration 0 = M 0 ~ M 1 ~ . . . ~ M' =
M by graded submodules, such that for each i, Mi/Mi- 1 ~ (S/p;)(l;),
where Pi is a homogeneous prime ideal of S, and liE Z. The filtration is
not unique, but for any such .filtration we do have:
(a) if p is a homogeneous prime ideal of S, then .p ~ Ann M <o> .p ~ .Pi
for some i. In particular, the minimal elements of the set {.p b . . . ,.p,} are
just the minimal primes of M, i.e., the primes which are minimal containing
AnnM;
(b) for each minimal prime of M, the number of times which .p occurs
in the set {p 1 , . . . ,p,} is equal to the length of Mp over the local ring Sp
(and hence is independent of the filtration).
PROOF. For the existence of the filtration, we consider the set of graded
submodules of M which admit such a filtration. Clearly, the zero module
does, so the set is nonempty. M is a noetherian module, so there is a maximal
such submodule M' ~ M. Now consider M" = MjM'. If M" = 0, we are
done. If not, we consider the set of ideals -3 = {Im = Ann(m)lm EM" is a
homogeneous element, m #- 0}. Each Im is a homogeneous ideal, and Jm #- S.
Since S is a noetherian ring, we can find an element mE M", m #- 0, such
that Im is a maximal element of the set -3. I claim that Im is a prime ideal.
Let a,b E S. Suppose that abE Jm, but b ¢ Im. We wish to show a E Jm. By
splitting into homogeneous components, we may assume that a,b are homo-
geneous elements. Now consider the element bm EM". Since b ¢ Im,

50
 7 Intersections in Projective Space

bm =1- 0. We have Im ~ Ibm' so by maximality of lm, Im = Ibm· But abE Im,

so abm = 0, so a E Ibm = Im as required. Thus Im is a homogeneous prime
ideal of S. Call it p. Let m have degree l. Then the module N ~ M" generated
by m is isomorphic to (S/p )( -1). Let N' ~ M be the inverse image of N in M.
Then M' ~ N', and N'/M' ~ (S/p)( -1). So N' also has a filtration of the
type required. This contradicts the maximality of M'. We conclude that M'
was equal to M, which proves the existence of the filtration.
Now suppose given such a filtration of M. Then it is clear that p 2
Ann M <o:> p 2 Ann(M;/M;- 1 ) for some i. But Ann((S/p;)(l)) = p; so this
proves (a).
To prove (b) we localize at a minimal prime p. Since p is minimal in the
set {PI> ... ,p,}, after localization, we will have M!, = M/,- 1 except in the
cases where P; = p. And in those cases M~/M/,- 1 ~ (S/p)p = k(p), the
quotient field of Sjp (we forget the grading). This shows that MP is an
Sp-module of finite length equal to the number of times p occurs in the set
{p1, ... ,p,}.

Definition. If p is a minimal prime of a graded S-module M, we define the

multiplicity of M at p, denoted flp(M), to be the length of MP over Sp.
Now we can define the Hilbert polynomial of a graded module Mover the
polynomial ring S = k[ x 0 , . . . ,xnJ. First, we define the Hilbert function
CfJM of M, given by

for each l E Z.

Theorem 7.5. (Hilbert-Serre). Let M be a finitely generated graded S =

k[ x 0 , . . . ,xnJ-module. Then there is a unique polynomial PM(z) E Q[ z]
such that cpM(l) = PM(l) for all l » 0. Furthermore, deg PM(z) =
dim Z(Ann M), where Z denotes the zero set in pn of a homogeneous
ideal (cf. §2).
PROOF. If 0---+ M' ---+ M ---+ M" ---+ 0 is a short exact sequence, then CfJM =
CfJM' + CfJM"' and Z(Ann M) = Z(Ann M') u Z(Ann M"), so if the theorem
is true forM' and M", it is also true forM. By (7.4), M has a filtration with
quotients of the form (Sjp)(l) where p is a homogeneous prime ideal, and
l E Z. So we reduce to M ~ (S/p)(l), The shift l corresponds to a change
of variables z f--+ z + l, so it is sufficient to consider the case M = Sjp. If
p = (x 0 , . , . ,xn), then CfJM(l) = 0 for l > 0, so PM = 0 is the corresponding
polynomial, and deg PM = dim Z(p), where we make the convention that
the zero polynomial has degree -1, and the empty set has dimension - 1.
If p =1- (x 0 , . . . ,xn), choose X; ¢ p, and consider the exact sequence
0---+ M ~ M---+ M" ---+ 0, where M" = M/x;M, Then CfJM"(l) = CfJM(l) -
CfJM(l- 1) = (LJcpM)(l- 1). On the other hand, Z(Ann M") = Z(p) n H,
where His the hyperplane X; = 0, and Z(p) '*' H by choice of X;, so by (7.2),
dim Z(Ann M") = dim Z(p) - 1. Now using induction on dim Z(Ann M),

51
I Varieties

we may assume that CfJM" is a polynomial function, corresponding to a poly-

nomial PM" of degree = dim Z(Ann M"). Now, by (7.3), it follows that CfJM
is a polynomial function, corresponding to a polynomial of degree =
dim Z(p). The uniqueness of PM is clear.

Definition. The polynomial PM of the theorem is the Hilbert polynomial of M.

Definition. If Y <:; P" is an algebraic set of dimension r, we define the Hilbert

polynomial of Y to be the Hilbert polynomial Py of its homogeneous
coordinate ring S(Y). (By the theorem, it is a polynomial of degree r.)
We define the degree of Y to be r! times the leading coefficient of Py.

Proposition 7.6.
(a) If Y <:; P", Y =I= 0, then the degree of Y is a positive integer.
(b) Let Y = Y1 u Y2 , where Y1 and Y2 have the same dimension r, and
where dim(Y1 n Y2 ) < r. Then deg Y = deg Y1 + deg Y2 •
(c) deg P" = 1.
(d) If H <:; P" is a hypersurface whose ideal is generated by a homo-
geneous polynomial of degree d, then deg H = d. (In other words, this
definition of degree is consistent with the degree of a hyperswface as defined
earlier (1.4.2).)
PROOF.
(a) Since Y =I= 0, Py is a nonzero polynomial of degree r = dim Y. By
(7.3a), deg Y = c0 , which is an integer. It is a positive integer because for
l » 0, Py(l) = CfJs1Al) ~ 0.
(b) Let It. I 2 be the ideals of Y1 and Y2 . Then I = I 1 n I 2 is the ideal of
Y. We have an exact sequence
0---> Sji ~ S/I 1 EB S/I 2 ---> S/(1 1 + I 2 )---> 0.
NowZ(I 1 + I 2 ) = Y1 n Y2 ,whichhassmallerdimension . HencePs;u,+J 2 l
has degree <r. So the leading coefficient of Ps 11 is the sum of the leading
coefficients of Ps 11 , and Ps;lo'
(c) We calculate the Hilbert polynomial of Pn. It is the polynomial Ps,
where s = k[xo, 0 ,xnJ. For l > 0, C(Js(l) = e~"), soPs = (Z~"). In partic-
0 0

ular, its leading coefficient is 1/n !, so deg P" = 1.

(d) Iff E Sis homogeneous of degree d, then we have an exact sequence
of graded S-modules
J
0 ---> S(- d) ---> S ---> S/(f) ---> 0.
Hence
CfJs;(fj(/) = CfJs(l) - CfJsU - d).
Therefore we can find the Hilbert polynomial of H, as

PH(z) = (z: n)- (z- ~ + n) = (n ~ 1)! zn-1 + ....

Thus deg H = d.

52
 7 Intersections in Projective Space

Now we come to our main result about the intersection of a projective

variety with a hypersurface, which is a partial gen-eralization of Bezout's
theorem to higher projective spaces. Let Y c:; P" be a projective variety
of dimension r. Let H be a hypersurface not containing Y Then, by (7.2),
Y n H = Z 1 u ... u Z, where Zi are varieties of dimension r - 1. Let
Pi be the homogeneous prime ideal of Zi. We define the intersection multi-
plicity of Yand H along Zi to be i(Y,H; Z) = f.lp,(Sf(/y + lu)). Here lrJH
are the homogeneous ideals of Y and H. The module M = Sj(/y + /H) has
annihilator /y + IH, and Z(/y + /H) = Y n H, so pi is a minimal prime of
M, and f.1 is the multiplicity introduced above.

Theorem 7.7. Let Y be a variety of dimension ~ 1 in P", and let H be a hyper-

surface not containing Y. Let Z 1, ... ,Zs be the irreducible components
of Yn H. Then
s
L i(Y,H; Z). deg zj = (deg Y)(deg H).
j= 1

PROOF.Let H be defined by the homogeneous polynomial f of degree d.

We consider the exact sequence of graded S-modules
0--> (Sjly)( -d)!.. Sflr--> M--> 0,

where M = Sj(I y + I H). Taking Hilbert polynomials, we find that

PM(z) = Py(z) - Py(z - d).

Our result comes from comparing the leading coefficients of both sides
of this equation. Let Y have dimension r and degree e. Then Py(z) =
(ejr!)z' + ... so on the right we have

(ejr!)z' + ... - [(ejr!)(z - d)' + ... ] = (dej(r - l)!)z'- 1 + ....

Now consider the module M. By (7.4), M has a filtration 0 = M 0 c:; M 1 c:;
... c:; Mq = M, whose quotients Mi/Mi- 1 are of the form (S/q;)(l;). Hence
PM = L1= 1 P;, where P; is the Hilbert polynomial of (S/q;)(l;). If Z(q;) is
a projective variety of dimension r; and degree/;, then
P; = (/;/r; !)z'' + ....
Note that the shift I; does not affect the leading coefficient of P;. Since we
are interested only in the leading coefficient of P;, we can ignore those P;
of degree < r - 1. We are left with those P;, where q; is a minimal prime of
M, namely, one of the primes p 1 , . . . ,ps corresponding to the Zi. Each one
of these occurs f.1p1 (M) times, so the leading coefficient of PM is

(t 1
i(Y,H; Zi) · deg zi)/(r- 1)!

Comparing with the above, we have our result.

53
I Varieties

Corollary 7.8 (Bezout's Theorem). Let Y,Z be distinct curves in P 2 , having

degrees d,e. Let Y n Z = {P 1 , . . . ,P5 }. Then
2)( Y,Z; Pi) = de.
PROOF. We have only to observe that a point has Hilbert polynomial 1,
hence degree 1. See (V, 1.4.2) for another proof.

Remark 7.8.1. Our definition of intersection multiplicity in terms of the

homogeneous coordinate ring is different from the local definition given
earlier (Ex. 5.4). However, it is easy to show that they coincide in the case of
intersections of plane curves.

Remark 7.8.2. The proof of (7.8) extends easily to the case where Y and Z are
"reducible curves," i.e., algebraic sets of dimension 1 in P 2 , provided they
have no irreducible component in common.

EXERCISES

7.1. (a) Find the degree of the d-uple embedding ofP" in pN (Ex. 2.12). [Answer: d"]
(b) Find the degree of the Segre embedding ofP' x ps in pN (Ex. 2.14). [Answer:
('~s)J

7.2. Let Y be a variety of dimension r in P", with Hilbert polynomial Py. We define
the arithmetic genus of Y to be p.(Y) = ( -1)'(Py(0) - 1). This is an important
invariant which (as we will see later in (Ill, Ex. 5.3)) is independent of the projective
embedding of Y
(a) Show that p.(P") = 0.
(b) If Y is a plane curve of degree d, show that p.(Y) = !(d - 1)(d - 2).
(c) More generally, if His a hypersurface of degree din P", then p.(H) = (d~ 1 ).
(d) If Y is a complete intersection (Ex. 2.17) of surfaces of degrees a,b in P 3 , then
p.(Y) = !ab(a + b - 4) + 1.
(e) Let Y' c:; P", zs c:; pm be projective varieties, and embed Y x Z c:; P" x
pm ..... pN by the Segre embedding. Show that

7.3. The Dual Curve. Let Y c:; P 2 be a curve. We regard the set oflines in P 2 as another
projective space, (P 2 )*, by taking (a 0 ,a 1 ,a 2 ) as homogeneous coordinates of the
line L:a 0 x 0 + a 1 x 1 + a2 x 2 = 0. For each nonsingular point P E Y, show that
there is a unique line Tp(Y) whose intersection multiplicity with Y at Pis > 1.
This is the tangent line to Y at P. Show that the mapping P r--> Tp(Y) defines a
morphism of Reg Y (the set of nonsingular points of Y) into (P 2 )*. The closure of
the image of this morphism is called the dual curve Y* c:; (P 2 )* of Y
7.4. Given a curve Y of degree din P 2 , show that there is a nonempty open subset U of
(P 2 )* in its Zariski topology such that for each L E U,L meets Yin exactly d points.
[Hint: Show that the set of lines in (P 2 )* which are either tangent to Y or pass
through a singular point of Y is contained in a proper closed subset.] This result
shows that we could have defined the degree of Y to be the number d such that
almost all lines in P 2 meet Y in d points, where "almost all" refers to a nonempty

54
 8 What Is Algebraic Geometry?

open set of the set of lines, when this set is identified with the dual projective space
(P2)*.

7.5. (a) Show that an irreducible curve Y of degree d > 1 in P 2 cannot have a point of
multiplicity ~ d (Ex. 5.3).
(b) If Y is an irreducible curve of degree d > 1 having a point of multiplicity
d - 1, then Y is a rational curve (Ex. 6.1).

7.6. Linear Varieties. Show that an algebraic set Y of pure dimension r (i.e., every
irreducible component of Y has dimension r) has degree 1 if and only if Y is a
linear variety (Ex. 2.11). [Hint: First, use (7.7) and treat the case dim Y = 1. Then
do the general case by cutting with a hyperplane and using induction.]
7.7. Let Ybe a variety of dimension rand degree d > 1 in P". Let P E Ybe a nonsingular
point. Define X to be the closure of the union of all lines PQ, where Q E Y, Q #- P.
(a) Show that X is a variety of dimension r + 1.
(b) Show that deg X < d. [Hint: Use induction on dim Y.J
7.8. Let Y' <;; P" be a variety of degree 2. Show that Y is contained in a linear subspace
L of dimension r + 1 in P". Thus Y is isomorphic to a quadric hypersurface in
pr+ 1 (Ex. 5.12).

8 What Is Algebraic Geometry?

Now that we have met some algebraic varieties, and have encountered some
of the main concepts about them, it is appropriate to ask, what is this subject
all about? What are the important problems in the field, and where is it
going?
To define algebraic geometry, we could say that it is the study of the
solutions of systems of polynomial equations in an affine or projective
n-space. In other words, it is the study of algebraic varieties.
In any branch of mathematics, there are usually guiding problems, which
are so difficult that one never expects to solve them completely, yet which
provide stimulus for a great amount of work, and which serve as yardsticks for
measuring progress in the field. In algebraic geometry such a problem is the
classification problem. In its strongest form, the problem is to classify all
algebraic varieties up to isomorphism. We can divide the problem into
parts. The first part is to classify varieties up to birational equivalence. As
we have seen, this is equivalent to the question of classifying function fields
(finitely generated extension fields) over k up to isomorphism. The second
part is to identify a good subset of a birational equivalence class, such as the
nonsingular projective varieties, and classify them up to isomorphism. The
third part is to study how far an arbitrary variety is from one of the good
ones considered above. In particular, we want to know (a) how much do you
have to add to a nonprojective variety to get a projective variety, and (b)
what is the structure of singularities, and how can they be resolved to give a
nonsingular variety?

55
I Varieties

Typically, the answer to any classification problem in algebraic geometry

consists of a discrete part and a continuous part. So we can rephrase the
problem as follows: define numerical invariants and continuous invariants
of algebraic varieties, which allow one to distinguish among nonisomorphic
varieties. Another special feature of the classification problem is that often
when there is a continuous family of nonisomorphic objects, the parameter
space can itself be given a structure of algebraic variety. This is a very power-
ful method, because then all the techniques of the subject can be applied to
the study of the parameter space as well as to the original varieties.

Let us illustrate these ideas by describing what is known about the classi-
fication of algebraic curves (over a fixed algebraically closed field k). First,
the birational classification. There is an invariant called the genus of a
curve, which is a birational invariant, and which takes on all nonnegative
values g ~ 0. For g = 0 there is exactly one birational equivalence class,
namely, that of the rational curves (i.e., those curves which are birationally
equivalent to P 1 ). For each g > 0 there is a continuous family of birational
equivalence classes, which can be parametrized by an irreducible algebraic
variety 9Jl9 , called the variety of moduli of curves of genus g, which has di-
mension 1 if g = 1, and dimension 3g - 3 if g ~ 2. Curves with g = 1 are
called elliptic curves. Thus for curves, the birational classification question
is answered by giving the genus, which is a discrete invariant, and a point on
the variety of moduli, which is a continuous invariant. See Chapter IV for
more details.
The second question for curves, namely, to describe all nonsingular pro-
jective curves in a given birational equivalence class, has a simple answer, as
we have seen, since there is exactly one.
For the third question, we know that any curve can be completed to a
projective curve by adding a finite number of points, so there is not much
more to say there. As for the classification of singularities of curves, see
(V, 3.9.4).

While we are discussing the classification problem, I would like to describe

another special case where a satisfactory answer is known, namely, the
classification of nonsingular projective surfaces within a given birational
equivalence class. In this case one knows that (1) every birational equivalence
class of surfaces has a nonsingular projective surface in it, (2) the set of
nonsingular projective surfaces with a given function field Kjk is a partially
ordered set under the relation given by the existence of a birational mor-
phism, (3) any birational morphism f: X --+ Y can be factored into a finite
number of steps, each of which is a blowing-up of a point, and (4) unless K is
rational (i.e., equal to K(P 2 )) or ruled (i.e., K is the function field of a product
P 1 x C, where C is a curve), there is a unique minimal element of this
partially ordered set, which is called the minimal model of the function field K.
(In the rational and ruled cases, there are infinitely many minimal elements,

56
 8 What Is Algebraic Geometry?

and their structure is also well-known.) The theory of minimal models is a

very beautiful branch of the theory of surfaces. The results were known to the
Italians, but were first proved in all characteristics by Zariski [ 5], [ 6]. See
Chapter V for more details.

From these remarks it should be clear that the classification problem is a

very fruitful problem to keep in mind while studying algebraic geometry.
This leads us to the next question: how does one go about defining invariants
of an algebraic variety? So far, we have defined the dimension, and for
projective varieties we have defined the Hilbert polynomial, and hence the
degree and the arithmetic genus Pa· Of course the dimension is a birational
invariant. But the degree and the Hilbert polynomial depend on the em-
bedding in projective space, so they are not even invariants under isomor-
phism of varieties. Now it happens that the arithmetic genus is an invariant
under isomorphism (III, Ex. 5.3), and is even a birational invariant in most
cases (curves, surfaces, nonsingular varieties in characteristic 0; see (V, 5.6.1) ),
but this is not at all apparent from our definition.
To go further, we must study the intrinsic geometry on a variety, which we
have not done at all yet. So, for example, we will study divisors on a variety X.
A divisor is an element ofthe free abelian group generated by the subvarieties
of codimension one. We will define linear equivalence of divisors, and then
we can form the group of divisors modulo linear equivalence, called the
Picard group of X. This is an intrinsic invariant of X. Another very important
notion is that of a differential form on a variety X. Using differential forms,
one can give an intrinsic definition of the tangent bundle and cotangent
bundle on an algebraic variety. Then one can carry over many constructions
from differential geometry to define numerical invariants. For example,
one can define the genus of a curve as the dimension of the vector space of
global differential forms on the nonsingular projective model. From this
definition it is clear that it is a birational invariant. See (II, §6,7,8).
Perhaps the most important modern technique for defining numerical
invariants is by cohomology. There are many cohomology theories, but we
will be principally concerned in this book with the cohomology of coherent
sheaves, which was introduced by Serre [3]. Cohomology is an extremely
powerful and versatile tool. Not only can it be used to define numerical
invariants (for example, the genus of a curve X can be defined as dim
H 1 (X,(()x) ), but it can be used to prove many important results which do not
apparently have any connection with cohomology, such as "Zariski's main
theorem," which has to do with the structure of birational transformations.
To set up a cohomology theory requires a lot of work, but I believe it is well
worth the effort. We will devote a whole chapter to cohomology later in the
book (Chapter III). Cohomology is also a useful vehicle for understanding
and expressing important results such as the Riemann-Roch theorem. This
theorem was known classically for curves and surfaces, but it was by using
cohomology that Hirzebruch [1 J and Grothendieck (see Borel and Serre

57
I Varieties

[1 ]) were able to clarify and generalize it to varieties of any dimension

(Appendix A).

Now that we have seen a little bit of what algebraic geometry is about, we
should discuss the degree of generality in which to develop the foundations
of the subject. In this chapter we have worked over an algebraically closed
field, because that is the simplest case. But there are good reasons for allowing
fields which are not algebraically closed. One reason is that the local ring
of a subvariety on a variety has a residue field which is not algebraically
closed (Ex. 3.13), and at times it is desirable to give a unified treatment of
properties which hold along a subvariety and properties which hold at a
point. Another strong reason for allowing non-algebraically closed fields
is that many problems in algebraic geometry are motivated by number
theory, and in number theory one is primarily concerned with solutions of
equations over finite fields or number fields. For example, Fermat's problem
is equivalent to the question, does the curve x" + y" = z" in P 2 for n ~ 3
have any points rational over Q (i.e., points whose coordinates are in Q),
with x,y,z # 0.
The need to work over arbitrary ground fields was recognized by Zariski
and Weil. In fact, perhaps one of the principal contributions of Weil's
"Foundations" [1] was to provide a systematic framework for studying
varieties over arbitrary fields, and the various phenomena which occur
with change of ground field. Nagata [2] went further by developing the
foundations of algebraic geometry over Dedekind domains.
Another direction in which we need to expand our foundations is to define
some kind of abstract variety which does not a priori have an embedding in an
affine or projective space. This is especially necessary in problems such as the
construction of a variety of moduli, because there one may be able to make
the construction locally, without knowing anything about a global em-
bedding. In §6 we gave a definition of an abstract curve. In higher dimen-
sions that method does not work, because there is no unique nonsingular
model of a given function field. However, we can define an abstract variety
by starting from the observation that any variety has an open covering by
affine varieties. Thus one can define an abstract variety as a topological
space X, with an open cover Ui, plus for each Ui a structure of affine variety,
such that on each intersection Ui n Ui the induced variety structures are
isomorphic. It turns out that this generalization of the notion of variety is
not illusory, because in dimension ~ 2 there are abstract varieties which
are not isomorphic to any quasi-projective variety (II, 4.10.2).
There is a third direction in which it is useful to expand our notion of
algebraic variety. In this chapter we have defined a variety as an irreducible
algebraic set in affine or projective space. But it is often convenient to allow
reducible algebraic sets, or even algebraic sets with multiple components.
For example, this is suggested by what we have seen of intersection theory
in §7, since the intersection of two varieties may be reducible, and the sum

58
 8 What Is Algebraic Geometry?

of the ideals of the two varieties may not be the ideal of the intersection. So
one might be tempted to define a "generalized projective variety" in P" to
be an ordered pair <V,J), where V is an algebraic set in P", and I s; S =
k[x 0 , . •• ,xnJ is any ideal such that V = Z(J). This is not in fact what we will
do, but it gives the general idea.
All three generalizations of the notion of variety suggested above are
contained in Grothendieck's definition of a scheme. He starts from the
observation that an affine variety corresponds to a finitely generated integral
domain over a field (3.8). But why restrict one's attention to such a special
class of rings? So for any commutative ring A, he defines a topological space
Spec A, and a sheaf of rings on Spec A, which generalizes the ring of regular
functions on an affine variety, and he calls this an affine scheme. An arbitrary
scheme is then defined by glueing together affine schemes, thus generalizing
the notion of abstract variety we suggested above.
One caution about working in extreme generality. There are many ad-
vantages to developing a theory in the most general context possible. In
the case of algebraic geometry there is no doubt that the introduction of
schemes has revolutionized the subject and has made possible tremendous
advances. On the other hand, the person who works with schemes has to
carry a considerable load of technical baggage with him: sheaves, abelian
categories, cohomology, spectral sequences, and so forth. Another more
serious difficulty is that some things which are always true for varieties may
no longer be true. For example, an affine scheme need not have finite di-
mension, even if its ring is noetherian. So our intuition must be supported
by a good knowledge of commutative algebra.
In this book we will develop the foundations of algebraic geometry using
the language of schemes, starting with the next chapter.

59
CHAPTER II

Schemes

This chapter and the next form the technical heart of this book. In this
chapter we develop the basic theory of schemes, following Grothendieck
[EGA]. Sections 1 to 5 are fundamental. They contain a review of sheaf
theory (necessary even to define a scheme), then the basic definitions of
schemes, morphisms, and coherent sheaves. This is the language that we use
for the rest of the book.
Then in Sections 6, 7, 8, we treat some topics which could have been done
in the language of varieties, but which are already more convenient to discuss
using schemes. For example, the notion of Cartier divisor, and of an in-
vertible sheaf, which belong to the new language, greatly clarify the dis-
cussion ofWeil divisors and linear systems, which belong to the old language.
Then in §8, the systematic use of nonclosed scheme points gives much more
flexibility in the discussion of sheaves of differentials and nonsingular
varieties, improving the treatment of (1, §5).
In §9 we give the definition of a formal scheme, which did not have an
analogue in the theory of varieties. It was invented by Grothendieck as a
good way of dealing with Zariski's theory of"holomorphic functions," which
Zariski regarded as an analogue in abstract algebraic geometry of the
holomorphic functions in a neighborhood of a subvariety in the classical case.

1 Sheaves

The concept of a sheaf provides a systematic way of keeping track of local

algebraic data on a topological space. For example, the regular functions
on open subsets of a variety, introduced in Chapter I, form a sheaf, as we will
see shortly. Sheaves are essential in the study of schemes. In fact, we cannot

60
 1 Sheaves

even define a scheme without using sheaves. So we begin this chapter with
sheaves. For additional information, see the book of Godement [1].

Definition. Let X be a topological space. A presheaf :Ji' of abelian groups on

X consists of the data
(a) for every open subset U s; X, an abelian group :Ji'(U), and
(b) for every inclusion V s; U of open subsets of X, a morphism of
abelian groups Puv::!i'(U)--+ :!i'(V),
subject to the conditions
(0) :!i'(0) = 0, where 0 is the empty set,
(1) Puu is the identity map :Ji'(U) --+ :Ji'(U), and
(2) if W s; V s; U are three open subsets, then Puw = Pvw o Puv·

The reader who likes the language of categories may rephrase this defi-
nition as follows. For any topological space X, we define a category 'Iop(X),
whose objects are the open subsets of X, and where the only morphisms are
the inclusion maps. Thus Hom(V,U) is empty if V <J._ U, and Hom(V,U)
has just one element if V s; U. Now a presheaf is just a contravariant
functor from the category 'Iop(X) to the category 'lib of abelian groups.
We define a presheaf of rings, a presheaf of sets, or a presheaf with values
in any fixed category <I, by replacing the words "abelian group" in the
definition by "ring", "set", or "object of <I" respectively. We will stick to
the case of abelian groups in this section, and let the reader make the necessary
modifications for the case of rings, sets, etc.
As a matter of terminology, if :Ji' is a presheaf on X, we refer to :Ji' ( U) as
the sections of the presheaf :Ji' over the open set U, and we sometimes use
the notation r(U,:Ji') to denote the group :Ji'(U). We call the maps Puv
restriction maps, and we sometimes write siv instead of Puv(s), if s E :Ji'(U).

A sheaf is roughly speaking a presheaf whose sections are determined by

local data. To be precise, we give the following definition.

Definition. A presheaf :Ji' on a topological space X is a sheaf if it satisfies

the following supplementary conditions:
(3) if U is an open set, if {V;} is an open covering of U, and if s E :Ji'(U) is
an element such that siv, = 0 for all i, then s = 0;
(4) if U is an open set, if {V;} is an open covering of U, and if we have
elements si E :Ji'( V;) for each i, with the property that for each i,j, s;jv,n vj =
sjiv,nvj' then there is an elements E :Ji'(U) such that siv, = si for each i.
(Note condition (3) implies that sis unique.)

Nate. According to our definition, a sheaf is a presheaf satisfying certain

extra conditions. This is equivalent to the definition found in some other

61
II Schemes

books, of a sheaf as a topological space over X with certain properties

(Ex. 1.13).

Example 1.0.1. Let X be a variety over the field k. For each open set U s; X,
let @(U) be the ring of regular functions from U to k, and for each V s; U, let
Puv:l9(U)--+ l9(V) be the restriction map (in the usual sense). Then (9 is a
sheaf of rings on X. It is clear that it is a presheaf of rings. To verify the
conditions (3) and (4), we note that a function which is 0 locally is 0, and a
function which is regular locally is regular, because of the definition of regular
function (1, §3). We call (9 the sheaf of regular functions on X.

Example 1.0.2. In the same way, one can define the sheaf of continuous real-
valued functions on any topological space, or the sheaf of differentiable
functions on a differentiable manifold, or the sheaf of holomorphic functions
on a complex manifold.

Example 1.0.3. Let X be a topological space, and A an abelian group. We

define the constant sheaf d on X determined by A as follows. Give A the
discrete topology, and for any open set U s; X, let d ( U) be the group of all
continuous maps of U into A. Then with the usual restriction maps, we
obtain a sheaf d. Note that for every connected open set U, d (U) ~ A,
whence the name "constant sheaf." If U is an open set whose connected
components are open (which is always true on a locally connected topological
space), then d ( U) is a direct product of copies of A, one for each connected
component of U.

Definition. If§' is a presheaf on X, and if P is a point of X, we define the

stalk ff'p of ff' at P to be the direct limit of the groups %( U) for all open
sets U containing P, via the restriction maps p.

Thus an element of ff'p is represented by a pair< U,s), where U is an open

neighborhood of P, and sis an element of ff'(U). Two such pairs <U,s) and
<V,t) define the same element of ff'p if and only ifthere is an open neighbor-
hood W of P with W s; U n V, such that siw = tlw· Thus we may speak of
elements of the stalk ff'p as germs of sections of§' at the point P. In the case
of a variety X and its sheaf of regular functions (9, the stalk (9 P at a point P
is just the local ring of P on X, which was defined in (1, §3).

Definition. If ff' and<§ are presheaves on X, a morphism cp:ff'--+ <§consists

of a morphism of abelian groups cp(U): ff'(U)--+ <§(U) for each open set
U, such that whenever V s; U is an inclusion, the diagram

ff'(U) _cp'--"(_U"-)___. <§(U)

]Puv jPuv
%( V) ------'cp_,_(V--')----+ <§ ( V)
62
 1 Sheaves

is commutative, where p and p' are the restriction maps in ff and '!J. If
ff and 'fJ are sheaves on X, we use the same definition for a morphism
of sheaves. An isomorphism is a morphism which has a two-sided inverse.

Note that a morphism cp: ff ~ 'fJ of presheaves on X induces a morphism

q>p:ffp ~ '!Jp on the stalks, for any point P EX. The following proposition
(which would be false for presheaves) illustrates the local nature of a sheaf.

Proposition 1.1. Let cp: ff ~ 'fJ be a morphism of sheaves on a topological

space X. Then cp is an isomorphism if and only if the induced map on the
stalk q>p:ffp ~ '!Jp is an isomorphism for every P EX.
PROOF. If cp is an isomorphism it is clear that each q>p is an isomorphism.
Conversely, assume q>p is an isomorphism for all P EX. To show that cp is
an isomorphism, it will be sufficient to show that cp(U):ff(U) ~ '!J(U) is
an isomorphism for all U, because then we can define an inverse morphism
tjJ by t/J(U) = cp(U)- 1 for each U. First we show cp(U) is injective. Let
s E ff(U), and suppose cp(s) E '!J(U) is 0. Then for every point P E U, the
image cp(s)p of cp(s) in the stalk '!Jp is 0. Since q>p is injective for each P, we
deduce that Sp = 0 in ffp for each P E U. To say that sp = 0 means that s
and 0 have the same image in ffp, which means that there is an open neigh-
borhood Wp of P, with Wp s; U, such that siwp = 0. Now U is covered by
the neighborhoods Wp of all its points, so by the sheaf property (3), s is 0
on U. Thus cp(U) is injective.
Next, we show that cp(U) is surjective. Suppose we have a section t E '!J(U).
For each P E U, let tp E '!Jp be its germ at P. Since q>p is surjective, we can
find Sp E ff P such that q>p(sp) = tp. Let sp be represented by a section s(P)
on a neighborhood Vp of P. Then cp(s(P)) and tlvp are two elements of
'!J(Vp), whose germs at P are the same. Hence, replacing Vp by a smaller
neighborhood of P if necessary, we may assume that cp(s(P)) = tlvp in
'!J(Vp). Now U is covered by the open sets Vp, and on each Vp we have a
section s(P) E ff(Vp). If P,Q are two points, then s(P)ivpnVQ and s(Q)IvpnVQ
are two sections of ff(Vp n VQ), which are both sent by cp to tlvpn vQ· Hence
by the injectivity of cp proved above, they are equal. Then by the sheaf
property (4), there is a section s E ff(U) such that sivp = s(P) for each P.
Finally, we have to check that cp(s) = t. Indeed, cp(s), t are two sections of
'!J(U), and for each P, cp(s)ivp = tlvP' hence by the sheaf property (3) applied
to cp(s) - t, we conclude that cp(s) = t.

Our next task is to define kernels, cokernels and images of morphisms

of sheaves.

Definition. Let cp:ff ~ 'fJ be a morphism of presheaves. We define the

presheaf kernel of cp, presheaf cokernel of cp, and presheaf image of cp to
be the presheaves given by U r-+ ker( cp( U) ), U r-+ coker( cp( U) ), and
U r-+ im(cp(U)) respectively.
63
II Schemes

Note that if qJ:ff---+ '§is a morphism of sheaves, then the presheafkernel

of qJ is a sheaf, but the presheaf cokernel and presheaf image of qJ are in
general not sheaves. This leads us to the notion of a sheaf associated to a
presheaf.

Proposition-Definition 1.2. Given a presheaf JF, there is a sheaf JF+ and a

morphism e:JF---+ JF+, with the property that for any sheaf'§, and any
morphism qJ:Ji!---+ '§, there is a unique morphism t/f:JF+ ---+ '§ such that
qJ = t/J a e. Furthermore the pair (JF+ ,e) is unique up to unique isomorphism.
JF+ is called the sheaf associated to the presheaf JF.
PROOF. We construct the sheaf ff+ as follows. For any open set U, let JF+(U)
be the set of functions s from U to the union UPE u Jl!p of the stalks of JF
over points of U, such that
(1) for each P E U, s(P) E Jl!p, and
(2) for each P E U, there is a neighborhood V of P, contained in U, and an
element t E JF(V), such that for all Q E V, the germ tQ oft at Q is equal
to s(Q).
Now one can verify immediately ( !) that ff+ with the natural restriction
maps is a sheaf, that there is a natural morphism e:JF---+ JF+, and that it
has the universal property described. The uniqueness of JF+ is a formal
consequence of the universal property. Note that for any point P, Ji!p = JF~.
Note also that if JF itself was a sheaf, then ff+ is isomorphic to JF via
e.

Definition. A subsheaf of a sheaf JF is a sheaf ff' such that for every open set
U s X, ff'( U) is a subgroup of ff( U), and the restriction maps of the
sheaf ff' are induced by those of JF. It follows that for any point P, the
stalk ff~ is a subgroup of Ji!p.
If qJ: JF ---+ '§ is a morphism of sheaves, we define the kernel of qJ,
denoted ker qJ, to be the presheaf kernel of qJ (which is a sheaf). Thus
ker qJ is a subsheaf of JF.
We say that a morphism of sheaves qJ:ff---+ '§is injective ifker qJ = 0.
Thus qJ is injective if and only if the induced map qJ(U): ff( U) ---+ '§(U) is
injective for every open set of X.
If qJ: JF ---+ '§ is a morphism of sheaves, we define the image of qJ,
denoted im qJ, to be the sheaf associated to the presheaf image of qJ. By
the universal property of the sheaf associated to a presheaf, there is a
natural map im qJ ---+ '§. In fact this map is injective (see Ex. 1.4), and thus
im qJ can be identified with a subsheaf of'§.
We say that a morphism qJ:Ji!---+ '§of sheaves is surjective ifim qJ = '§.

We say that a sequence ... ---+ Jl!i- 1 ~ Jl!i ~ Jl!i+ 1 ---+ ••• of sheaves
and morphisms is exact if at each stage ker qJi = im qJi- 1 . Thus a sequence

64
 1 Sheaves

0 --+ fi' ~ ':§ is exact if and only if q; is injective, and fi' ~ ':§ --+ 0 is exact
if and only if q; is surjective.
Now let ff'' be a subsheaf of a sheaf ff'. We define the quotient sheaf
fi' /fi'' to be the sheaf associated to the presheaf U --+ fi'(U)/ff'(U). It
follows that for any point P, the stalk (ff' /fi'')p is the quotient Ji'pjfi'~.
If q;: fi' --+ ':§ is a morphism of sheaves, we define the co kernel of q;,
denoted coker q;, to be the sheaf associated to the presheaf cokernel of q;.
Caution 1.2.1. We saw that a morphism q;:ff'--+ ':§of sheaves is injective if
and only if the map on sections q;(U):fi'(U)--+ ':§(U) is injective for each
U. The corresponding statement for surjective morphisms is not true: if
q;:ff'--+ ':§is surjective, the maps q;(U):fi'(U)--+ ':§(U) on sections need not
be surjective. However, we can say that q; is surjective if and only if the maps
q; P: ffp --+ ':§ P on stalks are surjective for each P. More generally, a sequence
of sheaves and morphisms is exact if and only if it is exact on stalks (Ex. 1.2).
This again illustrates the local nature of sheaves.

So far we have talked only about sheaves on a single topological space.

Now we define some operations on sheaves, associated with a continuous
map from one topological space to another.

Definition. Let f: X --+ Y be a continuous map of topological spaces. For

any sheaf fi' on X, we define the direct image sheaf f*Ji' on Y by
(f*fi')(V) = fi'(f- 1 (V)) for any open set V s; Y. For any sheaf ':§ on
Y, we define the inverse image sheaf f- 1 ':§ on X to be the sheaf associated
to the presheaf U ~---+ limv 2 f<UJ ':§(V), where U is any open set in X, and
the limit is taken over all open sets V of Y containingf(U). Do not confuse
f- 1':§ with the sheaf j*':§ which will be defined later for a morphism of
ringed spaces (§5).

Note that f* is a functor from the category ~b(X) of sheaves on X to

the category ~b(Y) of sheaves on Y. Similarly,f- 1 is a functor from ~b(Y)
to ~b(X).

Definition. If Z is a subset of X, regarded as a topological subspace with the

induced topology, if i: Z --+ X is the inclusion map, and if fi' is a sheaf
on X, then we call i- 1 fi' the restriction of fi' to Z, and we often denote
it by ff'lz· Note that the stalk of ff'lz at any point P E Z is just ffp.

EXERCISES

1.1. Let A be an abelian group, and define the constant presheaf associated to A on
the topological space X to be the presheaf U 1--+ A for all U i= 0, with restriction
maps the identity. Show that the constant sheaf sf defined in the text is the sheaf
associated to this presheaf.

65
II Schemes

1.2. (a) Foranymorphismofsheavescp:ff--> ~.showthatforeachpointP,(kercp)p =

ker(cpp) and (im cp)p = im(cpp).
(b) Show that cp is injective (respectively, surjective) if and only if the induced map
on the stalks qJp is injective (respectively, surjective) for all P.
(c) Show that a sequence ... g;•- 1 S ff' ~ g;i+ 1 --> ... of sheaves and mor-
phisms is exact if and only if for each P E X the corresponding sequence of
stalks is exact as a sequence of abelian groups.
1.3. (a) Let cp: ff --> ~ be a morphism of sheaves on X. Show that cp is surjective if
and only if the following condition holds: for every open set U c;; X, and for
every s E ~(U), there is a covering {U;} of U, and there are elements tiE ff(Ui),
such that cp(t;) = siu, for all i.
(b) Give an example of a surjective morphism of sheaves cp:ff--> ~. and an
open set U such that cp(U):ff(U)--> ~(U) is not surjective.

1.4. (a) Let cp:ff--> ~be a morphism ofpresheaves such that cp(U):ff(U)--> ~(U)
is injective for each U. Show that the induced map cp + : ff + --> ~ + of asso-
ciated sheaves is injective.
(b) Use part (a) to show that if cp.:ff--> ~is a morphism of sheaves, then im cp
can be naturally identified with a subsheaf of~. as mentioned in the text.
1.5. Show that a morphism of sheaves is an isomorphism if and only if it is both
injective and surjective.
1.6. (a) Let ff' be a subsheaf of a sheaf ff. Show that the natural map of ff to the
quotient sheaf ff Iff' is surjective, and has kernel ff'. Thus there is an exact
sequence
o ...... ff' ...... g; ...... ff Iff' ...... o.
(b) Conversely, if 0 --> ff' --> ff ...... ff" ...... 0 is an exact sequence, show that ff'
is isomorphic to a subsheaf of ff, and that ff" is isomorphic to the quotient of
ff by this subsheaf.

1.7. Let cp: ff --> ~ be a morphism of sheaves.

(a) Show that im cp ~ ff jker cp.
(b) Show that coker cp ~ ~lim cp.
1.8. For any open subset U <;; X, show that the functor r(U, ·) from sheaves on X to
abelian groups is a left exact functor, i.e., if 0 --> ff' --> ff --> ff" is an exact
sequence of sheaves, then 0 ...... r(U,ff') ...... r(U,ff) ...... r(U,ff") is an exact
sequence of groups. The functor r(U,·) need not be exact; see (Ex. 1.21) below.

1.9. Direct Sum. Let ff and~ be sheaves on X. Show that the presheaf U H ff( U) EB
~(U) is a sheaf. It is called the direct sum of ff and~. and is denoted by ff EB ~.
Show that it plays the role of direct sum and of direct product in the category of
sheaves of abelian groups on X.

1.10. Direct Limit. Let {ff;} be a direct system of sheaves and morphisms on X. We
define the direct limit of the system {ff,}, denoted lim
----+
ffi, to be the sheaf associated
to the presheaf U H ~ ffi(U). Show that this is a direct limit in the category
of sheaves on X, i.e., that it has the following universal property: given a sheaf~.
and a collection of morphisms ffi --> ~. compatible with the maps of the direct

66
 Sheaves

system, then there exists a unique map !i!!!

ff; ---+ '!J such that for each i, the original
map :Y'; ---+ '!J is obtained by composing the maps ff; ---+ !i!!!:Y'; ---+ '!J.
l.ll. Let {ff;} be a direct system of sheaves on a noetherian topological space X. In
this case show that the presheaf U f---+ lim :Y';(U) is already a sheaf. In particular,
T(X,lim :Y';) = lim T(X,:Y';). ----->
-----> ----->

1.12. lnrerse Limit. Let [ff;} be an inverse system of sheaves on X. Show that the pre-
sheaf U f---+ lim ff;(U) is a sheaf. It is called the inverse limit of the system {:Y';},
and is denoted by lim :Y';. Show that it has the universal property of an inverse
limit in the categorYof sheaves.
1.13. Espace Etale of a Presheaf. (This exercise is included only to establish the con-
nection between our definition of a sheaf and another definition often found in
the literature. See for example Godement [1, Ch. II, §1.2].) Given a presheaf §'
on X, we define a topological space SpeC~). called the espace hale of §', as
follows. As a set, Spe(.~) = UPEX :i'p. We define a projection map n:Spe(:Y')---+ X
by sending s E :Y'p to P. For each open set U ~ X and each sections E :Y'( U), we
obtain a maps: U ---+ Spe(:Y') by sending P f---+ sp, its germ at P. This map has the
property that n c s = idu, in other words, it is a "section" of n over U. We now
make SpeC~) into a topological space by giving it the strongest topology such that
all the maps s: U ---+ Spe(.¥) for all U, and all s E :Y'( U), are continuous. Now
show that the sheaf y;+ associated to F can be described as follows: for any
open set U ~ X, §' + ( U) is the set of continuous sections of Spe(.~) over U. In
particular, the original presheaf §'was a sheaf if and only if for each U, :Y'(U) is
equal to the set of all continuous sections of Spe(.~) over U.
1.14. Support. Let ff be a sheaf on X, and lets E ff(U) be a section over an open set U.
The support of s, denoted Supp s, is defined to be {P E Uisp ¥= 0}, where sp
denotes the germ of sin the stalk ffp. Show that Supp sis a closed subset of U.
We define the support of :Y', Supp ff, to be {P E Xiffp ¥= 0}. It need not be a
closed subset.
1.15. Sheaf X om. Let§', '!J be sheaves of abelian groups on X. For any open set U ~ X,
show that the set Hom(fflu,'!ilu) of morphisms of the restricted sheaves has a
natural structure of abelian group. Show that the presheaf U f---+ Hom(fflu,'!Jiu)
is a sheaf. It is called the sheaf of local morphisms of ff into '!J, "sheaf hom" for
short, and is denoted Xom(ff,'!J).
1.16. Flasque Sheaves. A sheaf ff on a topological space X is jlasque if for every in-
clusion V ~ U of open sets, the restriction map ff(U) ---+ :Y'(V) is surjective.
(a) Show that a constant sheaf on an irreducible topological space is flasque. See
(I, §I) for irreducible topological spaces.
(b) If 0 ---+ ff' ---+ ff ---+ §'" ---+ 0 is an exact sequence of sheaves, and if §'' is
ftasque, then for any open set U, the sequence 0---+ ff'(U)---+ ff(U)---+
§'"( U) ---+ 0 of abelian groups is also exact.
(c) If 0 ---+ ff' ---+ ff ---+ .~" ---+ 0 is an exact sequence of sheaves, and if ff' and§'
are ftasque, then §'" is ftasque.
(d) If f:X ---+ Y is a continuous map, and if ff is a ftasque sheaf on X, then f*:Y'
is a ftasque sheaf on Y.
(e) Let ff be any sheaf on X. We define a new sheaf'!J, called the sheaf of discon-
tinuous sections of ff as follows. For each open set U ~ X, '!J(U) is the set of

67
II Schemes

maps s: U ---> UPeu :!'p such that for each P E U, s(P) E ffp. Show that '§
is a fiasque sheaf, and that there is a natural injective morphism of:!' to '§.

1.17. Skyscraper Sheaves. Let X be a topological space, let P be a point, and let A be an
abelian group. Define a sheaf ip(A) on X as follows: ip(A)(U) = A if P E U, 0
otherwise. Verify that the stalk of ip(A) is A at every point Q E { P}-, and 0
elsewhere, where {P}- denotes the closure of the set consisting of the point P.
Hence the name "skyscraper sheaf." Show that this sheaf could also be described
as i*(A), where A denotes the constant sheaf A on the closed subspace {P}-, and
i: {P}- --->X is the inclusion.
1.18. Adjoint Property off- 1. Let f: X ---> Y be a continuous map of topological spaces.
Show that for any sheaf:!' on X there is a natural map f- 1f*:!' ---> :!', and for
any sheaf'§ on Y there is a natural map'§ ---> f*f- 1'§. Use these maps to show
that there is a natural bijection of sets, for any sheaves :!' on X and '§ on Y,
Homx(F 1 '§,:!') = Homy('§,f*ff).
Hence we say that f- 1 is a left adjoint off*' and that f* is a right adjoint of f- 1 •
1.19. Extending a Sheaf by Zero. Let X be a topological space, let Z be a closed subset,
let i:Z---> X be the inclusion, let U = X - Z be the complementary open subset,
and letj: U---> X be its inclusion.
(a) Let:!' be a sheaf on Z. Show that the stalk (i*ff)p of the direct image sheaf on
X is :l'p if P E Z, 0 if P r/= Z. Hence we call i*ff the sheaf obtained by extending
:!' by zero outside Z. By abuse of notation we will sometimes write :!' instead
of i*ff, and say "consider:!' as a sheaf on X," when we mean "consider i*ff."
(b) Now let:!' be a sheaf on U. LetNff) be the sheaf on X associated to the pre-
sheaf V c-+ ff(V) if V <;::: U, V c-+ 0 otherwise. Show that the stalk (j.(ff) )p is
equal to :l'p if P E U, 0 if P r/= U, and show thatj,:!' is the only sheaf on X which
has this property, and whose restriction to U is :!'. We call}!:!' the sheaf
obtained by extending :!' by zero outside U.
(c) Now let:!' be a sheaf on X. Show that there is an exact sequence of sheaves
on X,
0---> j,(fflul---> :!'---> i*(fflzl---> 0.
1.20. Subsheaf with Supports. Let Z be a closed subset of X, and let:!' be a sheaf on X.
We definerz(X,ff) to be the subgroup of T(X,ff) consisting of all sections whose
support (Ex. 1.14) is contained in Z.
(a) Show that the presheaf V c-+ r z n v(V.fflv) is a sheaf. It is called the subsheaf
of§' with supports in Z, and is denoted by J'f~(:!').
(b) Let U = X - Z, and let}: U ---> X be the inclusion. Show there is an exact
sequence of sheaves on X
0---> J'f~(:!')---> :!' ---> j*(fflul·
Furthermore, if:!' is fiasque, the map:!' ---> j*(fflul is surjective.
1.21. Some Examples of Sheaves on Varieties. Let X be a variety over an algebraically
closed field k, as in Ch. I. Let (Qx be the sheaf of regular functions on X (1.0.1).
(a) Let Y be a closed subset of X. For each open set U <;:::X, let §y( U) be
the ideal in the ring (1)x(U) consisting of those regular functions which vanish

68
 2 Schemes

at all points of Y n U. Show that the presheaf U H Jy(U) is a sheaf. It is

called the sheaf of ideals Jy of Y, and it is a subsheaf of the sheaf of rings {!) x·
(b) If Y is a subvariety, then the quotient sheaf I!Jx/ .Yy is isomorphic to i*(I!Jy),
where i: Y-> X is the inclusion, and I!Jy is the sheaf of regular functions on Y.
(c) Now let X= P 1 , and let Y be the union of two distinct points P,Q EX. Then
there is an exact sequence of sheaves on X, where:?= i*I!Jp EB i*I!JQ,
0-> fy ->~X -> :ff7-> 0.

Show however that the induced map on global sections r(X,I!Jx)-> r(X,$7)
is not surjective. This shows that the global section functor r(X, ·)is not exact
(cf. (Ex. 1.8) which shows that it is left exact).
(d) Again let X = P 1, and let{!) be the sheaf of regular functions. Let:£ be the
constant sheaf on X associated to the function field K of X. Show that there
is a natural injection 0 -> .ff. Show that the quotient sheaf:£ ji!J is isomorphic
to the direct sum of sheaves LPEX ip(Ip), where lp is the group Kji!Jp, and
ip(I p) denotes the skyscraper sheaf (Ex. 1.17) given by I P at the point P.
(e) Finally show that in the case of (d) the sequence
0 -> T(X,I!J) -> r(X,ff) -> T(X,ff j(!J) -> 0

is exact. (This is an analogue of what is called the "first Cousin problem" in

several complex variables. See Gunning and Rossi [1, p. 248].)

1.22. Glueing Sheaves. Let X be a topological space, let U = { Ui} be an open cover of
X, and suppose we are given for each i a sheaf :?i on Ui, and for each i,j an iso-
morphism <flii::?ilu,nu,.:::.. :?ilu,nuj such that (1) for each i, <flii = id, and (2) for
each i,j,k, <flik = <flik <flii on Ui n Ui n U k· Then there exists a unique sheaf
:? on X, together with isomorphisms 1/Ji::?iu, .:::.. :?i such that for each i,j, 1/Ji =
cpii o 1/Ji on Ui n Ui. We say loosely that.'#' is obtained by glueing the sheaves :?i
via the isomorphisms <flii·

2 Schemes

In this section we will define the notion of a scheme. First we define affine
schemes: to any ring A (recall our conventions about rings made in the
Introduction!) we associate a topological space together with a sheaf of
rings on it, called Spec A. This construction parallels the construction of
affine varieties (I, §1) except that the points of Spec A correspond to all prime
ideals of A, not just the maximal ideals. Then we define an arbitrary scheme
to be something which locally looks like an affine scheme. This definition
has no parallel in Chapter I. An important class of schemes is given by the
construction of the scheme Proj S associated to any graded ring S. This
construction parallels the construction of projective varieties in (1, §2).
Finally, we will show that the varieties of Chapter I, after a slight modification,
can be regarded as schemes. Thus the category of schemes is an enlargement
of the category of varieties.

69
II Schemes

Now we will construct the space Spec A associated to a ring A. As a set,

we define Spec A to be the set of all prime ideals of A. If a is any ideal of A,
we define the subset V(a) <:::::: Spec A to be the set of all prime ideals which
contain a.

Lemma 2.1.
(a) If a and bare two ideals of A, then V(ab) = V(a) u V(b).
(b) If {aJ is any set of ideals of A, then V(l:aJ = nv(aJ
(c) If a and b are two ideals, V(a) <:::::: V(b) if and only if JO. 2 Jh.
PROOF.
(a) Certainly if p 2 a or p 2 b, then p 2 ab. Conversely, if p 2 ab, and
if p "/2 b for example, then there is a bE b such that b ¢; p. Now for any
a E a, ab E p, so we must have a E p since p is a prime ideal. Thus p 2 a.
(b) p contains Ia; if and only if p contains each a;, simply because Ia;
is the smallest ideal containing all of the ideals a;.
(c) The radical of a is the intersection ofthe set of all prime ideals contain-
ing a. So JO. Jb
2 if and only if V( a) <:::::: V(b ).

Now we define a topology on Spec A by taking the subsets of the form

V(a) to be the closed subsets. Note that V(A) = 0; V((O)) =Spec A; and
the lemma shows that finite unions and arbitrary intersections of sets of the
form V(a) are again of that form. Hence they do form the set of closed sets
for a topology on Spec A.
Next we will define a sheaf of rings (!) on Spec A. For each prime ideal
p <:::::: A, let AP be the localization of A at p. For an open set U <:::::: Spec A,
we define (!)(U) to be the set of functions s: u ~ upEU Ap, such that s(p) E Ap
for each p, and such that sis locally a quotient of elements of A: to be precise,
we require that for each p E U, there is a neighborhood V of p, contained in
U, and elements a,f E A, such that for each q E V, f ¢; q, and s(q) = ajf in
Aq. (Note the similarity with the definition of the regular functions on a
variety. The difference is that we consider functions into the various local
rings, instead of to a field.)
Now it is clear that sums and products of such functions are again such,
and that the element 1 which gives 1 in each AP is an identity. Thus (!)(U) is a
commutative ring with identity. If V <:::::: U are two open sets, the natural
restriction map (!)(U) ~ (!)(V) is a homomorphism of rings. It is then clear
that(!) is a presheaf. Finally, it is clear from the local nature of the definition
that (!) is a sheaf.

Definition. Let A be a ring. The spectrum of A is the pair consisting of the

topological space Spec A together with the sheaf of rings (!) defined above.

Let us establish some basic properties of the sheaf(!) on Spec A. For any
element f E A, we denote by D(f) the open complement of V( (f)). Note
70
 2 Schemes

that open sets of the form D(f) form a base for the topology of Spec A.
Indeed, if V(a) is a closed set, and p ~ V(a), then p f2. a, so there is an f E a,
f ~ p. Then p E D(f) and D(f) n V(a) = 0-

Proposition 2.2. Let A be a ring, and (Spec A, @) its spectrum.

(a) For any p E Spec A, the stalk @P of the sheaf (1) is isomorphic to the
local ring AP.
(b) For any element f E A, the ring @(D(f)) is isomorphic to the localized
ring A 1 .
(c) In particular, r{Spec A,@) ~ A.
PROOF.
(a) First we define a homomorphism from (1)P to AP by sending any local
section s in a neighborhood of p to its value s(p) E AP. This gives a well-
defined homomorphism cp from (1) P to AP. The map cp is surjective, because
any element of AP can be represented as a quotient a/f, with a,f E A, f ~ p.
Then D(f) will be an open neighborhood of p, and a/f defines a section of (1)
over D(f) whose value at p is the given element. To show that cp is injective,
let U be a neighborhood ofp, and let s,t E @(U) be elements having the same
value s(p) = t(p) at p. By shrinking U if necessary, we may assume that
s = ajf, and t = bjg on U, where a,b,f,g E A, andf,g ~ p. Since a/f and bjg
have the same image in AP, it follows from the definition of localization that
there is an h ~ p such that h(ga - fb) = 0 in A. Therefore alf = bjg in every
local ring Aq such that f,g,h ~ q. But the set of such q is the open set D(f) n
D(g) n D(h), which contains p. Hence s = t in a whole neighborhood of p,
so they have the same stalk at p. So cp is an isomorphism, which proves (a).
(b) and (c). Note that (c) is the special case of (b) when f = 1, and D(f)
is the whole space. So it is sufficient to prove (b). We oefine a homomorphism
1/1: A J ~ @(D(f)) by sending a/f" to the section s E @(D(f)) which assigns to
each p the image of ajf" in Ap.
First we show t/J is injective. If 1/J(a/f") = t/J(b/fm), then for every p E
D(f), ajf" and b/fm have the same image in AP. Hence there is an element
h ~ p such that h(fma - f"b) = 0 in A. Let a be the annihilator of fma -
f"b. Then h E a, and h ~ p, so a rj;. p. This holds for any p E D(f), so we
conclude that V(a) n D(f) = 0. Therefore f E JO, so some power PEa,
so PUma - f"b) = 0, which shows that ajf" = b/fm in A 1 . Hence t/1 is
injective.
The hard part is to show that t/J is surjective. So let s E @(D(f) ). Then
by definition of@, we can cover D(f) with open sets v;, on which sis repre-
sented by a quotient a;/g;, with g; ~ p for all p E v;, in other words, Vi £ D(g;).
Now the open sets of the form D(h) form a base for the topology, so we may
assume that v; = D(h;) for some h;. Since D(h;) £ D(g;), we have V( (h;)) 2
V( (g;) ), hence by (2.1c), J(hJ £ J(iJ, and in particular, h? E (g;) for some n.
So h? = cg;, so a;/g; = cajh?. Replacing h; by h? (since D(h;) = D(h7)) and
a; by ca;, we may assume that D(f) is covered by the open subsets D(h;),
and that sis represented by a;/h; on D(h;).
71
II Schemes

Next we observe that D(f) can be covered by a finite number of the D(h;).
Indeed, D(f) c::::: UD(hJ if and only if V((f)) 2 n
V((hJ) = V(~)h;)).
By (2.lc) again, this is equivalent to saying f E ~.DhJ, or r E L,(hJ for
some n. This means that r can be expressed as a finite sum r = L,b;h;,
b; E A. Hence a finite subset of the h; will do. So from now on we fix a
finite set h 1, . . . A such that D(f) c::::: D(h 1 ) u ... u D(h,).
For the next step, note that on D(hJ n D(hj) = D(h;hj) we have two
elements of Ah,h1 , namely a;/h; and a)hj both of which represent s. Hence,
according to the injectivity of ljJ proved above, applied to D(h;h), we must
have a;/h; = a)hj in Ah,h1 • Hence for some n,
(h;h;t(h;a; - h;aj) = 0.
Since there are only finitely many indices involved, we may pick n so large
that it works for all i,j at once. Rewrite this equation as
h~+ 1 (h~a.)- hn+ 1 (h~a.) = 0
} l l
l J } •

Then replace each h; by h?+ 1 , and a; by h?a;. Then we still haves represented
on D(h;) by a;/h;, and furthermore, we have hjai = h;aj for all i,j.
Now write r = l);h; as above, which is possible for some n since the
D(hJ cover D(f). Let a = l);a;. Then for each j we have
hja = L: b;a;hj = I b;h;aj = raj.
i i

This says that afr = a)hj on D(h). So ljJ(a/r) = s everywhere, which

shows that ljJ is surjective, hence an isomorphism.

To each ring A we have now associated its spectrum (Spec A,(()). We

would like to say that this correspondence is functorial. For that we need a
suitable category of spaces with sheaves of rings on them. The appropriate
notion is the category of locally ringed spaces.

Definition. A ringed space is a pair (X,(Ox) consisting of a topological space

X and a sheaf of rings (Ox on X. A morphism of ringed spaces from (X,(()x)
to (Y,(()y) is a pair (f,f#) of a continuous map f:X--+ Y and a map
f# :(()y--+ j*(()x of sheaves of rings on Y. The ringed space (X,(Ox) is a
locally ringed space if for each point P EX, the stalk mx,P is a local ring.
A morphism of locally ringed spaces is a morphism (f,f#) of ringed
spaces, such that for each point P EX, the induced map (see below) of
local ringsfff:(OY,J(P)--+ mx,P is a local homomorphism oflocal rings. We
explain this last condition. First of all, given a point P E X, the morphism
of sheaves f# :(()y--+ j*(()x induces a homomorphism of rings (()y(V)--+
(Ox(j- 1 V), for every open set Vin Y. As Vranges over all open neighbor-
hoods of f(P), f- 1 (V) ranges over a subset of the neighborhoods of P.
Taking direct limits, we obtain a map
(()Y,J(P) = lim (()y(V) --+ lim mxu- 1 V),
~ ~

72
 2 Schemes

and the latter limit maps to the stalk mx,P· Thus we have an induced
homomorphism f: :(!)Y,J(P)--> mx,P· We require that this be a local
homomorphism: If A and Bare local rings with maximal ideals rnA and
mB respectively, a homomorphism cp:A --> B is called a local homo-
morphism if cp- 1 (mB) = mk
An isomorphism of locally ringed spaces is a morphism with a two-
sided inverse. Thus a morphism (f,f #) is an isomorphism if and only
iff is a homeomorphism of the underlying topological spaces, and f#
is an isomorphism of sheaves.

Proposition 2.3.
(a) If A is a ring, then (Spec A, (!)) is a locally ringed space.
(b) If cp:A --> B is a homomorphism of rings, then cp induces a nat-
ural morphism of locally ringed spaces
(f,f #):(Spec B, (!)Spec a) --> (Spec A, (!)Spec A).
(c) If A and B are rings, then any morphism of locally ringed spaces
from Spec B to Spec A is induced by a homomorphism of rings cp: A --> B
as in (b).
PROOF.
(a) This follows from (2.2a).
(b) Given a homomorphism cp:A --> B, we define a map f: Spec B-->
Spec A by f(p) = cp- 1(p) for any p E Spec B. If a is an ideal of A, then it is
immediate that f- 1(V(a)) = V(cp(a)), so f is continuous. For each p E
Spec B, we can localize cp to obtain a local homomorphism of local rings
cpP:A"'-'<Pl--> BP. Now for any open set V ~ Spec A we obtain a homo-
morphism of rings f#:(!JspecA(V)--> (!)specB(f- 1 (V)) by the definition of(!),
composing with the maps f and <fJp· This gives the morphism of sheaves
f # :@spec A --> f* ((!)spec B). The induced maps f # on the stalks are just the
local homomorphisms <pp, so (f,f #) is a morphism of locally ringed spaces.
(c) Conversely, suppose given a morphism of locally ringedspaces (f,f#)
from Spec B to Spec A. Taking global sections, f # induces a homomorphism
of rings cp: r{Spec A, (!)Spec A) --> T(Spec B, (!)Spec B)· By (2.2c), these rings
are A and B, respectively, so we have a homomorphism cp:A--> B. For any
p E Spec B, we have an induced local homomorphism on the stalks,
(!)spec A.f(P) --> (!)spec B,p or A f<Pl --> BP, which must be compatible with the
map cp on global sections and the localization homomorphisms. In other
words, we have a commutative diagram

A B

l
Af<pJ

73
II Schemes

Since f # is a local homomorphism, it follows that cp- 1 ( p) = f( p), which

shows that f coincides with the map Spec B ---+ Spec A induced by cp. Now
it is immediate that f# also is induced by cp, so that the morphism (f,f#)
of locally ringed spaces does indeed come from the homomorphism of
rmgs cp.

Caution 2.3.0. Statement (c) of the proposition would be false, if in the

definition of a morphism of locally ringed spaces, we did not insist that the
induced maps on the stalks be local homomorphisms of local rings (see
(2.3.2) below).

Now we come to the definition of a scheme.

Definition. An affine scheme is a locally ringed space (X,@x) which is iso-

morphic (as a locally ringed space) to the spectrum of some ring. A
scheme is a locally ringed space (X,@x) in which every point has an open
neighborhood U such that the topological space U, together with the
restricted sheaf (Dxlu, is an affine scheme. We call X the underlying topo-
logical space of the scheme (X,@x), and (Dx its structure sheaf By abuse
of notation we will often write simply X for the scheme (X,@x). If we
wish to refer to the underlying topological space without its scheme
structure, we write sp(X), read "space of X." A morphism of schemes is
a morphism as locally ringed spaces. An isomorphism is a morphism
with a two-sided inverse.

Example 2.3.1. If k is a field, Spec k is an affine scheme whose topological

space consists of one point, and whose structure sheaf consists of the field k.

Example 2.3.2. If R is a discrete valuation ring, then T = Spec R is an

affine scheme whose topological space consists of two points. One point
t 0 is closed, with local ring R; the other point t 1 is open and dense, with
local ring equal to K, the quotient field of R. The inclusion map R ---+ K
corresponds to the morphism Spec K ---+ T which sends the unique point
of Spec K to t 1 . There is another morphism of ringed spaces Spec K ---+ T
which sends the unique point of SpecK to t 0 , and uses the inclusion R ---+ K
to define the associated map f# on structure sheaves. This morphism is
not induced by any homomorphism R ---+ K as in (2.3b,c), since it is not a
morphism of locally ringed spaces.

Example 2.3.3. If k is a field, we define the affine line over k, Ai, to be

Spec k[ x]. It has a point ~' corresponding to the zero ideal, whose closure
is the whole space. This is called a generic point. The other points, which
correspond to the maximal ideals in k[ x ], are all closed points. They are

74
 2 Schemes

in one-to-one correspondence with the nonconstant monic irreducible poly-

nomials in x. In particular, if k is algebraically closed, the closed points of
Al are in one-to-one correspondence with elements of k.
Example 2.3.4. Let k be an algebraically closed field, and consider the
affine plane over k, defined as A~ = Spec k[x,y J (Fig. 6). The closed points
of A~ are in one-to-one correspondence with ordered pairs of elements of k.
Furthermore, the set of all closed points of M, with the induced topology,
is homeomorphic to the variety called A2 in Chapter I. In addition to the
closed points, there is a generic point ~' corresponding to the zero ideal of
k[x,y], whose closure is the whole space. Also, for each irreducible poly-
nomial f(x,y), there is a point 1J whose closure consists of 1J together with
all closed points (a,b) for which f(a,b) = 0. We say that 1J is a generic point
of the curve f(x,y) = 0.

~ generic
point of
curve
X

closed
points

Figure 6. Spec k [ x, y].

Example 2.3.5. Let X 1 and X 2 be schemes, let U 1 ~ X 1 and U 2 ~ X 2 be

open subsets, and let <p:(Ubmx,[u,)---> (U 2 ,(i)x 2 [u 2 ) be an isomorphism of
locally ringed spaces. Then we can define a scheme X, obtained by glueing
X 1 and X 2 along U 1 and U 2 via the isomorphism <p. The topological space
of X is the quotient of the disjoint union X 1 u X 2 by the equivalence
relation x 1 ~ <p(x 1 ) for each x 1 E U b with the quotient topology. Thus
there are maps i 1 : X 1 ---> X and i 2 : X 2 ---> X, and a subset V ~ X is open
if and only if ij 1 (V) is open in X 1 and i2 1 (V) is open in X 2 . The structure
sheaf mx is defined as follows: for any open set V ~ X,
(l)x(V) = { <s 1 ,sz)[s 1 E (17x,(ij 1 (V)) and s2 E (i)x 2 (i2 1 (V)) and
<p(sl[i~ 1 (V) n u,) = s2fi2 1 (V) n uJ·
Now it is clear that (l)x is a sheaf, and that (X,(I)x) is a locally ringed space.
Furthermore, since X 1 and X 2 are schemes, it is clear that every point of X
has a neighborhood which is affine, hence X is a scheme.

Example 2.3.6. As an example of glueing, let k be a field, let X 1 = X 2 =

AL let U 1 = U 2 = Al - {P}, where P is the point corresponding to the
75
II Schemes

maximal ideal (x), and let qJ: U 1 ~ U 2 be the identity map. Let X be ob-
tained by glueing X 1 and X 2 along U 1 and U 2 via ([J. We get an "affine
line with the point P doubled."

This is an example of a scheme which is not an affine scheme ( !). It is also

an example of a nonseparated scheme, as we will see later (4.0.1).

Next we will define an important class of schemes, constructed from

graded rings, which are analogous to projective varieties.
Let S be a graded ring. See (I, §2) for our conventions about graded
rings. We denote by s+ the ideal EBd>O Sd.
We define the set Proj S to be the set of all homogeneous prime ideals p,
which do not contain all of S +. If a is a homogeneous ideal of S, we define
the subset V(a) = {p E Proj Sip 2 a}.

Lemma 2.4.
(a) If a and bare homogeneous ideals inS, then V(ab) = V(a) u V(b).
(b) If {a;} is any family of homogeneous ideals of S, then V(Lai) =
nv(aJ
PROOF. The proofs are the same as for (2.1a,b), taking into account the
fact that a homogeneous ideal p is prime if and only if for any two homo-
geneous elements a,b E S, ab E p implies a E p or b E p.

Because of the lemma we can define a topology on Proj S by taking the

closed subsets to be the subsets of the form V( a).
Next we will define a sheaf of rings(!) on Proj S. For each p E Proj S, we
consider the ring S<vl of elements of degree zero in the localized ring T- 1 S,
where T is the multiplicative system consisting of all homogeneous elements
of S which are not in p. For any open subset U <;; Proj S, we define (!)(U)
to be the set of functions s: U ~ il S(p) such that for each p E U, s(.p) E S(p)•
and such that sis locally a quotient of elements of S: for each p E U, there
exists a neighborhood V of p in U, and homogeneous elements a,f in S,
of the same degree, such that for all q E V, f ¢ q, and s(q) = a/fin S(q)· Now
it is clear that (!) is a presheaf of rings, with the natural restrictions, and it is
also clear from the local nature of the definition that (!) is a sheaf.

Definition. If Sis any graded ring, we define (Proj S,(!)) to be the topological
space together with the sheaf of rings constructed above.

Proposition 2.5. LetS be a graded ring.

(a) For any p E Proj S, the stalk (!)v is isomorphic to the local ring S<vl·
(b) For any homogeneous fES+, let D+(f) = {.pEProj Slf¢p}.

76
 2 Schemes

Then D +(f) is open in Proj S. Furthermore, these open sets cover Proj S,
and for each such open set, we have an isomorphism of locally ringed
spaces

where s(f) is the sub ring of elements of degree 0 in the localized ring sJ·
(c) Proj Sis a scheme.

PROOF. Note first that (a) says that Proj S is a locally ringed space, and
(b) tells us it is covered by open affine schemes, so (c) is a consequence of
(a) and (b).
The proof of (a) is practically identical to the proof of (2.2a) above, so
is left to the reader.
To prove (b), first note that D +(f) = Proj S - V( (f)), so it is open.
Since the elements of Proj S are those homogeneous prime ideals p of S
which do not contain all of S +, it follows that the open sets D +(f) for homo-
geneous! E s+ cover Proj S. Now fix a homogeneous! E S+. We will define
an isomorphism (cp,cp#) of locally ringed spaces from D+(f) to Spec S<n·
There is a natural homomorphism of rings S -+ S I' and S<n is a subring of
S J· For any homogeneous ideal a <:; S, let cp(a) = (aS I) n S(f). In partic-
ular, if p E D+(f), then cp(p) E Spec S<n' so this gives the map cp as sets.
The properties of localization show that cp is bijective as a map from D +(f)
to Spec S(f). Furthermore, if a is a homogeneous ideal of S, then p 2 a
if and only if cp(p) 2 cp(a). Hence cp is a homeomorphism. Note also if
p ED +(f), then the local rings S<P> and (S<n),<P> are naturally isomorphic.
These isomorphisms and the homeomorphism cp induce a natural map of
sheaves cp# :(Ospecsul-+ cp*((OProjslv+(fl) which one recognizes immediately to
be an isomorphism. Hence (cp,cp#) is an isomorphism of locally ringed
spaces, as required.

Example 2.5.1. If A is a ring, we define projective n-space over A to be the

scheme PA. = Proj A[ x 0 , . . . ,xnJ. In particular, if A is an algebraically
closed field k, then P~ is a scheme whose subspace of closed points is naturally
homeomorphic to the variety called projective n-space-see (Ex. 2.14d)
below.

Next we will show that the notion of scheme does in fact generalize the
notion of variety. It is not quite true that a variety is a scheme. As we have
already seen in the examples above, the underlying topological space of a
scheme such as At or A~ has more points than the corresponding variety.
However, we will show that there is a natural way of adding generic points
(Ex. 2.9) for every irreducible subset of a variety so that the variety becomes
a scheme.
To state our result, we need a definition.

77
II Schemes

Definition. LetS be a fixed scheme. A scheme overS is a scheme X, together

with a morphism X --4 S. If X and Y are schemes over S, a morphism
of X to Y as schemes over S, (also called an S-morphism) is a morphism
f: X --4 Y which is compatible with the given morphisms to S. We denote
by 6c()(S) the category of schemes over S. If A is a ring, then by abuse of
notation we write 6c()(A) for the category of schemes over Spec A.

Proposition 2.6. Let k be an algebraically closed field. There is a natural

fully faithful functor t: IBar(k) --4 6c()(k) from the category of varieties over
k to schemes over k. For any variety V, its topological space is homeo-
morphic to the set of closed points of sp(t(V) ), and its sheaf of regular
functions is obtained by restricting the structure sheaf of t( V) via this
homeomorphism.
PROOF. To begin with, let X be any topological space, and let t(X) be the
set of (nonempty) irreducible closed subsets of X. If Y is a closed subset of
X, then t( Y) <;; t(X). Furthermore, t( Y1 u Y2 ) = t( Y1 ) u t( Y2 ) and t(n Y;) =
nt(Y;). So we can define a topology on t(X) by taking as closed sets the
subsets of the form t( Y), where Y is a closed subset of X. Iff: X 1 --4 X 2 is a
continuous map, then we obtain a map t(f): t(X 1 ) --4 t(X 2 ) by sending an
irreducible closed subset to the closure of its image. Thus t is a functor on
topological spaces. Furthermore, one can define a continuous map rx: X --4
t(X) by rx(P) = {P}-. Note that rx induces a bijection between the set of
open subsets of X and the set of open subsets of t(X).
Now let k be an algebraically closed field. Let V be a variety over
k, and let 0v be its sheaf of regular functions (1.0.1). We will show that
(t(V),rx*(Gv)) is a scheme over k. Since any variety can be covered by open
affine subvarieties (1, 4.3), it will be sufficient to show that if V is affine,
then (t(V),rx*(Gv)) is a scheme. So let V be an affine variety with affine
coordinate ring A. We define a morphism of locally ringed spaces
f3:(V, Gv) --4 X = Spec A
as follows. For each point P E V, let f3(P) = mp, the ideal of A consisting
of all regular functions which vanish at P. Then by (1, 3.2b), f3 is a bijection
of V onto the set of closed points of X. It is easy to see that f3 is a homeo-
morphism onto its image. Now for any open set U <;; X, we will define a
homomorphism of rings Gx( U) --4 {3*( Gv )( U) = Gv(/3- 1 U). Given a section
s E Gx(U), and given a point pEr 1(U), we define s(P) by taking the image
of sin the stalk Gx,p(PJ• which is isomorphic to the local ring A"'P' and then
passing to the quotient ring A"'Pjmp which is isomorphic to the field k. Thus
s gives a function from f3- 1 ( U) to k. It is easy to see that this is a regular
function, and that this map gives an isomorphism Gx( U) ~ Gv(/3- 1 U).
Finally, since the prime ideals of A are in 1-1 correspondence with the irre-
ducible closed subsets of V (see (1, 1.4) and proof), these remarks show that
(X,(Cx) is isomorphic to (t(V), rx*Gv), so the latter is indeed an affine scheme.

78
 2 Schemes

To give a morphism of (t(V),cx*(i]v) to Spec k, we have only to give a

homomorphism of rings k----> T(t(V),cx*CDv) = r(V, CDvl· We send I.E k to
the constant function I. on V. Thus t(V) becomes a scheme over k. Finally,
if V and Ware two varieties, then one can check (Ex. 2.15) that the natural
map
Homtlnr(ki(V,W)----> Hom 2 rt,(kl(t(V),t(W))
is bijective. This shows that the functor t:llJnr(k)----> Scl)(k) is fully faithful.
In particular it implies that t( V) is isomorphic to t( W) if and only if V is
isomorphic to W
It is clear from the construction that ex: V---+ t(V) induces a homeo-
morphism from V onto the set of closed points of t( V ), with the induced
topology.

Note. We will see later (4.10) what the image of the functor tis.

EXERCISES

2.1. Let A be a ring, let X = Spec A, let f E A and let D(f) <;; X be the open comple-
ment of V( (f)). Show that the locally ringed space (D(f), (f'xiD<fl) is isomorphic
to Spec A 1 .

2.2. Let (X,0x) be a scheme, and let U <;; X be any open subset. Show that (U,0xluJ
is a scheme. We call this the induced scheme structure on the open set U, and we
refer to (U,0xlul as an open subscheme of X.

2.3. Reduced Schemes. A scheme (X,(f;x) is reduced if for every open set U <;; X, the
ring (!) x( U) has no nilpotent elements.
(a) Show that (X,0x) is reduced if and only if for every P EX, the local ring crx.P
has no nilpotent elements.
(b) Let (X,0x) be a scheme. Let ((l:x)ced be the sheaf associated to the presheaf
U 1--+ (l)x(U),.d, where for any ring A, we denote by A,.d the quotient of A
by its ideal of nilpotent elements. Show that (X,((f x )"J) is a scheme. We call
it the reduced scheme associated to X, and denote it by X"J· Show that there is
a morphism of schemes X"d --->X, which is a homeomorphism on the under-
lying topological spaces.
(c) Letf: X ---> Y be a morphism of schemes, and assume that X is reduced. Show
that there is a unique morphism{]: X ---> Y,,d such that f is obtained by com-
posing {] with the natural map Y,,d ---> Y.

2.4. Let A be a ring and let (X,lPx) be a scheme. Given a morphism f:X---> Spec A,
we have an associated map on sheaves f #: cr:srccA ---> j*(!;x· Taking global sections
we obtain a homomorphism A ---> r(X,6x ). Thus there is a natural map
:x: Hom 2 ,,(X,Spec A)---> Hom"",, . (A,r(X,(' xl).
Show that :x is bijective (cf. (1, 3.5) for an analogous statement about varieties).

2.5. Describe Spec Z, and show that it is a final object for the category of schemes,
i.e., each scheme X admits a unique morphism to Spec Z.

79
II Schemes

2.6. Describe the spectrum of the zero ring, and show that it is an initial object for
the category of schemes. (According to our conventions, all ring homomorphisms
must take 1 to 1. Since 0 = 1 in the zero ring, we see that each ring R admits a
unique homomorphism to the zero ring, but that there is no homomorphism
from the zero ring toR unless 0 = 1 in R.)
2.7. Let X be a scheme. For any x EX, let (!)x be the local ring at x, and mx its maximal
ideal. We define the residue field of x on X to be the field k(x) = (!)x/mx. Now
let K be any field. Show that to give a morphism of SpecK to X it is equivalent
to give a point x EX and an inclusion map k(x)--+ K.
2.8. Let X be a scheme. For any point x EX, we define the Zariski tangent space Tx
to X at x to be the dual of the k(x)-vector space mx/m;. Now assume that X is
a scheme over a field k, and let k[ t:]/t: 2 be the ring of dual numbers over k. Show
that to give a k-morphism of Spec k[t:]/t: 2 to X is equivalent to giving a point
x EX, rational over k (i.e., such that k(x) = k), and an element of Tx.
2.9. If X is a topological space, and Z an irreducible closed subset of X, a generic
point for Z is a point ( such that Z = {0-. If X is a scheme, show that every
(nonempty) irreducible closed subset has a unique generic point.
2.10. Describe Spec R[ x]. How does its topological space compare to the set R? To C?
2.11. Let k = F P be the finite field with p elements. Describe Spec k[ x]. What are
the residue fields of its points? How many points are there with a given residue
field?
2.12. Clueing Lemma. Generalize the glueing procedure described in the text (2.3.5) as
follows. Let {X;} be a family of schemes (possible infinite). For each i of j,
suppose given an open subset Uij <;; Xi, and let it have the induced scheme
structure (Ex. 2.2). Suppose also given for each i of j an isomorphism of schemes
cp;i: Uu --+ Uii such that (1) for each i,j, cpii = ({J;j 1, and (2) for each i,j,k,
CfJu(U;i n U;k) = Uii n Uik• and CfJ;k = cpik o CfJu on Uii n Uik· Then show that
there is a scheme X, together with morphisms 1/J;:X;--+ X for each i, such that
(1) 1/Ji is an isomorphism of X; onto an open subscheme of X, (2) the 1/J;(X;) cover
X, (3) 1/J;(Ui) = 1/J;(X;) n 1/Ji(X) and (4) 1/J; = 1/Ji o CfJu on Uii. We say that X is
obtained by glueing the schemes Xi along the isomorphisms CfJu· An interesting
special case is when the family X; is arbitrary, but the Uii and cpii are all empty.
Then the scheme X is called the disjoint union of the X;, and is denoted UX;.
2.13. A topological space is quasi-compact if every open cover has a finite subcover.
(a) Show that a topological space is noetherian (I, §1) if and only if every open
subset is quasi-compact.
(b) If X is an affine scheme, show that sp(X) is quasi-compact, but not in general
noetherian. We say a scheme X is quasi-compact if sp(X) is.
(c) If A is a noetherian ring, show that sp(Spec A)is a noetherian topological space.
(d) Give an example to show that sp(Spec A) can be noetherian even when A is not.
2.14. (a) LetS be a graded ring. Show that Proj S = ¢ if and only if every element of
S + is nilpotent.
(b) Let cp: S --+ T be a graded homomorphism of graded rings (preserving degrees).
Let U = {p E Proj Tlpi12 q>(S+)}. Show that U is an open subset ofProj T,
and show that cp determines a natural morphism f: U --+ Proj S.

80
 2 Schemes

(c) The morphism f can be an isomorphism even when cp is not. For example,
suppose that cpd:Sd-> Td is an isomorphism for all d ;::, d 0 , where d0 is an
integer. Then show that U = Proj T and the morphism f: Proj T -> Proj S
is an isomorphism.
(d) Let V be a projective variety with homogeneous coordinate ringS (I, §2). Show
that t(V) ~ Proj S.

2.15. (a) Let V be a variety over the algebraically closed field k. Show that a point
P E t( V) is a closed point if and only if its residue field is k.
(b) If f:X-> Y is a morphism of schemes over k, and if P EX is a point with
residue field k, then f(P) E Y also has residue field k.
(c) Now show that if V,W are any two varieties over k, then the natural map

is bijective. (Injectivity is easy. The hard part is to show it is surjective.)

2.16. Let X be a scheme, let f E T(X,(IJx), and define X 1 to be the subset of points
x E X such that the stalk fx off at x is not contained in the maximal ideal mx
of the local ring (I) x·
(a) If U = Spec B is an open affine subscheme of X, and if J E B = r(U,(I)xlul is
the restriction off, show that U n X 1 = D(J). Conclude that X 1 is an open
subset of X.
(b) Assume that X is quasi-compact. Let A = r(X,(I)x), and let a E A be an
element whose restriction to X 1 is 0. Show that for some n > 0, f"a = 0.
[Hint:Use an open affine cover of X.]
(c) Now assume that X has a finite cover by open affines U; such that each inter-
section U; n Ui is quasi-compact. (This hypothesis is satisfied, for example,
if sp(X) is noetherian.) Let bE r(X1 ,(1)xJ Show that for some n > 0, f"b is
the restriction of an element of A.
(d) With the hypothesis of(c), conclude that r(X1 ,(1Jx1 ) ~ A 1 .

2.17. A Criterion for Affineness.

(a) Let f: X -> Y be a morphism of schemes, and suppose that Y can be covered
by open subsets U;, such that for each i, the induced map f- 1 (U;) -> U; is an
isomorphism. Then f is an isomorphism.
(b) A scheme X is affine if and only if there is a finite set of elements f 1 , •• . ,f,. E
A = r(X,(I)x), such that the open subsets X 1 , are affine, andf1 , ... ,f,. generate
the unit ideal in A. [Hint: Use (Ex. 2.4) and (Ex. 2.16d) above.]
2.18. In this exercise, we compare some properties of a ring homomorphism to the
induced morphism of the spectra of the rings.
(a) Let A be a ring, X = Spec A, andf EA. Show thatf is nilpotent if and only if
D(f) is empty.
(b) Let cp: A -> B be a homomorphism of rings, and let f: Y = Spec B -> X =
Spec A be the induced morphism of affine schemes. Show that cp is injective if
and only if the map of sheaves J# :(IJx -> f*(IJY is injective. Show furthermore
in that case f is dominant, i.e., f(Y) is dense in X.
(c) With the same notation, show that if cp is surjective, then f is a homeomor-
phism of Yonto a closed subset of X, andf# :(IJx-> j*(IJY is surjective.

81
II Schemes

(d) Prove the converse to (c), namely, iff: Y--> X is a homeomorphism onto a
closed subset, and f# :@x--> f*(r)Y is surjective, then qJ is surjective. [Hint:
Consider X' = Spec(A/ker ({J) and use (b) and (c).]
2.19. Let A be a ring. Show that the following conditions are equivalent:
(i) Spec A is disconnected;
(ii) there exist nonzero elements e~oe 2 E A such that e 1 e2 = 0, ei = e 1 , e~ = e2 ,
e 1 + e2 = 1 (these elements are called orthogonal idempotents);
(iii) A is isomorphic to a direct product A 1 x A 2 of two nonzero rings.

3 First Properties of Schemes

In this section we will give some of the first properties of schemes. In particu-
lar we will discuss open and closed subschemes, and products of schemes. In
the exercises we introduce the notion of constructible subsets, and study the
dimension of the fibres of a morphism.

Definition. A scheme is connected if its topological space is connected. A

scheme is irreducible if its topological space is irreducible.

Definition. A scheme X is reduced if for every open set U, the ring (l)x(U) has
no nilpotent elements. Equivalently (Ex. 2.3), X is reduced if and only if
the local rings @p, for all P EX, have no nilpotent elements.

Definition. A scheme X is integral if for every open set U <:; X, the ring
(l)x(U) is an integral domain.

Example 3.0.1. If X = Spec A is an affine scheme, then X is irreducible if

and only if the nilradical nil A of A is prime; X is reduced if and only if
nil A = 0; and X is integral if and only if A is an integral domain.

Proposition 3.1. A scheme is integral if and only if it is both reduced and ir-
reducible.
PROOF. Clearly an integral scheme is reduced. If X is not irreducible, then
one can find two nonempty disjoint open subsets U 1 and U 2 . Then
@(U 1 u U 2 ) = @(U 1 ) x @(U 2 ) which is not an integral domain. Thus
integral implies irreducible.
Conversely, suppose that X is reduced and irreducible. Let U <:; X be an
open subset, and suppose that there are elements f,g E @(U) with fg = 0.
Let Y = {x E Uifx E mx}, and let Z = {x E Ulgx E mx}· Then Y and Z are
closed subsets (Ex. 2.16a), and Y u Z = U. But X is irreducible, so U is
irreducible, so one of Y or Z is equal to U, say Y = U. But then the restric-
tion off to any open affine subset of U will be nilpotent (Ex. 2.18a), hence
zero, so f is zero. This shows that X is integral.

82
 3 First Properties of Schemes

Definition. A scheme X is locally noetherian if it can be covered by open affine

subsets Spec A;, where each A; is a noetherian ring. X is noetherian if
it is locally noetherian and quasi-compact. Equivalently, X is noetherian
if it can be covered by a finite number of open affine subsets Spec A;,
with each A; a noetherian ring.

Caution 3.1.1. If X is a noetherian scheme, then sp(X) is a noetherian topo-

logical space, but not conversely (Ex. 2.13) and (Ex. 3.17).

Note that in this definition we do not require that every open affine
subset be the spectrum of a noetherian ring. So while it is obvious from the
definition that the spectrum of a noetherian ring is a noetherian scheme, the
converse is not obvious. It is a question of showing that the noetherian
property is a "local property". We will often encounter similar situations
later in defining properties of a scheme or of a morphism of schemes, so we
will give a careful statement and proof of the local nature of the noetherian
property, to illustrate this type of situation.

Proposition 3.2. A scheme X is locally noetherian if and only if for every open
affine subset U = Spec A, A is a noetherian ring. In particular, an affine
scheme X = Spec A is a noetherian scheme if and only if the ring A is a
noetherian ring.
PROOF. The "if" part follows from the definition, so we have to show if X
is locally noetherian, and if U = Spec A is an open affine subset, then A is a
noetherian ring. First note that if B is a noetherian ring, so is any localization
B J· The open subsets D(f) ~ Spec B f form a base for the topology of Spec B.
Hence on a locally noetherian scheme X there is a base for the topology con-
sisting of the spectra of noetherian rings. In particular, our open set U can
be covered by spectra of noetherian rings.
So we have reduced to proving the following statement: let X = Spec A
be an affine scheme, which can be covered by open subsets which are spectra
of noetherian rings. Then A is noetherian. Let U = Spec B be an open
subset of X, with B noetherian. Then for some f E A, D(f) ~ U. Let
J be the image off in B. Then A f ~ B 1 , hence A f is noetherian. So we
can cover X by open subsets D(f) ~ Spec A J with A J noetherian. Since X
is quasi-compact, a finite number will do.
So now we have reduced to a purely algebraic problem: A is a ring,
/ 1 , . . . ,f.. are a finite number of elements of A, which generate the unit ideal,
and each localization AJ, is noetherian. We have to show A is noetherian.
First we establish a lemma. Let a ~ A be an ideal, and let ({J;: A --+ A J, be
the localization map, i = 1, ... ,r. Then
a= n({J;- 1 (cp;(a) · AJJ
The inclusion ~ is obvious. Conversely, given an element b E A contained

83
II Schemes

in this intersection, we can write cp;(b) = a;/fi' in AJ, for each i, where
a; E a, and n; > 0. Increasing the n; if necessary, we can make them all
equal to a fixed n. This means that in A we have
f'[''(fib - a;) = 0
for some m;. And as before, we can make all them; = m. Thus f'['+nb E a for
each i. Since f 1 , . . . ,f.. generate the unit ideal, the same is true of their Nth
powers for any N. Take N = n + m. Then we have 1 = 'f.cJf for suitable
c; EA. Hence
b = 'f.cJfb E a
as required.
Now we can easily show that A is noetherian. Let a 1 s; a 2 s; ... be an
ascending chain of ideals in A. Then for each i,
qJ;(a 1 )·AJ, s; o/;(a 2 )·AJ, s; ...

is an ascending chain of ideals in A f•' which must become stationary because

AJ, is noetherian. There are only finitely many AJ,, so from the lemma we
conclude that the original chain is eventually stationary, and hence A is
noetherian.

Definition. A morphism f: X ---+ Y of schemes is locally offinite type ifthere

exists a covering of Y by open affine subsets V; = Spec B;, such that for
each i, f- 1(V;) can be covered by open affine subsets U;i = Spec Aii, where
each A;i is a finitely generated B;-algebra. The morphism f is of finite
type if in addition each f- 1 (V;) can be covered by a finite number of the
Uii.

Definition. A morphism f: X ---+ Y is a finite morphism if there exists a

covering of Y by open affine subsets V; = Spec B;, such that for each i,
f- 1 (V;) is affine, equal to Spec A;, where A; is a B;-algebra which is a
finitely generated B;-module.

Note in each of these definitions that a property of a morphism f: X ---+ Y

is defined by the existence of an open affine cover of Y with certain properties.
In fact in each case it is equivalent to require the given property for every
open affine subset of Y (Ex. 3.1-3.4).

Example 3.2.1. If V is a variety over an algebraically closed field k, then the

assoc1ated scheme t(V) (see (2.6)) is an integral noetherian scheme of finite
type over k. Indeed, V can be covered by a finite number of open affine
subvarieties (I, 4.3), so t(V) can be covered by a finite number of open affines
of the form Spec A;, where each A; is an integral domain which is a finitely
generated k-algebra and hence noetherian.

84
 3 First Properties of Schemes

Example 3.2.2. If P is a point of a variety V, with local ring (!) p, then Spec(!) P
is an integral noetherian scheme, which is not in general of finite type over k.

Next we come to open and closed subschemes.

Definition. An open subscheme of a scheme X is a scheme U, whose topological

space is an open subset of X, and whose structure sheaf (!)u is isomorphic
to the restriction (!)xlu of the structure sheaf of X. An open immersion is a
morphism f:X--> Y which induces an isomorphism of X with an open
subscheme of Y.

Note that every open subset of a scheme carries a unique structure of

open subscheme (Ex. 2.2).

Definition. A closed immersion is a morphism f: Y--> X of schemes such that

f induces a homeomorphism of sp(Y) onto a closed subset of sp(X),
and furthermore the induced map f #: (!)x --> f * (!)Y of sheaves on X is
surjective. A closed subscheme of a scheme X is an equivalence class of
closed immersions, where we say f: Y--> X and f': Y' --> X are equi-
valent if there is an isomorphism i: Y'--> Y such that f' = f o i.

Example 3.2.3. Let A be a ring, and let a be an ideal of A. Let X = Spec A

and let Y = Spec A/a. Then the ring homomorphism A --> A/a induces a
morphism of schemes f: Y-> X which is a closed immersion. The map fis
a homeomorphism of Y onto the closed subset V(a) of X, and the map of
structure sheaves (!)x--> j*(!)Y is surjective because it is surjective on the
stalks, which are localizations of A and A/a,respectively (Ex. 2.18).
Thus for any ideal a s; A we obtain a structure of closed subscheme on the
closed set V(a) s; X. In particular, every closed subset Y of X has many closed
subscheme structures, corresponding to all the ideals a for which V(a) = Y.
In fact, every closed subscheme structure on a closed subset Y of an affine
scheme X arises from an ideal in this way (Ex. 3.11 b) or (5.10).

Example 3.2.4. For some more specific examples, let A = k[ x,y ], where k is
a field. Then Spec A = M is the affine plane over k. The ideal a = (xy)
gives a reducible subscheme, consisting of the union of the x and y axes. The
ideal a = (x 2 ) gives a subscheme structure with nilpotents on the y-axis.
The ideal a = (x 2 ,xy) gives another subscheme structure on they-axis, this
one having nilpotents only in the local ring at the origin. We say the origin
is an embedded point for this subscheme.

Example 3.2.5. Let V be an affine variety over the field k, and let W be a
closed subvariety. Then W corresponds to a prime ideal p in the affine co-
ordinate ring A of V (1, §1). Let X = t(V) and Y = t(W) be the associated
schemes. Then X = Spec A and Y is the closed subscheme defined by p.
For each n ;?: 1 let Y, be the closed subscheme of X corresponding to the

85
II Schemes

ideal pn. Then Y1 = Y, but for each n > 1, Y, is a nonreduced scheme struc-
ture on the closed set Y, which does not correspond to any subvariety of V.
We call Y, the nth irifinitesimal neighborhood of Yin X. The schemes Y, reflect
properties of the embedding of Yin X. Later (§9) we will study the "formal
completion" of Yin X, which is roughly the limit of the schemes Y, as n --+ oo.

Example 3.2.6. Let X be a scheme, and let Y be a closed subset. In general Y

will have many possible closed subscheme structures. However, there is one
which is "smaller" than any other, called the reduced induced closed subscheme
structure, which we now describe.
First let X = Spec A be an affine scheme, and let Y be a closed subset.
Let a <;; A be the ideal obtained by intersecting all the prime ideals in Y. This
is the largest ideal for which V(a) = Y. Then we take the reduced induced
structure on Y to be the one defined by a.
Now let X be any scheme, and let Y be a closed subset. For each open
affine subset Ui <;; X, consider the closed subset Y; = Y n Ui of Ui, and give
it the reduced induced structure just defined for affines (which may depend
on UJ I claim that for any i,j, the restrictions to Y; n lj of the two structure
sheaves just defined on Y; and lj are isomorphic, and furthermore, that the
three such isomorphisms on Y; n lj n 1k are compatible for all i,j,k. One
reduces easily to showing that if U = Spec A is an open affine, and iff E A,
and if V = D(f) = Spec A 1 , then the reduced induced structure on Y n U
obtained from A when restricted to Y n V agrees with the one obtained
from A 1 . This corresponds to the algebraic fact that if a is the intersection
of those prime ideals of A which are in Y, then aA 1 is the intersection of those
prime ideals of A 1 which are in Y n D(f).
So now we can glue the sheaves defined on the Y; to obtain a sheaf on Y
(Ex. 1.22}, which gives us the desired reduced induced subscheme structure
on Y. See (Ex. 3.11) below for a universal property of the reduced induced
subscheme structure.

Definition. The dimension of a scheme X, denoted dim X, is its dimension as a

topological space (I, §1). If Z is an irreducible closed subset of X, then the
codimension of Z in X, denoted codim(Z,X) is the supremum of integers n
such that there exists a chain
Z = Z 0 < Z 1 < ... < Zn
of distinct closed irreducible subsets of X, beginning with Z. If Y is any
closed subset of X, we define
codim(Y,X) = inf codim(Z,X)
Z"Y

where the infimum is taken over all closed irreducible subsets of Y.

Example 3.2.7. If X = Spec A is an affine scheme, then the dimension of X

is the same as the Krull dimension of A (1, §1).

86
 3 First Properties of Schemes

Caution 3.2.8. Be careful in applying the concepts of dimension and codi-

mension to arbitrary schemes. Our intuition is derived from working with
schemes of finite type over a field, where these notions are well-behaved.
For example, if X is an affine integral scheme of finite type over a field k,
and if Y c:; X is any closed irreducible subset, then (1, 1.8A) implies that
dim Y + codim( Y,X) = dim X. But on arbitrary (even noetherian) schemes,
funny things can happen. See (Ex. 3.20-3.22), and also Nagata [7], and
Grothendieck [EGA IV, §5].

Definition. Let S be a scheme, and let X, Y be schemes over S, i.e., schemes

with morphisms to S. We define the .fibred product of X and Y overS,
denoted X x s Y, to be a scheme, together with morphisms p 1 :X x s
Y --+ X and p 2 : X x s Y --+ Y, which make a commutative diagram with
the given morphisms X --+ S and Y --+ S, such that given any scheme Z
over S, and given morphisms f:Z--+ X and g:Z--+ Y which make a
commutative diagram with the given morphisms X --+ S and Y --+ S, then
there exists a unique morphism e: z--+ X X s y such that f = p 1 0 e, and
g = p 2 o e. The morphisms p 1 and p 2 are called the projection morphisms
of the fibred product onto its factors.

z ________ .. X X s y

~y
~/
s
If X and Y are schemes given without reference to any base scheme S,
we take S = Spec Z (Ex. 2.5) and define the product of X and Y, denoted
X X Y, to be X Xspec z Y.

Theorem 3.3. For any two schemes X and Y over a schemeS, the .fibred product
X x s Y exists, and is unique up to unique isomorphism.

PROOF. The idea is first to construct products for affine schemes and then
glue. We proceed in seven steps.
Step 1. Let X = Spec A, Y = Spec B, S = Spec R all be affine. Then
A and BareR-algebras, and I claim that Spec (A ® R B) is a product for X and
Y overS. Indeed, for any scheme Z, to give a morphism of Z to Spec (A ® R B)
is the same as to give a homomorphism of the ring A ® R B into the ring
r(Z,(!J 2 ), by (Ex. 2.4). But to give a homomorphism of A ®R B into any ring
is the sallie as to give homomorphisms of A and B into that ring, inducing
the same homomorphism on R. Applying (Ex. 2.4) again, we see that to give
a morphism of Z into Spec (A ® R B) is the same as giving morphisms of Z
into X and into Y, which give rise to the same morphism of Z into S. Thus
Spec (A ® R B) is the desired product.

87
II Schemes

Step 2. It follows immediately from the universal property of the product

that it is unique up to unique isomorphism, if it exists. We will need this
uniqueness for those products already constructed, as we go along.
Step 3. Glueing morphisms. We have already seen how to glue sheaves
(Ex. 1.22) and how to glue schemes (Ex. 2.12). Now we glue morphisms. If X
and Yare schemes, then to give a morphism f from X to Y, it is equivalent to
give an open cover { U;} of X, together with morphisms _[;: U; --+ Y, where U;
has the induced open subscheme structure, such that the restrictions of J;
and jj to U; n Ui are the same, for each i,j. The proof is straightforward.
Step 4. If X, Yare schemes over a schemeS, if U <:::; X is an open subset,
and if the product X x s Y exists, then p1 1( U) <:::; X x s Y is a product for U
and Y over S. Indeed, given a scheme Z, and morphisms f: Z --+ U and
g:Z--+ Y, f determines a map of Z to X by composing with the inclusion
U <:::; X. Hence there is a map B:Z --+X x s Y compatible with f,g and the
projections. But since f(Z) <:::; U, we have 8(Z) <:::; p1 1 ( U). So 8 can be
regarded as a morphism Z --+ p1 1(U). It is clearly unique, so p1 1 (U) is a
product U x s Y.
Step 5. Suppose given X,Y schemes over S, suppose {X;} is an open
covering of X, and suppose that for each i, X; x s Y exists. Then X x s Y
exists. Indeed, for each i,j, let Uii .:::; X; x s Y be p1 1 (X;i), where X;i =
X; n Xi. Then by Step 4, Uii is a product for Xii andY overS. Hence by the
uniqueness of products there are (unique) isomorphisms ({J;/ uij --+ uji for
each i,j compatible with all the projections. Furthermore, these isomor-
phisms are compatible with each other for each i,j,k, in the sense of(Ex. 2.12).
Thus we are in a position to glue the schemes X; x s Yvia the isomorphisms
({J;i· We obtain by (Ex. 2.12) a scheme X x s Y which I claim is a product for
X and Y overS. The projection morphisms p 1 and p 2 are defined by glueing
the projections from the pieces X; x s Y (Step 3). Given a scheme Z and
morphisms f:Z--+ X, g:Z--+ Y, let Z; = f~ 1 (X;). Then we get maps
8;: Z; --+ X; x s Y, hence by composition with the inclusions X; x s Y <:::;
X x s Y we get maps 8;: Z; --+ X x s Y. One verifies that these maps agree on
Z; n Zi, so we can glue the morphisms (Step 3) to obtain a morphism 8: Z --+
X x s Y, compatible with the projections and f and g. The uniqueness of 8
can be checked locally.
Step 6. We know from Step 1 that if X, Y, S are all affine, then X x s Y
exists. Thus using Step 5 we conclude that for any X, but Y, S affine, the
product exists. Using Step 5 again, with X and Y interchanged, we find that
the product exists for any X and any Y over an affine S.
Step 7. Given arbitrary X,Y, S, let q:X--+ S and r: Y--+ S. be the given
morphisms. Let S; be an open affine cover of S. Let X; = q~ 1 (S;) and let
li = r~ 1(S;). Then by Step 6, X; x s; li exists. Note that this same scheme is
a product for X; and Y overS. Indeed, given morphisms f:Z--+ X; and
g:Z--+ Y over S, the image of g must land inside l;. Thus X; x s Y exists
for each i, and one more application of Step 5 gives us X x s Y. This completes
the proof.

88
 3 First Properties of Schemes

Perhaps this is a good place to make some general remarks on the im-
portance and uses offibred products. To begin with, we can define the fibres
of a morphism.

Definition. Let f: X ---+ Y be a morphism of schemes, and let y E Y be a

point. Let k(y) be the residue field of y, and let Spec k(y) ---+ Y be the
natural morphism (Ex. 2.7). Then we define the fibre of the morphism
f over the point y to be the scheme
xy =X X y Spec k(y).

The fibre XY is a scheme over k(y), and one can show that its underlying
topological space is homeomorphic to the subset f- 1 (y) of X (Ex. 3.10).
The notion of the fibre of a morphism allows us to regard a morphism
as a family of schemes (namely its fibres) parametrized by the points of the
image scheme. Conversely, this notion of family is a good way of making
sense of the idea of a family of schemes varying algebraically. For example,
given a scheme X 0 over a field k, we define a family of deformations of X 0
to be a morphism f:X ---+ Y with Y connected, together with a point y 0 E Y,
such that k(y 0 ) = k, and XYo ~ X 0 . The other fibres XY off are called
deformations of X 0 .
An interesting kind offamily arises when we have a scheme X over Spec Z.
In this case, taking the fibre over the generic point gives a scheme XQ over
Q, while taking the fibre over a closed point, corresponding to a prime
number p, gives a scheme X P over the finite field F p· We say that X P arises
by reduction mod p of the scheme X.

Another important application of fib red products is to the notion of base

extension. Let S be a fixed scheme which we think of as a base scheme,
meaning that we are interested in the category of schemes over S. For
example, think of S = Spec k, where k is a field. If S' is another base scheme,
and if S' ----> S is a morphism, then for any scheme X over S, we let X' =
X x s S', which will be a scheme overS'. We say that X' is obtained from X
by making a base extension S' ---+ S. For example, think of S' = Speck'
where k' is an extension field of k. Note, by the way, that base extension is a
transitive operation: if S" ---+ S' ---+ S are two morphisms, then (X x s S') x s·
S" ~X Xs S".
This ties in with a general philosophy, emphasized by Grothendieck in
his "Elements de Geometrie Algebrique" ([EGA]), that one should try to
develop all concepts of algebraic geometry in a relative context. Instead of
always working over a fixed base field, and considering properties of one
variety at a time, one should consider a morphism of schemes f: X ---+ S,
and study properties of the morphism. It then becomes important to study
the behavior of properties off under base extension, and in particular, to
relate properties off to properties of the fibres off For example, iff: X ---+ S

89
II Schemes

is a morphism of finite type, and if S'----+ Sis any base extension, thenf' :X'----+ S'
is also a morphism of finite type, where X' = X x s S'. Hence we say the
property of a morphism f being of finite type is stable under base extension.
On the other hand, if for example f: X ----+ Sis a morphism of integral schemes,
the fibres off may be neither irreducible nor reduced. So the property of a
scheme being integral is not stable under base extension.

Example 3.3.1. Let k be an algebraically closed field, let

X = Spec k[x,y,t]/(ty- x 2 ),
let Y = Spec k[t], and let f: X ----+ Y be the morphism determined by the
natural homomorphism k[t] ----+ k[x,y,t]/(ty - x 2 ). Then X and Y are
integral schemes of finite type over k, and f is a surjective morphism. We
identify the closed points of Y with elements of k. For a E k, a i= 0, the fibre
Xa is the plane curve ay = x 2 in A~, which is an irreducible, reduced curve.
But for a = 0, the fibre X 0 is the nonreduced scheme given by x 2 = 0 in A 2 .
Thus we have a family (Fig. 7) in which most members are irreducible curves,
but one is nonreduced. This shows how nonreduced schemes occur naturally
even if one is primarily interested in varieties. We can say that the nonreduced
scheme x 2 = 0 in A2 is a deformation of the irreducible parabola ay = x 2
as a ----+ 0.

Figure 7. An algebraic family of schemes.

Example 3.3.2. Similarly, if X = Spec k[x,y,t]/(xy - t), we get a family

whose general member Xa is an irreducible hyperbola xy = a, when a i= 0,
but whose special member X 0 is the reducible scheme xy = 0 consisting
of two lines.

EXERCISES

3.1. Show that a morphism f:X-> Y is locally of finite type if and only if for every
open affine subset V = Spec B of Y, f- 1 (V) can be covered by open affine subsets
Ui = Spec Ai, where each Ai is a finitely generated B-algebra.

90
 3 First Properties of Schemes

3.2. A morphism f: X --> Y of schemes is quasi-compact if there is a cover of Y by open

affines V; such that f~ 1 (V;) is quasi-compact for each i. Show that f is quasi-
compact if and only if for every open affine subset V s; Y, f~ 1(V) is quasi-compact.
3.3. (a) Show that a morphism f: X --> Y is of finite type if and only if it is locally of
finite type and quasi-compact.
(b) Conclude from this that f is of finite type if and only if for er•ery open affine
subset V = Spec B of Y, f~ 1 ( V) can be covered by a finite number of open
affines U 1 = Spec A 1, where each A1 is a finitely generated B-algebra.
(c) Show also iff is of finite type, then for every open affine subset V = Spec B s;
Y, and for every open affine subset U = Spec A s; f~ 1 (V), A is a finitely gener-
ated B-algebra.
3.4. Show that a morphism f: X --> Y is finite if and only if for erery open affine subset
V = Spec B of Y, f~ 1(V) is affine, equal to Spec A, where A is a finite B-module.
3.5. A morphism f: X --> Y is quasi-finite if for every point y E Y, f~ 1(y) is a finite set.
(a) Show that a finite morphism is quasi-finite.
(b) Show that a finite morphism is closed, i.e., the image of any closed subset is
closed.
(c) Show by example that a surjective, finite-type, quasi-finite morphism need not
be finite.
3.6. Let X be an integral scheme. Show that the local ring @~ of the generic point ~
of X is a field. It is called the function field of X, and is denoted by K(X). Show
also that if U = Spec A is any open affine subset of X, then K(X) is isomorphic
to the quotient field of A.
3.7. A morphism f:X --> Y, with Y irreducible, is generically finite iff~ 1 (1'/) is a finite
set, where t) is the generic point of Y. A morphism f: X --> Y is dominant if f(X)
is dense in Y. Now let f:X--> Y be a dominant, generically finite morphism of
finite type of integral schemes. Show that there is an open dense subset U <;::: Y
such that the induced morphism f~ 1 ( U) --> U is finite. [Hint: First show that the
function field of X is a finite field extension of the function field of Y.J
3.8. Normalization. A scheme is normal if all of its local rings are integrally closed
domains. Let X be an integral scheme. For each open affine subset U = Spec A
of X, let A be the integral closure of A in its quotient field, and let U = Spec A.
Show that one can glue the schemes U to obtain a normal integral scheme X,
called the normalization of X. Show also that there is a morphism X--> X, having
the following universal property: for every normal integral scheme Z, and for every
dominant morphism f: Z--> X, f factors uniquely through X. If X is of finite type
over a field k, then the morphism X --> X is a finite morphism. This generalizes
(I, Ex. 3.1 7).

3.9. The Topological Space of a Product. Recall that in the category of varieties, the
Zariski topology on the product of two varieties is not equal to the product
topology (I, Ex. 1.4). Now we see that in the category of schemes, the underlying
point set of a product of schemes is not even the product set.
(a) Let k be a field, and let Ai = Spec k[ x J be the affine line over k. Show that
At x Speck Ai ~ Af, and show that the underlying point set of the product is
not the product of the underlying point sets of the factors (even if k is algebrai-
cally closed).

91
II Schemes

(b) Let k be a field, lets and t be indeterminates over k. Then Spec k(s), Spec k(t),
and Spec k are all one-point spaces. Describe the product scheme Spec
k(s) X Speck Spec k(t).
3.10. Fibres of a Morphism.
(a) Iff:X -+ Yis a morphism, andy E Ya point, show that sp(Xy) is homeomor-
phic tor 1 (y) with the induced topology.
(b) LetX = Speck[s,t]/(s- t2 ),let Y = Speck[s],andletf:X-+ Ybethemor-
phism defined by sending s -+ s. If y E Y is the point a E k with a =1= 0, show
that the fibre X Y consists of two points, with residue field k. If y E Y cor-
responds to 0 E k, show that the fibre X Y is a nonreduced one-point scheme.
If '1 is the generic point of Y, show that X q is a one-point scheme, whose residue
field is an extension of degree two of the residue field of '1· (Assume k alge-
braically closed.)
3.11. Closed Subschemes.
(a) Closed immersions are stable under base extension: iff: Y-+ X is a closed
immersion, and if X' -+ X is any morphism, then f': Y x x X' -+ X' is also a
closed immersion.
*(b) If Y is a closed subscheme of an affine scheme X = Spec A, then Y is also
affine, and in fact Y is the closed subscheme determined by a suitable ideal
a c;; A as the image of the closed immersion Spec A/a -+ Spec A. [Hints: First
show that Y can be covered by a finite number of open affine subsets of the
form D(h) n Y, with J; EA. By adding some more J; with D(J;) n Y = 0,
if necessary, show that we may assume that the D(J;) cover X. Next show that
f 1, . . . ,f.. generate the unit ideal of A. Then use (Ex. 2.17b) to show that Y
is affine, and (Ex. 2.18d) to show that Y comes from an ideal a c;; A.] Note: We
will give another proof of this result using sheaves of ideals later (5.10).
(c) Let Y be a closed subset of a scheme X, and give Y the reduced induced sub-
scheme structure. If Y' is any other closed subscheme of X with the same
underlying topological space, show that the closed immersion Y -+ X factors
through Y'. We express this property by saying that the reduced induced
structure is the smallest subscheme structure on a closed subset.
(d) Let f:Z-+ X be a morphism. Then there is a unique closed subscheme Y of
X with the following property: the morphism f factors through Y, and if Y'
is any other closed subscheme of X through which f factors, then Y -+ X
factors through Y' also. We call Y the scheme-theoretic image off If Z is a
reduced scheme, then Y is just the reduced induced structure on the closure of
the image f(Z).
3.12. Closed Subschemes of Proj S.
(a) Let cp: S -+ T be a surjective homomorphism of graded rings, preserving
degrees. Show that the open set U of (Ex. 2.14) is equal to Proj T, and the
morphism f: Proj T -+ Proj S is a closed immersion.
(b) If I c;; S is a homogeneous ideal, take T = Sjl and let Y be the closed sub-
scheme of X = Proj S defined as image of the closed immersion Proj Sjl-+ X.
Show that different homogeneous ideals can give rise to the same closed sub-
scheme. For example, let d0 be an integer, and let I' = EBd"'do !d. Show that
I and I' determine the same closed subscheme.
We will see later (5.16) that every closed subscheme of X comes from a ho-
mogeneous ideal I of S (at least in the case where Sis a polynomial ring over S0 ).

92
 3 First Properties of Schemes

3.13. Properties of Morphisms of Finite Type.

(a) A closed immersion is a morphism of finite type.
(b) A quasi-compact open immersion (Ex. 3.2) is of finite type.
(c) A composition of two morphisms of finite type is of finite type.
(d) Morphisms of finite type are stable under base extension.
(e) If X and Yare schemes of finite type overS, then X x s Y is of finite type over
S.
(f) If X !... Y .!!... Z are two morphisms, and iff is quasi-compact, and g c f is of
finite type, then f is of finite type.
(g) Iff: X --> Y is a morphism of finite type, and if Y is noetherian, then X is
noetherian.
3.14. If X is a scheme of finite type over a field, show that the closed points of X are
dense. Give an example to show that this is not true for arbitrary schemes.
3.15. Let X be a scheme of finite type over a field k (not necessarily algebraically closed).
(a) Show that the following three conditions are equivalent (in which case we say
that X is geometrically irreducible).
(i) X x k k is irreducible, where k denotes the algebraic closure of k. (By
abuse of notation, we write X x k k to denote X x Speck Speck.)
(ii) X x k ks is irreducible, where ks denotes the separable closure of k.
(iii) X x k K is irreducible for every extension field K of k.
(h) Show that the following three conditions are equivalent (in which case we say X
is geometrically reduced).
(i) X x k k is reduced.
(ii) X x k kP is reduced, where kP denotes the perfect closure of k.
(iii) X x k K is reduced for all extension fields K of k.
(c) We say that X is geometrically integral if X x k k is integral. Give examples of
integral schemes which are neither geometrically irreducible nor geometrically
reduced.
3.16. Noetherian Induction. Let X be a noetherian topological space, and let fJ' be a
property of closed subsets of X. Assume that for any closed subset Y of X, iffY
holds for every proper closed subset of Y, then fY holds for Y. (In particular, fY
must hold for the empty set.) Then fY holds for X.
3.17. Zariski Spaces. A topological space X is a Zariski space if it is noetherian and
every (nonempty) closed irreducible subset has a unique generic point (Ex. 2.9).
For example, let R be a discrete valuation ring, and let T = sp(Spec R). Then
T consists of two points t 0 = the maximal ideal, t 1 = the zero ideal. The open
subsets are 0. {ti}, and T. This is an irreducible Zariski space with generic point
t 1·
(a) Show that if X is a noetherian scheme, then sp(X) is a Zariski space.
(b) Show that any minimal nonempty closed subset of a Zariski space consists of
one point. We call these closed points.
(c) Show that a Zariski space X satisfies the axiom T 0 :given any two distinct
points of X, there is an open set containing one but not the other.
(d) If X is an irreducible Zariski space, then its generic point is contained in every
nonempty open subset of X.
(e) If x 0 ,x 1 are points of a topological space X, and if x 0 E {xd -, then we say
that x 1 specializes to x 0 , written x 1 ~"W+x 0 . We also say x 0 is a specialization

93
II Schemes

of x 1 , or that x 1 is a generization of x 0 . Now let X be a Zariski space. Show

that the minimal points, for the partial ordering determined by x 1 > x 0 if x 1 Nv->
x 0 , are the closed points, and the maximal points are the generic points of the
irreducible components of X. Show also that a closed subset contains every
specialization of any of its points. (We say closed subsets are stable under
specialization.) Similarly, open subsets are stable under generization.
(f) Let t be the functor on topological spaces introduced in the proof of (2.6).
If X is a noetherian topological space, show that t(X) is a Zariski space.
Furthermore X itself is a Zariski space if and only if the map r:x: X --+ t(X) is
a homeomorphism.

3.18. Constructible Sets. Let X be a Zariski topological space. A constructible subset

of X is a subset which belongs to the smallest family 0: of subsets such that (1) every
open subset is in 0:, (2) a finite intersection of elements of 0: is in 0:, and (3) the
complement of an element of 0: is in 0:.
(a) A subset of X is locally closed if it is the intersection of an open subset with a
closed subset. Show that a subset of X is constructible if and only if it can be
written as a finite disjoint union of locally closed subsets.
(b) Show that a constructible subset of an irreducible Zariski space X is dense if
and only if it contains the generic point. Furthermore, in that case it contains
a nonempty open subset.
(c) A subset S of X is closed if and only if it is constructible and stable under
specialization. Similarly, a subset T of X is open if and only if it is constructible
and stable under generization.
(d) If f:X --+ Y is a continuous map of Zariski spaces, then the inverse image of
any constructible subset of Y is a constructible subset of X.
3.19. The real importance of the notion of constructible subsets derives from the follow-
ing theorem of Chevalley-see Cartan and Chevalley [1, expose 7] and see also
Matsumura [2, Ch. 2, §6]: let f: X --+ Y be a morphism of finite type of noetherian
schemes. Then the image of any constructible subset of X is a constructible
subset of Y. In particular, f(X), which need not be either open or closed, is a
constructible subset of Y. Prove this theorem in the following steps.
(a) Reduce to showing that f(X) itself is constructible, in the case where X and Y
are affine, integral noetherian schemes, and f is a dominant morphism.
*(b) In that case, show that f(X) contains a nonempty open subset of Y by using
the following result from commutative algebra: let A <:; B be an inclusion of
noetherian integral domains, such that B is a finitely generated A-algebra.
Then given a nonzero element b E B, there is a nonzero element a E A with
the following property: if <p: A --+ K is any homomorphism of A to an algebrai-
cally closed field K, such that rp(a) oft 0, then <p extends to a homomorphism
<p' of B into K, such that <p'(b) oft 0. [Hint: Prove this algebraic result by
induction on the number of generators of B over A. For the case of one
generator, prove the result directly. In the application, take b = 1.]
(c) Now use noetherian induction on Y to complete the proof.
(d) Give some examples of morphisms f: X --+ Y of varieties over an algebraically
closed field k, to show that f(X) need not be either open or closed.
3.20. Dimension. Let X be an integral scheme of finite type over a field k (not necessarily
algebraically closed). Use appropriate results from (I, §1) to prove the following.

94
 4 Separated and Proper Morphisms

(a) For any closed point P EX, dim X = dim((} p, where for rings, we always mean
the Krull dimension.
(b) Let K(X) be the function field of X (Ex. 3.6). Then dim X = tr.d. K(X)jk.
(c) If Y is a closed subset of X, then codim(Y,X) = inf{ dim (I)P.xiP E Y}.
(d) If Y is a closed subset of X, then dim Y + codim(Y,X) = dim X.
(e) If U is a nonempty open subset of X, then dim U = dim X.
(f) If k,;; k' is a field extension, then every irreducible component of X'= X xk k'
has dimension = dim X.
3.21. Let R be a discrete valuation ring containing its residue field k. Let X =
Spec R[t] be the affine line over Spec R. Show that statements (a), (d), (e) of
(Ex. 3.20) are false for X.
*3.22. Dimension of the Fibres of a Morphism. Let f: X --+ Y be a dominant morphism
of integral schemes of finite type over a field k.
(a) Let Y' be a closed irreducible subset of Y, whose generic point rJ' is contained
in f(X). Let Z be any irreducible component of f- 1 ( Y'), such that IJ' E f(Z),
and show that codim(Z,X) ~ codim(Y',Y).
(b) Let e = dim X - dim Y be the relative dimension of X over Y. For any point
y E f(X), show that every irreducible component of the fibre Xy has dimen-
sion ~e. [Hint: Let Y' = {y}-, and use (a) and (Ex. 3.20b).]
(c) Show that there is a dense open subset U ,;; X, such that for any y E f(U),
dim UY = e. [Hint: First reduce to the case where X and Yare affine, say
X = Spec A and Y = Spec B. Then A is a finitely generated B-algebra.
Take t 1 , . . • ,teE A which form a transcendence base of K(X) over K(Y), and
let X 1 =Spec B[t 1 , •.• ,teJ. Then X 1 is isomorphic to affine e-space over Y,
and the morphism X --+ X 1 is generically finite. Now use (Ex. 3.7) above.]
(d) Going back to our original morphism f: X --+ Y, for any integer h, let Eh be
the set of points x EX such that, letting y = f(x), there is an irreducible com-
ponent Z of the fibre Xy, containing x, and having dim Z ~ h. Show that
(1) Ee = X (use (b) above); (2) if h > e, then Eh is not dense in X (use (c)
above); and (3) Eh is closed, for all h (use induction on dim X).
(e) Prove the following theorem of Chevalley-see Cartan and Chevalley [1,
expose 8]. For each integer h, let Ch be the set of points y E Y such that dim
XY = h. Then the subsets Ch are constructible, and Ce contains an open
dense subset of Y.
3.23. If V, W are two varieties over an algebraically closed field k, and if V x W is
their product, as defined in (1, Ex. 3.15, 3.16), and if t is the functor of (2.6),
then t(V X W) = t(V) X Speck t(W).

4 Separated and Proper Morphisms

We now come to two properties of schemes, or rather of morphisms between
schemes, which correspond to well-known properties of ordinary topological
spaces. Separatedness corresponds to the Hausdorff axiom for a topological
space. Properness corresponds to the usual notion of properness, namely
that the inverse image of a compact subset is compact. However, the usual
definitions are not suitable in abstract algebraic geometry, because the Zariski
topology is never Hausdorff, and the underlying topological space of a scheme

95
II Schemes

does not accurately reflect all of its properties. So instead we will use def-
initions which reflect the functorial behavior of the morphism within the
category of schemes. For schemes of finite type over C, one can show that
these notions, defined abstractly, are in fact the same as the usual notions if
we consider those schemes as complex analytic spaces in the ordinary
topology (Appendix B).
In this section we will define separated and proper morphisms. We will
give criteria for a morphism to be separated or proper using valuation rings.
Then we will show that projective space over any scheme is proper.

Definition. Let f: X -+ Y be a morphism of schemes. The diagonal morphism

is the unique morphism L1 :X -+ X x r X whose composition with both
projection maps p 1 ,p 2 :X x r X-+ X is the identity map of X-+ X. We
say that the morphism f is separated if the diagonal morphism L1 is a
closed immersion. In that case we also say X is separated over Y. A scheme
X is separated if it is separated over Spec Z.

Example 4.0.1. Let k be a field, and let X be the affine line with the origin
doubled (2.3.6). Then X is not separated over k. Indeed, X x k X is the
affine plane with doubled axes and four origins. The image of Ll is the usual
diagonal, with two of those origins. This is not closed, because all four
origins are in the closure of L1(X).

Example 4.0.2. We will see later (4.10) that if Vis any variety over an alge-
braically closed field k, then the associated scheme t(V) is separated over k.

Proposition 4.1. Iff: X -+ Y is any morphism of affine schemes, then f is

separated.
PROOF. Let X = Spec A, Y =.Spec B. Then A is a B-algebra, and X x r X
is also affine, given by Spec A @ 8 A. The diagonal morphism L1 comes from
the diagonal homomorphism A @ 8 A -+ A defined by a @a' -+ aa'. This is
a surjective homomorphism of rings, hence Ll is a closed immersion.

Corollary 4.2. An arbitrary morphism f: X -+ Y is separated if and only if

the image of the diagonal morphism is a closed subset of X x r X.
PROOF. One implication is obvious, so we have only to prove that if Ll(X) is
a closed subset, then Ll :X -+ X x r X is a closed immersion. In other words,
we have to check that L1: X -+ L1(X) is a homeomorphism, and that the
morphism of sheaves (!)XxyX-+ L1*(!)x is surjective. Let p 1 :X x r X-+ X be
the first projection. Since p 1 o L1 = idx, it follows immediately that L1 gives a
homeomorphism onto L1(X). To see that the map of sheaves (!)x x yX -+ Ll*(!)x
is surjective is a local question. For any point P EX, let U be an open affine

96
 4 Separated and Proper Morphisms

neighborhood of P which is small enough so that f(U) is contained m an

open affine subset V of Y. Then U x v U is an open affine neighborhood of
L1(P), and by the proposition, L1: U --+ U x v U is a closed immersion. So
our map of sheaves is surjective in a neighborhood of P, which completes
the proof.

Next we will discuss the valuative criterion of separatedness. The rough

idea is that in order for a scheme X to be separated, it should not contain
any subscheme which looks like a curve with a doubled point, as in the
example above. Another way of saying this is that if C is a curve, and P a
point of C, then given any morphism of C - P into X, it should admit at
most one extension to a morphism of all of C into X. (Compare (1, 6.8) where
we showed that a projective variety has this property.)
In practice, this rough idea has to be modified. The question is local, so
we replace the curve by its local ring at P, which is a discrete valuation ring.
Then since our schemes may be quite general, we must consider arbitrary
(not necessarily discrete) valuation rings. Finally, we make the criterion
relative over the image scheme Y of a morphism.
See (1, §6) for the definition and basic properties of valuation rings.

Theorem 4.3 (Valuative Criterion of Separatedness). Let f: X --+ Y be a mor-

phism of schemes, and assume that X is noetherian. Then f is separated if
and only if the following condition holds. For any field K, and for any
valuation ring R with quotient field K, let T = Spec R, let U = Spec K,
and let i: U --+ T be the morphism induced by the inclusion R c::::: K. Given
a morphism of T to Y, and given a morphism of U to X which makes a
commutative diagram

T Y,
there is at most one morphism of T to X making the whole diagram com-
mutative.

We will need two lemmas.

Lemma 4.4. Let R be a valuation ring of a field K. Let T = Spec R and let
U = SpecK. To give a morphism of U to a scheme X is equivalent to
giving a point x 1 EX and an inclusion of fields k(x 1) c::::: K. To give a
morphism of T to X is equivalent to giving two points x 0 ,x 1 in X, with x 0
a specialization (see Ex. 3.17e) ofx 1 , and an inclusion of fields k(xd c::::: K,

97
II Schemes

such that R dominates the local ring @ of x 0 on the sub scheme Z = { x 1 }-

of X with its reduced induced structure.
PROOF. U is a one-point scheme, with structure sheaf K. To give a local
homomorphism @x,.x --+ K is the same as giving an inclusion of k(x 1) <;; K,
so the first part is obvious. For the second part, let t 0 = mR be the closed
point ofT, and let t 1 = (0) be the generic point ofT. Given a morphism of
T to X, let x 0 and x 1 be the images of t 0 and t 1 . Since Tis reduced, the
morphism T--+ X factors through Z (Ex. 3.11). Furthermore, k(x 1 ) is the
function field of Z. So we have a local homomorphism of@ = @xo,z to R
compatible with the inclusion k(xd <;; K. In other words R dominates @.
Conversely, given the data consisting ofx 0 ,xt. and the inclusion k(x 1 ) <;; K
such that R dominates@, the inclusion@ --+ R gives a morphism T--+ Spec@,
which composed with the natural map Spec @ --+ X gives the desired
morphism T --+ X.
Lemma 4.5. Let f: X --+ Y be a quasi-compact morphism of schemes (see
Ex. 3.2). Then the subset f(X) of Y is closed if and only if it is stable under
specialization (Ex. 3.17e).
PROOF. One implication is obvious, so we have only to show that if f(X) is
stable under specialization, then it is closed. Clearly we may assume that
X and Y are both reduced, and that f(X)- = Y (replace Y by the reduced
induced structure on f(X)-). So let y E Y be a point. We wish to show that
y E f(X). Now we can replace Y by an affine neighborhood of y, and so
assume that Y is affine. Then since f is quasi-compact, X will be a finite
union of open affines Xi. We know that y E f(X)-. Hence y E f(XJ- for
some i. Let Y; = f(XJ- with the reduced induced structure. Then Y;
is also affine, and we will consider the dominant morphism Xi --+ Y; of
reduced affine schemes. Let Xi = Spec A and Y; = Spec B. Then the cor-
responding ring homomorphism B --+ A is injective, because the morphism
is dominant. The point y E Y; corresponds to a prime ideal p <;; B. Let
p' <;; p be a minimal prime ideal of B contained in p. (Minimal prime ideals
exist, by Zorn's lemma, because the intersection of any family of prime ideals,
totally ordered by inclusion, is again a prime ideal!) Then p' corresponds to
a point y' of Y; which specializes to y. I claim y' E f(XJ Indeed, let u~
localize A and B at p'. Localization is an exact functor, so BP' <;; A @ BP'·
Now Bp' is a field. Let q~ be any prime ideal of A @ BP'· Then q~ n BP' = (0).
Let q' <;; A be the inverse image of q~ under the localization map A --+ A @ BP'·
Then q' n B = p'. So q' corresponds to a point x' E Xi withf(x') = y'. Now
go back to the morphismf:X--+ Y. We have x' E X,f(x') = y', soy' E f(X).
But f(X) is stable under specialization by hypothesis, andy' /\1'-+ y, soy E f(X),
which is what we wanted to prove.
PROOF OF THEOREM 4.3. First suppose f is separated, and suppose given a
diagram as above where there are two morphisms h,h' of T to X making the
whole diagram commutative.

98
 4 Separated and Proper Morphisms

Then we obtain a morphism h": T --+ X x r X. Since the restrictions of h

and h' to U are the same, the generic point t 1 ofT has image in the diagonal
Ll(X). Since Ll(X) is closed, the image of t 0 is also in the diagonal. Therefore
h and h' both send the points t 0 ,t 1 to the same points x 0 ,x 1 of X. Since the
inclusions of k(x d. s K induced by h and h' are also the same, it follows
from (4.4) that hand h' are equal.
Conversely, let us suppose the condition of the theorem satisfied. To
show that f is separated, it is sufficient by (4.2) to show that Ll(X) is a closed
subset of X x r X. And since we have assumed that X is noetherian, the
morphism Ll is quasi-compact, so by (4.5) it will be sufficient to show that
Ll(X) is stable under specialization. So let ~ 1 E Ll(X) be a point, and let
~ 1 1V'-+ ~ 0 be a specialization. Let K = k(~ d and let (!) be the local ring of ~ 0
on the subscheme {~ 1 } - with its reduced induced structure. Then (!) is a
local ring contained in K, so by (I, 6.1A) there is a valuation ring R of K
which dominates(!). Now by (4.4) we obtain a morphism ofT = Spec R to
X x r X sending t 0 and t 1 to ~ 0 and ~ 1 . Composing with the projections
p 1 ,p 2 gives two morphisms ofT to X, which give the same morphism to Y,
and whose restrictions to U = Spec K are the same, since ~ 1 E Ll(X). So
by the condition, these two morphisms ofT to X must be the same. Therefore
the morphism T --+ X x r X factors through the diagonal morphism
Ll :X--+ X x r X, and so ~ 0 E Ll(X). This completes the proof. Note in the
last step it would not be sufficient to know only that p 1 (~ 0 ) = p 2 (~ 0 ). For in
general if~ EX x r X then pd~) = p 2(() does not imply¢ E Ll(X).

Corollary 4.6. Assume that all schemes are noetherian in the following state-
ments.
(a) Open and closed immersions are separated.
(b) A composition of two separated morphisms is separated.
(c) Separated morphisms are stable under base extension.
(d) If f:X--+ Y and f':X'--+ Y' are separated morphisms of schemes over
a base scheme S, then the product morphism f x f': X x s X'--+ Y x s Y'
is also separated.
(e) Iff: X --+ Y and g: Y --+ Z are two morphisms and if g o f is separated,
then f is separated.
(f) A morphism f:X--+ Y is separated if and only if Y can be covered by
open subsets V; such that f- 1 ( V;) --+ V; is separated for each i.

PROOF. These statements all follow immediately from the condition of the
theorem. We will give the proof of (c) to illustrate the method. Let f: X --+ Y

99
II Schemes

be a separated morphism, let Y' --+ Y be any morphism, and let X' =
X x y Y' be obtained by base extension. We must show that f':X'--+ Y'
is separated. So suppose we are given morphisms of T to Y' and U to X' as
in the theorem, and two morphisms of T to X' making the diagram
------+X

T ---------> Y' y
commutative. Composing with the map X' --+ X, we obtain two morphisms
of T to X. Since f is separated, these are the same. But X' is the fibred
product of X and Y' over Y, so by the universal property of the fib red prod-
uct, the two maps of T to X' are the same. Hence f' is separated.

Note on Noetherian Hypotheses. You have probably noticed that in order

to apply the theorem, it is not necessary to assume that all the schemes
mentioned in the corollary are noetherian. In fact, even in the theorem
itself, you can get by with assuming something less than X noetherian (see
Grothendieck [EGA I, new ed., 5.5.4]). My feeling is that if a noetherian
hypothesis will make statements and proofs substantially simpler, then I
will make that hypothesis, even though it may not be necessary. My justi-
fication for this attitude is that most of the motivation and examples in
algebraic geometry come from schemes of finite type over a field, and
constructions made from them, and practically all the schemes encountered
in this way are noetherian. This attitude will prevail in Chapter III, where
noetherian hypotheses are built into the very foundations of our treatment
of cohomology. The reader who wishes to avoid noetherian hypotheses is
advised to read [EGA], especially [EGA IV, §8].

Definition. A morphism f: X --+ Y is proper if it is separated, of finite type,

and universally closed. Here we say that a morphism is closed if the
image of any closed subset is closed. A morphism f: X --+ Y is universally
closed if it is closed, and for any morphism Y' --+ Y, the corresponding
morphism f': X' --+ Y' obtained by base extension is also closed.

Example 4.6.1. Let k be a field and let X be the affine line over k. Then X
is separated and of finite type over k, but it is not proper over k. Indeed,
take the base extension X --+ k. The map X x k X --+ X we obtain is the
projection map of the affine plane onto the affine line. This is not a closed
map. For example, the hyperbola given by the equation xy = 1 is a closed
subset of the plane, but its image under projection consists of the affine line
minus the origin, which is not closed.
Of course it is clear that what is missing in this example is the point at
infinity on the hyperbola. This suggests that the projective line would be

100
 4 Separated and Proper Morphisms

proper over k. In fact, we will see later (4.9) that any projective variety over
a field is proper.

Theorem 4.7 (Valuative Criterion of Properness). Let f:X--+ Y be a mor-

phism of finite type, with X noetherian. Then f is proper if and only if
for every valuation ring R and for every morphism of U to X and T to Y
forming a commutative diagram

u X

i j ..................................../jf
T y

(using the notation of (4.3) }, there exists a unique morphism T--+ X making
the whole diagram commutative.
PROOF. First assume that f is proper. Then by definition f is separated,
so the uniqueness of the morphism T--+ X will follow from (4.3), once we
know it exists. For the existence, we consider the base extension T--+ Y,
and let X r = X x y T. We get a map U --+ X T from the given maps U --+ X
and U--+ T.
U----~ Xr - - - - - - X

T y

Let ~ 1 EX r be the image of the unique point t 1 of U. Let Z = g d-.

Then Z is a closed subset of X T· Since f is proper, it is universally closed,
so the morphism f': X T --+ T must be closed, so f'(Z) is a closed subset
ofT. But f'(~ 1 } = tt> which is the generic point ofT, so in fact f'(Z) = T.
Hence there is a point ~ 0 E Z with f'(~ 0 } = t 0 . So we get a local homo-
morphism of local rings R --+ (l)~o.z corresponding to the morphism f'.
Now the function field of Z is k(~ 1 }, which is contained inK, by construc-
tion of~ 1 . By (1, 6.1A), R is maximal for the relation of domination between
local subrings of K. Hence R is isomorphic to (!) ~ 0 .z, and in particular R
dominates it. Hence by (4.4) we obtain a morphism ofT to Xr sending
t 0 ,t 1 to ~ 0 ,~ 1 . Composing with the map X r --+ X gives the desired morphism
ofT to X.
Conversely, suppose the condition of the theorem holds. To show f is
proper, we have only to show that it is universally closed, since it is of finite
type by hypothesis, and it is separated by (4.3). So let Y' --+ Y be any mor-
phism, and let f':X'--+ Y' be the morphism obtained from f by base ex-
tension. Let Z be a closed subset of X', and give it the reduced induced
structure.

101
II Schemes

Z s;: X' -----~ X

Y' - - - - - - - - + y

We need to show that f'(Z) is closed in Y'. Since f is of finite type, so is f'
and so is the restriction off' to Z (Ex. 3.13). In particular, the morphism
f':Z--+ Y' is quasi-compact, so by (4.5) we have only to show that f(Z) is
stable under specialization. So let z1 E Z be a point, let Y1 = f'(zd, and let
y 1 ~ y 0 be a specialization. Let (!) be the local ring of y 0 on { y 1 }- with its
reduced induced structure. Then the quotient field of(!) is k(y 1 ), which is a
subfield of k(z 1 ). Let K = k(z d, and let R be a valuation ring of K which
dominates(!) (which exists by (I, 6.1A) ).
From this data, by (4.4) we obtain morphisms U--+ Z and T--+ Y'
forming a commutative diagram
u --------+ z

j
T -------+ Y'.

Composing with the morphisms Z --+ X' --+ X and Y' --+ Y, we get mor-
phisms U --+ X and T --+ Y to which we can apply the condition of the
theorem. So there is a morphism of T --+ X making the diagram commute.
Since X' is a fibred product, it lifts to give a morphism T --+ X'. And since
Z is closed, and the generic point of T goes to z 1 E Z, this morphism factors
to give a morphism T--+ Z. Now let z 0 be the image of t0 . Then f'(z 0 ) =
y 0 , so Yo E f'(Z). This completes the proof.

Corollary 4.8. In the following statements, we take all schemes to be noetherian.

(a) A closed immersion is proper.
(b) A composition of proper morphisms is proper.
(c) Proper morphisms are stable under base extension.
(d) Products of proper morphisms are proper as in (4.6d).
(e) If f:X--+ Y and g: Y--+ Z are two morphisms, if go f is proper,
and if g is separated, then f is proper.
(f) Properness is local on the base as in (4.6£).
PROOF. These results follow immediately from the condition of the theorem,
taking into account (Ex. 3.13) which deals with the finite type property,
and (4.6). We will give the proof of (e) to illustrate the method. Assume
go f is proper and g is separated. Then f is of finite type by (Ex. 3.13). (We
have assumed that X is noetherian, so f is automatically quasi-compact.)
Also f is separated by (4.6). So we have to show that given a valuation
ring R, and morphisms U --+ X and T --+ Y making a commutative diagram,

102
 4 Separated and Proper Morphisms

----------~
T y

then there exists a morphism of T to X making the diagram commutative.

Let T ~ Z be the composed map. Then since g o f is proper, there is a
map of T to X commuting with the map of T ~ Z. By composing with f,
we get a second map of T to Y. But now since g is separated, the two maps
of T to Y are the same, so we are done.

Our next objective is to define projective morphisms and to show that

any projective morphism is proper. Recall that in Section 2 we defined
projective n-space PA over any ring A to be Proj A[ x0 , . . . ,xnJ. Note that
if A ~ B is a homomorphism of rings, and Spec B ~ Spec A is the corre-
sponding morphism of affine schemes, then P~ ;::;: PA x spec A Spec B. In
particular, for any ring A, we have PA ;::;: P~ x Spec z Spec A. This motivates
the following definition for any scheme Y.

Definition. If Y is any scheme, we define projective n-space over Y, denoted

P~, to be P~ x spec z Y. A morphism f:X ~ Y of schemes is projective
if it factors into a closed immersion i:X ~ P~ for some n, followed by
the projection P~ ~ Y. A morphism f:X ~ Y is quasi-projective if it
factors into an open immersionj:X ~X' followed by a projective mor-
phism g: X' ~ Y. (This definition of projective morphism is slightly
different from the one in Grothendieck [EGA II, 5.5]. The two definitions
are equivalent in case Y itself is quasi-projective over an affine scheme.)

Example 4.8.1. Let A be a ring, letS be a graded ring with S 0 = A, which

is finitely generated as an A-algebra by S 1 . Then the natural map Proj S ~
Spec A is a projective morphism. Indeed, by hypothesis S is a quotient of
a polynomial ring S' = A[x0 , . . . ,xnJ. The surjective homomorphism of
graded rings S' ~ S gives rise to a closed immersion Proj S ~ Proj S' =
PA, which shows that Proj Sis projective over A (Ex. 3.12).
Theorem 4.9. A projective morphism of noetherian schemes is proper. A quasi-
projective morphism of noetherian schemes is of finite type and separated.
PROOF. Taking into account the results of (Ex. 3.13) and (4.6) and (4.8), it
will be sufficient to show t~at X = P~ is proper over Spec Z. Recall by
(2.5) that X is a union of open affine subsets v; = D + (x;), and that v; is

103
II Schemes

isomorphic to Spec Z[ x 0/x;, . .. ,xn/xJ. Thus X is of finite type. To show

that X is proper, we will use the criterion of (4.7) and imitate the proof
of (I, 6.8). So suppose given a valuation ring R and morphisms U --+ X,
T --+ Spec Z as shown:

T - - - - - - - - + Spec Z.

Let ~ 1 EX be the image of the unique point of U. Using induction on n,

we may assume that ~ 1 is not contained in any of the hyperplanes X - V;,
which are each isomorphic to P"- 1 . In other words, we may assume that
~1 En J.'i, and hence all of the functions x;/xj are invertible elements of the
local ring @~~.
We have an inclusion k(~ 1 ) s;: K given by the morphism U --+X. Let
fu E K be the image of x;/xi. Then the fii are nonzero elements of K, and
hk = hi · jjk for all i,j,k. Let v: K --+ G be the valuation associated to the
valuation ring R. Let g; = v(/; 0 ) for i = 0, ... ,n. Choose k such that gk
is minimal among the set {g 0 , . . . ,gn}, for the ordering of G. Then for
each i we have

hence hk E R for i = 0, ... ,n. Then we can define a homomorphism

cp: Z[ x 0 /xk, ... ,xn/xk] -4 R
by sending x;/xk to hk· It is compatible with the given field inclusion k(~ d s;:
K. This homomorphism cp gives a morphism T --+ ~' and hence a mor-
phism ofT to X which is the one required. The uniqueness of this morphism
follows from the construction and the way the V; patch together.

Proposition 4.10. Let k be an algebraically closed field. The image of the

functor t: l!Jar(k) --+ 6cl)(k) of (2.6) is exactly the set of quasi-projective
integral schemes over k. The image of the set of projective varieties is the
set of projective integral schemes. In particular, for any variety V, t(V)
is an integral, separated scheme of finite type over k.
PROOF. We have already seen in Section 3 that for any variety V, the asso-
ciated scheme t( V) is integral and of finite type over k. Since varieties were
defined as locally closed subsets of projective space (I, §3), it is clear that
t(V) is also quasi-projective.
For the converse, it will be sufficient to show that any projective integral
scheme Y over k is in the image of t. Let Y be a closed subscheme of PZ,
and let V be the set of closed points of Y. Then V is a closed subset of the
variety P". Since V is dense in Y (Ex. 3.14) we see that V is irreducible, so
V is a projective variety, and we see also that t(V) and Y have the same

104
 4 Separated and Proper Morphisms

underlying topological space. But they are both reduced closed subschemes
of PZ, so they are isomorphic (Ex. 3.11).

Definition. An abstract variety is an integral separated scheme of finite type

over an algebraically closed field k. If it is proper over k, we will also
say it is complete.

Remark 4.10.1. From now on we will use the word "variety" to mean
"abstract variety" in the sense just defined. We will identify the varieties
of Chapter I with their associated schemes, and refer to them as quasi-
projective varieties. We will use the words "curve," "surface," "three-fold,"
etc., to mean an abstract variety of dimension 1, 2, 3, etc.

Remark 4.10.2. The concept of an abstract variety was invented by Weil

[1]. He needed it to provide a purely algebraic construction of the Jacobian
variety of a curve, which at first appeared only as an abstract variety
(Weil [2]). Then Chow [3] gave a different construction of the Jacobian
variety showing that it was in fact a projective variety. Later Weil [6]
himself showed that all abelian varieties were projective.
Meanwhile Nagata [1] found an example of a complete abstract non-
projective variety, showing that in fact the new class of abstract varieties
is larger than the class of projective varieties.
We can sum up the present state of knowledge of this subject as follows.
(a) Every complete curve is projective (III, Ex. 5.8).
(b) Every nonsingular complete surface is projective (Zariski [5]). See also
Hartshorne [ 5, 11.4.2].
(c) There exist singular nonprojective complete surfaces (Nagata [3]). See
also (Ex. 7.13) and (III, Ex. 5.9).
(d) There exist nonsingular complete nonprojective three-folds (Nagata
[ 4], Hironaka [2], and (Appendix B)).
(e) Every variety can be embedded as an open dense subset of a complete
variety (Nagata [6]).

The following algebraic result will be used in (Ex. 4.6).

Theorem 4.11A. 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 which contain A.
PROOF. Bourbaki [1, Ch. VI, §1, no. 3, Thm. 3, p. 92].

EXERCISES

4.1. Show that a finite morphism is proper.

4.2. Let S be a scheme, let X be a reduced scheme over S, and let Y be a separated
scheme over S. Let f and g be two S-morphisms of X to Y which agree on an
open dense subset of X. Show that f = g. Give examples to show that this

105
II Schemes

result fails if either (a) X is nonreduced, or (b) Y is nonseparated. [Hint: Consider

the map h:X-> Y x s Y obtained from f and g.]
4.3. Let X be a separated scheme over an affine scheme S. Let U and V be open
affine subsets of X. Then U n Vis also affine. Give an example to show that this
fails if X is not separated.
4.4. Let f: X -> Y be a morphism of separated schemes of finite type over a noetherian
scheme S. Let Z be a closed subscheme of X which is proper over S. Show that
f(Z) is closed in Y, and that f(Z) with its image subscheme structure (Ex. 3.11d)
is proper over S. We refer to this result by saying that "the image of a proper
scheme is proper." [Hint: Factor f into the graph morphism r 1 :X-> X Xs Y
followed by the second projection p 2 , and show that r 1 is a closed immersion.]
4.5. Let X be an integral scheme of finite type over a field k, having function field K.
We say that a valuation of Kjk (see I, §6) has center x on X if its valuation ring R
dominates the local ring (!Jx,x·
(a) If X is separated over k, then the center of any valuation of Kjk on X (if it
exists) is unique.
(b) If X is proper over k, then every valuation of Kjk has a unique center on X.
*(c) Prove the converses of(a) and (b). [Hint: While parts (a) and (b) follow quite
easily from (4.3) and (4.7), their converses will require some comparison of
valuations in different fields.]
(d) If X is proper over k, and if k is algebraically closed, show that r(X,(!Jx) = k.
This result generalizes (1, 3.4a). [Hint: Let a E r(X,(!Jx), with a¢ k. Show that
there is a valuation ring R of Kjk with a- 1 E mR. Then use (b) to get a con-
tradiction. J
Note. If X is a variety over k, the criterion of (b) is sometimes taken as the de-
finition of a complete variety.
4.6. Let f: X -> Y be a proper morphism of affine varieties over k. Then f is a finite
morphism. [Hint: Use (4.11A).]
4.7. Schemes Over R. For any scheme X 0 over R, let X= X 0 x R C. Let ocC-> C be
complex conjugation, and let a: X -> X be the automorphism obtained by keeping
X 0 fixed and applying r:t. to C. Then X is a scheme over C, and a is a semi-linear
automorphism, in the sense that we have a commutative diagram
X _______a____~
X

l
Spec C
l
Spec C.

Since a 2 = id, we call a an involution.

(a) Now let X be a separated scheme of finite type over C, let a be a semilinear
involution on X, and assume that for any two points xl>x 2 EX, there is an
open affine subset containing both of them. (This last condition is satisfied
for example if X is quasi-projective.) Show that there is a unique separated
scheme X 0 of finite type over R, such that X 0 x R C ~ X, and such that this
isomorphism identifies the given involution of X with the one on X 0 x R C
described above.

106
 4 Separated and Proper Morphisms

For the following statements, X 0 will denote a separated scheme of finite

type over R, and X,a will denote the corresponding scheme with involution
over C.
(b) Show that X 0 is affine if and only if X is.
(c) If X 0 ,Y0 are two such schemes over R, then to give a morphism f 0 :X 0 -+ Y0
is equivalent to giving a morphism f:X-+ Y which commutes with the in-
volutions, i.e., f o ax = ay of
(d) If X~ A~, then X 0 ~ A~.
(e) If X ~ P~, then either X 0 ~ P~, or X 0 is isomorphic to the conic in Pi given
by the homogeneous equation x~ + xf + x~ = 0.
4.8. Let [l}J be a property of morphisms of schemes such that:
(a) a closed immersion has&;
(b) a composition of two morphisms having [l}J has&;
(c) [l}J is stable under base extension.
Then show that:
(d) a product of morphisms having & has&;
(e) if f:X-+ Y and g: Y-+ Z are two morphisms, and if go f has & and g is
separated, then f has&;
(f) If f:X -+ Y has&, then feed :X"d -+ Yced has&.
[Hint: For (e), consider the graph morphism r 1 :X-+ X x z Y and note that
it is obtained by base extension from the diagonal morphism Ll: Y -+ Y x z Y.]
4.9. Show that a composition of projective morphisms is projective. [Hint: Use the
Segre embedding defined in (1, Ex. 2.14) and show that it gives a closed immersion
P' x ps -+ prs+r+s.J Conclude that projective morphisms have properties
(a)-(f) of (Ex. 4.8) above.
*4.10. Chow's Lemma. This result says that proper morphisms are fairly close to pro-
jective morphisms. Let X be proper over a noetherian scheme S. Then there is
a scheme X' and a morphism g: X' -+ X such that X' is projective over S, and
there is an open dense subset U ~ X such that g induces an isomorphism of
g- 1 (U) to U. Prove this result in the following steps.
(a) Reduce to the case X irreducible.
(b) Show that X can be covered by a finite number of open subsets V;, i = 1, . .. ,n,
each of which is quasi-projective over S. Let Vi -+ Pi be an open immersion
of Vi into a scheme Pi which is projective overS.
n
(c) Let V = Vi, and consider the map
f:V-+X x 5 P 1 x 5 • .. x 5 Pn
deduced from the given maps V -+ X and V -+ Pi. Let X' be the closed image
subscheme structure (Ex. 3.1ld) f( V) -. Let g: X' -+ X be the projection onto
the first factor, and let h: X' -+ P = P 1 x s . . . x s P" be the projection onto
the product of the remaining factors. Show that h is a closed immersion,
hence X' is projective over S.
(d) Show that g- 1(V) -+ Vis an isomorphism, thus completing the proof.
4.11. If you are willing to do some harder commutative algebra, and stick to noetherian
schemes, then we can express the valuative criteria of separatedness and properness
using only discrete valuation rings.
(a) If (IJ,m is a noetherian local domain with quotient field K, and if Lis a finitely
generated field extension of K, then there exists a discrete valuation ring R of

107
II Schemes

L dominating @. Prove this in the following steps. By taking a polynomial

ring over (1), reduce to the case where Lis a finite extension field of K. Then
show that for a suitable choice of generators x 1, . . . ,x" of m, the ideal a = (x 1 )
in (!!' = (I [x 2 /x 1, ••. ,x"/x 1] is not equal to the unit ideal. Then let p be a
minimal prime ideal of a, and let (!!'~'be the localization of(!!' at p. This is a
noetherian local domain of dimension 1 dominating@. Let lP'~'be the integral
closure of (!!'~'in L. Use the theorem of Krull-Akizuki (see Nagata [7, p. 115])
to show that if·., is noetherian of dimension 1. Finally, take R to be a local-
ization of if'., at one of its maximal ideals.
(b) Let f: X -+ Y be a morphism of finite type of noetherian schemes. Show that
f is separated (respectively, proper) if and only if the criterion of (4.3) (respec-
tively, (4.7)) holds for all discrete valuation rings.
4.12. Examples of Valuation Rings. Let k be an algebraically closed field.
(a) If K is a function field of dimension I over k (I, §6), then every valuation ring
of Kjk (except for K itself) is discrete. Thus the set of all of them is just the
abstract nonsingular curve CK of (1, §6).
(b) If K/k is a function field of dimension two, there are several different kinds of
valuations. Suppose that X is a complete·nonsingular surface with function
field K.
(I) If Y is an irreducible curve on X, with generic point X~o then the local ring
R = (!! x,.x is a discrete valuation ring of Kjk with center at the (nonclosed)
point x 1 on X.
(2) Iff: X' -+ X is a birational morphism, and if Y' is an irreducible curve in
X' whose image in X is a single closed point x 0 , then the local ring R of
the generic point of Y' on X' is a discrete valuation ring of Kjk with center
at the closed point x 0 on X.
(3) Let x 0 E X be a closed point. Let f: X 1 -+ X be the blowing-up of x 0
(I, §4) and let E 1 = f- \x 0 ) be the exceptional curve. Choose a closed
point x 1 E £ 1 , let / 2 : X 2 -+ X 1 be the blowing-up of X~o and let £ 2 =
/2 1(x 1 ) be the exceptional curve. Repeat. In this manner we obtain a
sequence of varieties X; with closed points X; chosen on them, and for
each i, the local ring (lx,+,.x,., dominates (!ix,,x,. Let R 0 = Ui~o (!ix,.x,·
Then R 0 is a local ring, so it is dominated by some valuation ring R of
Kjk by (I, 6.1A). Show that R is a valuation ring of Kjk, and that it has
center x 0 on X. When is R a discrete valuation ring?
Note. We will see later (V, Ex. 5.6) that in fact the R 0 of(3) is already a valuation
ring itself, so R 0 = R. Furthermore, every valuation ring of Kjk (except for K
itself) is one of the three kinds just described.

5 Sheaves of Modules
So far we have discussed schemes and morphisms between them without
mentioning any sheaves other than the structure sheaves. We can increase
the flexibility of our technique enormously by considering sheaves of modules
on a given scheme. Especially important are quasi-coherent and coherent
sheaves, which play the role of modules (respectively, finitely generated
modules) over a ring.

108
 5 Sheaves of Modules

In this section we will develop the basic properties of quasi-coherent and

coherent sheaves. In particular we will introduce the important "twisting
sheaf" @(1) of Serre on a projective scheme.
We will start by defining sheaves of modules on a ringed space.

Definitions. Let (X,@x) be a ringed space (see §2). A sheaf of @x-modules

(or simply an @x-module) is a sheaf !F on X, such that for each open set
U s;:: X, the group !F(U) is an @x(U)-module, and for each inclusion of
open sets V s;:: U, the restriction homomorphism !F(U) --+ !F(V) is com-
patible with the module structures via the ring homomorphism @x(U) --+
@x(V). A morphism !F --+ ':§ of sheaves of @x-modules is a morphism of
sheaves, such that for each open set U s;:: X, the map !F(U) --+ ':#(U) is a
homomorphism of @x(U)-modules.
Note that the kernel, co kernel, and image of a morphism of @x-modules
is again an @x-module. If !F' is a subsheaf of @x-modules of an @x-module
!F, then the quotient sheaf !Fj!F' is an @x-module. Any direct sum,
direct product, direct limit, or inverse limit of (9 x-modules is an (9 x-module.
If !F and':§ are two @x-modules, we denote the group ofmorphisms from
!F to ':§ by Homlllx(!F,':#), or sometimes Homx(ff,':#) or Hom(!#','§) if no
confusion can arise. A sequence of @x-modules and morphisms is exact
if it is exact as a sequence of sheaves of abelian groups.
If U is an open subset of X, and if !F is an @x-module, then :Flu is an
@xlu-module. If !F and ':§ are two @x-modules, the presheaf
U f--+ Homlllxlu(!Fiu,':#lu)
is a sheaf, which we call the sheaf J'fom (Ex. 1.15), and denote by
J'fomlllx(!F,':#). It is also an @x-module.
We define the tensor product !F ®lllx ':§ of two @x-modules to be the
sheaf associated to the presheaf U f--+ !F(U) ®lllx(UJ ':#(U). We will often
write simply !F 0 ':#,with (Qx understood.
An @x-module !F is free if it is isomorphic to a direct sum of copies of
@x. It is locally free if X can be covered by open sets U for which :Flu
is a free @xlu-module. In that case the rank of !F on such an open set is
the number of copies of the structure sheaf needed (finite or infinite).
If X is connected, the rank of a locally free sheaf is the same everywhere.
A locally free sheaf of rank 1 is also called an invertible sheaf
A sheaf of ideals on X is a sheaf of modules f which is a subsheaf
of (Qx· In other words, for every open set U, f(U) is an ideal in @x(U).
Let f:(X,@x)--+ (Y,@y} be a morphism of ringed spaces (see §2). If
!F is an @x-module, then f*!F is an f*@x-module. Since we have the
morphism f# :@y--+ f*(Qx of sheaves of rings on Y, this gives f*!F a
natural structure of @y-module. We call it the direct image of !F by the
morphism!
Now let':§ be a sheaf of @y-modules. Then f- 1 '§ is an f- 1 @y-module.
Because of the adjoint property off - l (Ex. 1.18) we have a morphism

109
II Schemes

f- 1 (!Jy--+ (!Jx of sheaves of rings on X. We define f*<§ to be the tensor

product
f-l<g ®J-•i!!y (!Jx·
Thus f*<§ is an @x-module. We call it the inverse image of <§ by the
morphism!
As in (Ex. 1.18) one can show that f* and f* are adjoint functors
between the category of (!Jx-modules and the category of @y-modules.
To be precise, for any @x-module ff and any @y-module <§, there is a
natural isomorphism of groups
Homl!!x(f*<§,ff) ~ Homi!Jy(<§,f*ff).
Now that we have the general notion of a sheaf of modules on a ringed
space, we specialize to the case of schemes. We start by defining the sheaf
of modules M on Spec A associated to a module M over a ring A.

Definition. Let A be a ring and let M be an A-module. We define the sheaf

associated to M on Spec A, denoted by M, as follows. For each prime
ideal p s;:: A, let MP be the localization of M at p. For any open set
U s;:: Spec A we define the group M(U) to be the set of functions s: U --+
Upe u MP such that for each p E U, s(p) E MP, and such that sis locally
a fraction m/f with mE M and f EA. To be precise, we require that for
each p E U, there is a neighborhood V of p in U, and there are elements
mE M and f E A, such that for each q E V, f 4 q, and s(q) = m/f in Mq.
We make Minto a sheaf by using the obvious restriction maps.

Proposition 5.1. Let A be a ring, let M be an A-module, and let M be the

sheaf on X = Spec A associated toM. Then:
(a) M is an (!Jx-module;
(b) for each p EX, the stalk (M)v of the sheaf M at p is isomorphic to
the localized module Mv;
(c) for any f E A, the A rmodule M(D(f)) is isomorphic to the localized
module M 1 ;
(d) in particular, r(X,M) = M.
PROOF. Recalling the construction of the structure sheaf (!Jx from §2, it is
clear that M is an @x-module. The proofs of (b), (c), (d) are identical to the
proofs of (a), (b), (c) of (2.2), replacing A by Mat appropriate places.

Proposition 5.2. Let A be a ring and let X = Spec A. Also let A --+ B be a
ring homomorphism, and let f: Spec B --+ Spec A be the corresponding
morphism of spectra. Then:
(a) the map M --+ M gives an exact, fully faithful functor from the
category of A-modules to the category of @x-modules;
(b) if M and N are two A-modules, then (M ®A Nf ~ M ®l!!x N;
(c) if {M;} is any family of A-modules, then (ffiM;f ~ ffiMi;

110
 5 Sheaves of Modules

(d) for any B-module N we have f*(il) ~ (AN)-, where AN means N

considered as an A -module;
(e) for any A-module M we have f*(M) ~ (M ®A B)-.
PROOF. The map M ...... M is clearly functorial. It is exact, because localiza-
tion is exact, and exactness of sheaves can be measured at the stalks (use
(Ex. 1.2) and (5.lb)). It commutes with direct sum and tensor product,
because these commute with localization. To say it is fully faithful means
that for any A-modules M and N, we have HomA(M,N) = Hom(l)x(M,N).
The functor ~ gives a natural map HomA(M,N) ...... Hom(l)x(M,N). Applying
r and using (5.ld) gives a map the other way. These two maps are clearly
inverse to each other, hence isomorphisms. The last statements about f*
and f* follow directly from the definitions.

These sheaves of the form M on affine schemes are our models for quasi-
coherent sheaves. A quasi-coherent sheaf on a scheme X will be an mx-
module which is locally of the form M. In the next few lemmas and propo-
sitions, we will show that this is a local property, and we will establish some
facts about quasi-coherent and coherent sheaves.

Definition. Let (X,(!)x) be a scheme. A sheaf of mx-modules ~ is quasi-

coherent if X can be covered by open affine subsets U; = Spec A;, such
that for each i there is an A;-module M; with ~lu, ~ M;. We say that
~ is coherent if furthermore each M; can be taken to be a finitely gen-
erated A;-module.

Although we have just defined the notion of quasi-coherent and coherent

sheaves on an arbitrary scheme, we will normally not mention coherent
sheaves unless the scheme is noetherian. This is because the notion of
coherence is not at all well-behaved on a nonnoetherian scheme.

Example 5.2.1. On any scheme X, the structure sheaf (!)x is quasi-coherent

(and in fact coherent).

Example 5.2.2. If X = Spec A is an affine sc.heme, if Y ~ X is the closed

subscheme defined by an ideal a s; A (3.2.3), and if i: Y ...... X is the in-
clusion morphism, then i*(!)y is a quasi-coherent (in fact coherent) mx-
module. Indeed, it is isomorphic to (A/ar.

Example 5.2.3. If U is an open subscheme of a scheme X, with inclusion

map j: U ...... X, then the sheaf j!((!)u) obtained by extending (!)u by zero
outside of U (Ex. 1.19), is an mx-module, but it is not in general quasi-
coherent. For example, suppose X is integral, and V = Spec A is any
open affine subset of X, not contained in U. Then j!((!)u )lv has no global

111
II Schemes

sections over V, and yet it is not the zero sheaf. Hence it cannot be of the
form M for any A-module M.

Example 5.2.4. If Y is a closed subscheme of a scheme X, then the sheaf

(()xlr is not in general quasi-coherent on Y. In fact, it is not even an @y-
module in general.

Example 5.2.5. Let X be an integral noetherian scheme, and let X be the

constant sheaf with group K equal to the function field of X (Ex. 3.6). Then
X is a quasi-coherent (()x-module, but it is not coherent unless X is reduced
to a point.

Lemma 5.3. Let X ;= Spec A be an affine scheme, let f E A, let D(f) s;: X
be the corresponding open set, and let §' be a quasi-coherent sheaf on X.
(a) If s E r(X,ff') is a global section of§' whose restriction to D(f)
is 0, then for some n > 0, rs = 0.
(b) Given a section t E §'(D(f)) of §' over the open set D(f), then
for some n > 0, rt extends to a global section of §' over X.
PROOF. First we note that since §' is quasi-coherent, X can be covered by
open affine subsets of the form V = Spec B, such that ff'lv ~ M for some
B-module M. Now the open sets ofthe form D(g) form a base for the topology
of X (see §2), so we can cover V by open sets of the form D(g), for various
g EA. An inclusion D(g) s;: V corresponds to a ring homomorphism
B---+ A 9 by (2.3). Hence 9""lv<o> ~ (M 0B A 9 ) - by (5.2). Thus we have shown
that if§' is quasi-coherent on X, then X can be covered by open sets of the
form D(g;) where for each i, ff'lv(g;) ~ Mi for some module Mi over the ring
A 9 ,. Since X is quasi-compact, a finite number of these open sets will do.
(a) Now suppose givens E r(X,9"") with siv<n = 0. For each i, s restricts
to give a section si of§' over D(g;), in other words, an element si E Mi (using
(5.ld) ). Now D(f) n D(g;) = D(fg;), so 9""lv(Jgd = (MJj using (5.lc). Thus
the image of si in (MJ1 is zero, so by the definition of localization, rsi = 0
for some n. This n may depend on i, but since there are only finitely many i,
we can pick n large enough to work for them all. Then since the D(g;) cover
X, we have rs = 0.
(b) Given an element t E :#'(D(f) ), we restrict it for each ito get an element
t of :#'(D(fg;)) = (M;) 1 . Then by the definition of localization, for some
n > 0 there is an element ti E Mi = :#'(D(g;)) which restricts to rt on
D(fgJ The integer n may depend on i, but again we take one large enough
to work for all i. Now on the intersection D(g;) n D(gj) = D(gigj) we have
two sections ti and tj of §', which agree on D(fgigj) where they are both
equal to rt. Hence by part (a) above, there is an integer m > 0 such that
fm(ti - tj) = 0 on D(gigj). This m depends on i and j, but we take one m
large enough for all. Now the local sections fmti of§' on D(gJ glue together
to give a global sections of:#', whose restriction to D(f) is r+mt.

112
 5 Sheaves of Modules

Proposition 5.4. Let X be a scheme. Then an (l}x-module ff is quasi-coherent if

and only if for every open affine subset U = Spec A of X, there is an A-
module M such that fflu ~ M. If X is noetherian, then ff is coherent if
and only if the same is true, with the extra condition that M be a finitely
generated A-module.
PROOF. Let ff be quasi-coherent on X, and let U = Spec A be an open
affine. As in the proof of the lemma, there is a base for the topology consisting
of open affines for which the restriction of ff is the sheaf associated to a
module. It follows that.fffu is quasi-coherent, so we can reduce to the case
X affine = Spec A. Let M = r(X,ff). Then in any case there is a natural
map rx: M --. ff (Ex. 5.3). Since ff is quasi-coherent, X can be covered by
open sets D(g;) with fflv(g;) ~ M; for some A9 ,-module M;. Now the lemma,
applied to the open set D(g;), tells us exactly that ff(D(g;)) ~ M 9 ,, so
M; = M 9 ,. It follows that the map rx, restricted to D(g;), is an isomorphism.
The D(g;) cover X, so rx is an isomorphism.
Now suppose that X is noetherian, and ff coherent. Then, using the
above notation, we have the additional information that each M 9 , is a
finitely generated A9 ,-module, and we want to prove that M is finitely
generated. Since the rings A and A 9 , are noetherian, the modules M 9 , are
noetherian, and we have to prove that M is noetherian. For this we just use
the proof of (3.2) with A replaced by M in appropriate places.

Corollary 5.5. Let A be a ring and let X = Spec A. The functor M r--+ M
gives an equivalence of categories between the category of A-modules and
the category of quasi-coherent (!} x-modules. Its inverse is the functor
ff r--+ r(X,ff). If A is noetherian, the same functor also gives an equiv-
alence of categories between the category of finitely generated A-modules
and the category of coherent (l}x-modules.
PROOF. The only new information here is that ff is quasi-coherent on X if
and only if it is of the form M, and in that case M = r(X,ff). This follows
from (5.4).

Proposition 5.6. Let X be an affine scheme, let 0 --. ff' --. ff --. ff" --. 0 be
an exact sequence of (l}x-modules, and assume that ff' is quasi-coherent.
Then the sequence
0 --. r(X,ff') --. r(X,ff) --. r(X,ff") --. 0
is exact.
PROOF. We know already that r is a left-exact functor (Ex. 1.8) so we have
only to show that the last map is surjective. Let s E r(X,ff") be a global
section of ff". Since the map of sheaves ff --. ff" is surjective, for any
x E X there is an open neighborhood D(f) of x, such that sfv<n lifts to a
section t E ff(D(f)) (Ex. 1.3). I claim that for some n > 0, f"s lifts to a
global section of ff. Indeed, we can cover X with a finite number of open

I I3
II Schemes

sets D(g;), such that for each i, siv<9 il lifts to a section t; E ff(D(g;) ). On
D(f) n D(g;) = D(fg;), we have two sections t,t; E ff(D(fg;)) both lifting s.
Therefore t - t; E ff'(D(fg;) ). Since.?!'' is quasi-coherent, by (5.3b) for some
n > 0, r(t - t;) extends to a section U; E ff'(D(g;) ). As usual, we pick one
n to work for all i. Let t; = rti + U;. Then t; is a lifting of rs on D(g;),
and furthermore andt; rr agree on D(fg;). Now on D(g;gj) we have two
sections ti and tj of.?!', both of which lift rs, so ti - tj E ff'(D(g;g) ). Further-
more, t; and tj are equal on D(fg;gj), so by (5.3a) we have fm(t; - tj) = 0 for
some m > 0, which we may take independent of i and j. Now the sections
fmt; of.?!' glue to give a global section t" of.?!' over X, which lifts r+ms.
This proves the claim.
Now cover X by a finite number of open sets D(/;), i = 1, ... ,r, such that
slvuil lifts to a section of.?!' over D(/;) for each i. Then by the claim, we can
find an integer n (one for all i) and global sections t; E r(X,ff) suchthat t;
is a lifting of fis. Now the open sets D(/;) cover X, so the ideal (f~, . .. ,f~)
is the unit ideal of A, and we can write 1 = Lt=
1 aJi, with a; EA. Let
t = Ia;t;. Then t is a global section of .?!' whose image in r(X,ff") is
IaJis = s. This completes the proof.

Remark 5.6.1. When we have developed the techniques of cohomology, we

will see that this proposition is an immediate consequence of the fact that
H 1 (X,ff') = 0 for any quasi-coherent sheaf .?!'' on an affine scheme X
(III, 3.5).

Proposition 5.7. Let X be a scheme. The kernel, cokernel, and image of any
morphism of quasi-coherent sheaves are quasi-coherent. Any extension of
quasi-coherent sheaves is quasi-coherent. If X is noetherian, the same is
true for coherent sheaves.
PROOF. The question is local, so we may assume X is affine. The statement
about kernels, cokernels and images follows from the fact that the functor
M r--+ M is exact and fully faithful from A-modules to quasi-coherent sheaves
(5.2a and 5.5). The only nontrivial part is to show that an extension of quasi-
coherent sheaves is quasi-coherent. So let 0 --+ .?!'' --+ .?!' --+ :JP' --+ 0 be an
exact sequence of l'Dx-modules, with :F and $'" quasi-coherent. By (5.6),
the corresponding sequence of global sections over X is exact, say
0 --+ M' --+ M --+ M" --+ 0. Applying the functor "', we get an exact com-
mutative diagram

0 --+ $'' --+ $' --+ $'" --+ 0.

The two outside arrows are isomorphisms, since §'' and §'" are quasi-
coherent. So by the 5-lemma, the middle one is also, showing that .?!' is
quasi-coherent.

114
 5 Sheaves of Modules

In the noetherian case, if :F' and :#'" are coherent, then M' and· M"
are finitely generated, so M is also finitely generated, and hence :F is
coherent.

Proposition 5.8. Let f: X --+ Y be a morphism of schemes.

(a) If '!I is a quasi-coherent sheaf of @y-modules, then f*'!l is a quasi-
coherent sheaf of (!) x-modules.
(b) If X andY are noetherian, and if '!I is coherent, then f*'!l is coherent.
(c) Assume that either X is noetherian, or f is quasi-compact (Ex. 3.2)
and separated. Then if :F is a quasi-coherent sheaf of {!}x-modules, f*:F
is a quasi-coherent sheaf of @y-modules.
PROOF.
(a) The question is local on both X and Y, so we can assume X and Y both
affine. In this case the result follows from (5.5) and (5.2e).
(b) In the noetherian case, the same proof works for coherent sheaves.
(c) Here the question is local on Y only, so we may assume that Y is affine.
Then X is quasi-compact (under either hypothesis) so we can cover X with
a finite number of open affine subsets U;. In the separated case, U; n Uj is
again affine (Ex. 4.3). Call it Uijk· In the noetherian case, U; n Uj is at least
quasi-compact, so we can cover it with a finite number of open affine subsets
Uijk· Now for any open subset V of Y, giving a section s of :F over f- 1 Vis
the same thing as giving a collection of sections S; of :F over u-
1 V) (\ U;

whose restrictions to the open subsets f- (V) n Uijk are all equal. This is
1

just the sheaf property (§1). Therefore, there is an exact sequence of sheaves
on Y,
0 --+ f*:F --+ EB f*(:Fiu) --+ EB f*(:Fiu, J,
1
i i,j,k

where by abuse of notation we denote also by f the induced morphisms

U; --+ Y and Uijk --+ Y. Now !*(:Flu) and f*(:FiuijJ are quasi-coherent by
(5.2d). Thus f*:F is quasi-coherent by (5.7).

Caution 5.8.1. If X and Yare noetherian, it is not true in general that f* of

a coherent sheaf is coherent (Ex. 5.5). However, it is true iff is a finite
morphism (Ex. 5.5) or a projective morphism (5.20) or (III, 8.8), or more
generally, a proper morphism: see Grothendieck [EGA III, 3.2.1].

As a first application of these concepts, we will discuss the sheaf of ideals

of a closed subscheme.

Definition. Let Y be a closed subscheme of a scheme X, and let i: Y --+ X be

the inclusion morphism. We define the ideal sheaf of Y, denoted § y, to
be the kernel of the morphism i# :(!)x--+ i*@y.

115
II Schemes

Proposition 5.9. Let X be a scheme. For any closed subscheme Y of X, the

corresponding ideal sheaf 5 y is a quasi-coherent sheaf of ideals on X. If
X is noetherian, it is coherent. Conversely, any quasi-coherent sheaf of
ideals on X is the ideal sheaf of a uniquely determined closed subscheme
of X.
-
PROOF. If Yis a closed subscheme of X, then the inclusion morphism i: Y -+ X
is quasi-compact (obvious) and separated (4.6), so by (5.8), i*(!)Y is quasi-
coherent on X. Hence 5 y, being the kernel of a morphism of quasi-coherent
sheaves, is also quasi-coherent. If X is noetherian, then for any open affine
subset U = Spec A of X, the ring A is noetherian, so the ideal I = r(U,5 rlu ),
is finitely generated, so 5 y is coherent.
Conversely, given a scheme X and a quasi-coherent sheaf of ideals /,
let Y be the support of the quotient sheaf(!) xl f. Then Y is a subspace of X,
and (I;(!)x//) is the unique closed subscheme of X with ideal sheaf f. The
unicity is clear, so we have only to check that ( Y, (!)x/f) is a closed sub-
scheme. This is a local question, so we may assume X = Spec A is affine.
Since/ is quasi-coherent, cf = a for some ideal a <;; A. Then (Y,(Ilx//) is
just the closed subscheme of X determined by the ideal a (3.2.3).

Corollary 5.10. If X = Spec A is an affine scheme, there is a 1-1 correspon-

dence between ideals a in A and closed subschemes Y of X, given by a~
image of Spec A/a in X (3.2.3). In particular, every closed subscheme of
an affine scheme is affine.
PROOF. By (5.5) the quasi-coherent sheaves of ideals on X are in 1-1 corre-
spondence with the ideals of A.

Our next concern is to study quasi-coherent sheaves on the Proj of a

graded ring. As in the case of Spec, there is a connection between modules
over the ring and sheaves on the space, but it is more complicated.

Definition. Let S be a graded ring and let M be a graded S-module. (See

(1, §7) for generalities on graded modules.) We define the sheaf associated
toM on Proj S, denoted by M, as follows. For each p E Proj S, let M(p)
be the group of elements of degree 0 in the localization y-t M, where T
is the multiplicative system of homogeneous elements of S not in p
(cf. definition of Proj in §2). For any open subset U <;; Proj S we define
M(U) to be the set of functions s from u to upEU M(p) which are locally
fractions. This means that for every p E U, there is a neighborhood V of
p in U, and homogeneous elements m E M and f E S of the same degree,
such that for every q E V, we have f ¢ q, and s(q) = m/f in M(q)· We make
M into a sheaf with the obvious restriction maps.
Proposition 5.11. Let S be a graded ring, and M a graded S-module. Let
X= Proj S.

116
 5 Sheaves of Modules

(a) For any p EX, the stalk (M)v = M<vJ· _

(b) For any homogeneous f E S+, we have Mln+<fl ~ (Mur via the
isomorphism of D +(f) with Spec S<n (see (2.5b) ), where M<n denotes the
group o[ elements of degree 0 in the localized module M J·
(c) M is a quasi-coherent {!}x-module. If S is noetherian and M is
finitely generated, then M is coherent.

PROOF. For (a) and (b), just repeat the proof of (2.5), with M in place of S.
Then (c) follows from (b).

Definition. LetS be a graded ring, and let X = Proj S. For any n E Z, we

define the sheaf (!Jx(n) to be S(nr. We call (!Jx(l) the twisting sheaf of
Serre. For any sheaf of {!}x-modules, :F, we denote by :F(n) the twisted
sheaf :F ®mx (!Jx(n).

Proposition 5.12. Let S be a graded ring and let X = Proj S. Assume that S
is generated by S 1 as an S 0 -algebra.
(a) The sheaf(!Jx(n) is an invertible sheaf on X.
(b) For any graded S-module M, M(n) ~ (M(n) r.
In particular,
(!Jx(n) ® (!Jx(m) ~ (!Jx(n + m).
(c) Let T be another graded ring, generated by T 1 as a T 0 -algebra,
let q>: S --+ T be a homomorphism preserving degrees, and let U <;; Y =
Proj T and f: U --+X be the morphism determined by q> (Ex. 2.14). Then
f*((!Jx(n)) ~ (!Jy(n)lu and f*((!Jy(n)iu) ~ (f*(!Ju)(n).
PROOF.
(a) Recall that invertible means locally free of rank 1. Let f E S 1, and
consider the restriction (!Jx(n)ln+ <fl· By the previous proposition this is
isomorphic to S(n)(jl on Spec S<n· We will show that this restriction is free
of rank 1. Indeed, S(n)<n is a free S<n-module of rank 1. For S<n is the group
of elements of degree 0 in S1 , and S(n)<n is the group of elements of degree n
in S1 . We obtain an isomorphism of one to the other by sending s tors.
This makes sense, for any n E Z, because f is invertible inS1 . Now since S
is generated by S 1 as an S0 -algebra, X is covered by the open sets D +(f)
for f E S 1. Hence (!J(n) is invertible.
(b) This follows from the fact that (M ®s Nr ~ M ®mx N for any two
graded S-modules M and N, when Sis generated by S 1 . Indeed, for any
f E S 1 we have (M ®s N)<n = M<n ®s(f> N<n·
(c) More generally, for any graded S-module M, f*(M) ~ (M ®s Tr lu
and for any graded T-module N, f*(Niu) ~ (8 N)-. Furthermore, the sheaf
T on X is just f*( (!Ju ). The proofs are straightforward (cf. (5.2) for the affine
case).

The twisting operation allows us to define a graded S-module associated

to any sheaf of modules on X = Proj S.

117
II Schemes

Definition. Let S be a graded ring, let X = Proj S, and let !!1' be a sheaf of
(()x-modules. We define the graded S-module associated to !!1' as a group, to
be r *(.~) = ffin E z T(X,!!i'(n) ). We give it a structure of graded S-module
as follows. If s E Sd, then s determines in a natural way a global section
s E T(X,(()x(d) ). Then for any t E T(X,!!i'(n)) we define the products· tin
T(X,!!i'(n + d)) by taking the tensor products ® t and using the natural
map !!i'(n) ® (()x(d) ~ !!i'(n + d).

Proposition 5.13. Let A be a ring, let S = A[ x 0 , . .. ,xr], r ~ 1, and let

X = Proj S. (This is just projective r-space over A.) Then T *((()x) ~ S.
PROOF. We cover X with the open sets D+(x;). Then to give a section
t E T(X,Gx(n)) is the same as giving sections t; E (()x(n)(D +(x;)) for each i,
which agree on the intersections D +(x;xi). Now t; is just a homogeneous
element of degree n in the localization Sx,, and its restriction to D+(x;x) is
just the image of that element in Sx,x1 . Summing over all n, we see that
r *(0x) can be identified with the set of (r + 1)-tuples (t 0 , . . . ,t,) where for
each i, t; E S,,, and for each i,j, the images oft; and ti in Sx,x1 are the same.
Now the X; are not zero divisors in S, so the localization maps S ~ Sx,
and Sx, ~ Sx,x1 are all injective, and these rings are all subrings of S' =
Sxo···xr· Hence T*(lDx) is the intersection nsx, taken insideS'. Now
any homogeneous element of S' can be written uniquely as a product
x~ · · · x~f(x 0 , . . . ,x,), where the ii E Z, and f is a homogeneous polynomial
not divisible by any X;. This element will be in Sx, if and only if ii ~ 0 for
j =1 i. It follows that the intersection of all the Sx, (in fact the intersection of
any two of them) is exactly S.

Caution 5.13.1. If S is a graded ring which is not a polynomial ring, then it

is not true in general that r *(lDx) = S (Ex. 5.14).

Lemma 5.14. Let X be a scheme, let 2! be an invertible sheaf on X, let f E

T(X,ff!), let X f be the open set of points x E X where fx ¢; mxff! x• and let
'"W be a quasi-coherent sheaf on X.
(a) Suppose that X is quasi-compact, and let s E T(X,!!i') be a global
section of !!1' whose restriction to X f is 0. Then for some n > 0, we have
fns = 0, where f"s is considered as a global section of !!1' ® ff!!?;~n_
(b) Suppose furthermore that X has a finite covering by open affine
subsets U;, such that 2iu, is free for each i, and such that U; n Ui is quasi-
compact for each i,j. Given a section t E T(X 1 ,!!1'), then for some n > 0,
the section f"t E T(X 1 ,!!1' ® ff!®n) extends to a global section of !!1' ® ff!®n.
PROOF. This lemma is a direct generalization of (5.3), with an extra twist due
to the presence of the invertible sheaf 2!. It also generalizes (Ex. 2.16). To
prove (a), we first cover X with a finite number (possible since X is quasi-
compact) of open affines U = Spec A such that 2lu is free. Let 1/f: 2lu ~ lDu
be an isomorphism expressing the freeness of 2lu· Since !!1' is quasi-coherent,

118
 5 Sheaves of Modules

by (5.4) there is an A-module M with ~lu ~ M. Our section s E F(X,~)

restricts to give an elements EM. On the other hand, our section/ E F(X,!f)
restricts to give a section of !flu, which in turn gives rise to an element
g = t/J(f) EA. Clearly X 1 n U = D(g). Now sixr is zero, so g"s = 0 in M
for some n > 0, just as in the proof of (5.3). Using the isomorphism
id x t/1°":~ ® !E"Iu ~ ~lu.
we conclude that f"s E F(U,~ ® !£") is zero. This statement is intrinsic
(i.e., independent of t/1). So now we do this for each open set of the covering,
pick one n large enough to work for all the sets of the covering, and we
find f"s = 0 on X.
To prove (b), we proceed as in the proof of(5.3), keeping track of the twist
due to !f as above. The hypothesis Vi n Vi quasi-compact is used to be
able to apply part (a) there.

Remark 5.14.1. The hypotheses on X made in the statements (a) and (b)
above are satisfied either if X is noetherian (in which case every open set is
quasi-compact) or if X is quasi-compact and separated (in which case the
intersection of two open affine subsets is again affine, hence quasi-compact).

Proposition 5.15. Let S be_ a graded ring, which is finitely generated by S 1 as

an S 0 -algebra. Let X = Proj S, and let~ be a quasi-coherent sheaf on X.
Then there is a natural isomorphism {J: r *(~)- --+ ~-
PROOF. First let us define the morphism {J for any (!) x-module ~- Let f E S 1 .
Since r *(~)- is quasi-coherent in any case, to define {J, it is enough to give
the image of a section of r *(~)- over D +(f) (see Ex. 5.3). Such a section is
represented by a fraction m/fd, where mE r(X,~(d) ), for some d ~ 0. We
can think of f-d as a section of (!)x(- d), defined over D +(f). Taking their
tensor product, we obtain m ® f-d as a section of ~ over D +(f). This
defines {J.
Now let~ be quasi-coherent. To show that {J is an isomorphism we have
to identify the module r *(~)<n with the sections of~ over D +(f). We apply
(5.14), considering f as a global section of the invertible sheaf !f = (!)(1).
Since we have assumed that Sis finitely generated by S 1 as an S0 -algebra, we
can find finitely many elements / 0 , . . . ,fr E S 1 such that X is covered by the
open affine subsets D+(JJ The intersections D+(JJ n D+(fJ) are also affine,
and !Eiv+ <JJ is free for each i, so the hypotheses of (5.14) are satisfied. The
conclusion of(5.14) tells us that ~(D+(f)) ~ r*(~)<n• which is just what
we wanted.

Corollary 5.16. Let A be a ring.

(a) If Y is a closed subscheme of P~, then there is a homogeneous ideal
I c;: S = A[ x 0 , . .. ,x,] such that Y is the closed subscheme determined by I
(Ex. 3.12).

119
II Schemes

(b) A scheme Y over Spec A is projective if and only if it is isomorphic to

Proj S for some graded ring S, where S 0 = A, and S is finitely generated by
S 1 as an S 0 -algebra.
PROOF.
(a) Let f y be the ideal sheaf of Y on X = PA. Now f y is a subsheaf of
lD x; the twisting functor is exact; the global section functor r is left exact;
hence r*(fy) is a submodule of r*(lDx). But by (5.13), r*(lDx) = S. Hence
r *(f y) is a homogeneous ideal of S, which we will call I. Now I determines
a closed subscheme of X (Ex. 3.12), whose sheaf of ideals will be I. Since f y
is quasi-coherent by (5.9), we have f y ~ l by (5.15), and hence Y is the sub-
scheme determined by I. In fact, r *(f y) is the largest ideal in S defining Y
(Ex. 5.10).
(b) Recall that by definition Y is projective over Spec A if it is isomorphic
to a closed subscheme ofPA for some r (§4). By part (a), any such Y is isomor-
phicto Proj Sji, and we can take I to be contained inS+ = ffid >O Sd (Ex. 3.12),
so that (S//) 0 = A. Conversely, any such graded ring S is a quotient of a
polynomial ring, so Proj S is projective.

Definition. For any scheme Y, we define the twisting sheaf lD(1) on P~ to be

g*(lD(1) ), where g:P~ -+ Pz is the natural map (recall that P~ was defined
as Pz X z Y).
Note that if Y = Spec A, this is the same as the lD(1) already defined on
PA = Proj A[ x 0 , ..• ,x,], by (5.12c).

Definition. If X is any scheme over Y, an invertible sheaf !l' on X is very ample

relative to Y, if there is an immersion i:X -+ P~ for some r, such that
i*(lD(1)) ~ !l'. We say that a morphism i:X-+ Z is an immersion if it
gives an isomorphism of X with an open subscheme of a closed subscheme
of Z. (This definition of very ample differs slightly from the one in
Grothendieck [EGA II, 4.4.2].)

Remark 5.16.1. Let Y be a noetherian scheme. Then a scheme X over Y is

projective if and only if it is proper, and there exists a very ample sheaf on X
relative to Y. Indeed, if X is projective over Y, then X is proper by (4.9). On
the other hand, there is a closed immersion i:X -+ P~ for some r, so i*lD(1)
is a very ample invertible sheaf on X. Conversely, if X is proper over Y, and
!l' is a very ample invertible sheaf, then fil ~ i*(lD(1)) for some immersion
i:X -+ P~. But by (Ex. 4.4) the image of X is closed, so in facti is a closed
immersion, so X is projective over Y.
Note however that there may be several nonisomorphic very ample
sheaves on a projective scheme X over Y. The sheaf !l' depends on the
embedding of X into P~ (Ex. 5.12). If Y = Spec A, and if X = Proj S, where S
is a graded ring as in (5.16b), then the sheaf lD(1) on X defined earlier is a very
ample sheaf on X. However, there may be nonisomorphic graded rings
having the same Proj and the same very ample sheaf lD(1) (Ex. 2.14).

120
 5 Sheaves of Modules

We end this section with some special results about sheaves on a projective
scheme over a noetherian ring.

Definition. Let X be a scheme, and let ff' be a sheaf of C9x-modules. We say

that ff' is generated by global sections if there is a family of global sections
{sJiEI• s; E r(X,ff'), such that for each x EX, the images of s; in the stalk
ff'x generate that stalk as an {9x-module.
Note that ff' is generated by global sections if and only if ff' can be
written as a quotient of a free sheaf. Indeed, the generating sections
{s;};EI define a surjective morphism of sheaves EBiEI C9x---> ff', and
conversely.

Example 5.16.2. Any quasi-coherent sheaf on an affine scheme is generated

by global sections. Indeed, if ff' = M on Spec A, any set of generators for M
as an A-module will do.

Example 5.16.3. Let X = Proj S, where Sis a graded ring which is generated
by S1 as an S0 -algebra. Then the elements of S1 give global sections of (9 x(l)
which generate it.

Theorem 5.17 (Serre). Let X be a projective scheme over a noetherian ring A,

let (9(1) be a very ample invertible sheaf on X, and let ff' be a coherent
C9x-module. Then there is an integer n0 such that for all n ~ n0 , the sheaf
ff'(n) can be generated by a .finite number of global sections.
PROOF. Let i:X---> P~ be a closed immersion of X into a projective space over
A, such that i*(CD(l)) = C9x(l). Then i*ff' is coherent on P~ (Ex. 5.5), and
i*(JF(n)) = (i*JF)(n) (5.12) or (Ex. 5.1d), and JF(n) is generated by global
sections if and only if i*(ff'(n)) is (in fact, their global sections are the same),
so we reduce to the case X = P~ = Proj A[ x 0 , . . . ,xrJ.
Now cover X with the open sets D +(x;), i = 0, ... ,r. Since ff' is coherent,
for each i there is a finitely generated module M; over B; =A[ x 0 /x;, ... ,xn/x;]
such that ff'lv+ <xil ~ M;. For each i, take a finite number of elements
sii EM; which generate this module. By (5.14) there is an integer n such that
x7s;j extends to a global section t;j of ff'(n). As usual, we take one n to work
for all i,j. Now ff'(n) corresponds to a B;-module Mi on D +(x;), and the map
x7:ff'---> ff'(n) induces an isomorphism of M; to Mi. So the sections x7s;j
generate M;, and hence the global sections tij E F(X,ff'(n)) generate the sheaf
ff'(n) everywhere.

Corollary 5.18. Let X be projective over a noetherian ring A. Then any

coherent sheaf ff' on X can be written as a quotient of a sheaf <ff, where <ff
is a finite direct sum of twisted structure sheaves CD(n;) for various integers
n;.

121
II Schemes

PROOF. Let .?(n) be generated by a finite number of global sections. Then

we have a surjection EBf= 1 (!Jx---+ .?(n)---+ 0. Tensoring with (!Jx(-n) we
obtain a surjection EBf= 1 (!Jx( -n)---+.?---+ 0 as required.

Theorem 5.19. Let k be afield, let A be a finitely generated k-algebra, let X be

a projective scheme over A, and let .? be a coherent {!}x-module. Then
r(X,.?) is a finitely generated A-module. In particular, if A = k, r(X,.?)
is a finite-dimensional k-vector space.
PROOF. First we write X = Proj S, where Sis a graded ring with S0 = A which
is finitely generated by S 1 as an S 0 -algebra (5.16b). Let M be the graded S-
module r *(.?). Then by (5.15) we have M ~ .?. On the other hand, by (5.17),
for n sufficiently large, .?(n) is generated by a finite number of global sections
in r(X,.?(n) ). Let M' be the submodule of M generated by these sections.
Then M' is a finitely generated S-module. Furthermore, the inclusion
M' ~ M induces an inclusion of sheaves M' ~ M = .?. Twisting by n we
have an inclusion M'(n) ~ .?(n) which is actually an isomorphism, because
.?(n) is generated by global sections in M'. Twisting by - n we find that
M' ~ .?. Thus .? is the sheaf associated to a finitely generated S-module,
and so we have reduced to showing that if M is a finitely generated S-module,
then r(X,M) is a finitely generated A-module.
Now by (1, 7.4), there is a finite filtration
0 = M0 ~ M1 ~ ... ~ Mr = M
of M by graded submodules, where for each i, Mi/Mi- 1 ~ (S/p;}(n;) for some
homogeneous prime ideal Pi ~ S, and some integer ni. This filtration gives
a filtration of M, and the short exact sequences
o ---+ Mi- 1 ---+ Mi ---+ Mi I Mi- 1 ---+ o
give rise to left-exact sequences
o---+ r(X,Mi- 1 )---+ r(X,Mi)---+ r(X,Mi/Mi- 1 ).
Thus to show that r(X,M) is finitely generated over A, it will be sufficient to
show that r(X,(S/p)-(n)) is finitely generated, for each p and n. Thus we have
reduced to the following special case: Let S be a graded integral domain,
finitely generated by S 1 as an S0 -algebra, where S 0 = A is a finitely generated
integral domain over k. Then r(X,(!Jx(n)) is a finitely generated A-module,
for any n E Z.
Let x 0 , . .. ,xr E S 1 be a set of generators of S 1 as an A-module. Since S
is an integral domain, multiplication by x 0 gives an injection S(n) ---+ S(n + 1)
for any n. Hence there is an injection r(X,(!Jx(n)) ---+ r(X,(!Jx(n + 1)) for
any n. Thus it is sufficient to prove r(X,(!Jx(n)) finitely generated for all
sufficiently large n, say n ~ 0.
LetS'= ffin;.o r(X,(!Jx(n) ). Then S' is a ring, containing S, and contained
in the intersection nsx, of the localizations of Sat the elements Xo, . .. ,Xr.

122
 5 Sheaves of Modules

(Use the same argument as in the proof of (5.13).) We will show that S' is
integral over S.
Let s' E S' be homogeneous of degree d ~ 0. Since s' E Sx, for each i, we
can find an integer n such that x7s' E S. Choose one n that works for all i.
Since the X; generate s 1' the monomials in the X; of degree m generate sm for
any m. So by taking a larger n, we may assume that ys' E S for ally E S". In fact,
since s' has positive degree, we can say that for any yES ;,n = EBe;,n Se,
ys' E S;,n· Now it follows inductively, for any q ~ 1 that y · (s')q E S;,n for
any y E S;,"' Take for example y = x~. Then for every q ~ 1 we have
(s')q E (1/x~)S. This is a finitely generated sub-S-module of the quotient field
of S'. It follows by a well-known criterion for integral dependence (Atiyah-
Macdonald [1, p. 59]), that s' is integral overS. Thus S' is contained in the
integral closure of Sin its quotient field.
To complete the proof, we apply the theorem of finiteness of integral
closure (I, 3.9A). Since Sis a finitely generated k-algebra, S' will be a finitely
generated S-modqle. It follows that for every n, S~ is a finitely generated
S0 -module, which is what we wanted to prove. In fact, our proof shows that
S~ = S" for all sufficiently large n (Ex. 5.9) and (Ex. 5.14).

Remark 5.19.1. This proof is a generalization of the proof of (1, 3.4a). We

will give another proof of this theorem later, using cohomology (III, 5.2.1).

Remark 5.19.2. The hypothesis "A is a finitely generated k-algebra" is used

only to be able to apply (1, 3.9A). Thus it would be sufficient to assume only
that A is a "Nagata ring" in the sense of Matsumura [2, p. 231 ]-see also
[Joe. cit., Th. 72, p. 240].

Corollary 5.20. Let f: X ---+ Y be a projective morphism of schemes of finite

type over afield k. Let:#' be a coherent sheaf on X. Then f*:F is coherent
on Y.
PROOF. The question is local on Y, so we may assume Y = Spec A, where A
is a finitely generated k-algebra. Then in any case, f*:F is quasi-coherent
(5.8c), so f*:F = T( Y,f*:F)- = T(X,:Fr. But T(X,ff) is a finitely generated
A-module by the theorem, so f*:F is coherent. See (III, 8.8) for another proof
and generalization.

EXERCISES

5.1. Let (X,((x) be a ringed space, and let tC be a locally free 0x-module of finite rank.
We define the dual of tff, denoted i, to be the sheaf Jffom(l)x(tff,C9x).
(a) Show that (ir ~ tff.
(b) For any C9x-module :#', Jffomf'x(tff,:J') ~ i ®(l)x :F.
(c) For any crx-modules .'#','§, Hom~x(tff ® :#','§) ~ Hom(l)x(:J',Jffom(')x(tff,'§) ).

123
II Schemes

(d) (Projection Formula). If f:(X/!Jx)-+ (Y,@y) is a morphism of ringed spaces, if

:F is an @x-module, and if<! is a locally free @y-module of finite rank, then there
is a natural isomorphism f*(:F ®<'!x f*<!) ~ f*(:F) ®<'!y <!.
5.2. Let R be a discrete valuation ring with quotient field K, and let X= Spec R.
(a) To give an @x-module is equivalent to giving an R-module M, a K-vector
space L, and a homomorphism p:M ®R K ..... L.
(b) That @x-module is quasi-coherent if and only if pis an isomorphism.
5.3. Let X = Spec A be an affine scheme. Show that the functors -and rare adjoint,
in the following sense: for any A-module M, and for any sheaf of @x-modules :F,
there is a natural isomorphism

5.4. Show that a sheaf of @x-modules :F on a scheme X is quasi-coherent if and only

if every point of X has a neighborhood U, such that :Flu is isomorphic to a
cokernel of a morphism of free sheaves on U. If X is noetherian, then :F is co-
herent if and only if it is locally a co kernel of a morphism of free sheaves of finite
rank. (These properties were originally the definition of quasi-coherent and
coherent sheaves.)
5.5. Let f: X ..... Y be a morphism of schemes.
(a) Show by example that if :F is coherent on X, then f*:F need not be coherent
on Y, even if X and Y are varieties over a field k.
(b) Show that a closed immersion is a finite morphism (§3).
(c) Iff is a finite morphism of noetherian schemes, and if :F is coherent on X,
then f*:F is coherent on Y.
5.6. Support. Recall the notions of support of a section of a sheaf, support of a sheaf,
and subsheafwith supports from (Ex. 1.14) and (Ex. 1.20).
(a) Let A be a ring, let M be an A-module, let X = Spec A, and let :F = M.
For any mE M = r(X,:F), show that Supp m = V(Ann m), where Ann m is
the annihilator ofm = {a E Alam = 0}.
(b) Now suppose that A is noetherian, and M finitely generated. Show that
Supp :F = V(Ann M).
(c) The support of a coherent sheaf on a noetherian scheme is closed.
(d) For any ideal a <;; A, we define a submodule T0 (M) of M by T.,(M) =
{mE Mla"m = 0 for some n > 0}. Assume that A is noetherian, and Many
A-module. Show that r.,(Mr ~ £'~(/F), where Z = V(a) and :F = M.
[Hint: Use (Ex. 1.20) and (5.8) to show a priori that £'~(/F) is quasi-coherent.
Then show that Fa(M) ~ Tz(ff).]
(e) Let X be a noetherian scheme, and let Z be a closed subset. If :F is a quasi-
coherent (respectively, coherent) @x-module, then £'~(/F) is also quasi-
coherent (respectively, coherent).
5.7. Let X be a noetherian scheme, and let :F be a coherent sheaf.
(a) If the stalk ffx is a free ((!x-module for some point x EX, then there is a neigh-
borhood U of x such that :Flu is free.
(b) :F is locally free if and only if its stalks .'f'x are free @x-modules for all x EX.
(c) :F is invertible (i.e., locally free of rank 1) if and only if there is a coherent sheaf
'!J such that :F ® '!J ~ ((!x· (This justifies the terminology invertible: it means

124
 5 Sheaves of Modules

that :F is an invertible element of the monoid of coherent sheaves under the

operation ®.)
5.8. Again let X be a noetherian scheme, and :F a coherent sheaf on X. We will
consider the function
rp(x) = dimk(x) ffx ®mx k(x),
where k(x) = (l)xfmx is the residue field at the point x. Use Nakayama's lemma
to prove the following results.
(a) The function rp is upper semi-continuous, i.e., for any nEZ, the set {xEXIrp(x) ;;> n}
is closed.
(b) If :F is locally free, and X is connected, then rp is a constant function.
(c) Conversely, if X is reduced, and rp is constant, then :F is locally free.
5.9. Let S be a graded ring, generated by S 1 as an S0 -algebra, let M be a graded S-
module, and let X = Proj S.
(a) Show that there is a natural homomorphism cx:M---> r *(M).
(b) Assume now that S 0 = A is a finitely generated k-algebra for some field k,
that S 1 is a finitely generated A-module, and that M is a finitely generated
S-module. Show that the map ex is an isomorphism in all large enough
degrees, i.e., there is a d0 E Z such that for all d ;;> d0 , cxd:Md---> T(X,M(d))
is an isomorphism. [Hint: Use the methods of the proof of (5.19).]
(c) With the same hypotheses, we define an equivalence relation ~ on graded
S-modules by saying M ~ M' if there is an integer d such that M :3d ~ M':3d·
Here M :3d = ffin:3d Mn. We will say that a graded S-module M is quasi-
finitely generated if it is equivalent to a finitely generated module. Now show
that the functors - and r * induce an equivalence of categories between the
category of quasi-finitely generated graded S-modules modulo the equivalence
relation ~, and the category of coherent (l)x-modules.
5.10. Let A be a ring, letS = A[x 0 , . . . ,x,] and let X = Proj S. We have seen that a
homogeneous ideal I in S defines a closed subscheme of X (Ex. 3.12), and that
conversely every closed subscheme of X arises in this way (5.16).
(a) For any homogeneous ideal I c;; S, we define the saturation 1 of I to be
{s E Slfor each i = 0, ... ,r, there is ann such that x7s E I}. We say that I is
saturated if I = 1. Show that 1 is a homogeneous ideal of S.
(b) Two homogeneous ideals I 1 and I 2 of S define the same closed subscheme of
X if and only if they have the same saturation.
(c) If Y is any closed subscheme of X, then the ideal r *( J\) is saturated. Hence
it is the largest homogeneous ideal defining the subscheme Y.
(d) There is a 1-1 correspondence between saturated ideals of Sand closed sub-
schemes of X.
5.11. Let S and T be two graded rings with S 0 = T 0 = A. We define the Cartesian
product s X A T to be the graded ring EBd:30 sd ®A Td. If X = Proj s and
Y = Proj T, show that Proj(S x A T) ~ X x A Y, and show that the sheaf (1)(1)
on Proj(S x A T) is isomorphic to the sheaf pf((l)x(1)) ® p!((l)y(1)) on X x Y.
The Cartesian product of rings is related to the Segre embedding of projective
spaces (1, Ex. 2.14) in the following way. If x 0 , . . . ,x, is a set of generators for S 1
over A, corresponding to a projective embedding X 4 PA., and if y 0 , . .• ,y, is
a set of generators for T 1 , corresponding to a projective embedding Y 4 P~,
then {xi® yj} is a set of generators for (S x A T) 1, and hence defines a projective

125
II Schemes

embedding Proj(S x A T) c. P~, with N = rs + r + s. This is just the image

of X x Y c:; P' x P' in its Segre embedding.
5.12. (a) Let X be a scheme over a scheme Y, and let 2', A be two very ample invertible
sheaves on X. Show that fi' ®A is also very ample. [Hint: Use a Segre
embedding.]
(b) Let f:X -+ Y and g: Y-+ Z be two morphisms of schemes. Let fi' be a very
ample invertible sheaf on X relative to Y, and let A be a very ample invertible
sheaf on Y relative to Z. Show that fi' ® f*jt is a very ample invertible sheaf
on X relative to Z.
5.13. Let S be a graded ring, generated by S 1 as an S0 -algebra. For any integer d > 0,
let s<d) be the graded ring EBn:.
0 s~d) where s~d) = snd· Let X = Proj S. Show

that Proj s<dl ~ X, and that the sheaf CD(l) on Proj s<dl corresponds via this
isomorphism to CDx(d).
This construction is related to the d-uple embedding (1, Ex. 2.12) in the fol-
lowing way. If x 0 , . . . ,x, is a set of generators for S 1 , corresponding to an em-
bedding X c. PA, then the set of monomials of degree d in the xi is a set of
generators for S\dl = Sd. These define a projective embedding of Proj s<dl which
is none other than the image of X under the d-uple embedding of P~.
5.14. Let A be a ring, and let X be a closed subscheme of P~. We define the homo-
geneous coordinate ring S(X) of X for the given embedding to be A[x 0 , . . . ,x,]/1,
where I is the ideal r *(.I x) constructed in the proof of (5.16). (Of course if A is
a field and X a variety, this coincides with the definition given in (1, §2) !) Recall
that a scheme X is normal if its local rings are integrally closed domains. A closed
subscheme X c:; P~ is projectively normal for the given embedding, if its homo-
geneous coordinate ring S(X) is an integrally closed domain (cf. (1, Ex. 3.18) ).
Now assume that k is an algebraically closed field, and that X is a connected,
normal closed subscheme ofP~. Show that for some d > 0, the d-uple embedding
of X is projectively normal, as follows.
(a) LetS be the homogeneous coordinate ring of X, and letS' = EBn:.o T(X,CDx(n) ).
Show that S is a domain, and that S' is its integral closure. [Hint: First show
that X is integral. Then regard S' as the global sections of the sheaf of rings
!/' = EBn:.o CDx(n) on X, and show that !/' is a sheaf of integrally closed
domains.]
(b) Use (Ex. 5.9) to show that Sd = S~ for all sufficiently large d.
(c) Show that s<dl is integrally closed for sufficiently large d, and hence conclude
that the d-uple embedding of X is projectively normal.
(d) As a corollary of (a), show that a closed subscheme X c:; P~ is projectively
normal if and only if it is normal, and for every n ~ 0 the natural map
T(P',@p,(n))-+ r(X,CDx(n)) is surjective.

5.15. Extension of Coherent Sheaves. We will prove the following theorem in several
steps: Let X be a noetherian scheme, let U be an open subset, and let.? be a
coherent sheaf on U. Then there is a coherent sheaf.?' on X such that .?'lu ~ .?.
(a) On a noetherian affine scheme, every quasi-coherent sheaf is the union of its
coherent subsheaves. We say a sheaf.? is the union of its subsheaves Sf
if for every open set U, the group .?(V) is the union of the subgroups Sf (U).
(b) Let X be an affine noetherian scheme, U an open subset, and.? coherent on
U. Then there exists a coherent sheaf.?' on X with .?'lu ~ .?. [Hint: Let
i: U -+ X be the inclusion map. Show that i*.? is quasi-coherent, then use(a).]

126
 5 Sheaves of Modules

(c) With X,U,:F as in (b), suppose furthermore we are given a quasi-coherent

sheaf'§ on X such that :F ~ '§lu· Show that we can find ff' a coherent sub-
sheaf of'§, with :F'lu ~ :F. [Hint: Use the same method, but replace i*ff
by p- 1 (i*ff), where pis the natural map'§-.. i*('§lul-]
(d) Now let X be any noetherian scheme, U an open subset, :Fa coherent sheaf
on U, and'§ a quasi-coherent sheaf on X such that :F ~ '§lu· Show that there
is a coherent subsheaf :F ~ '§on X with :F'lu ~ :F. Taking'§ = i*:F proves
the result announced at the beginning. [Hint: Cover X with open affines, and
extend over one of them at a time.]
(e) As an extra corollary, show that on a noetherian scheme, any quasi-coherent
sheaf :F is the union of its coherent subsheaves. [Hint: If sis a section of :F
over an open set U, apply (d) to the subsheaf of :Flu generated by s.]

5.16. Tensor Operations on Sheaves. First we recall the definitions of various tensor
operations on a module. Let A be a ring, and let M be an A-module. Let T"(M)
be the tensor product M ® ... ® M of M with itself n times, for n ;, 1. For
n = 0 we put T 0 (M) = A. Then T(M) = EB.~o T"(M) is a (noncommutative)
A-algebra, which we call the tensor algebra of M. We define the symmetric
algebra S(M) = ffi.~o S"(M) of M to be the quotient of T(M) by the two-sided
ideal generated by all expressions x ® y - y ® x, for all x,y EM. Then S(M)
is a commutative A-algebra. Its component S"(M) in degree n is called the nth
symmetric product of M. We denote the image of x ® y in S(M) by xy, for any
x,y EM. As an example, note that if M is a free A-module of rank r, then S(M) ~
A[x 1, ••• ,x,].
We define the exterior algebra 1\(M) = EB.~o /\"(M) of M to be the quo-
tient of T(M) by the two-sided ideal generated by all expressions x ® x for
x E M. Note that this ideal contains all expressions of the form x ® y + y ® x,
so that /\(M) is a skew commutative graded A-algebra. This means that if u E
/\'(M) and v E f\•(M), then u 1\ v = ( -l)"v 1\ u (here we denote by 1\ the
multiplication in this algebra; so the image of x ®yin f\ 2 (M) is denoted by
x 1\ y). The nth component /\"(M) is called the nth exterior power of M.
Now let (X,(!Jx) be a ringed space, and let :F be a sheaf of (!Jx-modules. We
define the tensor algebra, symmetric algebra, and exterior algebra of :F by taking
the sheaves associated to the presheaf, which to each open ·set U assigns the
corresponding tensor operation applied to ff(U) as an (!Jx(U)-module. The
results are (!Jx-algebras, and their components in each degree are (!Jx-modules.
(a) Suppose that :F is locally free of rank n. Then T'(ff), S'(ff), and /\'(ff) are
also locally free, of ranks n', ("~.':! 1 ), and G) respectively.
(b) Again let :F be locally free of rank n. Then the multiplication map/\':F ®
1\" -,:F -.. 1\ ".~ is a perfect pairing for any r, i.t,., it induces an isomorphism
of /\' :F with ( /\"- ':F r ® 1\" :F. As a special case, note if :F has rank 2,
then :F ~ :F- ® 1\1 :?.
(c) Let 0 -.. :F' -.. :F -.. :F" -.. 0 be an exact sequence of locally free sheaves.
Then for any r there is a finite filtration of S'(ff),

S'(ff) = F0 2 F 1 2 ... 2 F' 2 pr+ 1 = 0

with quotients

for each p.

127
II Schemes

(d) Same statement as (c), with exterior powers instead of symmetric powers. In
particular, if ff',ff,ff" have ranks n',n,n" respectively, there is an isomorphism
1\"ff ~ 1\"'ff' ® 1\""ff".
(e) Let f:X --+ Y be a morphism of ringed spaces, and let ff be an (l)y-module.
Then f* commutes with all the tensor operations on ff, i.e., f*(S"(ff)) =
S"(f* ff) etc.
5.17. Affine M orphisms. A morphism f: X --+ Y of schemes is affine ifthere is an open
affine cover {V;} of Y such that f- 1(V;) is affine for each i.
(a) Show that f: X --+ Y is an affine morphism if and only if for every open affine
V <;::; Y,f- 1 (V) is affine. [Hint: Reduce to the case Yaffine, and use (Ex. 2.17).]
(b) An affine morphism is quasi-compact and separated. Any finite morphism is
affine.
(c) Let Y be a scheme, and let d be a quasi-coherent sheaf of (l)y-algebras (i.e., a
sheaf of rings which is at the same time a quasi-coherent sheaf of (l)y-modules).
Show that there is a unique scheme X, and a morphism f:X --+ Y, such that
for every open affine V <;::; Y, f- 1(V) ~ Spec d(V), and for every inclusion
U 4 V of open affines of Y, the morphism f- 1 (U) 4 f- 1 (V) corresponds to
the restriction homomorphism d(V)--+ d(U). The scheme X is called
Spec d. [Hint: Construct X by glueing together the schemes Spec d(V),
for V open affine in Y.J
(d) If d is a quasi-coherent (l)y-algebra, then f:X = Spec d--+ Y is an affine
morphism, and d ~ j*(l)x· Conversely, if f:X--+ Y is an affine morphism,
then d = j*(l)x is a quasi-coherent sheaf of (l)y-algebras, and X ~ Spec d.
(e) Letf:X--+ Ybe an affine morphism, and let d = j*(l)x· Show thatf* induces
an equivalence of categories from the category of quasi-coherent (l)x-modules
to the category of quasi-coherent d-modules (i.e., quasi-coherent (l)y-modules
having a structure of d-module). [Hint: For any quasi-coherent d-module
.A, construct a quasi-coherent (l)x-module .ii, and show that the functors f*
and- are inverse to each other.
5.18. Vector Bundles. Let Y be a scheme. A (geometric) vector bundle of rank n over
Y is a scheme X and a morphism f:X--+ Y, together with additional data con-
sisting of an open covering {U;} of Y, and isomorphisms lj;i:f- 1 (U;)--+ Au,,
such that for any i,j, and for any open affine subset V = Spec A <;::; Ui n Ui,
the automorphism lj; = lj;ioi/Ji 1 of A~= SpecA[x 1 , . . . ,x"] is given by a
linear automorphism() of A[ x 1 , . . . ,x"], i.e., O(a) = a for any a E A, and O(x;) =
L.Giixi for suitable aii EA.
An isomorphism g:(X,f,{Ui},{lj;i})--+ (X',f',{U;},{tf;;}) of one vector bundle
of rank n to another one is an isomorphism g:X--+ X' of the underlying schemes,
such that f = f' o g, and such that X,f, together with the covering of Y con-
sisting of all the Ui and u;, and the isomorphisms lj;i and tf;; o g, is also a vector
bundle structure on X.
(a) Let@' be a locally free sheaf of rank non a scheme Y. Let S(t&') be the symmetric
algebra on t&', and let X = Spec S(t&'), with projection morphism f:X--+ Y.
For each open affine subset U <;::; Y for which t&'lu is free, choose a basis of@',
and let lj;:f- 1 (U)--+ Au be the isomorphism resulting from the identification
of S(t&'(U)) with (I)(U)[x 1, . . . ,x"]. Then (X,f,{U},{Ij;}) is a vector bundle of
rank n over Y, which (up to isomorphism) does not depend on the bases of
@' u chosen. We call it the geometric vector bundle associated to@', and denote
it by V(t&').

128
 6 Divisors

(b) For any morphism f: X --+ Y, a section off over an open set U s Y is a mor-
phism s: U--+ X such that f o s = idu. It is clear how to restrict sections to
smaller open sets, or how to glue them together, so we see that the presheaf
U H {set of sections off over U} is a sheaf of sets on Y, which we denote by
Y'(X /Y). Show that iff: X --+ Y is a vector bundle of rank n, then the sheaf
of sections Y'(X/Y) has a natural structure of @y-module, which makes it a
locally free @y-module of rank n. [Hint: It is enough to define the module
structure locally, so we can assume Y = Spec A is affine, and X = A~. Then a
sections: Y--+ X comes from an A-algebra homomorphism !:I:A[x 1 , •.. ,x.]--+
<
A, which in turn determines an ordered n-tuple !:l(x 1 ), . . . , !:l(x.)) of elements
of A. Use this correspondence between sections s and ordered n-tuples of
elements of A to define the module structure.]
(c) Again let tff be a locally free sheaf of rank non Y, let X = V(tff), and let Y' =
Y'(X/Y) be the sheaf of sections of X over Y. Show that Y' ~ tff~, as follows.
Given a sections E r(V,tff~) over any open set V, we think of s as an element of
Hom(tff!v,@v). So s determines an @v-algebra homomorphism S(t!lvl--+ @y.
This determines a morphism of spectra V = Spec (Dv--+ Spec S(t!lvl =
f- 1(V), which is a section of X /Y. Show that this construction gives an iso-
morphism of ,g~ to Y'.
(d) Summing up, show that we have established a one-to-one correspondence
between isomorphism classes of locally free sheaves of rank n on Y, and iso-
morphism classes of vector bundles of rank n over Y. Because of this, we
sometimes use the words "locally free sheaf" and "vector bundle" inter-
changeably, if no confusion seems likely to result.

6 Divisors
The notion of divisor forms an important tool for studying the intrinsic
geometry on a variety or scheme. In this section we will introduce divisors,
linear equivalence and the divisor class group. The divisor class group is
an abelian group which is an interesting and subtle invariant of a variety.
In §7 we will see that divisors are also important for studying maps from a
given variety to a projective space.
There are several different ways of defining divisors, depending on the
context. We will begin with Weil divisors, which are easiest to understand
geometrically, but which are only defined on certain noetherian integral
schemes. For more general schemes there is the notion of Cartier divisor
which we treat next. Then we will explain the connection between Weil
divisors, Cartier divisors, and invertible sheaves.

We start with an informal example. Let C be a nonsingular projective

curve in Pf, the projective plane over an algebraically closed field k. For
each line Lin P 2 , we consider L n C, which is a finite set of points on C.
If C is a curve of degree d, and if we count the points with proper multi-
plicity, then L n C will consist of exactly d points (I, Ex. 5.4). We write
L n C = l,n;P;, where P; E Care the points, and n; the multiplicities, and
we call this formal sum a divisor on C. As L varies, we obtain a family of
divisors on C, parametrized by the set of all lines in P 2 , which is the dual

129
II Schemes

projective space (P 2 )*. We call this set of divisors a linear system of divisors
on C. Note that the embedding of C in P 2 can be recovered just from
knowing this linear system: if Pis a point of C, we consider the set of divisors
in the linear system which contain P. They correspond to the lines L E (P 2 )*
passing through P, and this set of lines determines P uniquely as a point
of P 2 . This connection between linear systems and embeddings in pro-
jective space will be studied in detail in §7.
This example should already serve to illustrate the importance of divi-
sors. To see the relation among the different divisors in the linear system,
let Land L' be two lines in P 2 , and let D = L n C and D' = L' n C be the
corresponding divisors. If L and L' are defined by linear homogeneous
equations f = 0 and f' = 0 in P 2 , then f/f' gives a rational function on
P 2 , which restricts to a rational function g on C. Now by construction, g
has zeros at the points of D, and poles at the points of D', counted with
multiplicities, in a sense which will be made precise below. We say that
D and D' are linearly equivalent, and the existence of such a rational func-
tion can be taken as an intrinsic definition of the linear equivalence. We
will make these concepts more precise in our formal discussion, starting now.

Wei! Divisors
Definition. We say a scheme X is regular in codimension one (or sometimes
nonsingular in codimension one) if every local ring (iJ x of X of dimension
one is regular.

The most important examples of such schemes are nonsingular varieties

over a field (1, §5) and noetherian normal schemes. On a nonsingular
variety the local ring of every closed point is regular (1, 5.1), hence all the
local rings are regular, since they are localizations of the local rings of
closed points. On a noetherian normal scheme, any local ring of dimen-
sion one is an integrally closed domain, hence is regular (1, 6.2A).
In this section we will consider schemes satisfying the following condition:
(*)X is a noetherian integral separated scheme which is regular in
codimension one.

Definition. Let X satisfy (*). A prime divisor on X is a closed integral sub-

scheme Y of codimension one. A Weil divisor is an element of the free
abelian group Div X generated by the prime divisors. We write a divisor
as D = L\ }i, where the 1i are prime divisors, the n; are integers, and
only finitely many n; are different from zero. If all the n; ~ 0, we say
that D is effective.
If Y is a prime divisor on X, let lJ E Y be its generic point. Then the
local ring (iJ~.x is a discrete valuation ring with quotient field K, the
function field of X. We call the corresponding discrete valuation vy
the valuation of Y. Note that since X is separated, Y is uniquely deter-

130
 6 Divisors

mined by its valuation (Ex. 4.5). Now let f E K* be any nonzero rational
function on X. Then vy(f} is an integer. If it is positive, we say f has
a zero along Y, of that order; if it is negative, we say f has a pole along Y,
of order - Vy(f).

Lemma 6.1. Let X satisfy (*), and let f E K* be a nonzero function on X.

Then vy(f) = 0 for all except finitely many prime divisors Y.
PROOF. Let U = Spec A be an open affine subset of X on which f is regular.
Then Z = X - U is a proper closed subset of X. Since X is noetherian,
Z can contain at most finitely many prime divisors of X; all the others
must meet U. Thus it will be sufficient to show that there are only finitely
many prime divisors Y of U for which vy(f) # 0. Since f is regular on U,
we have vy(f} :;?: 0 in any case. And vy(f) > 0 if and only if Y is contained
in the closed subset of U defined by the ideal Af in A. Since f # 0, this is
a proper closed subset, hence contains only finitely many closed irreducible
subsets of codimension one of U.

Definition. Let X satisfy (*) and let f E K*. We define the divisor off,
denoted (f), by
(f) = Ivy(f) . Y,
where the sum is taken over all prime divisors of X. By the lemma, this
is a finite sum, hence it is a divisor. Any divisor which is equal to the
divisor of a function is called a principal divisor.

Note that if f,g E K*, then (fjg) = (f) - (g) because of the properties
of valuations. Therefore sending a function f to its divisor (f) gives a
homomorphism of the multiplicative group K* to the additive group
Div X, and the image, which consists of the principal divisors, is a sub-
group of Div X.

Definition. Let X satisfy(*). Two divisors D and D' are said to be linearly
equivalent, written D ,...., D', if D - D' is a principal divisor. The group
Div X of all divisors divided by the subgroup of principal divisors is
called the divisor class group of X, and is denoted by Cl X.

The divisor class group of a scheme is a very interesting invariant. In

general it is not easy to calculate. However, in the following propositions
and examples we will calculate a number of special cases to give some
idea of what it is like.

Proposition 6.2. Let A be a noetherian domain. Then A is a unique factor-

ization domain if and only if X = Spec A is normal and Cl X = 0.

131
II Schemes

PROOF. (See also Bourbaki [1, Ch. 7, §3]). It is well-known that a UFD
is integrally closed, so X will be normal. On the other hand, A is a UFD if
and only if every prime ideal of height 1 is principal (I, 1.12A). So what we
must show is that if A is an integrally closed domain, then every prime
ideal of height 1 is principal if and only if Cl(Spec A) = 0.
One way is easy: if every prime ideal of height 1 is principal, consider a
prime divisor Y <;; X = Spec A. Y corresponds to a prime ideal p of
height 1. If p is generated by an element f E A, then clearly the divisor
off is 1 · Y. Thus every prime divisor is principal, so Cl X = 0.
For the converse, suppose Cl X = 0. Let p be a prime ideal of height 1,
and let Y be the corresponding prime divisor. Then there is an f E K, the
quotient field of A, with (f) = Y. We will show that in fact f E A and
f generates p. Since vy(f) = 1, we have f E AP, and f generates pAP. If
p' <;; A is any other prime ideal of height 1, then p' corresponds to a prime
divisor Y' of X, and Vy·(f) = 0, so f E Ap'· Now the algebraic result (6.3A)
below implies that f EA. In fact, f E A n pAv = p. Now to show that f
generates p, let g be any other element of p. Then Vy(g) ? 1 and Vy·(g) ? 0
for all Y' i= Y. Hence vy.(g/f) ? 0 for all prime divisors Y' (including Y).
Thus gjf E Ap' for all p' ofheight 1, so by(6.3A) again,g/f EA. In other words,
g E Af, which shows that p is a principal ideal, generated by f

Proposition 6.3A. Let A be an integrally closed noetherian domain. Then

A= n Av
ht p =1
where the intersection is taken over all prime ideals of height 1.
PROOF. Matsumura [2, Th. 38, p. 124].

Example 6.3.1. If X is affine n-space AJ: over a field k, then Cl X = 0.

Indeed, X = Spec k[x 1 , ... ,xn], and the polynomial ring is a UFD.

Example 6.3.2. If A is a Dedekind domain, then Cl(Spec A) is just the ideal

class group of A, as defined in algebraic number theory. Thus (6.2) generalizes
the fact that A is a UFD if and only if its ideal class group is 0.

Proposition 6.4. Let X be the projective space PJ: over a field k. For any
divisor D = In;¥;, define the degree of D by deg D = In; deg Y;, where
deg Y; is the degree of the hypersurface ¥;. Let H be the hyperplane x 0 =
0. Then:
(a) if D is any divisor of degree d, then D ~ dH;
(b) for any f E K*, deg(f) = 0;
(c) the degree function gives an isomorphism deg:Cl X ---+ Z.
PROOF. Let S = k[ x 0 , .•. ,xn] be the homogeneous coordinate ring of X.
If g is a homogeneous element of degree d, we can factor it into irreducible

132
 6 Divisors

polynomials g = g1' · · · g~r. Then g; defines a hypersurface Y; of degree

d; = deg g;, and we can define the divisor of g to be (g) = In; Y;. Then
deg(g) = d. Now a rational function f on X is a quotient gjh of homo-
geneous polynomials of the same degree. Clearly (f) = (g) - (h), so we
see that deg(f) = 0, which proves (b).
If D is any divisor of degree d, we can write it as a difference D 1 - D 2
of effective divisors of degrees d 1 ,d2 with d 1 - d2 = d. Let D 1 = (g 1) and
D 2 = (g 2 ). This is possible, because an irreducible hypersurface in pn
corresponds to a homogeneous prime ideal of height 1 in S, which is prin-
cipal. Taking power products we can get any effective divisor as (g) for
some homogeneous g. Now D - dH = (f) where f = gtfx~g 2 is a ra-
tional function on X. This proves (a). Statement (c) follows from (a), (b),
and the fact that deg H = 1.

Proposition 6.5. Let X satisfy(*), let Z be a proper closed subset of X, and

let U = X - Z. Then:
(a) there is a surjective homomorphism Cl X -> Cl U defined by D =
In; Y; f---+ In;( Y; n U), where we ignore those Y; n U which are empty;
(b) if codim(Z,X) ;?; 2, then Cl X -> Cl U is an isomorphism;
(c) if Z is an irreducible subset of codimension 1, then there is an exact
sequence
Z -> Cl X -> Cl U -> 0,
where the first map is defined by 1 f---+ 1 · Z.
PROOF.
(a) If Y is a prime divisor on X, then Y n U is either empty or a prime
divisor on U. Iff E K*, and (f) = In;Y;, then considering f as a rational
function on U, we have (f)u = In;(Y; n U), so indeed we have a homo-
morphism Cl X -> Cl U. It is surjective because every prime divisor of U
is the restriction of its closure in X.
(b) The groups Div X and Cl X depend only on subsets of codimension
1, so removing a closed subset Z of codimension ;?; 2 doesn't change anyr l~ing.
(c) The kernel of Cl X -> Cl U consists of divisors whose support is
contained in Z. If Z is irreducible, the kernel is just the subgroup of Cl X
generated by 1 · Z.

Example 6.5.1. Let Y be an irreducible curve of degree d in Pl. Then

Cl(P 2 - Y) = Z/dZ. This follows immediately from (6.4) and (6.5).

Example 6.5.2. Let k be a field, let A = k[x,y,z]/(xy - z 2 ), and let X =

Spec A. Then X is an affine quadric cone in A~. We will show that Cl X =
Zj2Z, and that it is generated by a ruling of the cone, say Y:y = z = 0
(Fig. 8).
First note that Y is a prime divisor, so by (6.5) we have an exact sequence
Z -> Cl X -> Cl(X - Y) -> 0,

133
II Schemes

Figure 8. A ruling on the quadric cone.

where the first map sends 1~---> 1 · Y. Now Y can be cut out set-theoretically
by the function y. In fact, the divisor of y is 2 · Y, because y = 0 =z 2 = 0,
and z generates the maximal ideal of the local ring at the generic point of Y.
Hence X - Y = Spec AY. NowAY = k[x,y,y-l,z]/(xy - z2 ). In this ring
x = y- 1 z 2 , so we can eliminate x, and find AY ~ k[y,y- 1 ,z]. This is a UFD,
so by (6.2), Cl(X - Y) = 0.
Thus we see that Cl X is generated by Y, and that 2 · Y = 0. It remains
to show that Y itself is not a principal divisor. Since A is integrally closed
(Ex. 6.4), it is equivalent to show that the prime ideal of Y, namely p =
(y,z), is not principal (cf. proof of (6.2) ). Let m = (x,y,z), and note that
m/m 2 is a 3-dimensional vector space over k generated by x,y;z, the images
of x,y,z. Now p ~ m, and the image of pin m/m 2 contains y and z. Hence
p cannot be a principal ideal.

Proposition 6.6. Let X satisfy (*). Then X x A 1 (=X x specz Spec Z[t])
also satisfies (*), and Cl X ~ Cl(X x A 1 ).
PROOF. Clearly X x A 1 is noetherian, integral, and separated. To see that
it is regular in codimension one, we note that there are two kinds of points
of codimension one on X x A 1 . Type 1 is a point x whose image in X
is a pointy of codimension one. In this case xis the generic point of n- 1(y),
where n:X x A 1 --+ X is the projection. Its local ring is (!)x ~ (!)y[t]my'
which is clearly a discrete valuation ring, since (!) Y is. The corresponding
prime divisor {x}- isjustn- 1 ({y}-).
Type 2 is a point x E X x A 1 of codimension one, whose image in X
is the generic point of X. In this case (!)xis a localization of K[t] at some
maximal ideal, where K is the function field of X. It is a discrete valuation
ring because K[t] is a principal ideal domain. Thus X x A 1 also satisfies(*).
We define a map Cl X--+ Cl(X x A 1 ) by D =In)';~---> n*D = Inin- 1(¥;).
Iff E K*, then n*( (f)) is the divisor off considered as an element of K(t),
the function field of X x A 1 . Thus we have a homomorphism n*: Cl X --+
Cl(X X A 1).

134
 6 Divisors

To show n* is injective, suppose D E Div X, and n* D = (f) for some

f EK(t). Since n*D involves only prime divisors of type 1, f must be inK.
For otherwise we could write f = g/h, with g,h E K[t], relatively prime.
If g,h are not both in K, then (f) will involve some prime divisor of type 2
on X x A 1 . Now iff E K, it is clear that D = (f), son* is injective.
To show that n* is surjective, it will be sufficient to show that any prime
divisor of type 2 on X x A 1 is linearly equivalent to a linear combination
of prime divisors of type 1. So let Z s; X x A 1 be a prime divisor of type 2.
Localizing at the generic point of X, we get a prime divisor in Spec K[t],
which corresponds to a prime ideal p s; K[t]. This is principal, so let f
be a generator. Then f E K(t), and the divisor off consists of Z plus perhaps
something purely of type 1. It cannot involve any other prime divisors of
type 2. Thus Z is linearly equivalent to a divisor purely of type 1. This
completes the proof.

Example 6.6.1. Let Q be the nonsingular quadric surface xy = zw in Pi.

We will show that Cl Q ~ Z EB Z. We use the fact that Q is isomorphic
to Pl x k Pl (1, Ex. 2.15). Let p 1 and p 2 be the projections of Q onto the
two factors. Then as in the proof of (6.6) we obtain homomorphisms
Pi ,p~: Cl P 1 - t Cl Q. First we show that Pi and p~ are injective. Let Y =
pt x P 1 . Then Q - Y = A 1 x P 1 , and the composition
Cl P 1 ~ Cl Q - t Cl(A 1 X P 1)
is the isomorphism of (6.6). Hence p~ (and similarly pi) is injective.
Now consider the exact sequence of(6.5) for Y:
z -t Cl Q - t Cl(A 1 X P 1) -t 0.
In this sequence the first map sends 1 to Y. But if we identify Cl P 1 with
Z by letting 1 be the class of a point, then this first map is just Pi, hence is
injective. Since the image of p~ goes isomorphically to Cl(A 1 x P 1 ) as we
have just seen, we conclude that Cl Q ~ Im Pi EB Im p~ = Z EB Z. If D
is any divisor on Q, let (a,b) be the ordered pair of integers in Z EB Z cor-
responding to the class of D under this isomorphism. Then we say D is
of type (a,b) on Q.

Example 6.6.2. Continuing with the quadric surface Q s; P 3 , we will show

that the embedding induces a homomorphism Cl P 3 - t Cl Q, and that the
image of a hyperplane H, which generates Cl P 3 , is the element (1,1) in Cl Q =
Z EB Z. Let Y be any irreducible hypersurface of P 3 which does not con-
tain Q. Then we can assign multiplicities to the irreducible components
of Y n Q so as to obtain a divisor Y · Q on Q. Indeed, on each standard
open set U; of P 3 , Y is defined by a single function f; we can take the value
of this function (restricted to Q) for each valuation of a prime divisor of Q
to define the divisor Y · Q. By linearity we extend this map to define a

135
II Schemes

divisor D · Q on Q, for each divisor D = In;¥; on P 3 , such that no ¥; con-

tains Q. Clearly linearly equivalent divisors restrict to linearly equivalent
divisors. Since any divisor on P 3 is linearly equivalent to one whose prime
divisors don't contain Q by (6.4), we obtain a well-defined homomorphism
Cl P 3 ---+ Cl Q. Now if His the hyperplane w = 0, then H n Q is the divisor
consisting of the two lines x = w = 0 and y = w = 0. One is in each
family (I, Ex. 2.15) so H n Q is of type (1,1) in Cl Q = Z 6::> Z. Note that
the two families of lines correspond to pt x P 1 and P 1 x pt, so they are
of type (1,0) and (0,1).

Example 6.6.3. Carrying this example one step further, let C be the twisted
cubic curve x = t 3 , y = u 3 , z = t 2 u, w = tu 2 which lies on Q. If Y is the
quadric cone yz = w 2 , then Y n Q = C u L where Lis the line y = w = 0.
Since Y ~ 2H on P 3 , Y n Q is a divisor of type (2,2). The line L has type
(1,0), so C is of type (1,2). It follows that there does not exist any surface
Y s; P\ not containing Q, such that Y n Q = C, even set-theoretically!
For in that case the divisor Y n Q would be rC for some integer r > 0.
This is a divisor of type (r,2r) in Cl Q. But if Y is a surface of degree d, then
Y n Q is of type (d,d), which can never equal (r,2r). Thus Y does not exist.

Example 6.6.4. We will see later (V, 4.8) that if X is a nonsingular cubic
surface in P 3 , then Cl X ~ Z 7 .

Divisors on Curves
We will illustrate the notion of the divisor class group further by paying
special attention to the case of divisors on curves. We will define the degree
of a divisor on a curve, and we will show that on a complete nonsingular
curve, the degree is stable under linear equivalence. Further study of
divisors on curves will be found in Chapter IV.
To begin with, we need some preliminary information about curves and
morphisms of curves. Recall our conventions about terminology from the
end of Section 4:

Definition. Let k be an algebraically closed field. A curve over k is an

integral separated scheme X of finite type over k, of dimension one.
If X is proper over k, we say that X is complete. If all the local rings
of X are regular local rings, we say that X is nonsingular.

Proposition 6.7. Let X be a nonsingular curve over k with function field K.

Then the following conditions are equivalent:
(i) X is projective;
(ii) X is complete;
(iii) X ~ t(Cx), where Cx is the abstract nonsingular curve of (1, §6),
and t is the functor from varieties to schemes of (2.6).

136
 6 Divisors

PROOF.
(i) => (ii) follows from (4.9).
(ii) => (iii). If X is complete, then every discrete valuation ring of Kjk
has a unique center on X (Ex. 4.5). Since the local rings of X at the closed
points are all discrete valuation rings, this implies that the closed points
of X are in 1-1 correspondence with the discrete valuation rings of Kjk,
namely the points of Cx. Thus it is clear that X ~ t(Cx).
(iii) => (i) follows from (1, 6.9).

Proposition 6.8. Let X be a complete nonsingular curve over k, let Y be any

curve over k, and let f:X ~ Y be a morphism. Then either (1) f(X) =
a point, or (2) f(X) = Y. In case (2), K(X) is a finite extension field of
K(Y), f is a finite morphism, and Y is also complete.
PROOF. Since X is complete, f(X) must be closed in Y, and proper over
Spec k (Ex. 4.4). On the other hand, f(X) is irreducible. Thus either
(1) f(X) = pt, or (2) f(X) = Y, and in case (2), Y is also complete.
In case (2), f is dominant, so it induces an inclusion K(Y) ~ K(X) of
function fields. Since both fields are finitely generated extension fields of
transcendence degree 1 of k, K(X) must be a finite algebraic extension of
K(Y). To show that f is a finite morphism, let V = Spec B be any open
affine subset of Y. Let A be the integral closure of B in K(X). Then A is a
finite B-module (I, 3.9A), and Spec A is isomorphic to an open subset U of
X (I, 6. 7). Clearly U = f- 1 V, so this shows that f is a finite morphism.

Definition. Iff: X~ Y is a finite morphism of curves, we define the degree

off to be the degree of the field extension [ K(X):K(Y)].

Now we come to the study of divisors on curves. If X is a nonsingular

curve, then X satisfies the condition (*) used above, so we can talk about
divisors on X. A prime divisor is just a closed point, so an arbitrary divisor
can be written D = I,n;P;, where the P; are closed points, and n; E Z. We
define the degree of D to be :Ln;.

Definition. If f: X ~ Y is a finite morphism of nonsingular curves, we

define a homomorphism f*: Div Y ~ Div X as follows. For any point
Q E Y, let t E (!)Q be a local parameter at Q, i.e., tis an element of K(Y)
with vQ(t) = 1, where vQ is the valuation corresponding to the discrete
valuation ring (!)Q· We define f*Q = LJ<PJ;Q vp(t) · P. Since f is a
finite morphism, this is a finite sum, so we get a divisor on X. Note
that f*Q is independent of the choice of the local parameter t. Indeed,
if t' is another local parameter at Q, then t' = ut where u is a unit in
(!)Q· For any point P EX with f(P) = Q, u will be a unit in @p, so vp(t) =
vp(t'). We extend the definition by linearity to all divisors on Y. One
sees easily that f* preserves linear equivalence, so it induces a homo-
morphism f*: Cl Y ~ Cl X.

137
II Schemes

Proposition 6.9. Let f: X --+ Y be a finite morphism of nonsingular curves.

Then for any divisor Don Y we have deg f*D = deg f · deg D.
PROOF. It will be sufficient to show that for any closed point Q E Y we have
deg f*Q = deg f Let V = Spec B be an open affine subset of Y containing
Q. Let A be the integral closure of B in K(X). Then, as in the proof of
(6.8), U =Spec A is the open subset f- 1 v of X. Let mQ be the maximal
ideal of Q in B. We localize both Band A with respect to the multiplicative
system S = B - mQ, and we obtain a ring extension (DQ 4 A', where A'
is a finitely generated {DQ-module. Now A' is torsion-free, and has rank
equal to r = [K(X):K(Y)], so A' is a free {DQ-module of rank r = deg f
If t is a local parameter at Q, it follows that A'/tA' is a k-vector space of
dimension r.
On the other hand, the points Pi of X such that f(P;) = Q are in 1-1
correspondence with the maximal ideals mi of A', and for each i, A;" =
(DP,· Clearly tA' = ni(tA;,, 11 A'), so by the Chinese remainder theor~m,
dimk A'/tA' = L dimk A'/(tA;n, 11 A').
i
But
A'/(tA;", 11 A') ~ A;"jtA;", = (Dpjt@p,,
so the dimensions in the sum above are just equal to vp.(t). But f*Q =
l:vp,(t) · Pi, so we have shown that deg f*Q = deg f as required.

Corollary 6.10. A principal divisor on a complete nonsingular curve X has

degree zero. Consequently the degree function induces a surjective homo-
morphism deg: Cl X --+ Z.
PROOF. Let f E K(X)*. Iff E k, then (f) = 0, so there is nothing to prove.
Iff ¢= k, then the inclusion of fields k(f) <;; K(X) induces a finite morphism
cp:X --+ P 1 . It is a morphism by (1, 6.12), and it is finite by (6.8). Now (f) =
cp*( {0} - {oo }). Since {0} - {oo} is a divisor of degree 0 on P\ we conclude
that (f) has degree 0 on X.
Thus the degree of a divisor on X depends only on its linear equivalence
class, and we obtain a homomorphism Cl X --+ Z as stated. It is surjective,
because the degree of a single point is 1.

Example 6.10.1. A complete nonsingular curve X is rational if and only if

there exist two distinct points P,Q E X with P "' Q. Recall that rational
means birational to P 1 . If X is rational, then in fact it is isomorphic to P 1
by (6.7). And on P 1 we have already seen that any two points are linearly
equivalent (6.4). Conversely, suppose X has two points P =1- Q with P "' Q.
Then there is a rational function f E K(X) with (f) = P - Q. Consider
the morphism cp: X --+ P 1 determined by f as in the proof of (6.10). We
have cp*( {0}) = P, so cp must be a morphism of degree 1. In other words,
cp is birational, so X is rational.

138
 6 Divisors

Example 6.10.2. Let X be the nonsingular cubic curve y 2 z = x 3 - xz 2 in

P~, with char k =!= 2. We have already seen that X is not rational (I, Ex. 6.2).
Let Cl 0 X be the kernel of the degree map Cl X ~ Z. Then from the previous
example we know that Cl 0 X =!= 0. We will show in fact that there is a
natural 1-1 correspondence between the set of closed points of X and the
elements of the group Clo X. On the one hand this elucidates the structure
of the group Clo X. On the other hand it gives us a group structure on the
set of closed points of X, which makes X into a group variety (Fig. 9).

Figure 9. The group law on a cubic curve.

Let P 0 be the point (0,1,0) on X. It is an inflection point, so the tangent

line z = 0 at that point meets the curve in the divisor 3P 0 . If Lis any other
line in P 2 , meeting X in three points P,Q,R (which may coincide), then since L
is linearly equivalent to the line z = 0 in P 2 , we have P + Q + R ~ 3P 0
on X, as in (6.6.2) above.
Now to any closed point P EX, we associate the divisor P - P0 E Cl" X.
This map is injective, because if P - P0 ~ Q - P 0 , then P ~ Q, and X
would be rational by the previous example, which is impossible.
To show that the map from the closed points of X to Clo X is surjective,
we proceed in several steps. Let D E Clo X. Then D = I.niPi, with I.ni = 0.
Hence we can also write D = I.ni(Pi - P 0 ). Now for any point R, let the
line P 0 R meet X further in the point T (always counting intersections with
multiplicities-for example, if R = P 0 , we take the line P 0 R to be the tangent
line at P 0 , and then the third intersection Tis alsoP 0 ). Then P 0 + R + T ~
3P 0 , so R - P 0 ~ -(T - P 0 ). If i is an index such that ni < 0 in D, we
take Pi = R. Then replacing Pi by T, we get a linearly equivalent divisor
with the ith coefficient - ni > 0. Repeating this process, we may assume that
D = I.n;(P; - P 0 ) with all n; > 0. We now show by induction on I.ni that
D ~ P - P 0 for some point P. If In; = 1, there is nothing to prove. So
suppose In; ~ 2, and let P,Q be two of the points P; (maybe the same) which
occur in D. Let the line PQ meet X in R, and let the line P 0 R meet X in T.

139
II Schemes

Then we have
P + Q+ R ~ 3P0 and P0 +R + T ~ 3P 0
so
(P- P 0 ) + (Q - P0 ) ~ (T- P 0 ).
Replacing P and Q by T, we get D linearly equivalent to another divisor of the
same form whose In; is one less, so by induction D ~ P - P 0 for some P.
Thus we have shown that the group Clo X is in 1-1 correspondence with
the set of closed points of X. One can show directly that the addition law
determines a morphism of X x X --+X, and the inverse law determines a
morphism X--+ X (see for example Olson [1]). Thus X is a group variety
in the sense of (1, Ex. 3.21). See (IV, 1.3.7) for a generalization.

Remark 6.10.3. This example of the cubic curve illustrates the general fact
that the divisor class group of a variety has a discrete component (in this
case Z) and a continuous component (in this case Clo X) which itself has the
structure of an algebraic variety.
More specifically, if X is any complete nonsingular curve, then the group
Clo X is isomorphic to the group of closed points of an abelian variety called
the Jacobian variety of X. An abelian variety is a complete group variety
over k. The dimension of the Jacobian variety is the genus of the curve.
Thus the whole divisor class group of X is an extension of Z by the group
of closed points of the Jacobian variety of X.
If X is a nonsingular projective variety of dimension ~ 2, then one can
define a subgroup Clo X of Cl X, namely the subgroup of divisor classes
algebraically equivalent to zero, such that Cl X /Clo X is a finitely generated
abelian group, called the Neron-Severi group of X, and CloX is isomorphic
to the group of closed points of an abelian variety called the Picard variety
of X.
Unfortunately we do not have space in this book to develop the theory
of abelian varieties and to study the Jacobian and Picard varieties of a
given variety. For more information and further references on this beautiful
subject, see Lang [1], Mumford [2], Mumford [5], and Hartshorne [6].
See also (IV, §4), (V, Ex. 1.7), and Appendix B.

Cartier Divisors
Now we want to extend the notion of divisor to an arbitrary scheme. It
turns out that using the irreducible subvarieties of codimension one doesn't
work very well. So instead, we take as our point of departure the idea that
a divisor should be something which locally looks like the divisor of a
rational function. This is not exactly a generalization of the Weil divisors
(as we will see), but it gives a good notion to use on arbitrary schemes.

Definition. Let X be a scheme. For each open affine subset U = Spec A,

letS be the set of elements of A which are not zero divisors, and let K ( U) be

140
 6 Divisors

the localization of A by the multiplicative systemS. We call K(U) the total

quotient ring of A. For each open set U, let S(U) denote the set of elements of
r(U,f!'x) which are not zero divisors in each local ring {!!x for x E U. Then the
rings S(U)- 1 r(U,(!Jx) form a presheaf, whose associated sheaf of rings X we
call the sheaf of total quotient rings of(!). On an arbitrary scheme, the sheaf
f replaces the concept of function field of an integral scheme. We denote
by f* the sheaf (of multiplicative groups) of invertible elements in the
sheaf of rings f. Similarly {!!* is the sheaf of invertible elements in (!).

Definition. A Cartier divisor on a scheme X is a global section of the sheaf

f* j(!J*. Thinking of the properties of quotient sheaves, we see that a
Cartier divisor on X can be described by giving an open cover {U;} of X,
and for each i an element J; E r(U;,f*), such that for each i,j, J;!jj E
r( U; n Ui,{!!*). A Cartier divisor is principal if it is in the image of the
natural map r(X,f*) --> T(X,f* j{!!*). Two Cartier divisors are linearly
equivalent if their difference is principal. (Although the group operation
on x* j{!!* is multiplication, we will use the language of additive groups
when speaking of Cartier divisors, so as to preserve the analogy with
Wei! divisors.)

Proposition 6.11. Let X be an integral, separated noetherian scheme, all of

whose local rings are unique factorization domains (in which case we say
X is locally factorial). Then the group Div X of Wei! divisors on X is
isomorphic to the yroup of Cartier divisors r(X,f* j(!J*), and furthermore,
the principal Wei/ divisors correspond to the principal Cartier divisors
under this isomorphism.
PROOF. First note that X is normal, hence satisfies(*), since a UFD is inte-
grally closed. Thus it makes sense to talk about Weil divisors. Since X is
integral, the sheaf ff is just the constant sheaf corresponding to the function
field K of X. Now let a Cartier divisor be given by {(U;,J;)} where {U;} is
an open cover of X, and J; E r(U;,f*) = K*. We define the associated
Weil divisor as follows. For each prime divisor Y, take the coefficient of Y
to be ry(J;), where i is any index for which Y n U; i= 0. If j is another
such index, then J;/jj is invertible on U; n Ui, so vy(J;/jj) = 0 and vy(f) =
t'r(jj). Thus we obtain a well-defined Weil divisor D = Ivr(J;)Y on X.
(The sum is finite because X is noetherian!)
Conversely, if D is a Wei! divisor on X, let x EX be any point. Then D
induces a Weil divisor Dx on the local scheme Spec {!!x· Since {!!xis a UFD,
Dx is a principal divisor, by (6.2), so let Dx = Ux) for some fx E K. Now
the principal divisor Ux) on X has the same restriction to Spec {!!x as D,
hence they differ only at prime divisors which do not pass through x. There
are only finitely many of these which have a non-zero coefficient in D or
Ux), so there is an open neighborhood U x of x such that D and Ux) have the
same restriction to U x· Covering X with such open sets U x• the functions
fx give a Cartier divisor on X. Note that if f,f' give the same Weil divisor

141
II Schemes

on an open set U, then f/f' E r(U,lD*), since X is normal (cf. proof of (6.2) ).
Thus we have a well-defined Cartier divisor.
These two constructions are inverse to each other, so we see that the
groups of Wei! divisors and Cartier divisors are isomorphic. Furthermore
it is clear that the principal divisors correspond to each other.
Remark 6.11.1A. Since a regular local ring is UFO (Matsumura [2, Th. 48,
p. 142]), this proposition applies in particular to any regular integral sepa-
rated noetherian scheme. A scheme is regular if all of its local rings are
regular local rings.

Remark 6.11.2. If X is a normal scheme, which is not necessarily locally

factorial, we can define a subgroup of Div X consisting of the locally prin-
cipal Wei! divisors: Dis locally principal if X can be covered by open sets
U such that Diu is principal for each U. Then the above proof shows that
the Cartier divisors are the same as the locally principal Wei! divisors.

Example 6.11.3. Let X be the affine quadric cone Spec k[ x,y,z ]/(xy - z 2 )
treated above (6.5.2). The ruling Y is a Wei! divisor which is not locally
principal in the neighborhood of the vertex of the cone. Indeed, our earlier
proof shows that its prime ideal pAm is not a principal ideal even in the
local ring Am. Thus Y does not correspond to a Cartier divisor. On the
other hand 2 Y is locally principal, and in fact principal. So in this case the
group of Cartier divisors modulo principal divisors is 0, whereas Cl X ~
Z/2Z.
Example 6.11.4. Let X be the cuspidal cubic curve y 2 z = x 3 in P~, with
char k # 2. In this case X does not satisfy (*), so we cannot talk about
Wei! divisors on X. However, we can talk about the group CaCl X of
Cartier divisor classes modulo principal divisors. Imitating the case of the
nonsingular cubic curve (6.10.2) we will show:
(a) there is a surjective degree homomorphism deg:CaCl X --+ Z;
(b) there is a 1-1 correspondence between the set of nonsingular closed
points of X and the kernel CaClo X of the degree map, which makes it into
a group variety; and in fact
(c) there is a natural isomorphism of group varieties between CaCloX
and the additive group Ga of the field k (1, Ex. 3.21a).
To define the degree of a Cartier divisor on X, note that any Cartier
divisor is linearly equivalent to one whose local function is invertible in some
neighborhood of the singular point Z = (0,0,1). Then this Cartier divisor
corresponds to a Weil divisor D = 'IniPi on X - Z, and we define the
degree of the original divisor to be deg D = 'Ini. The proof of(6.10) shows
that iff E K is invertible at Z, then the principal divisor (f) on X - Z has
degree 0. Thus the degree of a Cartier divisor on X is well-defined, and it
passes to linear equivalence classes to give a surjective homomorphism
deg: CaCl X --+ Z.

142
 6 Divisors

Now let P 0 be the point (0,1,0) as in the case of the nonsingular cubic
curve. To each closed point P EX - Z, we associate the Cartier divisor
Dp which is 1 in a neighborhood of Z, and which corresponds to the Wei!
divisor P - P 0 on X - Z. First note this map is injective: if P # Q are
two points in X - Z, and if Dp - DQ, then there is an f E K*, which is
invertible at Z, and such that (f) = P - Q on X - Z. Then f gives a
morphism of X to P 1 , which must be birational. But then the local ring of
Z on X would dominate some discrete valuation ring of P 1 , and this is
impossible, because Z is a singular point.
To show that every divisor in CaClo X is linearly equivalent to Dp for
some closed point P E X - Z, we proceed exactly as in the case of the non-
singular cubic curve above. The only difference is to note that the geometric
constructions R H T and P,Q H R, T described above remain inside of
X - Z. Thus the group CaCloX is in 1-1 correspondence with the set of
closed points of X - Z, making it into a group variety.
In this case we are able to identify the group variety as G.. Of course,
we know that X is a rational curve, and so X - Z ~ Af (1, Ex. 3.2). But
in fact, if we use the right parametrization, the group law corresponds. So
define a morphism of G. = Spec k[t] to X - Z by t H (t,1,t 3 ). This is
clearly an isomorphism of varieties. Using a little elementary analytic
geometry (left to reader!) one shows that if P = (t,1,t 3 ) and if Q = (u,1,u 3 ),
then the point T constructed above is just (t + u,1,(t + u) 3 ). So we have an
isomorphism of group varieties of G. to X - Z with the group structure
ofCaCl 0 X.

Invertible Sheaves
Recall that an invertible sheaf on a ringed space X is defined to be a locally
free C9x-module of rank 1. We will see now that invertible sheaves on a
scheme are closely related to divisor classes modulo linear equivalence.

Proposition 6.12. If f£ and .A are invertible sheaves on a ringed space X,

so is f£ ® .A. Iff£ is any invertible sheaf on X, then there exists an
invertible sheaf y- 1 on X such that f£ ® y- 1 ~ C9x.
PROOF. The first statement is clear, since f£ and .A are both locally free of
rank 1, and (!Jx ® (!Jx ~ (!Jx· For the second statement, let f£ be any in-
vertible sheaf, and take f£- 1 to be the dual sheaf y~ = Yfom(fi',(!Jx). Then
!£~ ® f£ ~ Yfom(f£,!£) = (!Jx by (Ex. 5.1).

Definition. For any ringed space X, we define the Picard group of X, Pic X,
to be the group of isomorphism classes of invertible sheaves on X, under
the operation ®. The proposition shows that in fact it is a group.

Remark 6.12.1. We will see later (III, Ex. 4.5) that Pic X can be expressed as
the cohomology group H 1 (X,(!Jk).

143
II Schemes

Definition. Let D be a Cartier divisor on a scheme X, represented by {(Ui,.t;)}

as above. We define a subsheaf Y(D) of the sheaf of total quotient rings
%by taking Y(D) to be the sub-lDx-module of% generated by fi- 1 on
Ui. This is well-defined, since.f;/./j is invertible on Ui n Ui, so fi- 1 andfj 1
generate the same lDx-module. We callY(D) the sheaf associated to D.

Proposition 6.13. Let X be a scheme. Then:

(a) for any Cartier divisor D, Y(D) is an invertible sheaf on X. The
map D ~ Y(D) gives a 1-1 correspondence between Cartier divisors on X
and invertible sub sheaves of %;
(b) 2(D 1 - D 2 ) ~ 2(D 1 ) ® 2(D 2 )- 1 ;
(c) D 1 ~ D 2 if and only if 2(D 1) ~ 2(D 2 ) as abstract invertible
sheaves (i.e., disregarding the embedding in %).
PROOF.
(a) Since each/; E r(Ui,%*), the map lDu,-+ Y(D)Iu, defined by 1 ~ fi- 1
is an isomorphism. Thus Y(D) is an invertible sheaf. The Cartier divisor D
can be recovered from Y(D) together with its embedding in %, by taking
/; on Ui to be the inverse of a local generator of Y(D). For any invertible
subsheaf of %, this construction gives a Cartier divisor, so we have a 1-1
correspondence as claimed.
(b) If D 1 is locally defined by /; and D 2 is locally defined by gi, then
2(D 1 - D 2 ) is locally generated by fi 1 gi, so 2(D 1 - D 2 ) = 2(D 1 ) · 2(D 2 )- 1
as subsheaves of %. This product is clearly isomorphic to the abstract
tensor product 2'(D 1 ) ® 2'(D 2 )- 1 .
(c) Using (b), it will be sufficient to show that D = D 1 - D 2 is principal
if and only if Y(D) ~ lDx. If Dis principal, defined by f E F(X,%*), then
Y(D) is globally generated by f- 1 , so sending 1 ~ f- 1 gives an isomor-
phism lDx ~ Y(D). Conversely, given such an isomorphism, the image of 1
gives an element of r(X,%*) whose inverse will define D as a principal
divisor.

Corollary 6.14. On any scheme X, the map D ~ Y(D) gives an injective

homomorphism of the group CaCl X of Cartier divisors modulo linear
equivalence to Pic X.

Remark 6.14.1. The map CaCl X -+ Pic X may not be surjective, because
there may be invertible sheaves on X which are not isomorphic to any
invertible subsheaf of %. For an example of Kleiman, see Hartshorne
[5, 1.1.3, p. 9]. On the other hand, this map is an isomorphism in most
common situations. Nakai [2, p. 301] has shown that it is an isomorphism
whenever X is projective over a field. We will show now that it is an isomor-
phism if X is integral.

144
 6 Divisors

Proposition 6.15. If X is an integral scheme, the homomorphism CaCl X ~

Pic X of (6.14) is an isomorphism.
PROOF. We have only to show that every invertible sheaf is isomorphic to a
subsheaf of x, which in this case is the constant sheaf K, where K is the
function field of X. So let 2 be any invertible sheaf, and consider the sheaf
2 ®(/)x X. On any open set U where 2 ~ (!Jx, we have 2 @ X ~ X, so
it is a constant sheaf on U. Now because X is irreducible, it follows that any
sheaf whose restriction to each open set of a covering of X is constant, is
in fact a constant sheaf. Thus 2 @ X is isomorphic to the constant sheaf
X, and the natural map 2 ~ 2 @X ~ X expresses 2 as a subsheaf
of X.

Corollary 6.16. If X is a noetherian, integral, separated locally factorial

scheme, then there is a natural isomorphism Cl X ~ Pic X.
PROOF. This follows from (6.11) and (6.15).

Corollary 6.17. If X = PI: for some field k, then every invertible sheaf on X
is isomorphic to (!)(I) for some l E Z.
PROOF. By (6.4), Cl X ~ Z, so by (6.16), Pic X ~ Z. Furthermore the gen-
erator of Cl X is a hyperplane, which corresponds to the invertible sheaf
(!)(1). Hence Pic X is the free group generated by (!)(1), and any invertible
sheaf 2 is isomorphic to (!J(l) for some l E Z.

We conclude this section with some remarks about closed subschemes of

codimension one of a scheme X.

Definition. A Cartier divisor on a scheme X is effective if it can be repre-

sented by {(Ui,.J;)} where all the .J; E r(Ui,(!Ju). In that case we define
the associated subscheme of codimension 1, Y, to be the closed subscheme
defined by the sheaf of ideals § which is locally generated by .J;.

Remark 6.17.1. Clearly this gives a 1-1 correspondence between effective

Cartier divisors on X and locally principal closed subschemes Y, i.e., sub-
schemes whose sheaf of ideals is locally generated by a single element. Note
also that if X is an integral separated noetherian locally factorial scheme,
so that the Cartier divisors correspond to Weil divisors by (6.11 ), then the
effective Cartier divisors correspond exactly to the effective Weil divisors.

Proposition 6.18. Let D be an effective Cartier divisor on a scheme X, and let

Y be the associated locally principal closed subscheme. Then§ y ~ 2(- D).

145
II Schemes

PROOF. 2(- D) is the subsheaf of :ff generated locally by J;. Since D is

effective, this is actually a subsheaf of (!Jx, which is none other than the ideal
sheaf§ y of Y.

EXERCISES

6.1. Let X be a scheme satisfying (*). Then X x P" also satisfies (*),and Cl(X x P") ~
(ClX) X Z.

*6.2. Varieties in Projective Space. Let k be an algebraically closed field, and let X
be a closed subvariety of P~ which is nonsingular in codimension one (hence
satisfies(*)). For any divisor D = IniY; on X, we define the degree of D to be
Ini deg 1;, where deg 1; is the degree of 1;, considered as a projective variety
itself (I, §7).
(a) Let V be an irreducible hypersurface in P" which does not contain X, and let
1; be the irreducible components of V n X. They all have codimension 1 by
(I, Ex. 1.8). For each i, let J; be a local equation for Von some open set Ui of
P" for which Y; n Ui =1= 0, and let ni = Vy (hJ, where J: is the restriction of
J; to ui " X. Then we define the divisor v.x to be Ini Y;. Extend by linearity,
and show that this gives a well-defined homomorphism from the subgroup of
Div P" consisting of divisors, none of whose components contain X, to Div X.
(b) If Dis a principal divisor on P", for which D.X is defined as in (a), show that D.X
is principal on X. Thus we get a homomorphism Cl P" --> Cl X.
(c) Show that the integer ni defined in (a) is the same as the intersection multiplicity
i(X,V; Y;) defined in (I, §7). Then use the generalized Bezout theorem (I, 7.7)
to show that for any divisor Don P", none of whose components contain X,
deg(D.X) = (deg D)· (deg X).
(d) If Dis a principal divisor on X, show that there is a rational function f on P"
such that D = (f).X. Conclude that deg D = 0. Thus the degree function
defines a homomorphism deg: Cl X --> Z. (This gives another proof of (6.10),
since any complete nonsingular curve is projective.) Finally, there is a com-
mutative diagram
Cl P" ClX

~jdeg
·(deg X)
z z
and in particular, we see that the map Cl P" --> Cl X is injective.

*6.3. Cones. In this exercise we compare the class group of a projective variety V to
the class group of its cone (I, Ex. 2.10). So let V be a projective variety in P", which
is of dimension ~ 1 and nonsingular in codimension 1. Let X = C(V) be the
affine cone over V in A"+ 1 , and let X be its projective closure in P" + 1 . Let P E X
be the vertex of the cone.
(a) Let n:X - P --> V be the projection map. Show that V can be covered by
open subsets Ui such that n- 1(U;) ~ Ui x A 1 for each i, and then show as
in (6.6) that n*:Cl V--> Cl(X - P) is an isomorphism. Since Cl X ~
Cl(X - P), we have also Cl V ~ Cl X.

146
 6 Divisors

(b) We have V s X as the hyperplane section at infinity. Show that the class of
the divisor V in Cl X is equal ton* (class of V.H) where His any hyperplane
of P" not containing V. Thus conclude using (6.5) that there is an exact sequence
0 --> Z --> Cl V --> Cl X --> 0,
where the first arrow sends 1 H V.H, and the second is n* followed by the
restriction to X - P and inclusion in X. (The injectivity of the first arrow
follows from the previous exercise.)
(c) Let S(V) be the homogeneous coordinate ring of V (which is also the affine
coordinate ring of X). Show that S(V) is a unique factorization domain if and
only if(1) Vis projectively normal (Ex. 5.14), and (2) Cl V :::::: Z and is generated
by the class of V.H.
(d) Let (!JP be the local ring of P on X. Show that the natural restriction map in-
duces an isomorphism Cl X--> Cl(Spec (!Jp).
6.4. Let k be a field of characteristic #2. Let f E k[x 1, ... ,xn] be a square{ree
nonconstant polynomial, i.e., in the unique factorization off into irreducible
polynomials, there are no repeated factors. Let A = k[x 1 , •.. ,x",z]/(z 2 -f).
Show that A is an integrally closed ring. [Hint: The quotient field K of A is
just k(xb ... ,xn)[ z]/(z 2 - f). It is a Galois extension of k(x 1 , . . . ,xn) with Galois
group Z/2Z generated by z H -z. If r:t. = g + hz E K, where g,h E k(x 1, ... ,xn),
then the minimal polynomial of r:t. is X 2 - 2gX + (g 2 - h 2f). Now show
that r:t. is integral over k[x 1 , . . . ,xn] if and only if g,h E k[x 1 , . . . ,xnJ. Conclude
that A is the integral closure of k[ x 1 , . . . ,xnJ in K.]
*6.5. Quadric H ypersurfaces. Let char k # 2, and let X be the affine quadric hypersurface
Spec k[x 0 , .•• ,xnJ/(x;i +xi + ... + x?)-cf. (I, Ex. 5.12).
(a) Show that X is normal if r ~ 2 (use (Ex. 6.4) ).
(b) Show by a suitable linear change of coordinates that the equation of X could
be written as x 0 x 1 = x~ + ... + x?. Now imitate the method of (6.5.2) to
show that:
(1) Ifr = 2, then Cl X :::::: Z/2Z;
(2) lfr = 3, then Cl X:::::: Z (use (6.6.1) and (Ex. 6.3) above);
(3) If r ~ 4 then Cl X = 0.
(c) Now let Q be the projective quadric hypersurface in P" defined by the same
equation. Show that:
(1) If r = 2, Cl Q :::::: Z, and the class of a hyperplane section Q.H is twice the
generator;
(2) Ifr = 3, Cl Q :::::: Z E8 Z;
(3) If r ~ 4, Cl Q :::::: Z, generated by Q.H.
(d) Prove Klein's theorem, which says that ifr ~ 4, and if Yis an irreducible sub-
variety of codimension 1 on Q, then there is an irreducible hypersurface
V s P" such that V n Q = Y, with multiplicity one. In other words, Y is a
complete intersection. (First show that for r ~ 4, the homogeneous coordi-
nate ring S(Q) = k[ x 0 , . .. ,xn]/(x6 + ... + x;) is a UFD.)

6.6. Let X be the nonsingular plane cubic curve y 2 z = x 3 - xz 2 of (6.10.2).

(a) Show that three points P,Q,R of X are collinear if and only if P + Q + R = 0
in the group law on X. (Note that the point P 0 = (0,1,0) is the zero element
in the group structure on X.)

147
II Schemes

(b) A point P EX has order 2 in the group law on X if and only if the tangent
line at P passes through P0 .
(c) A point P E X has order 3 in the group law on X if and only if Pis an inflection
point. (An inflection point of a plane curve is a nonsingular point P of the
curve, whose tangent line (1, Ex. 7.3) has intersection multiplicity ~ 3 with
the curve at P.)
(d) Let k = C. Show that the points of X with coordinates in Q form a subgroup
ofthe group X. Can you determine the structure of this subgroup explicitly?

*6.7. Let X be the nodal cubic curve y 2 z = x 3 + x 2 z in P 2 . Imitate (6.11.4) and show
that the group of Cartier divisors of degree 0, CaClo X, is naturally isomorphic
to the multiplicative group Gm.

6.8. (a) Let f: X--+ Y be a morphism of schemes. Show that ff! r--+ f* ff! induces a
homomorphism of Picard groups, f*: Pic Y--+ Pic X.
(b) Iff is a finite morphism of nonsingular curves, show that this homomorphism
corresponds to the homomorphism f*: Cl Y--+ Cl X defined in the text, via
the isomorphisms of (6.16).
(c) If X is a locally factorial integral closed subscheme of Pi:, and iff: X --+ P" is
the inclusion map, then f* on Pic agrees with the homomorphism on divisor
class groups defined in (Ex. 6.2) via the isomorphisms of (6.16).

*6.9. Singular Curves. Here we give another method of calculating the Picard group
of a singular curve. Let X be a projective curve over k, let X be its normalization,
and let n: X--+ X be the projection map (Ex. 3.8). For each point P EX, let
@p be its local ring, and let (jjP be the integral closure of @p. We use a* to denote
the group of units in a ring.
(a) Show there is an exact sequence
0--+ EBPeX @tj(()J--+ Pic X~ Pic X--+ 0.
[Hint: Represent Pic X and Pic X as the groups of Cartier divisors modulo
principal divisors, and use the exact sequence of sheaves on X
0--+ n*@1/@l--+ %*/@l--+ %*/n*@!--+ 0.]
(b) Use (a) to give another proof of the fact that if X is a plane cuspidal cubic
curve, then there is an exact sequence
0--+ G. --+ Pic X --+ Z --+ 0,
and if X is a plane nodal cubic curve, there is an exact sequence
0--+ Gm--+ Pic X--+ Z--+ 0.

6.10. The Grothendieck Group K(X). Let X be a noetherian scheme. We define K(X)
to be the quotient of the free abelian group generated by all the coherent sheaves
on X, by the subgroup generated by all expressions:#'-:#''-:#'", whenever
there is an exact sequence 0--+ :#'' --+ :#' --+ :#'" --+ 0 of coherent sheaves on X.
If:#' is a coherent sheaf, we denote by y(ff) its image in K(X).
(a) If X= A~, then K(X) ~ Z.
(b) If X is any integral scheme, and:#' a coherent sheaf, we define the rank of:#'
to be dimK ~. where ~ is the generic point of X, and K = (()~is the function

148
 7 Projective Morphisms

field of X. Show that the rank function defines a surjective homomorphism

rank:K(X)--> Z.
(c) If Y is a closed subscheme of X, there is an exact sequence
K(Y)--> K(X)--> K(X - Y)--> 0,
where the first map is extension by zero, and the second map is restriction.
[Hint: For exactness in the middle, show that if !F is a coherent sheaf on X,
whose support is contained in Y, then there is a finite filtration !F = !F 0 2
$' 1 2 ... 2 !F" = 0, such that each !F;/!Fi+ 1 is an @y-module. To show
surjectivity on the right, use (Ex. 5.15).]
For further information about K(X), and its applications to the generalized
Riemann-Roch theorem, see Borel-Serre [1], Manin [1], and Appendix A.

*6.11. The Grothendieck Group of a Nonsingular Curve. Let X be a nonsingular curve

over an algebraically closed field k. We will show that K(X) ~ Pic X EB Z, in
several steps.
(a) For any divisor D = 'IniPi on X, let 1/J(D) = 'Iniy(k(PJ) E K(X), where k(PJ
is the skyscraper sheaf k at Pi and 0 elsewhere. If D is an effective divisor, let
(9 v be the structure sheaf of the associated subscheme of codimension 1, and
show that 1/J(D) = y((.Dv). Then use (6.18) to show that for any D, 1/J(D) depends
only on the linear equivalence class of D, so 1/J defines a homomorphism
1/J:Cl X--> K(X).
(b) For any coherent sheaf !F on X, show that there exist locally free sheaves~ 0
and ~ 1 and an exact sequence 0 --> ~ 1 --> ~ 0 --> !F --> 0. Let r0 = rank ~ 0 ,
r1 = rank ~ 1 , and define det !F = (/\' 0 ~ 0 ) ® (/\''~ 1 )- 1 E Pic X. Here 1\ de-
notes the exterior power (Ex. 5.16). Show that det !F is independent of the
resolution chosen, and that it gives a homomorphism det:K(X)--> Pic X.
Finally show that if Dis a divisor, then det(lji(D)) = !i'(D).
(c) If !F is any coherent sheaf of rank r, show that there is a divisor Don X and an
exact sequence 0 --> !i'(D)EB'--> !F --> :Y--> 0, where :Y is a torsion sheaf. Con-
clude that if !F is a sheaf of rank r, then y(!F) - ry(@x) Elm 1/J.
(d) Using the maps 1/J, det, rank, and 1 r-> y(@x) from Z --> K(X), show that K(X) ~
Pic X EB Z.

6.12. Let X be a complete nonsingular curve. Show that there is a unique way to define
the degree of any coherent sheaf on X, deg !FEZ, such that:
(1) If Dis a divisor, deg !i'(D) = deg D;
(2) If !F is a torsion sheaf( meaning a sheaf whose stalk at the generic point is zero),
then deg !F = LPEX length (!Fp); and
(3) If 0 --> !F' --> !F --> !F" --> 0 is an exact sequence, then deg !F = deg $'' +
deg $'".

7 Projective Morphisms

In this section we gather together several topics concerned with morphisms

of a given scheme to projective space. We will show how a morphism of a
scheme X to a projective space is determined by giving an invertible sheaf

149
II Schemes

5I! on X and a set of its global sections. We will give some criteria for this
morphism to be an immersion. Then we study the closely connected topic
of ample invertible sheaves. We also introduce the more classical language
of linear systems, which from the point of view of schemes is hardly more
than another set of terminology for dealing with invertible sheaves and their
global sections. However, the geometric understanding furnished by the
concept of linear system is often very valuable. At the end of this section
we define the Proj of a graded sheaf of algebras over a scheme X, and we
give two important examples, namely the projective bundle P(c&') associated
with a locally free sheaf c&', and the definition of blowing up with respect to
a coherent sheaf of ideals.

M orphisms to pn
Let A be a fixed ring, and consider the projective space PA = Proj A [ x 0 , ... ,xn]
over A. On PA we have the invertible sheaf (1)(1), and the homogeneous
coordinates x 0 , . . . ,xn give rise to global sections x 0 , . . . ,xn E r(PA,(I)(1) ).
One sees easily that the sheaf (1)(1) is generated by the global sections
x 0 , . . . ,xn, i.e., the images of these sections generate the stalk (1)(1)p of the
sheaf (1)(1) as a module over the local ring (l)p, for each point P EPA.
Now let X be any scheme over A, and let <p:X ~ PA be an A-morphism
of X to PA. Then 5I! = <p*((l)(1)) is an invertible sheaf on X, and the global
sections s0 , . . . ,sn, where s; = cp*(x;), s; E r(X,!I!), generate the sheaf 51!.
Conversely, we will see that 5I! and the sections s; determine <p.

Theorem 7.1. Let A be a ring, and let X be a scheme over A.

(a) If <p:X ~PAis an A-morphism, then <p*((l)(1)) is an invertible sheaf
on X, which is generated by the global sections s; = cp*(x;), i = 0,1, ... ,n.
(b) Conversely, if 5I! is an invertible sheaf on X, and if s 0 , . .. ,sn E
r(X,!I!) are global sections which generate 51!, then there exists a unique
A-morphisn1 r;0:X ~ PA such that 5I! ~ <p*((l)(1)) and s; = <p*(x;) under
this isomorphism.
PROOF. Part (a) is clear from the discussion above. To prove (b), suppose
given 5I! and the global sections s0 , . . . ,sn which generate it. For each i,
let X; = {P E Xi(s;)p ¢ mp!l! P}· Then (as we have seen before) X; is an
open subset of X, and since the s; generate 51!, the open sets X; must cover
X. We define a morphism from X; to the standard open set U; = {x; =I= 0}
of PA as follows. Recall that U; ~ Spec A[y 0 , . . . ,yn] where Yi = x)x;,
with Y; = 1 omitted. We define a ring homomorphism A[y 0 , . . . ,yn] ~
r(X;,(I)x,) by sending Yi ~ s)s; and making it A-linear. This makes sense,
because for each P EX;, (s;)p ¢ mp!l! p, and 5I! is locally free of rank 1, so
the quotient s)s; is a well-defined element of F(X;,(I)xJ Now by (Ex. 2.4)
this ring homomorphism determines a morphism of schemes (over A) X;~
U;. Clearly these morphisms glue (cf. Step 3 of proof of (3.3) ), so we obtain
a morphism cp: X ~ PA. It is clear from the construction that cp is an A-

150
 7 Projective Morphisms

morphism, that 5£ ~ CP.*(G(l) ), and that the sections s; correspond to cp*(x;)

under this isomorphism. It is clear that any morphism with these properties
must be the one given by the construction, so cp is unique.

Example 7.1.1 (Automorphisms of P;;J. If llaiJII x

is an invertible (n + 1) (n + 1)
matrix of elements of a field k, then x; = 'f.aiJxj determines an automor-
phism of the polynomial ring k[x 0 , . .. ,xnJ and hence also an automor-
phism of PZ. If I.E k is a nonzero element, then ll!caiJII
determines the same
automorphism of PZ. So we are led to consider the group PGL(n,k) =
GL(n + 1,k)/k*, which acts as a group of automorphisms of PJ:. By con-
sidering the points (1,0, ... ,0), (0,1,0, ... ,0), ... ,(0,0, ... ,1), and (1,1, ... ,1),
one sees easily that this group acts faithfully, i.e., if g E PGL(n,k) induces
the trivial automorphism of P;:, then g is the identity.
Now we will show conversely that every k-automorphism of PZ is an
element of PGL(n,k). This generalizes an earlier result for P~ {I, Ex. 6.6).
So let qJ be a k-automorphism of P;:. We have seen (6.17) that Pic PZ ~ Z
and is generated by 0(1). The automorphism cp induces an automorphism
of Pic pn, so cp*(0(1)) must be a generator of that group, hence isomorphic
to either 0(1) or 0( -1). But 0( -1) has no global sections, so we conclude
that cp*(0(1)) ~ 0(1). Now r(Pn,0(1)) is a k-vector space with basis
x 0 , . . . ,xn, by (5.13). Since qJ is an automorphism, the s; = cp*(x;) must be
another basis of this vector space, so we can write s; = Ia;jxj, where llaiJII
is an invertible matrix of elements of k. Since cp is uniquely determined by
the s; according to the theorem, we see that cp coincides with the auto-
morphism given by llaiJIIas an element of PGL(n,k).

Example 7.1.2. If X is a scheme over A, 5£ an invertible sheaf, and s0 , . .. ,sn

any set of global sections, which do not necessarily generate 5£, we can
always consider the open set U s; X (possibly empty) over which the s; do
generate 5£. Then Ylu and the s;lu give a morphism U -> P~. Such is the
case for example, if we take X = PZ+ 1 , 5£ = 0(1 ), and s; = X;, i = 0, ... ,n
(omitting xn+ tJ. These sections generate everywhere except at the point
{0,0, ... ,0,1) = P 0 . Thus U = pn+t - P 0 , and the corresponding mor-
phism U -> pn is nothing other than the projection from the point P 0 to
pn {1, Ex. 3.14).

Next we give some criteria for a morphism to a projective space to be a

closed immersion.

Proposition 7.2. Let qJ: X -> P~ he a morphism of schemes over A, corre-

sponding to an invertible sheaf!!' on X and sections s0 , . .. ,sn E T(X,!f')
as above. Then qJ is a closed immersion if and only if
(1) each open set X; = Xs, is affine, and
(2) for each i, the map of rings A[y 0 , ... ,yn] -> T(XJ!ix) defined by
yj ~---> s/\ is surjective.

151
II Schemes

PROOF. First suppose cp is a closed immersion. Then X; = X n U; is a

closed subscheme of U;. Therefore X; is affine and the corresponding map
of rings is surjective by (5.10). Conversely, suppose (1) and (2) satisfied.
Then each X; is a closed subscheme of U;. Since in any case X;= cp- 1 (U;),
and the X; cover X, it is clear that X is a closed subscheme ofP~.

With more hypotheses, we can give a more local criterion.

Proposition 7.3. Let k be an algebraically closed field, let X be a projective

scheme over k, and let cp: X -+ P~ be a morphism (over k) corresponding to
2 and s 0 , . . . ,sn E r(X,2) as above. Let V s; F(X,2) be the subspace
spanned by the s;. Then cp is a closed immersion if and only if
(1) elements of V separate points, i.e., for any two distinct closed points
P,Q E X, there is an s E V such that s E mp2P but s ¢' mQ2Q• or vice
versa, and
(2) elements of V separate tangent vectors, i.e.,for each closed point P EX,
the set {s E Vlsp E mp2 P} spans the k-vector space mp2 pjm;2 P·
PROOF. If cp is a closed immersion, we think of X as a closed subscheme of
P~. In this case 2 = llJx(1), and the vector space V s; r(X,llJx(1)) is just
spanned by the images of x 0 , . . . ,xn E r(Pn,(l)(1) ). Given closed points
P # Q in X, there is a hyperplane containing P but not Q. If its equation
is Ia;x; = 0, a; E k, then s = Ia;X; restricted to X has the right property
for (1). For (2), the hyperplanes passing through P give rise to sections
which generate mp2 pjm;2 P· For simplicity suppose that P is the point
(1,0,0, ... ,0). Then on the open affine U 0 ~ Spec k[y 1 , . .. ,yn], 2 istrivial,
P is the point (0, ... ,0), and mpjm; is exactly the vector space spanned by
y 1 , ••• ,yn- We use the hypothesis k algebraically closed to ensure that
every closed point of P~ is of the form (a 0 , . . . ,an) for suitable a; E k, hence
points can be separated by hyperplanes with coefficients in k.
For the converse, let cp: X -+ pn satisfy (1) and (2). Since the elements of
V are pull-backs of sections of (1)(1) on pn, it is clear from (1) that the map
(/) is injective as a map of sets. Since X is projective over k, it is proper over
k (4.9), so the image cp(X) in pn is closed (Ex. 4.4), and(/) is a proper morphism
(4.8e). In particular, (/) is a closed map. But, being a morphism, it is also
continuous, so we see that (/) is a homeomorphism of X onto its image (f)(X)
which is a closed subset of pn_ To show that (/) is a closed immersion, it
remains only to show that the morphism of sheaves llJpn -+ (/J*llJx is sur-
jective. This can be checked on the stalks. So it is sufficient to show, for
each closed point P, that (l)P",P -+ llJx,P is surjective. Both local rings have
the same residue field k, and our hypothesis (2) implies that the image of
the maximal ideal mP",P generates mx,P/mi,P· We also need to use (5.20),
which implies that cp*(l)x is a coherent sheaf on pn, and hence that llJx,P is
a finitely generated llJpn_p-module. Now our result is a consequence of the
following lemma.

152
 7 Projective Morphisms

Lemma 7.4. Let f: A --+ B be a local homomorphism of local noetherian rings,

such that
(1) AlmA --+ BlmB is an isomorphism,
(2) mA --+ mBim~ is surjective, and
(3) B is a finitely generated A -module.
Then f is surjective.

PROOF. Consider the ideal a = mAB of B. We have a ~ mB, and by (2), a

contains a set of generators for mBim~. Hence by Nakayama's lemma for
the local ring B and the B-module mB, we conclude that a = mB. Now
apply Nakayama's lemma to the A-module B. By (3), B is a finitely generated
A-module. The element 1 E B gives a generator for BlmAB = BlmB = AlmA
by (1), so we conclude that 1 also generates B as an A-module, i.e., f is
surjective.

Ample Invertible Sheaves

Now that we have seen that a morphism of a scheme X to a projective space
can be characterized by giving an invertible sheaf on X and a suitable set
of its global sections, we can reduce the study of varieties in projective space
to the study of schemes with certain invertible sheaves and given global
sections. Recall that in §5 we defined a sheaf !l' on X to be very ample
relative toY (where X is a scheme over Y) if there is an immersion i:X--+ P~
for some n such that !l' ~ i*lD(l). In case Y = Spec A, this is the same
thing as saying that !l' admits a set of global sections s0 , . .. ,sn such that
the corresponding morphism X--+ PAis an immersion. We have also seen
(5.17) that if !l' is a very ample invertible sheaf on a projective scheme X
over a noetherian ring A, then for any coherent sheaf.? on X, there is an
integer n0 > 0 such that for all n ;?: n 0 , .? ® !l'n is generated by global
sections. We will use this last property of being generated by global sections
to define the notion of an ample invertible sheaf, which is more general,
and in many ways is more convenient to work with than the notion of very
ample sheaf.

Definition. An invertible sheaf !l' on a noetherian scheme X is said to be

ample if for every coherent sheaf .? on X, there is an integer n0 > 0
(depending on .?) such that for every n ;::::: n0 , the sheaf .? @ !l'n is
generated by its global sections. (Here !l'n = !l' 0 n denotes the n-fold
tensor power of !l' with itself.)

Remark 7.4.1. Note that "ample" is an absolute notion, i.e., it depends only
on the scheme X, whereas "very ample" is a relative notion, depending on
a morphism X --+ Y.

153
II Schemes

Example 7.4.2. If X is affine, then any invertible sheaf is ample, because

every coherent sheaf on an affine scheme is generated by its global sections
(5.16.2).

Remark 7.4.3. Serre's theorem (5.17) asserts that a very ample sheaf !£:' on
a projective scheme X over a noetherian ring A is ample. The converse is
false, but we will see below (7.6) that if!£:' is ample, then some tensor power
!f:'m of!£:' is very ample. Thus "ample" can be viewed as a stable version of
"very ample."

Remark 7.4.4. In Chapter III we will give a characterization of ample in-

vertible sheaves in terms of the vanishing of certain cohomology groups
(III, 5.3).

Proposition 7.5. Let !£:' be an invertible sheaf on a noetherian scheme X.

Then the following conditions are equivalent:
(i) .!:£ is ample;
(ii) yin is ample for all m > 0;
(iii) !f:'m is ample for some m > 0.
PROOF. (i) = (ii) is immediate from the definition of ample; (ii) =(iii) is
trivial. To prove (iii) = (i), assume that !f:'m is ample. Given a coherent
sheaf :F on X, there exists an n 0 > 0 such that :F 0 (.!:Emr is generated by
global sections for all n ;?! n0 . Considering the coherent sheaf :F 0 !£:',
there. exists an 11 1 > 0 such that :#' 0 :E 0 (:Emt is generated by global
sections for alln ;?! 11 1 . Similarly, for each k = 1,2, ... ,m - 1, there is an
nk > 0 such that :F 0 !f:'k 0 (!f:'mr is generated by global sections for all
n ;?! nk. Now if we takeN = m · max{ndi = 0,1, ... ,m - 1}, then :F 0 !f:'n
is generated by global sections for all n ~ N. Hence !£:' is ample.

Theorem 7.6. Let X be a scheme of finite type over a noetherian ring A, and
let .!:£ be an invertible sheaf on X. Then !£:' is ample if and only if !f:'m is
very ample over Spec A for some m > 0.
PROOF. First suppose !f:'m is very ample for some m > 0. Then there is an
immersion i:X-+ P~ such that !f:'m ~ i*(0(1)). Let X be the closure of X
in P~. Then X is a projective scheme over A, so by (5.17), 0g{l) is ample
on X. Now given any coherent sheaf :F on X, it extends by (Ex. 5.15) to a
coherent sheaf~ on X. If~ 0 0g(/) is generated by global sections, then
a fortiori :F 0 0 xU) is also generated by global sections. Thus we see that
!f:'m is ample on X, and so by (7.5), !£:'is also ample on X.
For the converse, suppose that !£:' is ample on X. Given any P EX, let
U be an open affine neighborhood of P such that !!:'lu is free on U. Let Y
be the closed set X - U, and let 5 y be its sheaf of ideals with the reduced
induced scheme structure. Then 5 r is a coherent sheaf on X, so for some

154
 7 Projective Morphisms

n > 0, § y ® y;m is generated by global sections. In particular, there is a

sections E r(X,§y ®!.en) such that Sp ¢ mp(§y ® 5lm)p. Now §y ® ft'"
is a subsheaf of ft'", so we can think of s as an element of r(X,ft'"). If x.
is the open set {Q E XlsQ ¢ mQft'Q}, then it follows from our choice of s that
P EX. and that X. s;; U. Now U is affine, and fL'Iu is trivial, so s induces
an element! E r(U,@u), and then X.= U1 is also affine.
Thus we have shown that for any point P EX, there is an n > 0 and a
section s Er(X,ft'") such that P Ex. and x. is affine. Since X is quasi-
compact, we can cover X by a finite number of such open affines, corre-
sponding to sections s; E r(X,fi'"'). Replacing each s; by a suitable power
s~ E r(X,ft'kn,), which doesn't change x.,, we may assume that all n; are
equal to one n. Finally, since ft'n is also ample, and since we are only trying
to show that some power of 5l' is very ample, we may replace 5l' by ft'".
Thus we may assume now that we have global sections s 1, ... ,sk E r(X,ft')
such that each X; = x., is affine, and the X; cover X.
Now for each i, let B; = r(X;,@x,). Since X is a scheme of finite type over
A, each B; is a finitely generated A-algebra (Ex. 3.3). So let {biijj = 1, ... ,k;}
be a set of generators forB; as an A-algebra. By (5.14), for each i,j, there is
an integer n such that s?bij extends to a global section cii E r(X,ft'"). We
can take one n large enough to work for all i,j. Now we take the invertible
sheaf ft'" on X, and the sections {s?li = 1, ... ,k} and {ciili = 1, ... ,k;
j = 1, ... ,k;} and use all these sections to define a morphism (over A)
cp:X ~ P~ as in (7.1) above. Since X is covered by the X;, the sections s?
already generate the sheaf 2'", so this is indeed a morphism.
Let {x;ji = 1, ... ,k} and {x;ili = 1, ... ,k;j = 1, ... ,ki} be the homo-
geneous coordinates of P~ corresponding to the sections of ft'n mentioned
above. For each i = 1, ... ,k, let U; s;; P~ be the open subset xi =1- 0. Then
cp ~ 1 (U;) = Xi, and the corresponding map of affine rings

is surjective, because yij ~ cufs? = b;i, and we chose the bii so as to generate
B; as an A-algebra. Thus X; is mapped onto a closed subscheme of U;.
It follows that cp gives an isomorphism of X with a closed subscheme of
U~= 1 U; s;; P~, so cp is an immersion. Hence ft'" is very ample relative to
Spec A, as required.

Example 7.6.1. Let X = P~, where k is a field. Then (1)(1) is very ample by
definition. For any d > 0, @(d) corresponds to the d-uple embedding
(Ex. 5.13), so @(d) is also very ample. Hence @(d) is ample for all d > 0.
On the other hand, since the sheaf @(I) has no global sections for l < 0, one
sees easily that the sheaves @(/) for l ~ 0 cannot be ample. So on P~, we
have@(/) is ample~ very ample~ l > 0.

Example 7.6.2. Let Q be the nonsingular quadric surface xy = zw in P~

over a field k. We have seen (6.6.1) that Pic Q ~ Z Ef> Z, and so we speak

155
II Schemes

of the type (a,b), a,b E Z, of an invertible sheaf. Now Q ~ P 1 x P 1 . If

a,b > 0, then we consider an a-uple embedding P 1 ~ P"' and a b-uple
embedding P 1 -t P" 2 • Taking their product, and following with a Segre
embedding, we obtain a closed immersion

which corresponds to an invertible sheaf of type (a,b) on Q. Thus for any

a,b > 0, the corresponding invertible sheaf is very ample, and hence ample.
On the other hand, if 2? is of type (a,b) with either a < 0 orb < 0, then by
restricting to a fibre of the product P 1 x P\ one sees that 2? is not gen-
erated by global sections. Hence if a ~ 0 or b ~ 0, 2? cannot be ample.
So on Q, an invertible sheaf 2? of type (a,b) is ample-= very ample-= a,b > 0.

Example 7.6.3. Let X be the nonsingular cubic curve y 2 z = x 3 - xz 2 in

Pt, which was studied in (6.10.2). Let 2? be the invertible sheaf fi?(P 0 ). Then
2? is ample, because 2?(3P 0 ) ~ lDx(1) is very ample. On the other hand,
2? is not very ample, because fi?(P 0 ) is not generated by global sections. If
it were, then P 0 would be linearly equivalent to some other point Q E X,
which is impossible, since X is not rational (6.10.1). This shows that an ample
sheaf need not be very ample.

Example 7.6.4. We will see later (IV, 3.3) that if Dis a divisor on a complete
nonsingular curve X, then fi?(D) is ample if and only if deg D > 0. This is
a consequence of the Riemann-Roch theorem.

Linear Systems
We will see in a minute how global sections of an invertible sheaf correspond
to effective divisors on a variety. Thus giving an invertible sheaf and a set
of its global sections is the same as giving a certain set of effective divisors,
all linearly equivalent to each other. This leads to the notion of linear
system, which is the historically older notion. For simplicity, we will employ
this terminology only when dealing with nonsingular projective varieties
over an algebraically closed field. Over more general schemes the geo-
metrical intuition associated with the concept of linear system may lead one
astray, so it is safer to deal with invertible sheaves and their global sections
in that case.
So let X be a nonsingular projective variety over an algebraically closed
field k. In this case the notions of Weil divisor and Cartier divisor are
equivalent (6.11). Furthermore, we have a one-to-one correspondence be-
tween linear equivalence classes of divisors and isomorphism classes of
invertible sheaves (6.15). Another useful fact in this situation is that for any
invertible sheaf 2? on X, the global sections r(X,fl?) form a finite-dimen-
sional k-vector space (5.19).

156
 7 Projective Morphisms

Let !£ be an invertible sheaf on X, and let s E T(X,!£) be a nonzero

section of!£. We define an effective divisor D = (s) 0 , the divisor of zeros of
s, as follows. Over any open set U <::; X where!£ is trivial, let <p: !flu ..:::. (!Ju
be an isomorphism. Then <p(s) E r(U,(!Ju). As U ranges over a covering of
X, the collection {U,<p(s)} determines an effective Cartier divisor D on X.
Indeed, <p is determined up to multiplication by an element ofT( U,(!Jt), so
we get a well-defined Cartier divisor.

Proposition 7.7. Let X be a nonsingular projective variety over the algebrai-

cally closed field k. Let D0 be a divisor on X and let !£ ~ !£(D 0 ) be the'
corresponding invertible sheaf Then:
(a) for each nonzero s E T(X,!£), the divisor of zeros (s) 0 is an effective
divisor linearly equivalent to D0 ;
(b) every effective divisor linearly equivalent to D0 is (s) 0 for some
s E T(X,!£); and
(c) two sections s,s' E T(X,!£) have the same divisor of zeros if and only
if there is a A E k* such that s' = l.s.
PROOF.
(a) We may identify!£ with the subsheaf !£(D 0 ) of%. Then s corresponds
to a rational function f E K. If D 0 is locally defined as a Cartier divisor by
{U i,j;} with J; E K*, then !£(D 0 ) is locally generated by fi- 1 , so we get a
local isomorphism <p: !£(D 0 )-+ (!) by multiplying by fi· So D = (s) 0 is
locally defined by fJ Thus D = D0 + (f), showing that D ~ D0 .
(b) If D > 0 and D = D0 + (f), then (f) ~ - D0 . Thus f gives a global
section of !£(D 0 ) whose divisor of zeros is D.
(c) Still using the same construction, if (s) 0 = (s') 0 , then s and s' corre-
spond to rational functions f,f' E K such that (f/f') = 0. Therefore flf' E
r(X,(!J~). But since X is a projective variety over k algebraically closed,
r(X,(!Jx) = k, and so flf' E k* (1, 3.4).

Definition. A complete linear system on a nonsingular projective variety is

defined as the set (maybe empty) of all effective divisors linearly eqhiva-
lent to some given divisor D 0 . It is denoted by IDol·
We see from the proposition that the set IDol is in one-to-one corre-
spondence with the set (T(X,!£) - {0})/k*. This gives IDol a structure of
the set of closed points of a projective space over k.

Definition. A linear system b on X is a subset of a complete linear system

IDol which is a linear subspace for the projective space structure of IDol·
Thus b corresponds to a sub-vector space V <::; r(X,!£), where V =
{s E T(X,!£)1(s) 0 E b} u {0}. The dimension of the linear system b is its
dimension as a linear projective variety. Hence dim b = dim V - 1.
(Note these dimensions are finite because r(X,!£) is a finite-dimensional
vector space.)

157
II Schemes

Definition. A point P E X is a base point of a linear system b if P E Supp D

for all DE b. Here Supp D means the union of the prime divisors of D.

Lemma 7.8. Let b be a linear system on X corresponding to the subspace

V <:;::; T(X,!f). Then a point P EX is a base point of b if and only if sp E
mp!t? P for all s E V. In particular, b is base-point-free if and only if !t? is
generated by the global sections in V.
PROOF. This follows immediately from the fact that for any s E T(X,!f), the
support of the divisor of zeros (s) 0 is the complement of the open set X 8 •

Remark 7.8.1. We can rephrase (7.1) in terms of linear systems as follows:

to give a morphism from X to PZ it is equivalent to give a linear system b
without base points on X, and a set of elements s0 , . .. ,s" E V, which span
the vector space V. Often we will simply talk about the morphism to pro-
jective space determined by a linear system without base points b. In this
case we understand that s0 , . .. ,s" should be chosen as a basis of V. If we
chose a different basis, the corresponding morphism of X ~ P" would only
differ by an automorphism of P".

Remark 7.8.2. We can rephrase (7.3) in terms of linear systems as follows:

Let <p: X ~ P" be a morphism corresponding to the linear system (without
base points) b. Then <pis a closed immersion if and only if
(1) b separates points, i.e., for any two distinct closed points P,Q EX, there
is a D E b such that P E Supp D and Q ¢ Supp D, and
(2) b separates tangent vectors, i.e., given a closed point P E X and a tangent
vector t E Tp(X) = (mp/m~)', there is aD E b such that P E Supp D, but
t ¢ Tp(D). Here we think of D as a locally principal closed subscheme,
in which case the Zariski tangent space Tp(D) = (mp.v/m~.vY is naturally
a subspace of Tp(X).
The terminology of "separating points" and "separating tangent vectors"
is perhaps somewhat explained by this geometrical interpretation.

Definition. Let i: Y ~ X be a closed immersion of nonsingular projective

varieties over k. If b is a linear system on X, we define the trace of b on
Y, denoted bjy, as follows. The linear system b corresponds to an in-
vertible sheaf !f on X, and a sub-vector space V <:;::; T(X,!f). We take
the invertible sheaf i* !f = !f ® @yon Y, and we let W <:;::; T(Y,i* !f) be
the image of V under the natural map T(X,!f) ~ T(Y,i* !f). Then i* !f
and W define the linear system bjy.
One can also describe bjy geometrically as follows: it consists of all
divisors D. Y (defined as in (6.6.2) ), where DEb is a divisor whose support
does not contain Y.
Note that even ifb is a complete linear system, bjy may not be complete.

158
 7 Projective Morphisms

Example 7.8.3. If X = pn, then the set of all effective divisors of degree
d > 0 is a complete linear system of dimension (n~d) - 1. Indeed, it corre-
sponds to the invertible sheaf cP(d), whose global sections consist exactly of
the space of all homogeneous polynomials in x 0 , . .. ,xn of degree d. This is
a vector space of dimension (n~d), so the dimension of the complete linear
system is one less.

Example 7.8.4. We can rephrase (Ex. 5.14d) in terms of linear systems as

follows: a nonsingular projective variety X ~ Pi: is projectively normal if
and only if for every d > 0, the trace on X of the linear system of all divisors
of degree d on pn, is a complete linear system. By slight abuse of language,
we say that "the linear system on X, cut out by the hypersurfaces of degree
din pn, is complete."

Example 7.8.5. Recall that the twisted cubic curve in P 3 was defined by the
parametric equations x 0 = t 3 , x 1 = t 2 u, x 2 = tu 2 , x 3 = u 3 . In other words,
it is just the 3-uple embedding ofP 1 in P 3 (I, Ex. 2.9, Ex. 2.12). We will now
show that any nonsingular curve X in P 3 , of degree 3, which is not contained
in any P 2 , and which is abstractly isomorphic to P 1 , can be obtained from
the given twisted cubic curve by an automorphism of P 3 . So we will refer
to any such curve as a twisted cubic curve.
Let X be such a curve. The embedding of X in P 3 is determined by the
linear system b of hyperplane sections of X (7.1). This is a linear system on
X of dimension 3, because the planes in P 3 form a linear system of dimen-
sion 3, and by hypothesis X is not contained in any plane, so the map
r{P 3 ,cP(1)) --+ r(X,i*cP(1)) is injective. On the other hand, b is a linear
system of degree 3, since X is a curve of degree 3. By the degree of a linear
system on a complete nonsingular curve, we mean the degree of any of its
divisors, which is independent of the divisor chosen (6.10). Now thinking
of X as P 1 , the linear system b must correspond to a 4-dimensional subspace
V <;; r(P\cP(3) ). But r(P 1 ,@(3)) itself has dimension 4, so V = r(P 1 ,@(3))
and b is a complete linear system. Since the embedding is determined by
the linear system and the choice of basis of V by (7.1), we conclude that X
is the same as the 3-uple embedding of P 1 , except for the choice of basis of
V. This shows that there is an automorphism of P 3 sending the given twisted
cubic curve to X. (See (IV, Ex. 3.4) for generalization.)

Example 7.8.6. We define a nonsingular rational quartic curve in P 3 to be a

nonsingular curve X in P 3 , of degree 4, not contained in any P 2 , and which
is abstractly isomorphic to P 1. In this case we will see that two such curves
need not be obtainable one from the other by an automorphism of P 3 . To
give a morphism ofP 1 to P 3 whose image has degree 4 and is not contained
in any P 2 , we need a 4-dimensional subspace V <;; r(P\cP(4)). This latter
vector space has dimension 5. So if we choose two different subspaces V,V',

159
II Schemes

the corresponding curves in P 3 may not be related by an automorphism of

P 3 . To be sure the image is nonsingular, we use the criterion of (7.3). Thus
for example, one sees easily that the subspaces V = (t 4 ,t 3 u,tu 3 ,u 4 ) and V' =
(t 4 ,t 3 u + at 2 u2 ,tu 3 ,u 4 ) for a E k* give nonsingular rational quartic curves in
P 3 which are not equivalent by an automorphism of P 3 .

Proj, P(t&'), and Blowing Up

Earlier we have defined the Proj of a graded ring. Now we introduce a
relative version of this construction, which is the Proj of a sheaf of graded
algebras Y' over a scheme X. This construction is useful in particular
because it allows us to construct the projective space bundle associated to
a locally free sheaf 8, and it allows us to give a definition of blowing up
with respect to an arbitrary sheaf of ideals. This generalizes the notion of
blowing up a point introduced in (1, §4).
For simplicity, we will always impose the following conditions on a
scheme X and a sheaf of graded algebras Y' before we define a Proj :

(t) X is a noetherian scheme, Y' is a quasi-coherent sheaf of {!:x-modules,

which has a structure of a sheaf of graded Gx-algebras. Thus //' ~
ffid" 0.~, where .~ is the homogeneous part of degree d. We assume
furthermore that :1'0 = (!:'x, that Y'1 is a coherent ex-module, and that
Y' is locally generated by :1'1 as an Gx-algebra. (It follows that ~ is
coherent for all d ~ 0.)

Construction. Let X be a scheme and Y' a sheaf of graded (!) x-algebras

satisfying (t). For each open affine subset U = Spec A of X, let Y'( U) be
the graded A-algebra F( U,Y'Iul· Then we consider Proj 9"( U) and its
natural morphism nu: Proj Y'( U) ---> U. Iff E A, and U 1 = Spec A 1 , then
since//' is quasi-coherent, we see that Proj Y'( U1 ) ~ n[} 1( U1 ). It follows
that if U,V are two open affine subsets of X, then n[J 1( U n V) is naturally
isomorphic to nv
1(U n V)-here we leave some technical details to the

reader. These isomorphisms allow us to glue the schemes Proj Y'( U)

together (Ex. 2.12). Thus we obtain a scheme Proj Y' together with a mor-
phism n: Proj Y' ---> X such that for each open affine U c;; X, n- 1 ( U) ~
Proj Y'( U). Furthermore the invertible sheaves 0(1) on each Proj Y'( U)
are compatible under this construction (5.12c), so they glue together to
give an invertible sheaf (1)(1) on Proj //', canonically determined by this
construction.

Thus to any X, /1) satisfying (t), we have constructed the scheme Proj Y),
the morphism n:Proj //'--->X. and the invertible sheaf (1:(1) on Proj .'f.
Everything we have said about the Proj of a graded ringS can be extended
to this relative situation. We will not attempt to do this exhaustively, but
will only mention certain aspects of the new situation.

160
 7 Projective Morphisms

Example 7.8.7. If Y' is the polynomial algebra Y' = (9x[T 0 , . . . ,Tn], then
Proj Y' is just the relative projective space P~ with its twisting sheaf @(1)
defined earlier (§5).

Caution 7.8.8. In general, (9(1) may not be very ample on Proj Y' relative to
X. See (7.10) and (Ex. 7.14).

Lemma 7.9. Let Y' be a sheaf of graded algebras on a scheme X satisfying

(t). Let 2 be an invertible sheaf on X, and define a new sheaf of graded
algebras Y'' = Y'*2 by Y'~ = .5f'a 0 2a for each d ~ 0. Then Y'' also
satisfies (t), and there is a natural isomorphism cp: P' = Proj Y'' .; P =
Proj Y', commuting with the projections n and n' to X, and having the
property that

PROOF. Let 8:(9u ~ 2lu be a local isomorphism of (9u with 2lu over a
small open affine subset U of X. Then 8 induces an isomorphism of graded
rings Y'(U) ~ Y''(U) and hence an isomorphism()*: Proj Y''(U) ~ Proj Y'(U).
21
If 8 1 : (9 u ~ u is a different local isomorphism, then 8 and 8 1 differ by an
element f E r(U,(9tJ), and the corresponding isomorphism Y'(U) ~ Y''(U)
differs by an automorphism 1/1 of Y'(U) which consists of multiplying by fa
in degree d. This does not affect the set of homogeneous prime ideals in
Y'( U), and furthermore, since the structure sheaf of Proj Y'( U) is formed
by elements of degree zero in various localizations of Y'(U), the automor-
phism 1/1 of Y'(U) induces the identity automorphism of Proj Y'(U). In other
words, the isomorphism 8* is independent of the choice of 8. So these local
isomorphisms 8* glue together to give a natural isomorphism cp:Proj Y'' ~
Proj Y', commuting with nand n'. When we form the sheaf @(1), however,
the automorphism ljJ of Y'(U) induces multiplication by f in @(1). Thus
(9r(1) looks like @p(1) modified by the transition functions of 2. Stated
precisely, this says @p.(1) ~ cp*@p(1) 0 n'* 2.

Proposition 7.10. Let X,Y' satisfy (t), let P = Proj Y', with projection
n:P --+X and invertible sheaf @p(1) constructed above. Then:
(a) n is a proper morphism. In particular, it is separated and of finite
type;
(b) if X admits an ample invertible sheaf 2, then n is a projective
morphism, and we can take @p(1) 0 n* 2" to be a very ample invertible
sheaf on P over X, for suitable n > 0.

PROOF.
(a) For each open affine U ~ X, the morphism nu: Proj Y'(U) --+ U is a
projective morphism (4.8.1), hence proper (4.9). But the condition for a
morphism to be proper is local on the base (4.8f), son is proper.

161
II Schemes

(b) Let 2 be an ample invertible sheaf on X. Then for some n > 0,

9"1 @ 2" is generated by global sections. Since X is noetherian and 9"1 @
2" is coherent, we can find a finite number of global sections which generate
it, in other words we can find a surjective morphism of sheaves m~+ 1 ~
9"1 @ 2" for some N. This allows us to define a surjective map of sheaves
of graded lDx-algebras lDx[T 0 , . . . ,TN] ~ 9"*2", which gives rise to a
closed immersion Proj 9"*2" 4 Proj lDx[T0 , . . . ,TN] = P~ (Ex. 3.12).
But Proj 9"*2" ~ Proj 9" by (7.9), and the very ample invertible sheaf
induced by this embedding is just lDp(1)@ n* 2".

Definition. Let X be a noetherian scheme, and lett! be a locally free coherent

sheaf on X. We define the associated projective space bundle P(t!) as
follows. Let 9" = S(t!) be the symmetric algebra oft!, 9" = ffia>o Sa(t!)
(Ex. 5.16). Then Y' is a sheaf of graded lDx-algebras satisfying (t), and we
define P(t!) = Proj 9". As such, it comes with a projection morphism
n:P(t!) ~X, and an invertible sheaf lD(1).
Note that if t! is free of rank n + 1 over an open set U, then n- 1 (U) ~
P~, so P(t!) is a "relative projective space" over X.

Proposition 7.11. Let X,t!,P(t!) be as in the definition. Then:

(a) if rank t! ?: 2, there is a canonical isomorphism of graded lDx-
algebras 9" ~ ffi 1e z n*( lD(l) ), with the grading on the right hand side given
by l. In particular,for l < 0, n*(lD(l)) = O;for l = 0, n*((r)P(CJ) = lDx, and
for l = 1, n*(lD(1)) = t!;
(b) there is a natural surjective morphism n*t! ~ lD(l).

PROOF.
(a) is just a relative version of (5.13), and follows immediately from it.
(b) is a relative version of the fact that (9(1) on P" is generated by the
global sections x 0 , . . . ,xn (5.16.2).

Proposition 7.12. Let X,t!,P(t!) be as above. Let g: Y ~X be any morphism.

Then to give a morphism of Y to P(t!) over X, it is equivalent to give an
invertible sheaf 2 on Yanda surjective map of sheaves on Y, g*t! ~ 2.
PROOF. This is a local version of (7.1). First note that iff: Y ~ P(t!) is a
morphism over X, then the surjective map n*t! ~ lD(1) on P(t!) pulls back
to give a surjective map g*t! = f*n*t! ~ f*lD(1), so we take 2 = f*lD(1).
Conversely, given an invertible sheaf 2 on Y, and a surjective morphism
g*t! ~ 2, I claim there is a unique morphism f: Y ~ P(t!) over X, such
that 2 ~ f*lD(1), and the map g*t! ~ 2 is obtained from n*t! ~ lD(1) by
applying f*. In view of the claimed uniqueness off, it is sufficient to verify
this statement locally on X. Taking open affine subsets U = Spec A of X
which are small enough so that t!lu is free, the statement reduces to (7.1).
Indeed, if t! ~ m~+ 1 , then to give a surjective morphism g*t! ~ 2 is the
same as giving n + 1 global sections of 2 which generate.

162
 7 Projective Morphisms

Note. We refer to the exercises for further properties of P( t&") and for the
general notion of projective space bundle over a scheme X. Cf. (Ex. 5.18)
for the notion of a vector bundle associated to a locally free sheaf.

Now we come to the generalized notion of blowing up. In (1, §4) we

defined the blowing-up of a variety with respect to a point. Now we will
define the blowing-up of a noetherian scheme with respect to any closed
subscheme. Since a closed subscheme corresponds to a coherent sheaf of
ideals (5.9), we may as well speak of blowing up a coherent sheaf of ideals.

Definition. Let X be a noetherian scheme, and let § be a coherent sheaf of

ideals on X. Consider the sheaf of graded algebras !/ = ffid 0
~ §d, where
§d is the dth power of the ideal §, and we set § 0 = mx. Then X,!/
clearly satisfy (t), so we can consider X = Proj !/. We define X to be
the blowing-up of X with respect to the coherent sheaf of ideals §. If Y
is the closed subscheme of X corresponding to §, then we also call X
the blowing-up of X along Y, or with center Y

Example 7.12.1. If X is Ak and P EX is the origin, then the blowing-up of

P just defined is isomorphic to the one defined in (1, §4). Indeed, in this
case X = Spec A, where A = k[xb ... ,xn], and P corresponds to the ideal
I = (xi> ... ,xn). So X = Proj S, where S = EBDo
Id. We can define a
surjective map of graded rings cp:A[y 1, . . . ,yn] --+ S by sending Y; to the
element X; E I considered as an element of S in degree 1. Thus X is isomor-
phic to a closed subscheme of Proj A[yb ... ,ynJ = p~- 1 . It is defined by
the homogeneous polynomials in the Y; which generate the kernel of q>, and
one sees easily that {x;yi- XiYili,j = 1, ... ,n} will do.

Definition. Let f: X --+ Y be a morphism of schemes, and let § be a

~ (!) r
sheaf of ideals on Y We define the inverse image ideal sheaf§' as~ (!)x
follows. First consider f as a continuous map of topological spaces
X --+ Y and let f- 1 § be the inverse image of the sheaf§, as defined in
§1. Then f- 1 § is a sheaf of ideals in the sheaf of rings f- 1 (!)y on the
topological space X. Now there is a natural homomorphism of sheaves
of rings on X, f- 1 (!)r --+ mx, so we define§' to be the ideal sheaf in mx
generated by the image off- 1 §. We will denote §' by f- 1 § · (!) x or
simply§· mx, if no confusion seems likely to result.

Caution 7.12.2. If we consider § as a sheaf of my-modules, then in §5 we

have defined the inverse image f* §as a sheaf of mx-modules. It may happen
that f* § =1= f- 1 § · mx. The reason is that f* §is defined as
f-1§ @J-li!Jy (!)X·
Since the tensor product functor is not in general left exact, f* § may not
be a subsheaf of mx. However, there is a natural map f* § --+ (!)x coming

163
II Schemes

from the inclusion 5 c. @y, and f- 1 5 · 0x is just the image off* 5 under
this map.

Proposition 7.13. Let X be a noetherian scheme, 5 a coherent sheaf of ideals,

and let n: X --+ X be the blowing-up off Then:
(a) the inverse image ideal sheaf .J = n:- 1 5 · (!)x is an invertible sheaf
on X.
(b) if Y is the closed subscheme corresponding to 5, and if U = X - Y,
then n:n:- 1 (U)--+ U is an isomorphism.
PROOF.
(a) Since X is defined as Proj !:/', where !:1' = EBH 0 5d, it comes
equipped with a natural invertible sheaf 0(1). For any open affine U <;: X,
this sheaf (!)(1) on Proj !:/'( U) is the sheaf associated to the graded !:/'( U)-
module !:f'(U)(l) = EBHo 5d+ 1 (U). But this is clearly equal to the ideal
5 · !:f'(U) generated by 5 in !:f'(U), so we see that the inverse image ideal
sheaf .} = n- 1 5 · (!) x is in fact equal to (!) :x( 1). Hence it is an invertible
sheaf.
(b) If U = X - Y, then 5lu ;:; 0u, so n- 1 U = Proj 0u[T] = U.

Proposition 7.14 (Universal Property of Blowing Up). Let X be a noetherian

scheme, 5 a coherent sheaf of ideals, and n: X --+ X the blowing-up with
respect to f If f:Z--+ X is any morphism such that f- 1 5 · 0z is an in-
vertible sheaf of ideals on Z, then there exists a unique morphism g: Z --+ X
factoring f

z ___{/_ __ -+ x

PROOF. In view of the asserted uniqueness of g, the question is local on X.

So we may assume that X = Spec A is affine, A is noetherian, and that 5
corresponds to an ideal I <;: A. Then X = Proj S, where S = EBd;.o Id. Let
a 0 , . . . ,an E I be a set of generators for the ideal I. Then we can define a
surjective map of graded rings <p: A[ x 0 , . . . ,xn] --+ S by sending X; to a; E I,
considered as an element of degree one in S. This homomorphism gives rise
to a closed immersion X c. ·p~· The kernel of <p is the homogeneous ideal
in A[x 0 , . . . ,xn] generated by all homogeneous polynomials. F(x 0 , . .. ,xn)
such that F(a 0 , .•• ,an) = 0 in A.
Now let f: Z --+ X be a morphism such that the inverse image ideal sheaf
f- 15 · (!)z is an invertible sheaf Yon Z. Since I is generated by a0 , . . . ,an,
the inverse images of these elements, considered as global sections of 5, give
global sections s,, . ,s" of Y which generate. Then by (7.1) there is a unique

164
 7 Projective Morphisms

morphismg:Z--+ P~withthepropertythat2' ~ g*@(l)andthats; = g- 1 x;

under this isomorphism. Now I claim that g factors through the closed
subscheme X of P~. This follows easily from the fact that if F(x 0 , . . . ,xn)
is a homogeneous element of degree d of ker cp, where ker cp is the homoge-
neous ideal described above which determines X, then F(a 0 , . .. ,an) = 0 in A
and so F(s 0 , . . . ,sn) = 0 in r(Z,2'd).
Thus we have constructed a morphism g:Z --+X factoring f For any
such morphism, we must necessarily have f- 1 § · (!) 2 = g- 1 (n- 1 § · @g) · (!) 2
which is just g- 1 (@g(1)) · @2 . Therefore we have a surjective map g*@g(l) --+
f- 1 § · (!) 2 = 2'. Now a surjective map of invertible sheaves on a locally
ringed space is necessarily an isomorphism(Ex. 7.1), so we have g*@g(1) ~ 2'.
Clearly the sections s; of 2' must be the pull-backs of the sections X; of @(1)
on P~. Hence the uniqueness of g under our conditions follows from the
uniqueness assertion of (7.1).

Corollary 7.15. Let f: Y--+ X be a morphism of noetherian schemes, and let§

be a coherent sheaf of ideals on X. Let X be the blowing-up of 5, and let Y
be the blowing-up of the inverse image ideal sheaf,$ = f- 1 §· CVy on Y.
Then there is a unique morphism .f: Y --+ X

y- l
-----------> X-

j j
y --"---f_ _____. X
making a commutative diagram as shown. Moreover, if f is a closed im-
mersion, so is J
PROOF. The existence and uniqueness of J follow immediately from the
J
proposition. To show that is a closed immersion iff is, we go back to the
definition of blowing up. X = Proj // where Y = ffid, 0 §d, and Y =
Proj Y', where Y' = ffiDo Jd. Since Y is a closed subscheme of X, we
can consider Y' as a sheaf of graded algebras on X. Then there is a natural
surjective homomorphism of graded rings Y --+ Y', which gives rise to the
closed immersion J

Definition. In the situation of (7.15), if Y is a closed subscheme of X, we call

the closed subscheme Yof X the strict transform of Y under the blowing-up
n:X--+ X.

Example 7.15.1. If Y is a closed subvariety of X = A~ passing through the

origin P, then the strict transform Y of Yin X is a closed subvariety. Hence,
provided Y is not just P itself, we can recover Y as the closure of n- 1( Y - P),

165
II Schemes

where n:n- 1 (X- P) ~X- Pis the isomorphism of(7.13b). This shows
that our new definition of blowing up coincides with the one given in (I, §4)
for any closed subvariety of AZ. In particular, this shows that blowing up as
defined earlier is intrinsic.

Now we will study blowing up in the special case that X is a variety.

Recall (§4) that a variety is defined to be an integral separated scheme of
finite type over an algebraically closed field k.

Proposition 7.16. Let X be a variety over k, let J £ {!Jx be a nonzero coherent

sheaf of ideals on X, and let n: X ~ X be the blowing-up with respect to J.
Then:
(a) X is also a variety;
(b) n is a birational, proper, surjective morphism;
(c) if X is quasi-projective (respectively, projective) over k, then X is
also, and n is a projective morphism.
PROOF. First of all, since X is integral, the sheaf Y = EBd;;,o Jd is a sheaf of
integral domains on X, so X is also integral. Next, we have already seen that
n is proper (7.10). In particular, n is separated and of finite type, so it follows
that X is also separated and of finite type, i.e., X is a variety. Now since
J =1= 0, the corresponding closed subscheme Y is not all of X, and so the
open set U = X - Y is nonempty. Since n induces an isomorphism from
n- 1 U to U (7.13), we see that n is birational. Since n is proper, it is a closed
map, so the image n(X) is a closed set containing U, which must be all of X
since X is irreducible. Thus n is surjective. Finally, if X is quasi-projective
(respectively, projective), then X admits an ample invertible sheaf (7.6), so by
(7.10b) n is a projective morphism. It follows that X is also quasi-projective
(respectively, projective) (Ex. 4.9).

Theorem 7.17. Let X be a quasi-projective variety over k. If Z is another

variety and f:Z ~X is any birational projective morphism, then there
exists a coherent sheaf of ideals J on X such that Z is isomorphic to the
blowing-up X of X with respect to J, and f corresponds ton: X ~ X under
this isomorphism.
PROOF. The proof is somewhat difficult, so we divide it into steps.
Step 1. Since f is assumed to be a projective morphism, there exists a
closed immersion i:Z ~ P~ for some n.

z c ) p~

~j X

166
 7 Projective Morphisms

Let !£ be the invertible sheaf i*(l)(1) on Z. Now we consider the sheaf of

graded (l)x-algebras !/ = (l)x E9 ffid~ 1 f*!Ed. Each f*!Ed is a coherent sheaf
on X, by (5.20), so!/ is quasi-coherent. However,!/ may not be generated by
!/ 1 as an (I) x-algebra.
Step 2. For any integer e > 0, let y<eJ = EBd ~ 0 Y<;>, where Y<;> = ~e
(cf. Ex. 5.13). I claim that for e sufficiently large, y<e> is generated as an
(l)x-algebra by y~>. Since X is quasi-compact, this question is local on X,
so we may assume X = Spec A is affine, where A is a finitely generated
k-algebra. Then Z is a closed subscheme of P~, and !/ corresponds to the
graded A-algebraS= A E9 ffid~ 1 T(Z,(I) 2 (d)). Let T = A[x 0 , . . . ,xn]/lz,
where I z is a homogeneous ideal defining Z. Then, using the technique of
(Ex. 5.9, Ex. 5.14), one can show that.the A-algebras S,T agree in all large
enough degrees (details left to reader). But Tis generated as an A-algebra
by T 1 , so y<e> is generated by T~>, and this is the same as s<eJ fore sufficiently
large.
Step 3. Now let us replace our original embedding i:Z -+ P~ by i followed
by an e-uple embedding fore sufficiently large. This has the effect of replacing
!£ by fEe and !/ by y<eJ (Ex. 5.13). Thus we may now assume that !/ is
generated by !/1 as an (l)x-algebra. Note also by construction that Z ~
Proj !/ (cf. (5.16) ). So at least we have Z isomorphic to Proj of something.
If !/1 = f*!£ were a sheaf of ideals in (l)x we would be done. So in the next
step, we try to make it into one.
Step 4. Now!£ is an invertible sheaf on the integral scheme Z, so we can
find an embedding !£ ~ % 2 where % 2 is the constant sheaf of the function
field of Z (proof of 6.15). Hence f*!£ s:: f*%2 . But since f is assumed to be
birational, we have f*%2 = Xx, and so f*!£ s:: Xx. Now let A be an ample
invertible sheaf on X, which exists because X is assumed to be quasi-pro-
jective. Then I claim that there is an n > 0 and an embedding A-n s:: Xx
such that .A-n· f*!£ s:: (l)x· Indeed, let f be the ideal sheaf of denominators
off*!£, defined locally as {a E (l)xla · f*!£ s:: (l)x}- This is a nonzero coherent
sheaf of ideals on X, because f*!£ is a coherent subsheaf of Xx, so locally
one can just take common denominators for a set of generators of the cor-
responding finitely generated module. Since A is ample, f ® A" is gener-
ated by global sections for n sufficiently large. In particular, for suitable
n > 0, there is a nonzero map (l)x -+ f ®A", and hence a nonzero map
A-n -+ f. Then by construction A-n· f*!£ S:: (l)x·
Step 5. Since .A-n· f*!£ <;; (l)x, it is a coherent sheaf of ideals on X, which
we call f!. This is the required ideal sheaf, as we will now show that Z is
isomorphic to the blowing up of X with respect to f!. We already know that
Z ~ Proj !/. Therefore by (7.9) Z is also isomorphic to Proj ff*A-".
So to complete the proof, it will be sufficient to identify (Y*A-")d =
.A-dn ® f*!Ed with §d for any d ~ 1. First note that f*!Ed <;; Xx for any d
(same reason as above for d = 1), and since A is invertible, we can write
A-dn · f*!Ed instead of (8). Now since !/ is locally generated by !/1 as an
(l)x-algebra, we have a natural surjective map §d -+ A-dn · f*!Ed for each

167
II Schemes

d ? 1. It must also be injective, since both are subsheaves of Xx, so it is an

isomorphism. This shows finally that Z ~ Proj ffiDo 5d, which completes
the proof.

Remark 7.17.1. Of course the sheaf of ideals 5 in the theorem is not unique.
This is clear from the construction, but see also (Ex. 7.11).

Remark 7.17.2. We see from this theorem that blowing up arbitrary coherent
sheaves of ideals is a very general process. Accordingly in most applications
one learns more by blowing up only along some restricted class of sub-
varieties. For example, in his paper on resolution of singularities [ 4],
Hironaka uses only blowing up along a nonsingular subvariety which is
"normally flat" in its ambient space. In studying birational geometry of
surfaces in Chapter V, we will use only blowing up at a point. In fact one of
our main results there will be that any birational transformation of non-
singular projective surfaces can be factored into a finite number of blowings
up (and blowings down) of points. One important application of the more
general blowing-up we have been studying here is Nagata's theorem [6] that
any (abstract) variety can be embedded as an open subset of a complete
variety.

Example 7.17.3. As an example of the general concept of blowing up a co-

herent sheaf of ideals, we show how to eliminate the points of indeterminacy
of a rational map determined by an invertible sheaf. So Jet A be a ring, let X
be a noetherian scheme over A, let !£ be an invertible sheaf on X, and Jet
s 0 , . . . ,s" E r(X,!£) be a set of global sections of !£. Let U be the open
subset of X where the si generate the sheaf!£. Then according to (7.1) the
invertible sheaf !flu on U and the global sections s0 , . . . ,s" determine an
A-morphism cp: U --+ P~. We will now show how to blow up a certain sheaf
of ideals 5 on X, whose corresponding closed subsheme Y has support equal
to X - U (i.e., the underlying topological space of Y is X - U), so that the
morphism cp extends to a morphism if> of X to P~.

j ',, -
x
1t ' . . . .q>
... ,
''
"
X ._____) U L P~

So let$' be the coherent subsheaf of!£ generated by s0 , • .. ,s"" We define

a coherent sheaf of ideals 5 on X as follows: for any open set V <;; X, such
that !Eiv is free, let if;: !Eiv ..::::. (9v be an isomorphism, and take 5lv = 1/J(.?Ivl·
Clearly the ideal sheaf 5lv is independent of the choice ofljl, so we get a well-
defined coherent sheaf of ideals 5 on X. Note also that 5 x = (Ox if and only

168
 7 Projective Morphisms

E U, so the corresponding closed subscheme Y has support equal to

if x
X - U. Let n:X ~X be the blowing-up of .f. Then by (7.13a), n- 1 f · I!Jg
is an invertible sheaf of ideals, so we see that the global sections n*s; of n* £"
generate an invertible coherent subsheaf £"' ofn* £". Now£"' and the sections
n*s; define a morphism ip:X ~ P~ whose restriction to n- 1 (U) corresponds
to cp under the natural isomorphism n:n- 1 (U)..:::. U (7.13b).
In case X is a nonsingular projective variety over a field, we can rephrase
this example in terms oflinear systems. The given£" and sections s; determine
a linear system b on X. The base points of b are just the points of the closed
set X - U, and cp: U ~ P;; is the morphism determined by the base-point-
free linear system blu on U. We call Y the scheme of base points of b. So our
example shows that if we blow up Y, then b extends to a base-point-free linear
system bon all of X.

EXERCISES

7.1. Let (X,@x) be a locally ringed space, and let f:!l! ->At be a surjective map of
invertible sheaves on X. Show that f is an isomorphism. [Hint: Reduce to a
question of modules over a local ring by looking at the stalks.]
7.2. Let X be a scheme over a field k. Let !I! be an invertible sheaf on X, and let
{s 0 , . . . ,sn} and {t 0 , . . . ,tm} be two sets of sections of !1!, which generate the
same subspace V <;; T(X,fl!), and which generate the sheaf !I! at every point.
Suppose n ~ m. Show that the corresponding morphisms <p: X ..... P~ and 1/t: X ->
P;;' differ by a suitable linear projection pm - L ..... P" and an automorphism of
P", where Lis a linear subspace ofPrn of dimension m - n - 1.
7.3. Let <p:P~ ..... P;;' be a morphism. Then:
(a) either <p(P") = pt or m ;;, nand dim <p(P") = n;
(b) in the second case, <p can be obtained as the composition of (1) a d-uple em-
bedding P" ..... pN for a uniquely determined d ;;, 1, (2) a linear projection
pN - L ..... pm, and (3) an automorphism of pm_ Also, <p has finite fibres.

7.4. (a) Use (7.6) to show that if X is a scheme of finite type over a noetherian ring A,
and if X admits an ample invertible sheaf, then X is separated.
(b) Let X be the affine line over a field k with the origin doubled (4.0.1 ). Calculate
Pic X, determine which invertible sheaves are generated by global sections,
and then show directly (without using (a)) that there is no ample invertible
sheaf on X.
7.5. Establish the following properties of ample and very ample invertible sheaves on
a noetherian scheme X. fi!,At will denote invertible sheaves, and for (d), (e) we
assume furthermore that X is of finite type over a noetherian ring A.
(a) If !I! is ample and At is generated by global sections, then !I! ® At is ample.
(b) If !I! is ample and At is arbitrary, then At ® !I!" is ample for sufficiently large n.
(c) If fi!,At are both ample, so is !I! ® At.
(d) If !I! is very ample and At is generated by global sections, then !I! ® At is
very ample.
(e) If !I! is ample, then there is an n0 > 0 such that !I!" is very ample for all n ;;, n0 .

169
II Schemes

7.6. The Riemann-Roch Problem. Let X be a nonsingular projective variety over an

algebraically closed field, and let D be a divisor on X. For any n > 0 we consider
the complete linear system lnDI. Then the Riemann-Roch problem is to deter-
mine dimlnDI as a function of n, and, in particular, its behav.ior for large n. If
!i' is the corresponding invertible sheaf, then dimlnDI = dim T(X,!i'") - 1, so
an equivalent problem is to determine dim T(X,!i'") as a function of n.
(a) Show that if D is very ample, and if X c. P~ is the corresponding embedding
in projective space, then for all n sufficiently large, dimlnDI = Px(n) - 1,
where Px is the Hilbert polynomial of X (1, §7). Thus in this case dimlnDI is a
polynomial function of n, for n large.
(b) If D corresponds to a torsion element of Pic X, of order r, then dimlnDI = 0
if r In, - 1 otherwise. In this case the function is periodic of period r.
It follows from the general Riemann-Roch theorem that dimlnD I is a polyno-
mial function for n large, whenever Dis an ample divisor. See (IV, 1.3.2), (V, 1.6),
and Appendix A. In the case of algebraic surfaces, Zariski [7] has shown for any
effective divisor D, that there is a finite set of polynomials P 1 , . . . ,P., such that for
alln sufficiently large, dimlnDI = Pi(n)(n), where i(n) E { 1,2, ... ,r} is a function of n.
7.7. Some Rational Swfaces. Let X = Pt, and let IDI be the complete linear system
of all divisors of degree 2 on X (conics). D corresponds to the invertible sheaf
,z
@(2), whose space of global sections has a basis x 2 ,y 2 2 ,xy,xz,yz, where x,y,z
are the homogeneous coordinates of X.
(a) The complete linear system IDI gives an embedding ofP 2 in P 5 , whose image
is the Veronese surface (1, Ex. 2.13).
(b) Show that the subsystemdefined by x 2 ,yl,z 2 , y(x - z), (x - y)z gives a closed
immersion of X into P4 . The image is called the Veronese surface in P4 .
Cf. (IV, Ex. 3.11).
(c) Let b ~ IDI be the linear system of all conics passing through a fixed point P.
Then b gives an immersion of U = X - Pinto P 4 . Furthermore, if we blow
up P, to get a surface X, then this map extends to give a closed immersion of
X in P 4 . Show that X is a surface of degree 3 in P4 , and that the lines in X
through Pare transformed into straight lines in X which do not meet. X is the
union of all these lines, so we say X is a ruled surface (V, 2.19.1).
7.8. Let X be a noetherian scheme, let fff be a coherent locally free sheaf on X, and
let n: P(fff) -+ X be the corresponding projective space bundle. Show that there
is a natural 1-1 correspondence between sections of n (i.e., morphisms rJ:X -+
P(fff) such that no rJ = idx) and quotient invertible sheaves fff-+ !i' -+ 0 of fff.
7.9. Let X be a regular noetherian scheme, and fff a locally free coherent sheaf of rank
~ 2 on X.
(a) Show that Pic P(fff) ~ Pic X x Z.
(b) If fff' is another locally free coherent sheaf on X, show that P(fff) ~ P(fff')(over X)
if and only if there is an invertible sheaf !i' on X such that fff' ~ fff@ !i'.
7.10. P"-Bundles Over a Scheme. Let X be a noetherian scheme.
(a) By analogy with the definition of a vector bundle (Ex. 5.18), define the notion
of a projective n-space bundle over X, as a scheme P with a morphism n:P-+ X
such that Pis locally isomorphic to U x P", U ~ X open, and the transition
automorphisms on Spec A x P" are given by A-linear automorphisms of the
homogeneous coordinate ring A[ x 0 , . .. ,x.] (e.g., x; = ~)ijx 1 , aij E A).
(b) If fff is a locally free sheaf of rank n + 1 on X, then P(fff) is a P"-bundle over X.

170
 7 Projective Morphisms

*(c) Assume that X is regular, and show that every P"-bundle P over X is iso-
morphic to P(n') for some locally free sheaf 8 on X. [Hint: Let U c;: X be an
open set such that n- 1( U) ~ U x P", and let Y 0 be the invertible sheaf(': (I)
on U x P". Show that~ 0 extends to an invertible sheaf!!' on P. Then show
that n*!i' = 6 is a locally free sheaf on X and that P ~ P(8).] Can you
weaken the hypothesis "'X regular'"?
(d) Conclude (in the case X regular) that we have a 1-1 correspondence between
P"-bundles over X, and equivalence classes of locally free sheaves 6 of rank
11 + I under the equivalence relation&' - 6 if and only if 6' ~ 8 ®.If for

some invertible sheaf .It on X.

7.11. On a noetherian scheme X, different sheaves of ideals can give rise to isomorphic
blown up schemes.
(a) If.~ is any coherent sheaf of ideals on X, show that blowing up .~d for any
d ~ I gives a scheme isomorphic to the blowing up oL~ (cf. Ex. 5.13).
(b) If§ is any coherent sheaf of ideals, and if ,I is an invertible sheaf of ideals,
then .~ and § · f give isomorphic blowings-up.
(c) If X is regular, show that (7.17) can be strengthened as follows. Let U c;: X
be the largest open set such that f :f- 1 U -+ U is an isomorphism. Then 1
can be chosen such that the corresponding closed subscheme Y has support
equal to X- U
7.12. Let X be a noetherian scheme, and let Y, Z be two closed subschemes, neither
one containing the other. Let X be obtained by blowing up Y n Z (defined by
the ideal sheaf§ y + § z). Show that the strict transforms Y and Z of Y and Z
in X do not meet.
*7.13. A Complete Nonprojective Variety. Let k be an algebraically closed field of
char i= 2. Let C c;: Pt be the nodal cubic curve y 2 z = x 3 + x 2 z. If P 0 = (0,0,1)
is the singular point, then C- P 0 is isomorphic to the multiplicative group
Gm = Spec k[t,t- 1 ] (Ex. 6.7). For each a E k, a i= 0, consider the translation of
Gm given by tHat. This induces an automorphism of C which we denote by q>•.
Now consider C x (P 1 ~ {0}) and C x (P 1 - { oo }). We glue their open
subsets C x (P 1 - {0, oo}) by the isomorphism q>: <P,u)~---+<q>JP),u) for
P E C, u E Gm = P 1 - {O,oo }. Thus we obtain a scheme X, which is our example.
The projections to the second factor are compatible with q>, so there is a natural
morphism n: X-+ P 1 .
(a) Show that n is a proper morphism, and hence that X is a complete variety
over k.
(b) Use the method of (Ex. 6.9) to show that Pic(C x A 1) ~ Gm x Z and
Pic(C x (A 1 - {0})) ~ Gm x Z x Z. [Hint: If A is a domain and if *
denotes the group of units, then (A[u])* ~A* and (A[u,u- 1 ])* ~A* x Z.]
(c) Now show that the restriction map Pic(C x A 1)-+ Pic(C x (A 1 - {0}))
isoftheform <t,n) H <t,O,n),and that the automorphism q>ofC x (A 1 - {0})
induces a map of the form <t,d,n) H <t,d + n,n) on its Picard group.
(d) Conclude that the image of the restriction map Pic X-+ Pic(C x {0})
consists entirely of divisors of degree 0 on C. Hence X is not projective
over k and n is not a projective morphism.
7.14. (a) Give an example of a noetherian scheme X and a locally free coherent sheaf 8,
such that the invertible sheaf (11(1) on P(8) is not very ample relative to X.

171
II Schemes

(b) Let f: X--+ Y be a morphism of finite type, let .!f be an ample invertible
sheaf on X, and let Sf' be a sheaf of graded C9x-algebras satisfying (t). Let
P = Proj Y', let n: P --+ X be the projection, and let (9 p( 1) be the associated
invertible sheaf. Show that for all n » 0, the sheaf (9 p(1) ® n* !£'"is very ample
on P relative toY. [Hint: Use (7.10) and (Ex. 5.12).]

8 Differentials

In this section we will define the sheaf of relative differential forms of one
scheme over another. In the case of a nonsingular variety over C, which is
like a complex manifold, the sheaf of differential forms is essentially the same
as the dual of the tangent bundle defined in differential geometry. However,
in abstract algebraic geometry, we will define the sheaf of differentials first,
by a purely algebraic method, and then define the tangent bundle as its dual.
Hence we will begin this section with a review of the module of differentials of
one ring over another. As applications of the sheaf of differentials, we will
give a characterization of nonsingular varieties among schemes of finite
type over a field. We will also use the sheaf of differentials on a nonsingular
variety to define its tangent sheaf, its canonical sheaf, and its geometric genus.
This latter is an important numerical invariant of a variety.

Kahler Differentials
Here we will review the algebraic theory of Kahler differentials. We will use
Matsumura [2, Ch. 10] as our main reference, but proofs can also be found in
the exposes of Cartier and Godement in Cartan and Chevalley [I, exposes 13,
17], or in Grothendieck [EGA 01v, §20.5].
Let A be a ring (commutative with identity as always), let B be an A-
algebra, and let M be a B-module.

Definition. An A -derivation of B into M is a map d: B -+ M such that ( 1) d is

additive, (2) d(bb') = bdb' + b'db, and (3) da = 0 for all a EA.

Definition. We define the module of relative differential forms of B over A to

be a B-module Q 81 A, together with an A-derivation d:B-+ Q 81 A, which
satisfies the following universal property: for any B-module M, and for
any A-derivation d': B -+ M, there exists a unique B-module homomor-
phism f:Q 81 A-+ M such that d' = f o d.

Clearly one way to construct such a module Q 81 A is to take the free

B-module F generated by the symbols {dblb E B}, and to divide out by the
submodule generated by all expressions of the form (1) d(b + b') - db - db'
for h,b' E B, (2) d(bh') - bdb' - h'db for b,b' E B, and (3) da for a EA. The
derivation d:B -+ Q 81 A is defined by sending b to db. Thus we see that Q81 A
exists. It follows from the definition that the pair (Q 81 A,d) is unique up to

172
 8 Differentials

unique isomorphism. As a corollary of this construction, we see that QB/A is

generated as a B-module by {dblb E B}.

Proposition 8.1A. Let B be an A-algebra. Let f:B ®A B --+ B be the "diago-

nal" homomorphism defined by f(b ® b') = bb', and let I = kerf Consider
B ®A B as a B-module by multiplication on the left. Then I/I 2 inherits a
structure of B-module. Define a map d:B --+ I/I 2 by db = 1 ® b - b ® 1
(modulo I 2). Then <I/I 2,d) is a module of relative differentials forB/A.
PROOF. Matsumura [2, p. 182].

Proposition 8.2A. If A' and B are A-algebras, let B' = B ®A A'. Then
QB'JA' ~ QB/A ®B B'. Furthermore, if S is a multiplicative system in B,
then QS-IBjA ~ s-lQB/A-
PROOF. Matsumura [2, p. 186].

Example 8.2.1 If B = A[x 1, . . . ,xn] is a polynomial ring over A, then QBJA

is the free B-module of rank n generated by dx 1 , . . . ,dxn (Matsumura
[2, p. 184]).

Proposition 8.3A (First Exact Sequence). Let A --+ B--+ C be rings and homo-
morphisms. Then there is a natural exact sequence of C-modules
QB/A @B C --+ QCJA --+ QC/B --+ 0.
PROOF. Matsumura [2, Th. 57 p. 186].

Proposition 8.4A (Second Exact Sequence). Let B be an A-algebra, let I be

an ideal of B, and let C = B/I. Then there is a natural exact sequence of
C-modules
I/I2 ~ QB/A @B C--+ QCjA --+ 0,
where for any bE I, iflJ is its image in I/I 2, then (57)= db® 1. Note in
particular that I/I 2 has a natural structure of C-module, and that (5 is a C-
linear map, even though it is defined via the derivation d.
PROOF. Matsumura [2, Th. 58, p. 187].

Corollary 8.5. If B is a finitely generated A-algebra, or if B is a localization

of a finitely generated A-algebra, then QBJA is a finitely generated B-module.
PROOF. Indeed, B is a quotient of a polynomial ring (or its localization) so the
result follows from (8.4A), (8.2A), and the example of the polynomial ring
itself.

Now we will consider the module of differentials in the case of field

extensions and local rings. Recall (I, §4) that an extension field K of a field k

173
II Schemes

is separably generated if there exists a transcendence base {x,J for K/k such
that K is a separable algebraic extension of k( {x "} ).

Theorem 8.6A. Let K be a finitely generated extension field of afield k. Then

dimK QK!k ~ tr.d. K/k, and equality holds if and only if K is separably
generated over k. (Here dimK denotes the dimension as a K-vector space.)
PROOF. Matsumura [2, Th. 59, p. 191]. Note in particular that if K/k is a
finite algebraic extension, then QK/k = 0 if and only if K/k is separable.

Proposition 8.7. Let B be a local ring which contains afield k isomorphic to its
residue field B/m. Then the map c5: m/m 2 ---+ QB/k 0 B k of (8.4A) is an
isomorphism.
PROOF. According to (8.4A), the cokernel of c) is Qk/k = 0, so c) is surjective.
To show that c5 is injective, it will be sufficient to show that the map
c5':Homk(QB/k 0 k, k)---+ Homk(m/m 2 , k)
of dual vector spaces is surjective. The term on the left is isomorphic to
HomB(QB!k•k), which by definition of the differentials, can be identified with
the set Derk(B,k) of k-derivations of B to k. If d:B ---+ k is a derivation, then
c)'(d) is obtained by restricting to m, and noting that d(m 2 ) = 0. Now, to
show that c)' is surjective, lethE Hom(m/m 2 ,k). For any bE B, we can write
b = A + c, A E k, c E m, in a unique way. Define db = h(c), where c E m/m 2
is the image of c. Then one verifies immediately that d is a k-derivation of B
to k, and that c)'(d) = h. Thus c)' is surjective, as required.

Theorem 8.8. Let B be a local ring containing afield k isomorphic to its residue
field. Assume furthermore that k is perfect, and that B is a localization of a
finitely generated k-algebra. Then QB/k is a free B-module of rank equal to
dim B if and only if B is a regular local ring.
PRooF. First suppose QB/k is free of rank = dim B. Then by (8.7) we have
dimk m/m 2 = dim B, which says by definition that B is a regular local ring
(I, §5). Note in particular that this implies that B is an integral domain.
Now conversely, suppose that B is regular local of dimension r. Then
dimk m/m 2 = r, so by (8.7) we have dimk QB/k 0 k = r. On the other hand,
let K be the quotient field of B. Then by (8.2A) we have QB/k 0B K = QKfk·
Now since k is perfect, K is a separably generated extension field of k (I, 4.8A),
and so dimK QK!k = tr.d. K/k by (8.6A). But we also have dim B = tr.d. K/k
by (I, 1.8A). Finally, note that by (8.5), QB/k is a finitely generated B-module.
We conclude that QB/k is a free B module of rank r by using the following
well-known lemma.

Lemma 8.9. Let A be a noetherian local integral domain, with residue field k
and quotient field K. If M is a finitely generated A-module and if
dimk M 0 A k = dimK M 0 A K = r, then M is free of rank r.

174
 8 Differentials

PROOF. Since dimk M ® k = r, Nakayama's lemma tells us that M can be

generated by r elements. So there is a surjective map cp: Ar --+ M --+ 0. Let
R be its kernel. Then we obtain an exact sequence
0 --+ R ® K --+ Kr --+ M ® K --+ 0,
and since dimK M ® K = r, we have R ® K = 0. But R is torsion-free,
so R = 0, and M is isomorphic to Ar.

Sheaves of Differentials
We now carry the definition of the module of differentials over to schemes.
Let f:X--+ Y be a morphism of schemes. We consider the diagonal mor-
phism Ll :X--+ X x y X. It follows from the proof of (4.2) that Ll gives an
isomorphism of X onto its image Ll{X), which is a locally closed subscheme
of X x y X, i.e., a closed subscheme of an open subset W of X x y X.

Definition. Let J be the sheaf of ideals of Ll(X) in W. Then we define the

sheaf of relative differentials of X over Y to be the sheaf Qx;r = Ll*(J /.f 2 )
on X.

Remark 8.9.1. First note that JjJ 2 has a natural structure of @A <Xl-modure.
Then since Ll induces an isomorphism of X to Ll{X), Qx;r has a natural
structure of CDx-module. Furthermore, it follows from (5.9) that Qx;r is
quasi-coherent; if Y is noetherian and f is a morphism of finite type, then
X x y X is also noetherian, and so Qx;r is coherent.

Remark 8.9.2. Now if U = Spec A is an open affine subset of Y and V =

Spec B is an open affine subset of X such that f(V) s; U, then V x u Vis an
open affine subset of X x y X isomorphic to Spec (B ®A B), and Ll(X) n
(V xu V) is the closed subscheme defined by the kernel of the diagonal
homomorphism B ®A B--+ B. Thus .f/.§ 2 is the sheaf associated to the
module 1/1 2 of (8.1A). It follows that Qv;u ~ (QB 1A)-. Thus our definition
of the sheaf of differentials of X /Y is compatible, in the affine case, with the
module of differentials defined above, via the functor ~ . This also shows
that we could have defined Qx;r by covering X and Y with open affine sub-
sets Vand U as above, and glueing the corresponding sheaves (QB/A)-. The
derivations d:B--+ QB/A glue together to give a map d:CDx--+ Qx;r of sheaves
of abelian groups on X, which is a derivation of the local rings at each point.
Therefore, we can carry over our algebraic results to sheaves, and we
obtain the following results.

Proposition 8.10. Let f: X --+ Y be a morphism, let g: Y' --+ Y be another

morphism, and let f': X' = X x y Y' --+ Y' be obtained by base extension.
Then Qx·;r· ~ g'*(Qx;r) where g':X'--+ X is the .first projection.
PROOF. Follows from (8.2A).

175
II Schemes

Proposition 8.11. Let f:X ~ Y and g: Y ~ Z be morphisms of schemes.

Then there is an exact sequence of sheaves on X,
f*QY/Z ~ QX/Z ~ QX/Y ~ 0.
PROOF. Follows from (8.3A).

Proposition 8.12. Let f: X ~ Y be a morphism, and let Z be a closed sub-

scheme of X, with ideal sheaf .f!. Then there is an exact sequence of sheaves
onZ,
.f!/.f! 2 ~ QX/Y@ {!)z ~ QZ/Y ~ 0.
PROOF. Follows from (8.4A).

Example 8.12.1. If X = A~, then QX!Y is a free (!)x-module of rank n, gener-

ated by the global sections dx 1 , • •• ,dxn, where x 1, . . . ,xn are affine coordi-
nates for A".

Next we will give an exact sequence relating the sheaf of differentials on

a projective space to sheaves we already know. This is a fundamental result,
upon which we will base all future calculations involving differentials on
projective varieties.

Theorem 8.13. Let A be a ring, let Y = Spec A, and let X = P~. Then there
is an exact sequence of sheaves on X,
0 ~ QX/Y ~ {!)x(-lt+ 1 ~{!)X~ 0.
(The exponent n + 1 in the middle means a direct sum of n + 1 copies of
(!)x( -1).)

PROOF. Let S = A[ x 0 , . . . ,xn] be the homogeneous coordinate ring of X.

Let E be the graded S-module S( -1)"+ 1 , with basis e0 , . . . ,en in degree 1.
Define a (degree 0) homomorphism of graded S-modules E ~ S by sending
e; r--+ x;, and let M be the kernel. Then the exact sequence

o~M~E~s

of graded S-modules gives rise to an exact sequence of sheaves on X,

0~ M~ (!)x(-1)"+ 1 ~ (!)x ~ 0.
Note that E ~ S is not surjective, but it is surjective in all degrees ): 1, so
the corresponding map of sheaves is surjective.
We will now proceed to show that M ~ Qx1y. First note that if we localize
at X;, then Ex, ~ sx, is a surjective homomorphism of free sx,-modules, so
M x; is free of rank n, generated by {ej - (xj/x;)e; jj '!: i}. It follows that if U;
is the standard open set of X defined by X;, then Mlu, is a free (!)u,-module
generated by the sections (1/x;)ej - (x/xf)e; for j #- i. (Here we need the
additional factor 1/x; to get elements of degree 0 in the module Mx,.)

176
 8 Differentials

We define a map <p;:Qx1Yiu, --+ Mlu, as follows. Recall that U; ~

Spec A[x 0/X;, ... ,xnfxJ, so Qx1ylu, is a free (i)u,-module generated by
d(x 0 jxJ, ... ,d(xnfx;). So we define (/J; by
cpM(x/xJ) = (1/xr)(x;ej - xie;).
Thus (/J; is an isomorphism. I claim now that the isomorphisms (/J; glue
together to give an isomorphism <p: Qx;Y --+ M on all of X. This is a simple
calculation. On U; n Vi, we have, for any k, (xkjx;) = (xk/x) · (x/x;).
Hence in Qju,nu1 we have

Now applying (/J; to the left-hand side and <pi to the right-hand side, we get
the same thing both ways, namely (1/x;x)(xiek - xkeJ Thus the isomor-
phisms (/J; glue, which completes our proof.

N onsingular Varieties
Our principal application of the sheaf of differentials is to nonsingular
varieties. In (I, §5) we defined a nonsingular quasi-projective variety to be
one whose local rings were all regular local rings. Here we extend that def-
inition to abstract varieties.

Definition. An (abstract) variety X over an algebraically closed field k is

nonsingular if all its local rings are regular local rings.

Note that we are apparently requiring more here, because in Chapter I

we had only closed points, but now our varieties also have nonclosed points.
However, the two definitions are equivalent, because every local ring at a
nonclosed point is the localization of a local ring at a closed point, and we
have the following algebraic result.

Theorem 8.14A. Any localization of a regular local ring at a prime ideal is

again a regular local ring.
PROOF. Matsumura [2, p. 139].

The connection between nonsingularity and differentials is given by the

following result.

Theorem 8.15. Let X be an irreducible separated scheme of finite type over an

algebraically closed field k. Then QX/k is a locally free sheaf of rank
n = dim X if and only if X is a nonsingular variety over k.
PROOF. Ifx EX is a closed point, then the local ring B = (i)x,x has dimension n,
residue field k, and is a localization of a k-algebra of finite type. Furthermore

177
II Schemes

the module QB/k of differentials of B over k is equal to the stalk (Qx;dx of

the sheaf QX/k· Thus we can apply (8.8) and we see that (Qx 1dx is free of
rank n if and only if B is a regular local ring. Now the theorem follows in
view of (8.14A) and (Ex. 5.7).

Corollary 8.16. If X is a variety over k, then there is an open dense subset U

of X which is nonsingular.
PROOF. (This gives a new proof of (1, 5.3).) If n = dim X, then the function
field K of X has transcendence degree n over k, and it is a finitely generated
extension field, which is separably generated by (1, 4.8A). Therefore by (8.6A),
QK/k is a K-vector space of dimension n. Now QK!k is just the stalk of the
sheaf Qx;k at the generic point of X. Thus by (Ex. 5.7), QX!k is locally free of
rank n in some neighborhood of the generic point, i.e., on a nonempty open
set U. Then U is nonsingular by the theorem.

Theorem 8.17. Let X be a nonsingular variety over k. Let Y c;: X be an irre-

ducible closed subscheme defined by a sheaf of ideals J. Then Y is non-
singular if and only if
(1) QY/k is locally free, and
(2) the sequence of (8.12) is exact on the left also:
0--> 5/5 2 --> QX/k 0 {l)y--> QY/k--> 0.
Furthermore, in this case, 5 is locally generated by r = codim(Y,X)
elements, and 5/5 2 is a locally free sheaf of rank ron Y.
PROOF. First suppose (1) and (2) hold. Then QY/k is locally free, so by (8.15)
we have only to show that rank QY/k = dim Y. Let rank QY/k = q. We
know that QX/k is locally free of rank n, so it follows from (2) that 5/5 2 is
locally free on Y of rank n - q. Hence by Nakayama's lemma, 5 can be
locally generated by n - q elements, and it follows that dim Y ~ n -
(n - q) = q (1, Ex. 1.9). On the other hand, considering any closed point
y E Y, we have q = dimk(my/m;) by (8.7), and so q ~ dim Y by (1, 5.2A).
Thus q = dim Y. This shows that Y is nonsingular, and at the same time
establishes the statements at the end of the theorem, since we now have
n - q = codim(Y,X).
Conversely, assume that Y is nonsingular. Then QY!k is locally free of
rank q = dim Y, so (1) is immediate. From (8.12) we have the exact sequence
5/5 2 ~ QX/k 0 {l)y ~ QYjk--> 0.
We consider a closed pointy E Y. Then ker cp is locally free of rank r = n - q
at y, so it is possible to choose sections x 1, ... ,xr E 5 in a suitable neigh-
borhood of y, such that dx 1, ••• ,dxr generate ker cp. Let 5' be the ideal
sheaf generated by xI> ... ,x" and let Y' be the corresponding closed sub-
scheme. Then by construction, the dx 1 , . . . ,dxr generate a free subsheaf

178
 8 Differentials

of rank r of QX/k ® 0r· in a neighborhood of y. It follows that in the exact

sequence of (8.12) for Y',
§'jJ' 2 ~ QX/k @ @y, ~ QY'/k ~ 0,
we have 6 injective (since its image is free of rank r), and QY'/k is locally free
of rank n - r. The previous part of the proof now shows that Y' is irre-
ducible and nonsingular of dimension n - r (in a neighborhood of y).
But Y s Y', both are integral schemes of the same dimension, so we must
have Y = Y', J = J', and this shows that JjJ 2 ~ QX!k ® @y is injective,
as required.

Next we include a result which tells us that under suitable conditions, a

hyperplane section of a nonsingular variety in projective space is again
nonsingular. There is actually a large class of such results, which say that
if a projective variety has a certain property, then a sufficiently general
hyperplane section has the same property. The result we give here is not
the strongest, but it is sufficient for many applications. See also (III, 10.9) for
another version in characteristic 0.

Theorem 8.18 (Bertini's Theorem). Let X be a nonsingular closed subvariety

of PZ, where k is an algebraically closed field. Then there exists a hyperplane
H s PZ, not containing X, and such that the scheme H n X is regular at
every point. (In fact, we will see later (III, 7.9.1) that if dim X ;?: 2, then
H n X is connected, hence irreducible, and so H n X is a nonsingular
variety.) Furthermore, the set of hyperplanes with this property forms an
open dense subset of the complete linear system IHI, considered as a pro-
jective space.
PROOF. For a closed point x EX, let us consider the set Bx = {hyperplanes
HIH :::2 X or H ~ X but x E H n X, and xis not a regular point of H n X.}
(Fig. 10). These are the bad hyperplanes with respect to the point x. Now a

HeBx

HnX regular

Figure 10. Hyperplane sections of a nonsingular variety.

179
II Schemes

hyperplane His determined by a nonzero global section! E V = I'(P",cYpn(l) ).

Let us fix an foE V such that x ¢ H 0 , the hyperplane defined by f 0 . Then
we can define a map of k-vector spaces
<fJx: V--+ C9x,x/m~
as follows. Given f E V, then fifo is a regular function on P" - H 0 , which
induces a regular function on X - X n H 0 . We take <pjf) to be the image
of fifo in the local ring C9x,x modulo m~. Now the scheme H n X is defined
at x by the ideal generated by f/ fo in (9 x· Sox E H n X if and only if <fJxUl E mx,
and xis nonregular on H n X if and only if <fJAfl Em~, because in. that case,
the local ring 0xf(<p(f)) will not be regular. Thus we see that the hyperplanes
HE Bx correspond exactly to those f E ker <fJx (note also that <fJxUl = 0 ~
H 2 X.)
Since x is a closed point and k is algebraically closed, ntx is generated by
linear forms in the coordinates, so we see that <fJx is surjective. If dim X = r,
then dimk C9x/m~ = r + 1. We have dim V = n + 1, so dim ker <fJx =
n - r. This shows that Bx is a linear system of hyperplanes (in the sense
of §7) of dimension n - r - 1.
Now, considering the complete linear system IHI as a projective space,
consider the subset B <:; X x IHI consisting of all pairs <x,H) such that
x EX is a closed point and HE Bx. Clearly B is the set of closed points of
a closed subset of X x IHI, which we denote also by B, and which we give a
reduced induced scheme structure. We have just seen that the first projection
p 1 : B --+ X is surjective, with fibre a projective space of dimension 11 - r - 1.
Hence B is irreducible, and has dimension (n - r - 1) + r = 11 - 1.
Therefore, considering the second projection p 2 : B --+ IHI, we have
dim p2 (B) ~ n - 1. Since dimiHI = n, we conclude that p 2 (B) < IHI. If
H E IHI - p 2 (B), then H ~ X and every point of H n X is regular, so that
H satisfies the requirements of the theorem. Finally note that since X is
projective, p 2 : X x IHI --+ IHI is a proper morphism; B is closed in X x IHI,
so p 2 (B) is closed in IHI. Thus IHI - p 2 (B) is an open dense subset of IHI,
which proves the last statement of the theorem.

Remark 8.18.1. This result continues to hold even if X has a finite number
of singular points, because the set of hyperplanes containing any one of
them is a proper closed subset of IHI.

Applications
Now we will apply the preceding ideas to define some invariants of non-
singular varieties over a field.

Definition. Let X be a nonsingular variety over k. We define the tangent

sheaf of X to be :Yx = Yfomex(Qx 1k,cYx). It is a locally free sheaf of
rank n =dim X. We define the canonical sheaf of X to be Wx = (\"Qx;ko

180
 8 Differentials

the nth exterior power of the sheaf of differentials, where n = dim X.

It is an invertible sheaf on X. If X is projective and nonsingular, we de-
fine the yeometric yen us of X to be p9 = dimk F(X,wx ). It is a nonnegative
integer.

Remark 8.18.2. Earlier (I, Ex. 7.2) we defined the arithmetic genus Pa of a
variety in projective space. In the case of a projective nonsingular curve,
the arithmetic genus and the geometric genus coincide. This is a consequence
of the Serre duality theorem which we will prove later (III, 7.12.2). For
varieties of dimension ~ 2, however, Pa and p9 need not be equal (Ex. 8.3).
See also (I I I, 7.12.3).

Remark 8.18.3. Since the sheaf 9f differentials, the tangent sheaf, and the
canonical sheaf are all defined intrinsically, any numbers which we can
define from them, such as the geometric genus, are invariants of X up to
isomorphism. In 1 fact, we will now show that the geometric genus is a
hirational invariant of a nonsingular projective variety. This makes it ex-
tremely important for the classification problem.

Theorem 8.19. Let X and X' he two birationally equiralent nonsinyular pro-
jective rarieties over k. Then py(X) = p 9(X').
PROOF. Recall from (1, §4) that for X and X' to be birationally equivalent
means that there are rational maps from X to X' and from X' to X which
are inverses to each other. Considering the rational map from X to X',
let V <;: X be the largest open set for which there is a morphism f: V ---> X'
representing this rational map. Then from (8.11) we have a map f*Qx·;k --->
Qv;k· These are locally free sheaves of the same rank n = dim X, so we get
an induced map on the exterior powers: f*wx·---> wv. This map in turn
inducesamaponthespaceofglobalsectionsf*:T(X',wx.)---> T(V,wv). Now
since f is birational, by (I, 4.5), there is an open set U <;: V such that f( U)
is open in X', and f induces an isomorphism from U to/( U). Thus wvlu ~
Wx·IJ<Vl via f Since a nonzero global section of an invertible sheaf cannot
vanish on a dense open set, we conclude that the map of vector spaces
f*: T(X',wx·) ---> T( V,wv) must be injective.
Next we will compare T(V,wv) with T(X,wx). First I claim that X - V
has codimension ~ 2 in X. Indeed, this follows from the valuative criterion
of properness (4.7). If P EX is a point of codimension 1, then (!'P.x is a
discrete valuation ring (because X is nonsingular). We already have a map
of the generic point of X to X'; and X' is projective, hence proper over k, so
there exists a unique morphism Spec 0P.x ---> X' compatible with the given
birational map. This extends to a morphism of some neighborhood of P
to X', so we must have P E V by definition of V
Now we can show that the natural restriction map T(X,wx) ---> T( V,wv) is
bijective. It is enough to show, for any open affine subset U <:; X such that

181
II Schemes

wxlu ~ (!Ju, that r(U,(!Ju) ~ r(U n V,(!Ju,v) is bijective. Since X is non-

singular, hence normal, and since U - U n V has codimension ~ 2 in U,
this is an immediate consequence of (6.3A).
Combining our results, we see that pg(X') ~ pg(X). We obtain the reverse
inequality by symmetry, and thus conclude that p 9(X) = p 9(X').

Next we study the behavior of the tangent sheaf and the canonical sheaf
for a nonsingular subvariety of a variety X.

Definition. Let Y be a nonsingular subvariety of a nonsingular variety X

over k. The locally free sheaf f /5 2 of (8.17) we call the co normal sheaf of
Y in X. Its dual %y;x = Yf oml'!y(f /5 2 ,@y) is called the normal sheaf of
Yin X. It is locally free of rank r = codim( Y,X).

Note that if we take the dual on Y of the exact sequence of locally free
sheaves on Y given in (8.17), then we obtain an exact sequence
0 ~ !Ty ~ !Tx ® @y ~ Jlly;x ~ 0.
This shows that the normal sheaf we have just defined corresponds to the
usual geometric notion of normal vectors being tangent vectors of the
ambient space modulo tangent vectors of the subspace.

Proposition 8.20. Let Y be a nonsingular subvariety of codimension r in a non-

singular variety X over k. Then Wy ~ Wx ® N %y;x· In case r = 1,
consider Y as a divisor, and let !£' be the associated invertible sheaf on X.
Then Wy ~ Wx ® !l' ® @y.
PROOF. We take the highest exterior powers of the locally free sheaves in
the exact sequence
0 ~ f/5 2 ~ Qx ® @y ~ QY ~ 0
(Ex. 5.16d). Thus we find that wx ® @y ~ Wy ® N(f/5 2 ). Since formation
of the highest exterior power commutes with taking the dual sheaf, we find
Wy ~ Wx ® N %y;x· In the special case r = 1, we have f r ~ !£'- 1 by
(6.18). Thus f/5 2 ~ !l'- 1 ® @y, and %r;x ~ !£' ® @y. So applying the
previous result with r = 1, we obtain Wy ~ Wx ® !£' ® @y.

Example 8.20.1. Let X = P~. Taking the dual ofthe exact sequence of(8.13)
gives us this exact sequence involving the tangent sheaf of P":
0 ~ (!Jx ~ (!Jx(1)"+ 1 ~ !Tx ~ 0.
To obtain the canonical sheaf of P", we take the highest exterior powers of
the exact sequence of (8.13) and we find wx ~ (!Jx(- n - 1). Since @(f) has
no global sections for l < 0, we find that pg(P") = 0 for any n ~ 1. Recall
that a rational variety is defined as a variety birational to P" for some n

182
 8 Differentials

(I, Ex. 4.4). We conclude from (8.19) that if X is any nonsingular projective
rational variety, then p9 (X) = 0. This fact will enable us to demonstrate the
existence of nonrational varieties in all dimensions.

Example 8.20.2. Let X = P;;, with n ~ 2. For any integer d ~ 1, the divisor
dH, where H is a hyperplane, is a very ample divisor (7.6.1). Thus dH be-
comes a hyperplane section of X in a suitable projective embedding (the
d-uple embedding), and we can apply Bertini's theorem (8.18). We find that
there is a subscheme Y E ldHI which is regular at every one of its points. If
Y had at least two irreducible components, say Y1 and Y2 , then since n ~ 2,
their intersection Y1 n Y2 would be nonempty (I, 7.2). But this cannot happen
because Y would be singular at any point of Y1 n Y2 , so we conclude in fact
that Y is irreducible, hence a nonsingular variety. Thus we see for any d ~ 1
that there are nonsingular hypersurfaces of degree din P". In fact, they form
a dense open subset of the complete linear system ldHI. (This generalizes
(1, Ex. 5.5).)

Example 8.20.3. Let Y be a nonsingular hypersurface of degree d in P",

n ~ 2. Then from (8.20) and the first example above, we conclude that
Wy ~ (0y(d - n - 1). Let's look at some particular cases.
n = 2, d = 1. Y is a line in P 2 , so Y ~ Pi, and we have Wy ~ lPy( -2)
which we already knew.
n = 2, d = 2. Y is a conic in P 2 , and Wy ~ @y( -1). In this case Y is
the 2-uple embedding of Pi, so pulling Wy back to P 1 gives Wpt ~ lPpt(- 2),
which is again what we already knew.
n = 2, d = 3. Y is a nonsingular plane cubic curve, and Wy ~ lPy. There-
fore p9 (Y) =dim r(Y,lPy) = 1, and we see that Y is not rational! This
generalizes (1, Ex. 6.2), where we gave just one example of a nonsingular
cubic curve, and showed by a different method that it was not rational.
n = 2, d ~ 4. Y is a nonsingular plane curve of degree d, Wy ~ lPy(d - 3),
and d - 3 > 0. Hence p9 > 0, and Y is not rational. In fact, p9 =
!(d - 1)(d - 2) (Ex. 8.4f), so we see that plane curves of different degrees
d,d' ~ 3 are not birational to each other. Another way of seeing this is as
follows. For any nonsingular projective curve, we can consider the degree of
the canonical sheaf. Since a nonsingular projective curve is unique in its
birational equivalence class (1, §6), this number is in fact a birational invariant.
In the present case its value is d(d - 3), since @(1) has degree don Y. These
numbers are also distinct for different d,d' ~ 3. This shows the existence of
infinitely many mutually nonbirational curves.
n = 3, d = 1. This gives Y ~ P 2 , Wy ~ lPy(- 3) which we knew.
n = 3, d = 2. Here Y is a nonsingular quadric surface, and Wy ~ lPy(- 2).
We have p 9 (Y) = 0, which is consistent with the fact that Y is rational
(I, Ex. 4.5). In terms of the isomorphism Y ~ P 1 x P 1 , wy corresponds to a
divisor class of type (- 2,- 2)-see (6.6.1 ). This illustrates the general fact

183
II Schemes

(Ex. 8.3) that the canonical sheaf on a direct product of nonsingular varieties
is the tensor product of the pull-backs of the canonical sheaves on the two
factors.
n = 3, d = 3. Y is a nonsingular cubic surface in P 3 , wy ~ l'9y( -1) and
so p9 (Y) = 0. In this case also, Y is a rational surface, as we will see later
(Chapter V).
n = 3, d = 4. In this case Wy ~ l'9y. The canonical sheaf is trivial so
p 9 = 1. This is a nonrational surface which belongs to the class of "K3
surfaces."
n = 3, d ?: 5. Here Wy ~ l'9y(d - 4) with d - 4 > 0. Hence p9 > 0,
and Y is not rational. Surfaces such as these on which the canonical sheaf
is very ample belong to the class of "surfaces of general type."
n = 4, d = 3,4. The cubic and the quartic threefold in P 4 both have
p 9 = 0, but it has recently been shown (by different methods) that they are
not in general rational varieties. For the cubic threefold, see Clemens and
Griffiths [1]. For the quartic threefold, see lskovskih and Manin [1].
n arbitrary, d ?: n + 1. In this case we obtain a nonsingular hypersurface
Yin P", with Wy ~ l'9y(d - n - 1) and d - n - 1 ?: 0. Hence p9 (Y) ?: 1,
and so Y is not rational. This shows the existence of nonrational varieties
in all dimensions.

Some Local Algebra

Here we will gather some results from local algebra, mainly concerning
depth and Cohen-Macaulay rings, which are useful in algebraic geometry.
Then we relate them to the geometric notion of local complete intersection,
and give an application to blowing up. We refer to Matsumura [2, Ch. 6]
for proofs.
If A is a ring, and M is an A-module, recall that a sequence xt> ... ,x, of
elements of A is called a regular sequence for M if x 1 is not a zero divisor
in M, and for all i = 2, ... ,r, X; is not a zero divisor in Mj(x 1 , . .. ,x;_ 1 )M.
If A is a local ring with maximal ideal m, then the depth of M is the maximum
lengthofaregularsequencex 1, ... ,x,for Mwithallx; Em. These definitions
apply to the ring A itself, and we say that a local noetherian ring A is Cohen-
Macaulay if depth A = dim A. Now we list some properties of Cohen-
Macaulay rings.

Theorem 8.21A. Let A be a local noetherian ring with maximal ideal m.

(a) If A is regular, then it is Cohen-Macaulay.
(b) If A is Cohen-Macaulay, then any localization of A at a prime ideal
is also Cohen-M acaulay.
(c) If A is Cohen-Macaulay, then a set of elements x 1, ... ,x, E mforms
a regular sequence for A if and only if dim Aj(x 1 , • •• ,x,) = dim A - r.
(d) If A is Cohen-Macaulay, and x 1, ... ,x, Em is a regular sequence
for A, then Aj(x 1 , . . . ,x,) is also Cohen-Macaulay.

184
 8 Differentials

(e) If A is Cohen-Macaulay, and x 1 , . . . ,x, Em is a regular sequence,

let I be the ideal(x 1 , ... ,x,). Then the natural map(A/l)[t~> ... ,t,]-> gr1 A =
EBn"o 1"/1"+ 1 , defined by sending ti 1-> xi> is an isomorphism. In other words,
1/1 2 is a free A/1-module of rank r, and for each n ~ 1, the natural map
S"(/// 2 )-> I"/ I"+ 1 is an isomorphism, where S" denotes the nth symmetric
power.
PROOFS. Matsumura [2: (a) p. 121; (b) p. 104; (c) p. 105; (d) p. 104;
(e) p. 110].

In keeping with the terminology for schemes (Ex. 3.8), we will say that a
noetherian ring A is normal if for every prime ideal p, the localization AP
is an integrally closed domain. A normal ring is a finite direct product of
integrally closed domains.

Theorem 8.22A (Serre). A noetherian ring A is normal if and only if it satisfies

the following two conditions:
(1) for every prime ideal p ~ A of height :C 1, AP is regular (hence a
field or a discrete valuation ring); and
(2) for every prime ideal p ~ A of height ~ 2, we have depth A P ~ 2.
PROOF. Matsumura [2, Th. 39, p. 125]. Condition (1) is sometimes called
"R 1", or "regular in codimension
I
1". Condition (2), supplemented by the
requirement that for ht p = 1, depth AP = 1, which is a consequence of (1)
in our case, is called the "condition S 2 of Serre".

Now we apply these results to algebraic geometry. We will say that a

scheme is Cohen-M acaulay if all of its local rings are Cohen-Macaulay.

Definition. Let Y be a closed subscheme of a nonsingular variety X over k.

We say that Y is a local complete intersection in X if the ideal sheaf J r
of Yin X can be locally generated by r = codim(Y,X) elements at every
point.

Example 8.22.1. If Y itself is nonsingular, then by (8.17) it is a local complete

intersection inside any nonsingular X which contains it.

Remark 8.22.2. In fact, the notion of being a local complete intersection is

an intrinsic property of the scheme Y, i.e., independent of the nonsingular
variety containing it. This is proved using the cotangent complex of a
morphism, which extends the concept of relative differentials introduced
above-see Lichtenbaum and Schlessinger [ 1]. We will not use this fact in
the sequel.

185
II Schemes

Proposition 8.23. Let Y be a locally complete intersection subscheme of a

nonsingular variety X over k. Then:
(a) Y is Cohen-Macaulay;
(b) Y is normal if and only if it is regular in codimension 1.

PROOF.
(a) Since X is nonsingular, it is Cohen-Macaulay by (8.21Aa). Since J y
is locally generated by r = codim(Y,X) elements, those elements locally
form a regular sequence in (!Jx, by (8.21Ac), and so Y is Cohen-Macaulay
by (8.21Ad).
(b) We already know that normal implies regular in codimension 1
(1, 6.2A). For the converse, we use (8.22A) applied to the local rings of Y.
Condition (1) is our hypothesis, and condition (2) holds automatically
because Y is Cohen-Macaulay.

As our last application, we consider the blowing-up of a nonsingular

variety along a nonsingular subvariety (cf. §7 for definition of blowing-up).
The following theorem will be useful in comparing invariants of X and X
(Ex. 8.5).

Theorem 8.24. Let X be a nonsingular variety over k, and let Y !:;; X be a

nonsingular closed subvariety, with ideal sheaf J. Let n:X -->X be the
blowing-up of J, and let Y' !:;; X be the subscheme defined by the inverse
image ideal sheaf J' = n- 1 J ·@g. Then:
(a) X is also nonsingular;
(b) Y', together with the induced projection map n: Y' --> Y, is isomorphic
to P(JjJ 2 ), the projective space bundle associated to the (locally free)
sheaf JjJ 2 on Y;
(c) under this isomorphism, the normal sheaf JV Y'/X corresponds to
{!JP($/$2)( -1).

PROOF. We prove (b) first. Since X = EB Jd, we have

Proj
Y' ~ Proj EB (Jd@ (!Jx/J) = Proj EB Jd/Jd+ 1.

But Y is nonsingular, so J is locally generated by a regular sequence in (!Jx,

and we can apply (8.21Ae). This implies that JjJ 2 is locally free and that
for each n ~ 1, Jn;Jn+ 1 ~ sn(J/J 2 ). Thus Y' ~ Proj EB Sd(J/J 2 ), which
by definition is P(J/J ). 2

In particular, Y' is locally isomorphic to Y x P'- 1 , where r = codim( Y,X),

so Y' is also nonsingular. Since Y' is locally principal in X (7.13a), it follows
that X is also nonsingular: if a quotient of a noetherian local ring by an
element which is not a zero divisor is regular, then the local ring itself is regular.
To prove (c), we recall from the proof of (7.13) that J' = n- 1 (J) · (!Jg is
isomorphic to @g(1). It follows that f'/f' 2 ~ @y.(1), and hence JV Y'/X ~
(!Jy.(-1).

186
 8 Differentials

We will use the following algebraic result in the exercises.

Theorem 8.25A (I. S. Cohen). Let A be a complete local ring contaznzng a

field k. Assume that the residue field k(A) = A/m is a separably generated
extension of k. Then there is a subfield K s;; A, containing k, such that
K ~ A/m is an isomorphism. (The subfield K is called a field of repre-
sentatives for A.)

PROOF. Matsumura [2, p. 205].

EXERCISES

8.1 Here we will strengthen the results of the text to include information about the
sheaf of differentials at a not necessarily closed point of a scheme X.
(a) Generalize (8.7) as follows. Let B be a local ring containing a field k, and
assume that the residue field k(B) = B/m of B is a separably generated ex-
tension of k. Then the exact sequence of (8.4A),
0---> mjm 2 ~ QB/k@ k(B)---> Qk(B)/k---> 0
is exact on the left also. [Hint: In copying the proof of(8.7), first pass to Bjm 2 ,
which is a complete local ring, and then use (8.25A) to choose a field of repre-
sentatives for Bjm 2 .]
(b) Generalize (8.8) as follows. With B, k as above, assume furthermore that k is
perfect, and that B is a localization of an algebra of finite type over k. Then
show that B is a regular local ring if and only if QB!k is free of rank = dim B +
tr.d. k(B)/k.
(c) Strengthen (8.15) as follows. Let X be an irreducible scheme of finite type over
a perfect field k, and let dim X = n. For any point x EX, not necessarily
closed, show that the local ring (l)x.x is a regular local ring if and only if the
stalk (.Qx 1k)x of the sheaf of differentials at xis free of rank n.
(d) Strengthen (8.16) as follows. If X is a variety over an algebraically closed field
k, then U = {x E Xl(l)x is a regular local ring} is an open dense subset of X.

8.2. Let X be a variety of dimension n over k. Let r! be a locally free sheaf of rank > n
on X, and let V ~ r(X,r!) be a vector space of global sections which generate
<!. Then show that there is an element s E V, such that for each x E X, we have
sx rf= mxrffx. Conclude that there is a morphism (l)x---> r! giving rise to an exact
sequence
0 ---> (IJX ---> rff ---> rff' ---> 0

where<!' is also locally free. [Hint: Use a method similar to the proof of Bertini's
theorem (8.18).]

8.3. Product Schemes.

(a) Let X and Y be schemes over another schemeS. Use (8.10) and (8.11) to show
that Qx x, YJS ::::; p!QxJs EB p~QYJS·
(b) If X and Yare nonsingular varieties over a field k, show that Wx x y ::::; p!wx ®
p~Wy.

187
II Schemes

(c) Let Y be a nonsingular plane cubic curve, and let X be the surface Y x Y.
Show that pg(X) = 1 but p.(X) = -1 (I, Ex. 7.2). This shows that the arith-
metic genus and the geometric genus of a nonsingular projective variety may
be different.
8.4. Complete Intersections in P". A closed subscheme Y ofP~ is called a (strict, global)
complete intersection if the homogeneous ideal I of Y in S = k[ x0 , ..• ,x.J can
be generated by r = codim(Y,P") elements (I, Ex. 2.17).
(a) Let Y be a closed subscheme of codimension r in P". Then Y is a complete
intersection if and only if there are hypersurfaces (i.e., locally principal sub-
schemes of codimension 1) H 1, ..• ,H" such that Y = H 1 n ... n H, as schemes,
i.e.,...fy = ...fH, + ... + ...fn,· [Hint:Usethefactthattheunmixednesstheorem
holds inS (Matsumura [2, p. 107]).]
(b) If Y is a complete intersection of dimension ;;;, 1 in P", and if Y is normal, then
Y is projectively normal (Ex. 5.14). [Hint: Apply (8.23) to the affine cone over
Y.]
(c) With the same hypotheses as (b), conclude that for all/ ;;;, 0, the natural map
T(P",0pn(/)) --+ T(Y,0y(l)) is surjective. In particular, taking I = 0, show that
Y is connected.
(d) Now suppose given integers d 1, .•• ,d, ;;;, 1, with r < n. Use Bertini's theorem
(8.18) to show that there exist nonsingular hypersurfaces H 1 , .•• ,H, in P",
with deg Hi = di, such that the scheme Y = H 1 n ... n H, is irreducible and
nonsingular of codimension r in P".
(e) If Y is a nonsingular complete intersection as in (d), show that Wy ~
0r(Ldi - n - 1).
(f) If Y is a nonsingular hypersurface of degree din P", use (c) and (e) above to
show that p9 (Y) = (4 .- 1 ). Thus pg(Y) = P.(Y) (I, Ex. 7.2). In particular, if Y
is a nonsingular plane curve of degree d, then p9 (Y) = !(d - l)(d - 2).
(g) If Y is a nonsingular curve in P 3 , which is a complete intersection of nonsingular
surfacesofdegreesd,e,thenp9 (Y) = !de(d + e- 4) + 1. Againthegeometric
genus is the same as the arithmetic genus (I, Ex. 7.2).
8.5. Blowing up a Nonsingular Subvariety. As in (8.24), let X be a nonsingular variety,
let Y be a nonsingular subvariety of codimension r ;;;, 2, let n: X --+ X be the
blowing-up of X along Y, and let Y' = rr- 1 (Y).
(a) Show that the maps rr*: Pic X --+ Pic X, and Z --+ Pic X defined by n r--> class
of nY', give rise to an isomorphism Pic X ~ Pic X EB Z.
(b) Show that Wx ~ f*wx ® .st((r- 1)Y'). [Hint: By (a) we can write in any
casewx ~ f* At ® .sf(q Y')for some invertible sheaf A on X, and some integer
q. By restricting to X - Y' ~ X - Y, show that A ~ Wx. To determine q,
proceed as follows. First show that Wy· ~ f*wx ® @y.(- q - 1). Then take
a closed point y E Y and let Z be the fibre of Y' over y. Then show that Wz ~
@z(-q- 1). ButsinceZ ~ P'-l,wehavewz ~ @z(-r),soq = r- 1.]
8.6. The Infinitesimal Lifting Property. The following result is very important in study-
ing deformations of nonsingular varieties. Let k be an algebraically closed field,
let A be a finitely generated k-algebra such that Spec A is a nonsingular variety
over k. Let 0 --+ I --+ B' --+ B --+ 0 be an exact sequence, where B' is a k-algebra,
and I is an ideal with I 2 = 0. Finally suppose given a k-algebra homomorphism
f: A --+ B. Then there exists a k-algebra homomorphism g: A --+ B' making a
commutative diagram

188
 8 Differentials

0
~
I
~
B'
g ..--"
,.,. . . . . . . ....f --- ~
A B
~
0
We call this result the infinitesimal lifting property for A. We prove this result
in several steps.
(a) First suppose that g: A ..... B' is a given homomorphism lifting f If g': A ..... B'
is another such homomorphism, show that 8 = g - g' is a k-derivation of A
into I, which we can consider as an element of HomA(QA/k,l). Note that since
I 2 = 0, I has a natural structure of B-module and hence also of A-module.
Conversely, for any e E HomA(QA/k,l), g' = g + eis another homomorphism
lifting f (For this step, you do not need the hypothesis about Spec A being
nonsingular.)
(b) NowletP = k[x 1 , •.. ,x.] beapolynomialringoverkofwhichA is a quotient,
and let J be the kernel. Show that there does exist a homomorphism h: P ..... B'
making a commutative diagram,
0 0
~ ~
J I
~ ~
p h
B'
~ ~
A f 8
~ ~
0 0
and show that h induces an A-linear map Ti:J/] 2 ..... I.
(c) Now use the hypothesis Spec A nonsingular and (8.17) to obtain an exact
sequence
0 ..... J/]2 ..... QP/k@ A ..... QA/k ..... 0.
Show furthermore that applying the functor Hom A(·,/) gives an exact sequence
0-+ HomA(QA 1k,l)-+ Homp(QP1k,J) ..... HomA(J/] 2,1) ..... 0.
Let e E Homp(QP/k,/) be an element whose image gives TiE HomA(J/1 2 ,/).
Consider 8 as a derivation of P to B'. Then let h' = h - 8, and show that h'
is a homomorphism of P ..... B' such that h'(J) = 0. Thus h' induces the
desired homomorphism g: A-+ B'.
8.7. As an application of the infinitesimal lifting property, we consider the following
general problem. Let X be a scheme of finite type over k, and let :F be a coherent
sheaf on X. We seek to classify schemes X' over k, which have a sheaf of ideals
J such that J 2 = 0 and (X',@x.fJ) ~ (X,@x), and such that J with its resulting
structure of @x-module is isomorphic to the given sheaf :F. Such a pair X',J
we call an infinitesimal extension of the scheme X by the sheaf :F. One such

189
II Schemes

extension, the trivial one, is obtained as follows. Take @X' = @x EB ff as sheaves

of abelian groups, and define multiplication by (a EB f) · (a' EB f') = aa' EB
(af' + a'f). Then the topological space X with the sheaf of rings @X' is an in-
finitesimal extension of X by ff.
The general problem of classifying extensions of X by ff can be quite com-
plicated. So for now, just prove the following special case: if X is affine and
nonsingular, then any extension of X by a coherent sheaf ff is isomorphic to
the trivial one. See (III, Ex. 4.10) for another case.
8.8. Let X be a projective nonsingular variety over k. For any n > 0 we define the
nth plurigenus of X to be P" = dimkr(X,wj'"). Thus in particular P 1 = p9 •
Also, for any q, 0 :::;; q :::;; dim X we define an integer hq,o = dimkT(X,Q~1 k)
where m/k = 1\q.QX/k is the sheaf of regular q-forms on X. In particular, for
q = dim X, we recover the geometric genus again. The integers hq,o are called
Hodge numbers.
Using the method of (8.19), show that P" and hq,o are birational invariants of
X, i.e., if X and X' are birationally equivalent nonsingular projective varieties,
then Pn(X) = P.(X') and hq· 0 (X) = hq· 0 (X').

9 Formal Schemes
One feature which clearly distinguishes the theory of schemes from the older
theory of varieties is the possibility of having nilpotent elements in the struc-
ture sheaf of a scheme. In particular, if Y is a closed subvariety of a variety X,
defined by a sheaf of ideals f, then for any n ;?: 1 we can consider the closed
subscheme Y, defined by the nth power f" of the sheaf of ideals f. For n ;?: 2,
this is a scheme with nilpotent elements. It carries information about Y
together with the infinitesimal properties of the embedding of Y in X.
The formal completion of Yin X, which we will define precisely below, is
an object which carries information about all the infinitesimal neighborhoods
Y, of Y at once. Thus it is thicker than any Y,, but it is contained inside any
actual open neighborhood of Yin X. We might call it the formal neighbor-
hood of Yin X.
The idea of considering these formal completions is already implicit in the
memoir of Zariski [3], where he uses the "holomorphic functions along a
subvariety" for his proof of the connectedness principle. We will give different
proofs of some of Zariski's results, using cohomology, in (III, §11). A striking
application of formal schemes as something in between a subvariety and an
ambient variety is in Grothendieck's proof of the Lefschetz theorems on
Pic and rc 1 [SGA 2]. This material is also explained in Hartshorne [5, Ch. IV].
We will define an arbitrary formal scheme as something which looks
locally like the completion ofa usual scheme along a closed subscheme.

Inverse Limits of Abelian Groups

First we recall the notion of inverse limit. An inverse system of abelian groups
is a collection of abelian groups An, for each n ;?: 1, together with homomor-

190
 9 Formal Schemes

phisms <f>n·n: An' --+ An for each n' ? n, such that for each n" ? n' ? n we
have <f>n"n = <f>n·n o <f>n"n'· We will denote the inverse system by (An,<f>n•n),
or simply (An), with the q> being understood. If (An) is an inverse system of
abelian groups, we define the inverse limit A = fu!! An to be the set of se-
quences {an} E flAn such that <f>n·n(an.) = an for all n' ? n. Clearly A is a
group. The inverse limit A can be characterized by the following universal
property: given a group B, and homomorphisms 1/Jn:B--+ An for each n,
such that for any n' ? n, 1/Jn = <f>n·n o 1/Jn•, then there exists a unique homo-
morphism 1/1: B --+ A such that 1/Jn = Pn o 1/1 for each n, where Pn: A --+ An is
the restriction of the nth projection map fl An --+ A"'
Ifthe groups An have the additional structure of vector spaces over a field
k, or modules over a ring R, then the above discussion makes sense in the
category of k-vector spaces orR-modules.

Next we study exactness properties of the inverse limit (cf. Atiyah-

Macdonald [1, Ch.lO]). A homomorphism (An) --+ (Bn) of inverse systems of
abelian groups is a collection of homomorphisms fn: An --+ Bn for each n,
which are compatible with the maps of the inverse system, i.e., for each
n' ? n, we have a commutative diagram
A n' -----"-"~-~
j~. Bn'

l<f>n•n

---'h'-"n--~ Bn.
A sequence
0 --+ (An) --+ (Bn) --+ ( Cn) --+ 0
of homomorphisms of inverse systems is exact if the corresponding sequence
of groups is exact for each n. Given such a short exact sequence of inverse
systems, one sees easily that the sequence of inverse limits
0 --+ +-----
lim An --+ +-----
lim Bn --+ +--
lim Cn
is also exact. However, the last map need not be surjective. So we say that
lim is a left exact functor.
+--
To give a criterion for exactness of lim on the right, we make the following
+--
definition: an inverse system (An,<f>n·n) satisfies the M ittag-Le.ffler condition
(ML) iffor each n, the decreasing family {<r>n·n(An.) ~ Anln' ? n} of subgroups
of An is stationary. In other words, for each n, there is an n0 ? n, such that
for all n', n" ? n0 , <f>n·n(An.) = <f>n"n(An") as subgroups of An.
Suppose an inverse system (An) satisfies (ML). Then for each n, we let
A~ ~ An be the stable image <f>n·n(An.) for any n' ? n0 , which exists by the
definition. Then one sees easily that (A~) is also an inverse system, with the
induced maps, and that the maps of the new system (A~) are all surjective.
Furthermore, it is clear that +-----lim A~ = +-----
lim An. So we see that A = +--
lim An
maps surjectively to each A~.

191
II Schemes

Proposition 9.1. Let

be a short exact sequence of inverse systems of abelian groups. Then:

(a) if (Bn) satisfies (ML), so does (Cn).
(b) if (An) satisfies (ML), then the sequence of inverse limits
0 --+ lim An
f--
--+ lim Bn
f--
--+ lim C.
f--
--+ 0
is exact.
PROOF. (See also Grothendieck [EGA 0111, 13.2].)
(a) For each n' ~ n, the image of B •. in B. maps surjectively to the image of
c•. in c., so (ML) for (B.) implies (ML) for (Cn) immediately.
(b) The only nonobvious part is to show that the last map is surjective.
So let {c.} Ell!!! c.. For each n, let En= g- 1 (c.). Then E. is a subset of
B., and (E.) is an inverse system of sets. Furthermore, each E. is bijective,
in a noncanonical way, with An, because of the exactness of the sequence
0 --+ An --+ Bn --+ c. --+ 0. Thus since (An) satisfies (ML), one sees easily that
(E.) satisfies the Mittag-Leffier condition as an inverse system of sets (same
definition). Since each En is nonempty, it follows from considering the
inverse system of stable images as above, that flim E. is also nonempty. Taking
any element of this set gives an element of ll!!! B. which maps to {c.}.
--

Example 9.1.1. If all the maps IPn·n: A •. --+ A. are surjective, then (A.) satis-
fies (ML), so (9.lb) applies.
Example 9.1.2. If(An) is an inverse system of finite-dimensional vector spaces
over a field, or more generally, an inverse system of modules with descending
chain condition over a ring, then (A") satisfies (ML).

Inverse Limits of Sheaves

In any category <£, we define the notion of inverse limit by analogy with the
universal property of the inverse limit of abelian groups above. Thus if
(A.,<pn·n) is an inverse system of objects of(£ (same definition as above), then
an inverse limit A = lim f--
A. is an object A of<£, together with morphisms
p.:A --+An for each n, such that for each n' ~ n, Pn = IPn'n o p•. , satisfying the
following universal property: given any object B of<£, together with mor-
phisms t/J.:B--+ A. for each n, such that for each n' ~ n, t/1. = IPn·n a tfJ •. ,
there exists a unique morphism ljJ:B --+A such that for each n, t/1. = Pn a t/J.
Clearly the inverse limit is unique if it exists. But the question of existence
depends on the particular category considered.
Proposition 9.2. Let X be a topological space, and let (£ be the category of
sheaves of abelian groups on X. Then inverse limits exist in<£. Furthermore,
if (ff.) is an inverse system of sheaves on X, and ff = ll!!! ffn is its inverse
limit, then for any open set U, we have r(U,ff) = r(U,ffn) in the !!!!!
category of abelian groups.

192
 9 Formal Schemes

PROOF. Given an inverse system of sheaves (~n) on X, we consider the pre-

sheaf U ~ <--
lim r(U,~n), where this inverse limit is taken in the category of
abelian groups. Now using the sheaf property for each ~n, one verifies
immediately that this presheaf is a sheaf. Call it ~. Now given any other
sheaf<:#, and a system of compatible maps 1/Jn:'§ ~ ~n for each n, it follows
from the universal property of an inverse limit of abelian groups that we
obtain unique maps, for each U, r(U,<:#) ~ r(U,~). These give a sheaf map
<:§ ~ ~. thus verifying that ~ is the inverse limit of the ~n in <r.

Caution 9.2.1. Even though inverse limits exist in the category(£: of abelian
sheaves on a topological space, one must beware of using intuition derived
from the category of abelian groups. In particular, the statement of(9.lb) is
false in <r, even if all maps in the inverse system (An) are surjective. So in
studying exactness questions, we will always pass to sections over an open
set, and thus reduce to questions about abelian groups. For more details
about exactness of!!!!! in <r, see Hartshorne [7, I, §4].

Completion of a Ring
One important application of inverse limits is to define the completion of a
ring with respect to an ideal. This generalizes the notion of completion of a
local ring which was discussed in (1, §5). It also forms the algebraic model for
the completion of a scheme along a closed subscheme which will come next.
So let A be a commutative ring with identity (as always), and let I be an
ideal of A. We denote by r the nth power of the ideal I. Then we have
natural homomorphisms
... ~ A/I 3 ~ A/I 2 ~A/I,
which make (A/ r) into an inverse system of rings. The inverse limit ring
lim Ajr is denoted by A and is called the completion of A with respect to I
<--
or the 1-adic completion of A. For each n we have a natural map A ~ Ajr,
so by the universal property we obtain a homomorphism A ~ A.
Similarly, if M is any A-module, we define M = lim MjrM, and call it the
<---~
I-adic completion of M. It has a natural structure of A-module.

Theorem 9.3A. Let A be a noetherian ring, and I an ideal of A. We denote by ~

the I-adic completion as above. Then:
(a) i = +---
lim Ijr is an ideal of A. For any n, Jn = rA., and A.;Jn ~ Ajr;
(b) if M is a .finitely generated A-module, then M ~ M @A A;
~ ~

(c) the functor M H M is an exact functor on the category of finitely

generated A -modules;
(d) A is a noetherian ring;
(e) if(Mn) is an inverse system, where each Mn is a.finitely generated Ajr-
module, each <pn·n: M n' ~ M n is surjective, and ker <pn'n = r M n', then M =
lim Mn is a .finitely generated A-module, and for each n, Mn ~ MjrM.
+---

193
II Schemes

PROOFS.
(a) Atiyah-Macdonald [1, p. 109].
(b) [Ibid., p. 108].
(c) [Ibid., p. 108].
(d) [Ibid., p. 113].
(e) Bourbaki [1, Ch. III, §2, no. 11, Prop. & Cor. 14].

Formal Schemes
We begin by defining the completion of a scheme along a closed subscheme.
For technical reasons we will limit our discussion to noetherian schemes.

Definition. Let X be a noetherian scheme, and let Y be a closed subscheme,

defined by a sheaf of ideals .f. Then we define the formal completion of X
along Y, denoted (X,(I)x ), to be the following ringed space. We take the
lim (I) xl.f". Here we
topological space Y, and on it the sheaf of rings (!) x = +--
consider each(!) xl.f" as a sheaf of rings on Y, and make them into an inverse
system in the natural way.

Remark 9.3.1. The structure sheaf (!Jx of X actually depends only on the
closed subset Y, and not on the particular scheme structure on Y. For iff
is another sheaf of ideals defining a closed subscheme structure on Y, then
since X is a noetherian scheme, there are integers m,n such that .f 2 ;m and
f 2 .f". Thus the inverse systems ((l)x/f") and ((l)x/fm) are cofinal with
each other, and hence have the same inverse limit.
One sees easily that the stalks of the sheaf (l)g are local rings, so in fact
(X,(!)x) is a locally ringed space. If U = Spec A is an open affine subset of X,
and if I s; A is the ideal r( U ,f), then from (9.2) we see that r(X n U ,(I) x) =
A, the /-adic completion ~fA. Thus the process of completing X along Y
is analogous to the /-adic completion of a ring discussed above. However,
one should note that the local rings of X are in general not complete, and
their dimension ( = dim X) is not equal to the dimension of the underlying
topological space Y.

Definition. With X, Y,.f as in the previous definition, let :F be a coherent

sheaf on X. We define the completion of :F along Y, denoted#, to be the
lim :FI.f" :F on Y. It has a natural structure of(!) x-module.
sheaf +---

Definition. A noetherian formal scheme is a locally ringed space (X,@l') which

has a finite open cover {UJ such that for each i, the pair (U;,(I).<IuJ is
isomorphic, as a locally ringed space, to the completion of some noetherian
scheme X; along a closed subscheme Y;. A morphism of noetherian formal
schemes is a morphism as locally ringed spaces. A sheaf 0: of @..-modules is
said to be coherent if there is a finite open cover U; as above, with U; ~ X;,
and for each i there is a coherent sheaf :F; on!; such that O:lu, ~ #;as
&x,-modules via the given isomorphism U; ~ X;.

194
 9 Formal Schemes

Examples 9.3.2. If X is any noetherian scheme, and Y a closed subscheme,

then its completion X is a formal scheme. Such a formal scheme, which can be
obtained by completing a single noetherian scheme along a closed subscheme,
is called algebraizable. It is not so easy to give examples, but there are
nonalgebraizable noetherian formal schemes-see Hironaka and Matsumura
[1, §5] or Hartshorne [5, V, 3.3, p. 205].

Example 9.3.3. If X is a noetherian scheme, and we take Y = X, then X = X.

Thus the category of noetherian formal schemes includes all noetherian
schemes.

Example 9.3.4. If X is a noetherian scheme, and Y is a closed point P, then

X is a one point space {P} with the completion@P of the local ring at Pas its
structure sheaf. An &rmodule M, considered as a sheaf on X, is coherent
if and only if M is a finitely generated module. Indeed, clearly coherent
implies finitely generated. But the converse is also true since we can obtain
X by completing the scheme Spec@Pat its closed point, and any finitely gener-
ated &rmodule M corresponds to a coherent sheaf on Spec @p.

Next we will study the structure of coherent sheaves on a formal scheme.

As in the study of coherent sheaves on usual schemes in §5, we first analyze
what happens in the affine case.

Definition. An affine (noetherian) formal scheme is a formal scheme obtained

by completing a single affine noetherian scheme along a closed subscheme.
If X = Spec A, Y = V(I), and X = X, then for any finitely generated
A-module M, we define the sheaf M6. on X to be the completion of the
coherent sheaf M on X. Thus by definition, M6. is a coherent sheaf on X.

Proposition 9.4. Let A be a noetherian ring, I an ideal of A, let X = Spec A,

Y = V(I), and let X = X. Then:
(a) 3 = J6. is a sheaf of ideals in (1) 30 , and for any n, (1),J:5" ~ (Ajl")-
as sheaves on Y;
(b) if M is a .finitely generated A-module, then Mf!,. = M ®@x mx.
(c) The functor M ~ M6. is an exact functor from the category of
finitely generated A-modules to the category of coherent mx-modules.
PROOF. In each case we have a statement about sheaves on X. Since the open
affine subsets of X form a base for the topology of X, and their intersections
with Y a base for the topology of Y, it will be sufficient to establish the cor-
responding property of the sections over any such open set. So let U =
Spec B be an open affine subset of X, let J = r(U,l), and for any finitely
generated A-module M, let N = r(U,M). Then B is a noetherian ring (3.2),
N is a finitely generated B-module (5.4), and the functor M ~ N is an exact
functor from A-modules to B-modules (5.5).

195
II Schemes

--
We prove (c) first. So let M be a finitely generated A-module. Then

---
Mt> = lim MjlnM by definition, so by (9.2), r(U,Mt>) = lim r(U,Mj[nM).
lim NjrN = N,
But this is equal to +-- ~ where~ now denotes the J-adic comple-
tion of a B-module. Now M t--> N is exact as we saw above, and N t--> N is
~

exact by (9.3A). Thus M t--> r(U,Mt>) is exact for each U, and so M t--> Mt>
is exact.
(a) For any U as above, r(U,It>) = lim r(U,lj[n) = J. Furthermore
r(U,(!)x) = B similarly. But by (9.3A), J i~ ideal of B, so this shows that
3 = It> is a sheaf of ideals in (1)_,.
Now we consider the exact sequence of A-modules
o~ r ~ A ~ Afr ~ o.
According to (c) which we have already proved, this gives an exact sequence
of {!}x-modules
0 ~ 3n ~ (!):li ~ (Ajr)l> ~ 0.

Observe that the inverse system which defines (Ajr)t> as the completion of
(Ajrf is eventually stationary, since this sheaf is annihilated by Jn. Hence
(Ajr)t> = (A/ IT, and we conclude that (!)x j3n ~ (Afrr as required.
(b) We have a slight abuse of notation in our statement: since M and (!)x
are sheaves on X, we should actually write Mt> ~ Mly ®(f!xiY (!)x· But we
will simply regard Mt> and (!)x as sheaves on X, by extending by zero outside
of Y (Ex. 1.19). For any finitely generated A module M, and for U an open
set as above, we have r(U,Mt>) = N as before. On the other hand, M ®(fix
(!)x is the sheaf associated to the presheaf
U t--> r(U,M) ®r(U,(f!x) r(U,(!)x) = N ®B B.
Since N ~ N <8> B B by (9.3A), we conclude that the corresponding sheaves
are isomorphic too: Mt> ~ M ®(fix (!)x·

Definition. Let (X,@ x) be a noetherian formal scheme. A sheaf of ideals 3 s; (!)x

is called an ideal of definition for X if Supp @_,/3 = X and the locally
ringed space (X,@x/3) is a noetherian scheme.

Proposition 9.5. Let (X,(!)x) be a noetherian formal scheme.

(a) If 3 1 and 3 2 are two ideals of definition, then there are integers
m,n > 0 such that 3 1 2 3'2 and 3 2 2 31.
(b) There is a unique largest ideal of definition 3, characterized by the
fact that (X,@x/3) is a reduced scheme. In particular, ideals of definition
exist.
(c) If3 is an ideal of definition, so is 3n,for any n > 0.
PROOF.
(a) Let 3 1 and 3 2 be two ideals of definition. Then on the topological
space X, we have surjective maps of sheaves of rings f 1 :(!)x ~ @x/3 1 and
f 2 :(!)f. ~ (9x/3 2 • For any point P EX, the stalk (3 2 )p of 3 2 at Pis contained

196
 9 Formal Schemes

in mp, the maximal ideal of the local ring {!)<,P· Indeed, 0.,,P/(3 2 )p is the local
ring of P on the scheme (I,<'~\/3 2 ). In particular, it is nonzero, so (3 2 )p s;
mp. Now we consider the sheaf of ideals f 1(3 2 ) on the scheme (I,<'9x/3d.
For each point P, its stalk is contained inside the maximal ideal of the local
ring. Hence every local section of f 1(3 2 ) is nilpotent (Ex. 2.18), and since
(I,<'9x/3d is a noetherian scheme, f 1(3 2 ) itself is nilpotent. This shows that
for some m > 0, 3 1 2 3~. The other way follows by symmetry.
(b) Suppose (I,<'9.{/3d is a reduced scheme. Then in the proof of (a), we
find f 1(:J 2 ) = 0, so 3 1 2 :J 2 . Thus such an :J 1 is largest, if it exists. Since
it is unique, the existence becomes a local question. Thus we may assume
that I is the completion of an affine noetherian scheme X along a closed
subscheme Y. By (9.3.1) we may assume that Y has the reduced induced
structure. Let X = Spec A, Y = V{l). Then by (9.4), :J = 16. is an ideal in
<'9 1 , and <'9 1 /:J ~(A/I)- = <'9y. Thus 3 is an ideal of definition for which
(I,0x/3) is reduced. This shows the existence of the largest ideal of de.finition.
(c) Let 3 be any ideal of definition, and suppose given n > 0. Let 3 0 be
the unique largest ideal of definition. Then by (a), there is an integer r such
that :J 2 3~, and hence :J" 2 :J0. First note that 3 0 is an ideal of defini-
tion. Indeed, this can be checked locally. If 3 0 = 16. on an affine, using the
notation of (b), then <'91 /:J() ~ (A/l"T by (9.4), so (I,<'9.d3 0 ) is a scheme
with support Y. Let's call this scheme Y', and let f:<'9x-+ <'9y· be the corre-
sponding map of sheaves. Then (Y',<'9r·l.f(3)) = (I/Dx/:3) is a noetherian
scheme, by hypothesis, so f(:J) is a coherent sheaf. Therefore f(:J") = f(:J)"
is also coherent, and we conclude that ( Y',<'9r·l.f(:J")) = (I,<'9.{/:J") is also a
noetherian scheme.

Proposition 9.6. Let I be a noetherian formal scheme and let :J be an ideal of

definition. For each n > 0 we denote by Yn the scheme (I,<'9x/3").
(a) If (j is a coherent sheaf of (():cmodules, then for each n, .?n = g;j:J"(j
is a coherent sheaf of @y"-modules, and (j ~ fu!! .?".
(b) Conversely, suppose given for each n a coherent <'9y"-module .?",
together with surjective maps CfJn·n: .?n' --> .?, for each n' ~ n, making {.?n}
into an inverse system of sheaves. Assume furthermore that for each n' ~ n,
ker CfJn·n = 3"-?n'· Then (j = fu!! .?n is a coherent (Ocmodule, and for
each n, .?n ~ !J/3"(j.
PROOF.
(a) The question is local, so we may assume that l is affine, equal to the
completion of X = Spec A along Y = V(I), and that (j = Mt::, for some
finitely generated A-module M. Then as in the proof of (9.4a) we see that
(jj:J"(j ~ (Mjl"M)- for each n. Thus .?n is coherent on Y. = Spec(A/1"),
and (J ~ fu!! .?n.
(b) Again the question is local, so we may assume that I is affine as
above. Furthermore, we may assume that A is /-adically complete, because
replacing A by A does not change X. For each n, let M. = T( Y,, .?.). Then

197
II Schemes

(Mn) is an inverse system of modules satisfying the hypotheses of (9.3Ae).

Therefore we conclude that M = !!!!! Mn is a finitely generated A-module

--
(since A is complete), and that for each n, Mn ~ MjrM. But then ~ =
lim :Fn is just M", hence it is a coherent (!)I -module. Furthermore ~~~n~ ~
(MWMr as in (a), so ~~~n~ ~ :Fn.

Theorem 9.7. Let A be a noetherian ring, I an ideal, and assume that A is

I -adically complete. Let X = Spec A, Y = V(l), and X = X. Then the
functors M ~ Mt; and ~ ~ r(X,~) are exact, and inverse to each other,
on the categories of finitely generated A -modules and coherent (!) :cmodules
respectively. Thus they establish an equivalence of categories. In par-
ticular, every coherent {!}x-module ~ is of the form Mt; for some M.

--
PROOF. We have already seen that M ~ Mt; is exact (9.4). If M is an A-
module of finite type, then F(X,Mt;) = lim MjrM = M, and M = M be-
cause A is complete (9.3Ab). Thus one composition of our two functors is
the identity.

--
Conversely, let ~ be a coherent (!Jrmodule, and let ~ = It;. Then by
(9.6a), ~ ~ lim :Fn, where for each n > 0, :Fn = ~~~n~. Now the inverse
system of sheaves (:Fn) satisfies the hypotheses of (9.6b), and the proof of

--
(9.6b) shows in fact that~ ~ Mt;, for some finitely generated A-module M.

---
Furthermore, by (9.2), r(X,~) = limr(Y,(MWMr) = limMjrM = M,
which is equal to M since A is complete. This shows that r(X,m is a finitely
generated A-module, and~ ~ r(X,~)t;. Thus the other composition of our
two functors is the identity.
It remains to show that the functor r(X,-) is exact on the category of
coherent @_.-modules. So let
0 --+ ~1 --+ ~2 --+ ~3 --+ 0
be an exact sequence of coherent {!}:~:-modules. For each i, let M; = r(X,~J
Then the M; are finitely generated A-modules, and we have at least a left-
exact sequence
0--+ M 1 --+ M 2 --+ M 3 .
Let R be the cokernel on the right. Then applying the functor t; we obtain
an exact sequence
0 --+ Mf --+ M~ --+ M~ --+ Rt; --+ 0
on X. But for each i, Mf ~ ~i as we saw above, so we conclude that Rt; =
0. But also by the above, R = r(X,Rt;), so R = 0. This shows that r(X, ·)
is exact, which concludes the proof.

Corollary 9.8. If X is any noetherian scheme, Y a closed subscheme, and

X = X the completion along Y, then the functor :F ~ .# is an exact functor
from coherent (!)x-modules to coherent (!)rmodules. Furthermore, if~ is
the sheaf of ideals of Y, and .J its completion, then we have .#;.Jn,# ~
:F/~n:F for each n, and.#~ :F ®(!lx (!)x·

198
 9 Formal Schemes

PROOF. These questions are all local, in which case they reduce to (9.4).

Corollary 9.9. Any kernel, cokernel, or image of a morphism of coherent

sheaves on a noetherian formal scheme is again coherent.
PROOF. These questions are also local, in which case they follow from (9. 7).

Remark 9.9.1. It is also true that an extension of coherent sheaves on a

noetherian formal scheme is coherent (Ex. 9.4). On the other hand, some
properties of coherent sheaves on usual schemes do not carry over to formal
schemes. For example, if X is the completion of a projective variety X <;;
P~ along a closed subvariety Y, and if lDx(l) = lDx(lt, then there may be
nonzero coherent sheaves !j on X such that r(X,!j(v)) = 0 for all v E Z. In
particular, no twist of !j is generated by global sections (III, Ex. 11. 7).

EXERCISES

9.1. Let X be a noetherian scheme, Y a closed subscheme, and X the completion of

X along Y. We call the ring r(X,(I)x) the ring of formal-regular functions on X
along Y. In this exercise we show that if Y is a connected, nonsingular, positive-
dimensional subvariety of X = P~ over an algebraically closed field k, then
r(X,(I)x) = k.
(a) Let § be the ideal sheaf of Y. Use (8.13) and (8.17) to show that there is an
inclusion of sheaves on Y, §j§ 2 4 (l)y(-1)"+ 1 .
(b) Show that for any r ;;:: 1, T(Y,§'j§'+ 1 ) = 0.
(c) Use the exact sequences

and induction on r to show that F(Y,(I)x/§') = k for all r ;;:: 1. (Use (8.21Ae).)
(d) Conclude that r(X,(I)x) = k. (Actually, the same result holds without the
hypothesis Y nonsingular, but the proof is more difficult-see Hartshorne
[3, (7.3)].)
9.2. Use the result of (Ex. 9.1) to prove the following geometric result. Let Y c;; X =
P~ be as above, and let f:X--> Z be a morphism of k-varieties. Suppose that
f(Y) is a single closed point P E Z. Then f(X) = P also.
9.3. Prove the analogue of (5.6) for formal schemes, which says, if :tis an affine formal
scheme, and if
0 --> ~' ..... ~ ..... ~" ..... 0

is an exact sequence of {l)x-modules, and if ~, is coherent, then the sequence of

global sections

is exact. For the proof, proceed in the following steps.

(a) Let 3 be an ideal of definition for :t, and for each n > 0 consider the exact
sequence
0 ..... ~'/3"~'--> ~/3"~'--> ~" ..... 0.

199
II Schemes

Use (5.6), slightly modified, to show that for every open affine subset U s; l:,
the sequence

is exact.
(b) Now pass to the limit, using (9.1), (9.2), and (9.6). Conclude that~ ~ !!!!! ~/3"~'
and that the sequence of global sections above is exact.
9.4. Use (Ex. 9.3) to prove that if

0 --> ~' --> ~ --> ~" --> 0

is an exact sequence of {9x-modules on a noetherian formal scheme l:, and if~',~"
are coherent, then~ is coherent also.
9.5. If~ is a coherent sheaf on a noetherian formal scheme l:, which can be generated
by global sections, show in fact that it can be generated by a finite number of its
global sections.
9.6. Let l: be a noetherian formal scheme, let 3 be an ideal of definition, and for each
n, let Y, be the scheme (l:,@x/3"). Assume that the inverse system of groups
(r(Y,,@y.)) satisfies the Mittag-Leffler condition. Then prove that Pic l: =
!!!!! Pic Y,. As in the case of a scheme, we define Pic l: to be the group of locally
free {9x-modules of rank 1 under the operation (8). Proceed in the following steps.
(a) Use the fact that ker(r(Y,+ 1,@yn+,) --> r(Y,,@y.)) is a nilpotent ideal to show
that the inverse system (r(Y,,@t)) of units in the respective rings also satisfies
(ML).
(b) Let ~be a coherent sheaf of @,-modules, and assume that for each n, there is
some isomorphism cp.: ~p·~ ~ @y n' Then show that there is an isomorphism
~-~ (!)_,. Be careful, because the <p. may not be compatible with the maps in
the two inverse systems m/3"m and ((l)yJ! Conclude that the natural map
Pic l: --> !!!!! Pic Y, is injective.
(c) Given an invertible sheaf!£'. on Y, for each n, and given isomorphisms!£'.+ 1 (8)
@y" ~ !£'.,construct maps!£' •. -->!£'.for each n' ~ n so as to make an inverse
system, and show that £ = !!!!! !£'. is a coherent sheaf on l:. Then show that
£ is locally free of rank 1, and thus conclude that the map Pic l: --> lim Pic Y,
is surjective. Again be careful, because even though each !£'. is locilly free of
rank 1, the open sets needed to make them free might get smaller and smaller
with n.
(d) Show that the hypothesis "(r(Y,,@y.)) satisfies (ML)" is satisfied if either l: is
affine, or each Y, is projective over a field k.
Note: See (III, Ex. 11.5-11. 7) for further examples and applications.

200
CHAPTER III

Cohomology

In this chapter we define the general notion of cohomology of a sheaf of

abelian groups on a topological space, and then study in detail the coho-
mology of coherent and quasi-coherent sheaves on a noetherian scheme.
Although the end result is usually the same, there are many different ways
of introducing cohomology. There are the fine resolutions often used in
several complex variables-see Gunning and Rossi [1]; the Cech coho-
mology used by Serre [3], who first introduced cohomology into abstract
algebraic geometry; the canonical flasque resolutions ofGodement [1]; and
the derived functor approach of Grothendieck [ 1]. Each is important in its
own way.
We will take as our basic definition the derived functors of the global
section functor (§1, 2). This definition is the most general, and also best
suited for theoretical questions, such as the proof of Serre duality in §7.
However, it is practically impossible to calculate, so we introduce Cech
cohomology in §4, and use it in §5 to compute explicitly the cohomology of
the sheaves llJ(n) on a projective space P'. This calculation is the basis of
many later results on projective varieties.
In order to prove that the Cech cohomology agrees with the derived
functor cohomology, we need to know that the higher cohomology of a
quasi-coherent sheaf on an affine scheme is zero. We prove this in §3 in the
noetherian case only, because it is technically much simpler than the case
of an arbitrary affine scheme ([EGA III, §1]). Hence we are bound to in-
clude noetherian hypotheses in all theorems involving cohomology.
As applications, we show for example that the arithmetic genus of a
projective variety X, whose definition in (I, §7) depended on a projective
embedding of X, can be computed in terms of the cohomology groups
Hi(X,lDx), and hence is intrinsic (Ex. 5.3). We also show that the arithmetic
genus is constant in a family of normal projective varieties (9.13).

201
III Cohomology

Another application is Zariski's main theorem (11.4) which is important

in the birational study of varieties.
The latter part of the chapter (§8-12) is devoted to families of schemes,
i.e., the study of the fibres of a morphism. In particular, we include a section
on flat morphisms and a section on smooth morphisms. While these can
be treated without cohomology, it seems to be an appropriate place to
include them, because flatness can be understood better using cohomology
(9.9).

1 Derived Functors

In this chapter we will assume familiarity with the basic techniques of

homological algebra. Since notation and terminology vary from one source
to another, we will assemble in this section (without proofs) the basic defini-
tions and results we will need. More details can be found in the following
sources: Godement [1, esp. Ch. I, §1.1-1.8, 2.1-2.4, 5.1-5.3], Hilton and
Stammbach [1, Ch. II,IV,IX], Grothendieck [1, Ch. II, §1,2,3], Cartan and
Eilenberg [1, Ch. III,V], Rotman [1, §6].

Definition. An abelian category is a category m:, such that: for each A,B E
Ob m:, Hom(A,B) has a structure of an abelian group, and the composi-
tion law is linear; finite direct sums exist; every morphism has a kernel
and a cokernel; every monomorphism is the kernel of its co kernel, every
epimorphism is the co kernel of its kernel; and finally, every morphism
can be factored into an epimorphism followed by a monomorphism.
(Hilton and Stammbach [1, p. 78].)

The following are all abelian categories.

Example 1.0.1. m:b, the category of abelian groups.

Example 1.0.2. 9Jlob(A), the category of modules over a ring A (commutative

with identity as always).

Example 1.0.3. m:b(X), the category of sheaves of abelian groups on a

topological space X.

Example 1.0.4. 9Jlob(X), the category of sheaves of CDx-modules on a ringed

space (X,CDx).

Example 1.0.5 . .Qco(X), the category of quasi-coherent sheaves of CDx-

modules on a scheme X (II, 5.7).

Example 1.0.6. (tof)(X), the category of coherent sheaves of CDx-modules on

a noetherian scheme X (II, 5. 7).

202
 1 Derived Functors

Example 1.0.7. <rol)(.I), the category of coherent sheaves of mx-modules on

a noetherian formal scheme (3:,@ 30 ) (II, 9.9).

In the rest of this section, we will be stating some basic results of homo-
logical algebra in the context of an arbitrary abelian category. However, in
most books, these results are proved only for the category of modules over
a r!ag, and proofs are often done by "diagram-chasing": you pick an element
and chase its images and pre-images through a diagram. Since diagram-
chasing doesn't make sense in an arbitrary abelian category, the conscientious
reader may be disturbed. There are at least three ways to handle this difficulty.
(1) Provide intrinsic proofs for all the results, starting from the axioms of an
abelian category, and without even mentioning an element. This is cumber-
some, but can be done-see, e.g., Freyd [ 1]. Or (2), note that in each of the
categories we use (most of which are in the above list of examples), one can
in fact carry out proofs by diagram-chasing. Or (3), accept the "full embed-
ding theorem" (Freyd [1, Ch. 7]), which states roughly that any abelian
category is equivalent to a subcategory of'llb. This implies that any category-
theoretic statement (e.g., the 5-lemma) which can be proved in 'llb (e.g., by
diagram-chasing) also holds in any abelian category.

Now we begin our review of homological algebra. A complex A. in an

abelian category 'll is a collection of objects Ai, i E Z, and morphisms
di:Ai--> Ai+t, such that di+ 1 o di = 0 for all i. If the objects A; are specified
only in a certain range, e.g., i ~ 0, then we set A; = 0 for all other i. A
morphism of complexes, f:A'-+ B. is a set of morphisms P:Ai-+ Bi for
each i, which commute with the coboundary maps di.
The ith cohomology object hi(A") of the complex A" is defined to be
ker d; jim d;- 1 . Iff: A· -+ B" is a morphism of complexes, then f induces a
natural map hi(f):hi(A")-+ hi(B"). If 0-+ A. -+ B" -+ C -+ 0 is a short
exact sequence of complexes, then there are natural maps bi: hi( C) -+ hi+ 1 (A")
giving rise to a long exact sequence

Two morphisms of complexes f,g:A· -+ B" are homotopic (written f ~ g)

if there is a collection of morphisms ki:Ai-+ Bi- 1 for each i (which need
not commute with the di) such that f - g = dk + kd. The collection of mor-
phisms, k = (ki) is called a homotopy operator. Iff ~ g, then f and g induce
the same morphism hi(A") -+ hi(B") on the cohomology objects, for each i.
A covariant functor F: 'll -+ mfrom one abelian category to another is
additive if for any two objects A,A' in 'll, the induced map Hom(A,A') -+
Hom(F A,FA') is a homomorphism of abelian groups. F is left exact if it is
additive and for every short exact sequence

0 -+ A' --> A -+ A" -+ 0

203
III Cohomology

in~. the sequence

0--+ FA'--+ FA--+ FA"

is exact in '.8. If we can write a 0 on the right instead of the left, we say F is
right exact. If it is both left and right exact, we say it is exact. If only the
middle part FA' --+ FA --+ FA" is exact, we say F is exact in the middle.
For a contravariant functor we make analogous definitions. For example,
F: ~ --+ '.8 is left exact if it is additive, and for every short exact sequence as
above, the sequence
0--+ FA"--+ FA--+ FA'
is exact in '.8.

Example 1.0.8. If~ is an abelian category, and A is a fixed object, then the
functor B --+ Hom(A,B), usually denoted Hom(A, · ), is a covariant left exact
functor from ~ to ~b. The functor Hom(· ,A) is a contravariant left exact
functor from ~ to ~b.

Next we come to resolutions and derived functors. An object I of~ is

injective if the functor Hom(· ,1) is exact. An injective resolution of an object
A of~ is a complex r, defined in degrees i ;;::: 0, together with a morphism
e:A --+ 1°, such that Ji is an injective object of~ for each i ;;::: 0, and such
that the sequence

is exact.
If every object of~ is isomorphic to a subobject of an injective object of
~'then we say~ has enough injectives. If~ has enough injectives, then every
object has an injective resolution. Furthermore, a well-known lemma states
that any two injective resolutions are homotopy equivalent.
Now let~ be an abelian category with enough injectives, and let F: ~ --+ '.8
be a covariant left exact functor. Then we construct the right derived functors
RiF, i ;;::: 0, ofF as follows. For each object A of~, choose once and for all
an injective resolution r of A. Then we define RiF(A) = hi(F(r) ).

Theorem l.lA. Let ~ be an abelian category with enough injectives, and let
F: ~ --+ '.8 be a covariant left exact functor to another abelian category '.8.
Then
(a) For each i ;;::: 0, RiF as defined above is an additive functor from~
to '.8. Furthermore, it is independent (up to natural isomorphism of functors)
of the choices of injective resolutions made.
(b) There is a natural isomorphism F ~ R 0 F.
(c) For each short exact sequence 0--+ A' --+A --+A" --+ 0 and for each
i;;::: 0 there is a natural morphism Ji:RiF(A")--+ Ri+ 1 F(A'), such that we
obtain a long exact sequence

... --+ R;F(A') --+ R;F(A) --+ R;F(A") ~ Ri+ 1 F(A') --+ Ri+ 1 F(A) --+ ...

204
 1 Derived Functors

(d) Given a morphism of the exact sequence of (c) to another 0 ~ B' ~

B ~ B" ~ 0, the 6's give a commutative diagram
R;F(A") ~ Ri+ 1 F(A')
! !
R;F(B") ~ Ri+ 1 F(B').
(e) For each injective object I of m:, and for each i > 0, we have
RiF(I) = 0.

Definition. With F:m: ~mas in the theorem, an object J of m: is acyclic for

F if RiF(J) = 0 for all i > 0.

Proposition 1.2A. With F: m: ~ m as in (l.lA), suppose there is an exact

sequence
0 ~ A ~ Jo ~ 11 ~ ...
where each f is acyclic for F, i ): 0. (We say J' is an F-acyclic resolution
of A.) Then for each i ): 0 there is a natural isomorphism RiF(A) ~
h;(F(J') ).

We leave to the reader the analogous definitions of projective objects,

projective resolutions, an abelian category having enough projectives, and
the left derived functors of a covariant right exact functor. Also, the right
derived functors of a left exact contravariant functor (use projective resolu-
tions) and the left derived functors of a right exact contravariant functor
(use injective resolutions).
Next we will give a universal property of derived functors. For this
purpose, we generalize slightly with the following definition.

Definition. Let m: and m be abelian categories. A (covariant) 6-functor from

m: tom is a collection of functors T = (Ti);"o' together with a morphism
6;: Ti(A") ~ yi+ 1 (A') for each short exact sequence 0 ~A'~ A~ A"~ 0,
and each i ): 0, such that:
(1) For each short exact sequence as above, there is a long exact sequence
0 ~ T 0 (A') ~ T 0 (A) ~ T 0 (A") ~ T 1 (A') ~ .. .
. . . ~ Ti(A) ~ Ti(A") ~ yi+ 1(A') ~ yi+ 1(A) ~ ... ;
(2) for each morphism of one short exact sequence (as above) into another
0 ~ B' ~ B ~ B" ~ 0, the 6's give a commutative diagram
Ti(A") ~ yi+ l(A')
! . !
Ti(B") ~ yi+ 1(B').

= (Ti):m: ~ m is said to be universal if, given

Definition. The 6-functor T
any other 6-functor T' = (T'i):m: ~ m, and given any morphism of

205
III Cohomology

functors f 0 : T 0 --+ T' 0 , there exists a unique sequence of morphisms

f;: T;--+ T'; for each i ~ 0, starting with the given f 0 , which commute
with the fi for each short exact sequence.

Remark 1.2.1. If F:21--+ lB is a covariant additive functor, then by definition

there can exist at most one (up to unique isomorphism) universal <5-functor
T with T 0 = F. If T exists, the T; are sometimes called the right satellite
functors of F.

Definition. An additive functor F: 21 --+ lB is effaceable if for each object A

of21, there is a monomorphism u:A--+ M, for some M, such that F(u) =
0. It is coeffaceable if for each A there exists an epimorphism u:P --+ A
such that F(u) = 0.

Theorem 1.3A. Let T = (T;);;.o be a covariant <5-functor from 21 to lB. If

T; is effaceable for each i > 0, then T is universal.
PROOF. Grothendieck (1, II, 2.2.1]

Corollary 1.4. Assume that 21 has enough injectives. Then for any left exact
functor F:21 --+ !8, the derived functors (R;F);;.o form a universal <5-functor
with F ~ R 0 F. Conversely, if T = (T;);;.o is any universal <5-functor,
then T 0 is left exact, and the T; are isomorphic to R;To for each i ~ 0.
PROOF. IfF is a left exact functor, then the (R;Fb 0 form a c5-functor by
(l.lA). Furthermore, for any object A, let u:A --+I be a monomorphism of
A into an injective. Then R;F(I) = 0 for i > 0 by (l.lA), so R;F(u) = 0.
Thus R;F is effaceable for each i > 0. It follows from the theorem that
(R;F) is universal.
On the other hand, given a universal <5-functor T, we have T 0 left exact
by the definition of <5-functor. Since 21 has enough injectives, the derived
functors R;To exist. We have just seen that (R;T 0 ) is another universal
c5-functor. Since R 0 T 0 = T 0 , we find R;To ~ T; for each i, by (1.2.1).

2 Cohomology of Sheaves

In this section we define cohomology of sheaves by taking the derived

functors of the global section functor. Then as an application of general
techniques of cohomology we prove Grothendieck's theorem about the
vanishing of cohomology on a noetherian topological space. To begin with,
we must verify that the categories we use have enough injectives.

Proposition 2.1A. If A is a ring, then every A-module is isomorphic to a sub-

module of an injective A-module.

PROOF. Godement [1, I, 1.2.2] or Hilton and Stammbach [1, I, 8.3].

206
 2 Cohomology of Sheaves

Proposition 2.2. Let (X,C9x) be a ringed space. Then the category Wlob(X)
of sheaves of C9x-modules has enough injectives.
PROOF. Let ff be a sheaf of C9x-modules. For each point x EX, the stalk
ffx is an mx,x-module. Therefore there is an injection ffx--+ IX, where IX is
an injective (9 x x-module (2.1A). For each point x, let j denote the inclusion
of the one-poi~t space {x} into X, and consider the sheaf f = Oxed*(JJ.
Here we consider Ix as a sheaf on the one-point space {x}, and j* is the
direct image functor (II, §1).
Now for any sheaf r§ of C9x-modules, we have Hom(l)x(r§,f) =
0 Hom(l)x(r§,j*(JJ) by definition of the direct product. On the other hand,
for each point x EX, we have Hom(l)x(r§,j*(IJ) ~ Hom(l)x,Jr§x,JJ as one
sees easily. Thus we conclude first that there is a natural morphism of
sheaves of C9x-modules ff--+ f obtained from the local maps ffx--+ Ix. It
is clearly injective. Second, the functor Hom(l)x( ·,f) is the direct product
over all x EX of the stalk functor r§ ~---+ r§ x• which is exact, followed by
Hom(l)x,X(·,Jx), which is exact, since IX is an injective mx,x-module. Hence
Hom(· ,f) is an exact functor, and therefore f is an injective C9x-module.

Corollary 2.3. If X is any topological space, then the category mb(X) of

sheaves of abelian groups on X has enough injectives.
PROOF. Indeed, if we let C9x be the constant sheaf of rings Z, then (X,C9x) is
a ringed space, and Wlob(X) = mb(X).

Definition. Let X be a topological space. Let r(X, ·) be the global section

functor from mb(X) to mb. We define the cohomology functors H;(X, ·)
to be the right derived functors of r(X,- ). For any sheaf ff, the groups
Hi(X,ff) are the cohomology groups of !F. Note that even if X and ff
have some additional structure, e.g., X a scheme and ff a quasi-coherent
sheaf, we always take cohomology in this sense, regarding ff simply as
a sheaf of abelian groups on the underlying topological space X.
We let the reader write out the long exact sequences which follow from
the general properties of derived functors (l.lA).

Recall (II, Ex. 1.16) that a sheaf ff on a topological space X is fiasque if

for every inclusion of open sets V s; U, the restriction map ff(U) --+ ff(V)
is surjective.

Lemma 2.4. If (X,C9x) is a ringed space, any injective C9x-module is fiasque.

PROOF. For any open subset U s; X, let C9u denote the sheaf NC9xlu), which
is the restriction of C9x to U, extended by zero outside U (II, Ex. 1.19). Now
let f be an injective C9x-module, and let V s; U be open sets. Then we
have an inclusion 0--+ C9v--+ C9u of sheaves of C9x-modules. Since f is injec-
tive, we get a surjection Hom(C9u.f)--+ Hom(C9v,f)--+ 0. But Hom(C9u,f) =
f(U) and Hom(C9v,f) = f(V), so f is flasque.

207
III Cohomology

Proposition 2.5. If :#' is a fiasque sheaf on a topological space X, then

H;(X,ff) = 0 for all i > 0.
PROOF. Embed:#' in an injective object§ of 2lb(X) and let<§ be the quotient:
0 ~ g; ~ § ~ <;§ ~ 0.
Then :#' is ftasque by hypothesis, § is flasque by (2.4), and so <§ is ftasque
by (II, Ex. 1.16c). Now since :#' is flasque, we have an exact sequence
(II, Ex. 1.16b)
0 ~ r(X,:F) ~ r(X,§) ~ r{X,<§) ~ 0.
On the other hand, since § is injective, we have H;(X,§) = 0 for i > 0
(l.lAe). Thus from the long exact sequence of cohomology, we get
H 1 (X,ff) = 0 and H;(X,ff) ~ H;- 1 (X,<§) for each i ;:::,: 2. But <§ is also
flasque, so by induction on i we get the result.

Remark 2.5.1. This result tells us that flasque sheaves are acyclic for the
functor r(X, · ). Hence we can calculate cohomology using ftasque resolu-
tions (1.2A). In particular, we have the following result.

Proposition 2.6. Let (X,@x) be a ringed space. Then the derived functors of
the functor r(X, ·) from 9Rob(X) to 2lb coincide with the cohomology
functors Hi( X,·).
PROOF. Considering r{X, ·) as a functor from 9Rob(X) to 2lb, we calculate
its derived functors by taking injective resolutions in the category 9Rob(X).
But any injective is flasque (2.4), and flasques are acyclic (2.5) so this resolu-
tion gives the usual cohomology functors (1.2A).

Remark 2.6.1. Let (X,@x) be a ringed space, and let A = r{X,@x). Then
for any sheaf of @x-modules :#', r(X,:F) has a natural structure of A-module.
In particular, since we can calculate cohomology using resolutions in the
category 9Rob(X), all the cohomology groups of:#' have a natural structure
of A-module; the associated exact sequences are sequences of A-modules,
and so forth. Thus for example, if X is a scheme over Spec B for some ring
B, the cohomology groups of any @x-module :#'have a natural structure of
B-module.

A Vanishing Theorem ofGrothendieck

Theorem 2. 7 (Grothendieck [ 1] ). Let X be a noetherian topological space of
dimension n. Then for all i > n and all sheaves of abelian groups :#' on
X, we have W(X,:F) = 0.

Before proving the theorem, we need some preliminary results, mainly

concerning direct limits. If (!Fa) is a direct system of sheaves on X, indexed
by a directed set A, then we have defined the direct limit lim
---+
:Fa (II, Ex. 1.10).

208
 2 Cohomology of Sheaves

Lemma 2.8. On a noetherian topological space, a direct limit of jlasque

sheaves is fiasque.
PROOF. Let (ffJ be a directed system of flasque sheaves. Then for any
inclusion of open sets V £ U, and for each ex, we have ffi;.(U)--+ ffi;.(V) is
____. is an exact functor, we get
surjective. Since lim
____. ffa(V)
____. ffa(U)--+ lim
lim
is also surjective. But on a noetherian topological space, lim ffa(U) =
(!i!n ffa)(U) for any open set (II, Ex. 1.11).
~

So we have
____. ff a)( U)
(lim --+ ____. ffa)( V)
(lim
____. %;. is flasque.
is surjective, and so lim

Proposition 2.9. Let X be a noetherian topological space, and let (ffa) be a

direct system of abelian sheaves. Then there are natural isomorphisms,
for each i ;?: 0

PROOF. For each ex we have a natural map ffa --+ lim ffa. This induces a
map on cohomology, and then we take the direct limit of these maps. For
i = 0, the result is already known (II, Ex. 1.11). For the general case, we
consider the category inb A (mb(X)) consisting of all directed systems of
objects of m:b(X), indexed by A. This is an abelian category. Furthermore,
____. is an exact functor, we have a natural transformation of c:5-functors
since lim
____. Hi( X,·)
lim --+ ____. ·)
H;(X,lim
from inb A (mb(X)) to m:b. They agree for i = 0, so to prove they are the
same, it will be sufficient to show they are both effaceable for i > 0. For
in that case, they are both universal by (1.3A), and so must be isomorphic.
So let (ffa) E inb A(mb(X) ). For each ex, let '!J a be the sheaf of discon-
tinuous sections of ffa (II, Ex. 1.16e). Then '!Ja is flasque, and there is a
natural inclusion ffa --+ '!J a· Furthermore, the construction of '!J a is func-
torial, so the '!J a also form a direct system, and we obtain a monomorphism
u: (ffa) --+ ('!J a) in the category inb A(mb(X) ). Now the '!J a are all fiasque, so
H;(X,'!Ja) = 0 for i > 0 (2.5). Thus !!!n H;(X,'!Ja) = 0, and the functor on
the left-hand side is effaceable
. fori > 0. On the other hand, lim
____. '!Ja is also
flasque by (2.8). So H'(X,lim ____. '!Ja) = 0 fori > 0, and we see that the functor
on the right-hand side is also effaceable. This completes the proof.

Remark 2.9.1. As a special case we see that cohomology commutes with

infinite direct sums.

Lemma 2.10. Let Y be a closed subset of X, let ff be a sheaf of abelian

groups on Y, and let j: Y--+ X be the inclusion. Then H;(Y,ff) = H;(X,j*ff),
where j*ff is the extension of ff by zero outside Y (II, Ex. 1.19).

209
III Cohomology

PROOF. If/. is a fl.asque resolution of ff on Y, then j*/. is a fl.asque res-

olution of j*ff on X, and for each i, T(Y,/;) = T(X,j*/;). So we get the
same cohomology groups.

Remark 2.10.1. Continuing our earlier abuse of notation (II, Ex. 1.19), we
often write ff instead ofj*ff. This lemma shows there will be no ambiguity
about the cohomology groups.
PROOF OF (2.7). First we fix some notation. If Y is a closed subset of X,
then for any sheaf ff on X we let ff y = j *(,~ IY ), where j: Y -+ X is the
inclusion. If U is an open subset of X, we let ff u = i.(fflu), where i: U -+
X is the inclusion. In particular, if U = X - Y, we have an exact sequence
(II, Ex. 1.19)
0-+ ffu-+ ff-+ !i'y-+ 0.
We will prove the theorem by induction on n = dim X, in several steps.
Step 1. Reduction to the case X irreducible. If X is reducible, let Y be
one of its irreducible components, and let U = X - Y. Then for any ff
we have an exact sequence
0-+ ffu-+ ff-+ :#'y-+ 0.
From the long exact sequence of cohomology, it will be sufficient to prove
that Hi(X,!i'y) = 0 and H;(X,ffu) = 0 for i > n. But Y is closed and
irreducible, and ff u can be regarded as a sheaf on the closed subset a,
which has one fewer irreducible components than X. Thus using (2.10) and
induction on the number of irreducible components, we reduce to the case
X irreducible.
Step . Suppose X is irreducible of dimension 0. Then the only open
subsets of X are X and the empty set. For otherwise, X would have a
proper irreducible closed subset, and dim X would be ~ 1. Thus r(X, ·)
induces an equivalence of categories ~b(X) -+ ~b. In particular, r(X, ·)
is an exact functor, so H;(X,ff) = 0 for i > 0, and for all ff.
Step 3. Now let X be irreducible of dimension n, and let ff E ~b(X).
Let B = Uus:xff'(U), and let A be the set of all finite subsets of B. For
each a E A, let ffa be the subsheaf of:#' generated by the sections in a (over
various open sets). Then A is a directed set, and :#' = lim ffa. So by (2.9),
--->
it will be sufficient to prove vanishing of cohomology for each ffa. If a'
is a subset of a, then we have an exact sequence
0 -+ ff'a' -+ ffa -+ '1J -+ 0,
where '1J is a sheaf generated by # (a - a') sections over suitable open sets.
Thus, using the long exact sequence of cohomology, and induction on
#(a), we reduce to the case that ff is generated by a single section over
some open set U. In that case :#' is a quotient of the sheaf Zu (where Z
denotes the constant sheaf Z on X). Letting !!It be the kernel, we have an
exact sequence

210
 2 Cohomology of Sheaves

Again using the long exact sequence of cohomology, it will be sufficient to

prove vanishing for fJl and for Zu.
Step 4. Let U be an open subset of X and let fJl be a subsheaf of Zu.
For each x E U, the stalk f!llx is a subgroup of Z. If f!ll = 0, skip to Step 5.
If not, let d be the least positive integer which occurs in any of the groups f!llx.
Then there is a nonempty open subset V s:.:: U such that f!lllv ~ d · Zlv as a
subsheaf of Zlv· Thus f!llv ~ Zv and we have an exact sequence

0 ~ Zv ~ fJl ~ f!ll /Zv ~ 0.

Now the sheaf f!ll/Zv is supported on the closed subset (U - V)- of X,

which has dimension < n, since X is irreducible. So using (2.1 0) and the
induction hypothesis, we know Hi(X,f!ll/Zv) = 0 for i ~ n. So by the
long exact sequence of cohomology, we need only show vanishing for Zv.
Step 5. To complete the proof, we need only show that for any open
subset U s:.:: X, we have H;(X,Zu) = 0 fori> n. Let Y =X - U. Then
we have an exact sequence

0 ~ Zu ~ Z ~ Zy ~ 0.

Now dim Y < dim X since X is irreducible, so using (2.10) and the in-
duction hypothesis, we have Hi(X,Zy) = 0 for i ;:::;: n. On the other hand,
Z is fiasque, since it is a constant sheaf on an irreducible space (II, Ex. 1.16a).
Hence Hi(X,Z) = 0 for i > 0 by (2.5). So from the long exact sequence
of cohomology we have H;(X,Zu) = 0 fori > n. q.e.d.

Historical Note: The derived functor cohomology which we defined in

this section was introduced by Grothendieck [1]. It is the theory which is
used in [EGA]. The use of sheaf cohomology in algebraic geometry started
with Serre [3]. In that paper, and in the later paper [ 4], Serre used Cech
cohomology for coherent sheaves on an algebraic variety with its Zariski ·
topology. The equivalence of this theory with the derived functor theory
follows from the "theorem of Leray" (Ex. 4.11). The same argument, using
Cartan's "Theorem B" shows that the Cech cohomology of a coherent
analytic sheaf on a complex analytic space is equal to the derived functor
cohomology. Gunning and Rossi [1] use a cohomology theory computed
by fine resolutions of a sheaf on a paracompact Hausdorff space. The
equivalence of this theory with ours is shown by Godement [1, Thm. 4.7.1,
p. 181 and Ex. 7.2.1, p. 263], who shows at the same time that both theories
coincide with his theory which is defined by a canonical flasque resolution.
Godement also shows [1, Thm. 5.10.1, p. 228] that on a paracompact
Hausdorff space, his theory coincides with Cech cohomology. This provides
a bridge to the standard topological theories with constant coefficients, as
developed in the book of Spanier [1]. He shows that on a paracompact
Hausdorff space, Cech cohomology and Alexander cohomology and singular
cohomology all agree (see Spanier [1, pp. 314, 327, 334]).

211
III Cohomology

The vanishing theorem (2. 7) was proved by Serre [3] for coherent sheaves
on algebraic curves and projective algebraic varieties, and later [5] for
abstract algebraic varieties. It is analogous to the theorem that singular
cohomology on a (real) manifold of dimension n vanishes in degrees i > n.

ExERCISES

2.1. (a) Let X = Al be the affine line over an infinite field k. Let P,Q be distinct closed
points of X, and let U =X- [P,Q]. Show that H 1(X,Zu) i= 0.
*(b) More generally, let Y ~X = A~ be the union of n + 1 hyperplanes in suit-
ably general position, and let U = X - Y. Show that H"(X,Zu) i= 0. Thus the
result of (2. 7) is the best possible.

2.2. Let X = P~ be the projective line over an algebraically closed field k. Show that
the exact sequence 0---> (!' ---> .X---> ff/C!' ---> 0 of (II, Ex. 1.21d) is a ftasque res-
olution of(!. Conclude from (II, Ex. 1.21e) that Hi(X,{!!) = 0 for all i > 0.

2.3. Cohomology with Supports (Grothendieck [7]). Let X be a topological space, let
Y be a closed subset, and let::¥' be a sheaf of abelian groups. Let r y(X,ff) denote
the group of sections of ff with support in Y (II, Ex. 1.20).
(a) Show that r y(X, ·)is a left exact functor from ~b(X) to ~b.
We denote the right derived functors of r y(X, ·) by H~(X, · ). They are the
cohomology groups of X with supports in Y, and coefficients in a given sheaf.
(b) If 0---> 5'---> 5---> .'F"---> 0 is an exact sequence of sheaves, with .?' ftasque,
show that

is exact.
(c) Show that if::¥' is ftasque, then H~(X,ff) = 0 for all i > 0.
(d) If ff is ftasque, show that the sequence

0---> r r(X,ff)---> T(X,ff)---> r(X - Y,ff)---> 0

is exact.
(e) Let U = X - Y. Show that for any ff, there is a long exact sequence of
cohomology groups

0---> H~(X,ff)---> H 0 (X,ff)---> H 0 (U,fflul--->

---> m(X,ff) ---> H 1(X,ff) ---> H 1(U,fflul --->
---> H~(X,ff) ---> ....

(f) Excision. Let V be an open subset of X containing Y. Then there are natural
functorial isomorphisms, for all i and ff,

2.4. Muyer-Vietoris Seque11ce. Let Y1, Y2 be two closed subsets of X. Then there is a
long exact sequence of cohomology with supports

... ---> HL n y2(X,ff)---> H~JX,ff) EB HL(X,ff) ---> H~ 1 u y2 (X,ff) --->

---> H~~ ~ r,(X,ff) ---> ....

212
 3 Cohomology of a Noetherian Affine Scheme

2.5. Let X be a Zariski space (II, Ex. 3.17). Let P E X be a closed point, and let X P
be the subset of X consisting of all points Q E X such that P E { Q}-. We call X P
the local space of X at P, and give it the induced topology. Let j: X P ---+ X be the
inclusion, and for any sheaf$' on X, let $'p = j* $'. Show that for all i, $', we
have
H~(X,$') = H~(X p,$ip).

2.6. Let X be a noetherian topological space, and let {J,},eA be a direct system of
injective sheaves of abelian groups on X. Then ~ J. is also injective. [Hints:
First show that a sheaf J is injective if and only if for every open set U ~ X, and
for every subsheaf !1lt ~ Zu, and for every map f:!Jlt---+ J, there exists an ex-
tension off to a map of Zu ---+ J. Secondly, show that any such sheaf !1lt is finitely
generated, so any map !1lt ---+ ~ J, factors through one of the J,.]
2.7. Let S 1 be the circle (with its usual topology), and let Z be the constant sheaf Z.
(a) Show that H 1 (S 1 ,Z) ~ Z, using our definition of cohomology.
(b) Now let !1lt be the sheaf of germs of continuous real-valued functions on S 1 .
Show that H 1 (S 1 ,!Jlt) = 0.

3 Cohomology of a Noetherian Affine Scheme

In this section we will prove that if X = Spec A is a noetherian affine

scheme, then Hi(X,g;) = 0 for all i > 0 and all quasi-coherent sheaves g; of
CDx-modules. The key point is to show that if I is an injective A-module,
then the sheaf I on Spec A is flasque. We begin with some algebraic
preliminaries.

Proposition 3.1A (Krull's Theorem). Let A be a noetherian ring, let M ~ N

be finitely generated A-modules, and let a be an ideal of A. Then the
a-adic topology on M is induced by the a-adic topology on N. In particular,
for any n > 0, there exists an n' ?: n such that an M 2 M n an' N.
PROOF. Atiyah-Macdonald [1, 10.11] or Zariski-Samuel [1, vol. II, Ch. VIII,
Th. 4].

Recall (II, Ex. 5.6) that for any ring A, and any ideal a ~ A, and any
A-module M, we have defined the submodule F 0 (M) to be {mE Mlanm = 0
for some n > 0}.

Lemma 3.2. Let A be a noetherian ring, let a be an ideal of A, and let I be an

injective A -module. Then the submodule J = ra (I) is also an injective
A-module.
PROOF. To show that J is injective, it will be sufficient to show that for any
ideal b ~ A, and for any homomorphism cp: b -+ J, there exists a homo-
morphism l/1: A -+ J extending cp. (This is a well-known criterion for an
injective module-Godement [1, I, 1.4.1 ]). Since A is noetherian, b is finitely
generated. On the other hand, every element of J is annihilated by some

213
III Cohomology

power of a, so there exists an n > 0 such that ancp(b) = 0, or equivalently,

cp(anb) = 0. Now applying (3.1A) to the inclusion b <;;A, we find that there
is an n' ~ n such that anb 2 b n an'. Hence cp(b n an') = 0, and so the
map cp:b---+ J factors through b/(b nan} Now we consider the following
diagram:

J
<p

Since I is injective, the composed map of b/(b n an') to I extends to a map

1/J': A/an' ---+ I. But the image of 1/J' is annihilated by an', so it is contained in
J. Composing with the natural map A ---+ Ajan', we obtain the required map
1/J: A ---+ J extending <p.

Lemma 3.3. Let I be an injective module over a noetherian ring A. Then for
any f E A, the natural map of I to its localization I I is surjective.
PROOF. For each i > 0, let b; be the annihilator of fi in A. Then b 1 <;; b 2 <;;
... , and since A is noetherian, there is an r such that b, = b,+ 1 = .... Now
let e: I ---+ I J be the natural map, and let X E I J be any element. Then by
definition of localization, there is ayE I and ann ~ 0 such that x = 8(y)fr.
We define a map cp from the ideal (r+') of A to I by sending r+r to f'y.
This is possible, because the annihilator of r+r is bn+r = b, and b, anni-
hilates f'y. Since I is injective, <p extends to a map 1/f:A ---+I. Let 1/J(l) = z.
Then fn+rz = f'y. But this implies that 8(z) = 8(y)/r = x. Hence 8 is
surjective.

Proposition 3.4. Let I be an injective module over a noetherian ring A. Then

the sheaf 1 on X = Spec A is fiasque.

PROOF. We will use noetherian induction on Y = (Supp 1)-. See (II, Ex.
1.14) for the notion of support. If Y consists of a single closed point of X,
then 1 is a skyscraper sheaf (II, Ex. 1.17) which is obviously fiasque.
In the general case, to show that 1 is fiasque, it will be sufficient to show,
for any open set U <;; X, that T(X,l)---+ T(U,l) is surjective. If Y n U = 0,
there is nothing to prove. If Y n U =f. 0, we can find an f E A such that
the open set X I = D(f) (II, §2) is contained in U and X I n Y =f. 0. Let
Z = X - X I• and consider the following diagram:

T(X,l) ---+ F(U,l) ---+ T(X I})

j' j'

r z(X}) ---+ r z(U}),

214
 3 Cohomology of a Noetherian Affine Scheme

where r z denotes sections with support in Z (II, Ex. 1.20). Now given a
section s E r( u,l), we consider its image s~ in r(X 1 ,1). But r(X 1 ,1) = If
(II, 5.1), so by (3.3), there is a t E I = r(X,I) restricting to s'. Let t' be the
restriction oft to r(U,l). Then s - t' goes to 0 in r(X 1 ,1), so it has support
in Z. Thus to complete the proof, it will be sufficient to show that r z(X,l) -+
r z( u,l) is surjective.
Let J = r z(X,l). If a is the ideal generated by f, then J = r.(l) (II,
Ex. 5.6), so by (3.2), J is also an injective A-module. Furthermore, the
support of J is contained in Y n Z, which is strictly smaller than Y Hence
by our induction hypothesis, J is fiasque. Since r(U,J) = Tz(U}) (II,
Ex. 5.6), we conclude that r z(X,l) -+ r z( U,l) is surjective, as required.

Theorem 3.5. Let X = Spec A be the spectrum of a noetherian ring A. Then

for all quasi-coherent sheaves :#' on X, and for all i > 0, we have
H;(X,ff) = 0.
PROOF. Given :#', let M = r(X,ff), and take an injective resolution 0 -+
M -+ r
of M in the category of A-modules. Then we obtain an exact
sequence of sheaves 0 -+ M -+ T on X. Now :#' = M (II, 5.5) and each I;
is fiasque by (3.4), so we can use this resolution of:#' to calculate cohomology
(2.5.1). Applying the functor r, we recover the exact sequence of A-modules
0-+ M-+ r. Hence H 0 (X,ff) = M, and H;(X,ff) = 0 fori > 0.

Remark 3.5.1. This result is also true without the noetherian hypothesis, but
the proof is more difficult [EGA III, 1.3.1].

Corollary 3.6. Let X be a noetherian scheme, and let :#' be a quasi-coherent

sheaf on X. Then :#' can be embedded in a flasque, quasi-coherent sheaf '!J.
PROOF. Cover X with a finite number of open affines U; = Spec A;, and let
:Flu, M; for each i. Embed M; in an injective Acmodule I;. For each i,
=
let f: U; -+X be the inclusion, and let '!J = ffif*(I;). For each i we have
an injective map of sheaves :Flu, -+ I;. Hence we obtain a map :#' -+ f*(I;).
Taking the direct sum over i gives a map :#' -+ '!J which is clearly injective.
On the other hand, for each i, I; is fiasque (3.4) and quasi-coherent on U;.
Hence f*(l;) is also fiasque (II, Ex. 1.16d) and quasi-coherent (II, 5.8). Taking
the direct sum of these, we see that '!J is fiasque and quasi-coherent.

Theorem 3.7 (Serre [5]). Let X be a noetherian scheme. Then the following
conditions are equivalent:
(i) X is affine;
(ii) H;(X,ff) = 0 for all:#' quasi-coherent and all i > 0;
(iii) H 1(X,J) = 0 for all coherent sheaves of ideals J.
PROOF. (i) = (ii) =
is (3.5). (ii) (iii) is trivial, so we have only to prove
(iii) = (i). We use the criterion of (II, Ex. 2.17). First we show that X can

215
III Cohomology

be covered by open affine subsets of the form X 1 , with f E A = T(X,0x).

Let P be a closed point of X, let U be an open affine neighborhood of P,
and let Y = X - U. Then we have an exact sequence
0 -+ J Yu{PJ -+ J y -+ k(P) -+ 0,
where J y and J Yu{PJ are the ideal sheaves of the closed sets Y and Y u { P},
respectively. The quotient is the skyscraper sheaf k(P) = 0pjmp at P. Now
from the exact sequence of cohomology, and hypothesis (iii), we get an
exact sequence
T(X,J y) -+ T(X,k(P)) -+ H 1(X,J Yu{P)) = 0.
So there is an element f E F(X ,J y) which goes to 1 in k(P), i.e., fp 1=
(mod mp). Since J y s;: (0 X• we can consider f as an element of A. Then by
construction, we have P EX f s;: U. Furthermore, X f = U 1 , where J is
the image off in T(U,0u), so X f is affine.
Thus every closed point of X has an open affine neighborhood of the
form X J· By quasi-compactness, we can cover X with a finite number of
these, corresponding to f 1, ... ,f.. E A.
Now by (II, Ex. 2.17), to show that X is affine, we need only verify that
j 1 , . . . ,f.. generate the unit ideal in A. We use f 1 , . . . ,f.. to define a map
()(:(0~-+ (Ox by sending (a 1 , . . . ,a,) to "[j;a;. Since the Xf, cover X, this is
a surjective map of sheaves. Let :F be the kernel:
0 -+ :F -+ (0~ ~ (Ox -+ 0.
We filter :F as follows:

for a suitable ordering of the factors of @~. Each of the quotients of this
filtration is a coherent sheaf of ideals in 0x. Thus using our hypothesis (iii)
and the long exact sequence of cohomology, we climb up the filtration and
deduce that H 1 (X,:F) = 0. But then r(X,(O~) ~ T(X,0x) is surjective,
which tells us that f 1 , . . . ,f.. generate the unit ideal in A. q.e.d.

Remark 3.7.1. This result is analogous to another theorem of Serre in

complex analytic geometry, which characterizes Stein spaces by the vanishing
of coherent analytic sheaf cohomology.

ExERCISEs
3.1. Let X be a noetherian scheme. Show that X is affine if and only if X ced (II, Ex. 2.3)
is affine. [Hint: Use (3.7), and for any coherent sheaf ff on X, consider the filtra-
tion ff 2 .AI · ff 2 % 2 · ff 2 ... , where .AI is the sheaf of nilpotent elements
on X.]
3.2. Let X be a reduced noetherian scheme. Show that X is affine if and only if each
irreducible component is affine.

216
 3 Cohomology of a Noetherian Affine Scheme

3.3. Let A be a noetherian ring, and let a be an ideal of A.

(a) Show that r;, (·)(II, Ex. 5.6) is a left-exact functor from the category of A-modules
to itself. We denote its right derived functors, calculated in 9Jlob(A), by H;( · ).
(b) Now let X = Spec A, Y = V(a). Show that for any A-module M,
H~(M) = HHX,M),
where H~(X, ·) denotes cohomology with supports in Y (Ex. 2.3).
(c) For any i, show that r; (H,~(M)) = H,~(M).

3.4. Cohomological Interpretation of Depth. If A is a ring, a an ideal, and M an A-

module, then depth, M is the maximum length of an M -regular sequence x 1, ..• ,x,
with all xi E a. This generalizes the notion of depth introduced in (II, §8).
(a) Assume that A is noetherian. Show that if depth, M ;;, 1, then r; (M) = 0,
and the converse is true if M is finitely generated. [Hint: When M is finitely
generated, both conditions are equivalent to saying that a is not contained in
any associated prime of M.]
(b) Show inductively, for M finitely generated, that for any n ;;, 0, the following
conditions are equivalent:
(i) depth, M ;;, n;
(ii) H!(M) = 0 for all i < n.
For more details, and related results, see Grothendieck [7].

3.5. Let X be a noetherian scheme, and let P be a closed point of X. Show that the
following conditions are equivalent:
(i) depth (!JP ;;, 2;
(ii) if U is any open neighborhood of P, then every section of (!Jx over U - P
extends uniquely to a section of (!Jx over U.
This generalizes (I, Ex. 3.20), in view of (II, 8.22A).

3.6. Let X be a noetherian scheme.

(a) Show that the sheaf':§ constructed in the proof of (3.6) is an injective object in
the category .Oco(X) of quasi-coherent sheaves on X. Thus .Oco(X) has enough
injectives.
*(b) Show that any injective object of .Oco(X) is flasque. [Hints: The method of
proof of (2.4) will not work, because (!Ju is not quasi-coherent on X in general.
Instead, use (II, Ex. 5.15) to show that if J E .Oco(X) is injective, and if U ~ X
is an open subset, then Jlu is an injective object of .Qco(U). Then cover X
with open affines ... ]
(c) Conclude that one can compute cohomology as the derived functors of r(X, · ),
considered as a functor from .Oco(X) to tlb.

3.7. Let A be a noetherian ring, let X= Spec A, let a~ A be an ideal, and let U ~X
be the open set X - V(a).
(a) For any A-module M, establish the following formula of Deligne:
T(U,M) ~ lim HomA(a",M).
---->
n

(b) Apply this in the case of an injective A-module I, to give another proof of(3.4).

217
III Cohomology

3.8. Without the noetherian hypothesis, (3.3) and (3.4) are false. Let A= k[x 0 ,x 1 ,x 2 , ••• ]
with the relations x~x. = 0 for n = 1, 2, .... Let I be an injective A-module con-
taining A. Show that I -+ I xo is not surjective.

4 Cech Cohomology

In this section we construct the Cech cohomology groups for a sheaf of

abelian groups on a topological space X, with respect to a given open
covering of X. We will prove that if X is a noetherian separated scheme,
the sheaf is quasi-coherent, and the covering is an open affine covering,
then these Cech cohomology groups coincide with the cohomology groups
defined in §2. The value of this result is that it gives a practical method for
computing cohomology of quasi-coherent sheaves on a scheme.
Let X be a topological space, and let U = (U ;); e 1 be an open covering
of X. Fix, once and for all, a well-ordering of the index set I. For any
finite set of indices io, ... ,ip E I we denote the intersection uio n ... n uip
by uio, ... . ip"
Now let $' be a sheaf of abelian groups on X. We define a complex
C"(U,ff') of abelian groups as follows. For each p ): 0, let

cP(U,ff') = fl
io< .. . <.ip
ff'(U; 0 , • . . ,iJ

Thus an element IX E CP(U,ff') is determined by giving an element

for each (p + 1)-tuple i 0 < ... < iP of elements of I. We define the co-
boundary map d: CP--+ cp+ 1 by setting
p+1
(diX)·lQ, · · · ,1p
· +1 = ".i...J ( -1)k1X·lQ, ~ ·
· · · ,1k, · · · ,lp +1 lu io, .. ,lp+ 1
k=O

Here the notation ~ means omit ik. Then since IX;0 , . . . • h .... ,ip +1 is an ele-
ment of ff'(U; 0 , . . . • h .... ,ip+J, we restrict to U; 0 , •.. ,ip+l to get an element
of ff'(U; 0 , . . . ,ip+l). One checks easily that d2 = 0, so we have indeed de-
fined a complex of abelian groups.

Remark 4.0.1. If IX E CP(U,ff'), it is sometimes convenient to have the symbol

1X;0 , . . • • ip defined for all (p + 1)-tuples of elements of I. If there is a re-
peated index in the set {i 0 , ..• ,iP}, we define 1X;0 , . . . ,ip = 0. If the indices
are all distinct, we define IX;0 , . . . ,ip = ( -1)u1Xuio, ... ,uip' where (J is the per-
mutation for which (Ji 0 < ... < (JiP" With these conventions, one can
check that the formula given above for diX remains correct for any (p + 2)-
tuple i 0, . .. ,ip+ 1 of elements of I.

218
 4 Cech Cohomology

Definition. Let X be a topological space and let U be an open covering of

X. For any sheaf of abelian groups ff on X, we define the pth Cech
cohomology group of ff, with respect to the covering U, to be
HP(U,ff) = hP( C'(U,ff) ).

Caution 4.0.2. Keeping X and U fixed, if 0 -+ ff' -+ ff -+ ff" -+ 0 is a

short exact sequence of sheaves of abelian groups on X, we do not in general
get a long exact sequence of Cech cohomology groups. In other words,
the functors JlP(U,-) do not form a <5-functor (§1). For example, ifU consists
of the single open set X, then this results from the fact that the global section
functor r(X, ·) is not exact.

Example 4.0.3. To illustrate how well suited Cech cohomology is for com-
putations, we will compute some examples. Let X = Pf, let ff be the sheaf
of differentials Q (II, §8), and let U be the open covering by the two open
sets U = A1 with affine coordinate x, and V = A 1 with affine coordinate
y = 1/x. Then the Cech complex has only two terms:
C0 = r(U,Q) X r(V,Q)
C1 = F(U n V,Q).
Now
r(U,Q) = k[x] dx
r(V,Q) = k[y] dy

r( U n V,Q) = k [ x, ~ Jdx,
and the map d: C 0 -+ C 1 is given by

So ker dis the set of pairs <f(x)dx,g(y)dy) such that

f(x) = _ __!__2 g (~).

x x
This can happen only iff = g = 0, since one side is a polynomial in x and
the other side is a polynomial in 1/x with no constant term. So H0 (U,Q) = 0.
To compute H 1, note that the image of d is the set of all expressions

219
III Cohomology

where f and g are polynomials. This gives the subvector space of k[ x,ljx] dx
generated by all xn dx, n E Z, n =!= -1. Therefore H1(U,Q) ~ k, generated by
the image of x- 1 dx.

Example 4.0.4. Let S 1 be the circle (in its usual topology), let Z be the
constant sheaf Z, and let U be the open covering by two connected open
semi-circles U, V, which overlap at each end, so that U n V consists of two
small intervals. Then
C0 X r(V,Z) = z X z
= r(U,Z)
C1 = nun v,z) = z x z
and the map d: C 0 ~ C 1 takes (a,b) to (b- a, b- a). Thus H0 (U,Z) = Z
and H 1 (U,Z) = Z. Since we know this is the right answer (Ex. 2.7), this
illustrates the general principle that Cech cohomology agrees with the usual
cohomology provided the open covering is taken fine enough so that there
is no cohomology on any of the open sets (Ex. 4.11 ).

Now we will study some properties of the Cech cohomology groups.

Lemma 4.1. For any X,U,$' as above, we have H0 (U,$') ~ r(X,$').

PROOF. H0 (U,$') = ker(d: C 0 (U,$') ~ C 1 (U,$') ). If rt. E C 0 is given by

{rt.; E $'(U;)}, then for each i < j, (drt.)ii = rt.i - rt.;. So drt. = 0 says the
sections rt.; and rt.i agree on U; n Vi. Thus it follows from the sheaf axioms
that ker d = r(X,$').

Next we define a "sheafified" version of the Cech complex. For any

open set V <:::; X, let f: V ~ X denote the inclusion map. Now given X,U,$'
as above, we construct a complex ~·(U,$') of sheaves on X as follows.
For each p ): 0, let

~P(U,$') = n
io< .. . < ip
f*($'iu,o. .,)

and define
d:~P ~ ~p+l

by the same formula as above. Note by construction that for each p we

have r(X,~P(U,$')) = CP(U,$').

Lemma 4.2. For any sheaf of abelian groups$' on X, the complex~·(~,$')

is a resolution of $', i.e., there is a natural map e: $' ~ ~ 0 such that the
sequence of sheaves
0 ~ g; ~ ~ 0 (U,$') ~ ~ 1 (U,$') ~ ...
is exact.

220
 4 Cech Cohomology

PROOF. We define a:%--+ ~ 0 by taking the product of the natural maps

% --+ f*(ff'iu,) for i E /. Then the exactness at the first step follows from
the sheaf axioms for %.
To show the exactness of the complex~· for p ;:;:: 1, it is enough to check
exactness on the stalks. So let x EX, and suppose x E Vi. For each p ;:;:: 1,
we define a map
k:~P(U,ff')x--+ ~p- 1 (U,ff')x

as follows. Given rxx E ~P(U,ff')x, it is represented by a section rx E

r(V,~P(U,%)) over a neighborhood V of x, which we may choose so small
that V c:; Vi. Now for any p-tuple i0 < ... < iP_ 1 , we set
.
(krx).lQ, . . . ,lp-1 = rx..
J,lQ,
.
. . . ,lp-1'

using the notational convention of (4.0.1). This makes sense because V n

via, ... ,ip -1 = v (\ vj,ia .... ,ip -1' Then take the stalk of krx at X to get the
required map k. Now one checks that for any p ;:;:: 1, rx E ~~.
(dk + kd)(rx) = rx.
Thus k is a homotopy operator for the complex ~~' showing that the iden-
tity map is homotopic to the zero map. It follows (§1) that the cohomology
groups hP(~J of this complex are 0 for p ;:;:: 1.

Proposition 4.3. Let X be a topological space, let U be an open covering,

and let % be a fiasque sheaf of abelian groups on X. Then for all p > 0
we have HP(U,%) = 0.
PROOF. Consider the resolution 0 --+ % --+ ~·(U,%) given by (4.2). Since
% is flasque, the sheaves ~P(U,%) are flasque for each p ;:;:: 0. Indeed, for
any io, ... ,ip, ff'iu,o. .ip is a flasque sheaf on via .... .ip; f* preserves flasque
sheaves (II, Ex. 1.16d), and a product of flasque sheaves is flasque. So by
(2.5.1) we can use this resolution to compute the usual cohomology groups
of%. But% is flasque, so HP(X,%) = 0 for p > 0 by (2.5). On the other hand,
the answer given by this resolution is
hP(T(X,~'(U,%))) = HP(U,%).
So we conclude that HP(U,%) = 0 for p > 0.

Lemma 4.4. Let X be a topological space, and U an open covering. Then

for each p ;:?;: 0 there is a natural map, functorial in %,
HP(U,%) --+ HP(X,%).
PROOF. Let 0 --+ % --+ f be an injective resolution of % in m:b(X). Com-
paring with the resolution 0 --+ % --+ ~·(U,%) of (4.2), it follows from a
general result on complexes (Hilton and Stammbach [1, IV, 4.4]) that there
is a morphism of complexes ~·(U,$') --+ f , inducing the identity map on
%, and unique up to homotopy. Applying the functors r(X, ·) and hP,
we get the required map.

221
III Cohomology

Theorem 4.5. Let X be a noetherian separated scheme, let U be an open

affine cover of X, and let $' be a quasi-coherent sheaf on X. Then for
all p ;::: 0, the natural maps of (4.4) give isomorphisms
HP(U,ff') ~ HP(X,$').
PROOF. For p = 0 we have an isomorphism by (4.1). For the general case,
embed $'in a flasque, quasi-coherent sheaf'§ (3.6), and let !?It be the quotient:
0 --+ $' --+ '§ --+ !?It --+ 0.
For each i 0 < ... < iP, the open set U; 0 • . . . • ip is affine, since it is an inter-
section of affine open subsets of a separated scheme (II, Ex. 4.3). Since $'
is quasi-coherent, we therefore have an exact sequence
0 --+ ff'(U; 0 , . . . ,i) --+ t§(U; 0 , . . . ,i) --+ !?ll(U; 0 , . . . ,;P) --+ 0
of abelian groups, by (3.5) or (II, 5.6). Taking products, we find that the
corresponding sequence of Cech complexes
0 --+ C(U,ff') --+ C(U,'§) --+ C(U,!?ll) --+ 0
is exact. Therefore we get a long exact sequence of Cech cohomology
groups. Since '§ is flasque, its Cech cohomology vanishes for p > 0 by
(4.3), so we have an exact sequence
0 --+ H 0 (U,ff') --+ H0 (U,'§) --+ H 0 (U,!?ll) --+ H 1 (U,ff') --+ 0
and isomorphisms
HP(U,!?ll) ~ jfp+ 1 (U,ff')
for each p ;::: 1. Now comparing with the long exact sequence of usual
cohomology for the above short exact sequence, using the case p = 0,
and (2.5), we conclude that the natural map
H 1 (U,ff') --+ H 1 (X,ff')
is an isomorphism. But !?It is also quasi-coherent (II, 5. 7), so we obtain the
result for all p by induction.

EXERCISES

4.1. Letf:X -> Ybe an affine morphism of noetherian separated schemes (II, Ex. 5.17).
Show that for any quasi-coherent sheaf ff on X, there are natural isomorphisms
for all i ;;;. 0,

[Hint: Use (II, 5.8).]

4.2. Prove Chevalley's theorem: Let f:X-> Y be a finite surjective morphism of
noetherian separated schemes, with X affine. Then Y is affine.
(a) Let f: X -> Y be a finite surjective morphism of integral noetherian schemes.
Show that there is a coherent sheaf .A on X, and a morphism of sheaves
a: @j, -> f*.A for some r > 0, such that a is an isomorphism at the generic
point of Y

222
 4 Cech Cohomology

(b) For any coherent sheaf$' on Y, show that there is a coherent sheaf~ on X,
and a morphism f3: f* ~ ---> $'' which is an isomorphism at the generic point
of Y. [Hint: Apply .Yt'om( ·,$')to Cl. and use (II, Ex. 5.17e).]
(c) Now prove Chevalley's theorem. First use (Ex. 3.1) and (Ex. 3.2) to reduce to
the case X and Y integral. Then use (3.7), (Ex. 4.1), consider ker f3 and coker {3,
and use noetherian induction on Y.
4.3. Let X = Af = Spec k[x,y], and let U =X - {(0,0)}. Using a suitable cover of
U by open affine subsets, show that H 1(U,(9u) is isomorphic to the k-vector space
spanned by {xiyili,j < 0}. In particular, it is infinite-dimensional. (Using (3.5),
this provides another proof that U is not affine-d. (1, Ex. 3.6).)
4.4. On an arbitrary topological space X with an arbitrary abelian sheaf ff, Cech
cohomology may not give the same result as the derived functor cohomology. But
here we show that for H 1 , there is an isomorphism if one takes the limit over all
coverings.
(a) Let U = (U;)iei be an open covering of the topological space X. A refinement
ofU is a covering 5B = (J'})ieJo together with a map A.:J---> I of the index sets,
such that for each j E J, J'} ~ U ).(j)· If 5B is a refinement of U, show that there
is a natural induced map on Cech cohomology, for any abelian sheaf ff, and
for each i,
A.': ii'(U,ff)---> ii'(\B,ff).

The coverings of X form a partially ordered set under refinement, so we can

consider the Cech cohomology in the limit

lim H'(U,ff).
7
(b) For any abelian sheaf$' on X, show that the natural maps (4.4) for each
covering
H'(U,ff)---> Hi(X,ff)

are compatible with the refinement maps above.

(c) Now prove the following theorem. Let X be a topological space,$' a sheaf of
abelian groups. Then the natural map

~ H 1 (U,ff) ---> H 1 (X,ff)

u
is an isomorphism. [Hint: Embed$' in a flasque sheaf~, and let !3l = ~/ff,
so that we have an exact sequence

Define a complex D"(U) by

0 ---> C(U,ff) ---> C(U,~) ---> D"(U) ---> 0.

Then use the exact cohomology sequence of this sequence of complexes, and
the natural map of complexes

and see what happens under refinement.]

223
III Cohomology

4.5. For any ringed space (X,@x), let Pic X be the group of isomorphism classes of
invertible sheaves (II, §6). Show that Pic X ~ H 1 (X,@~:), where (Dk denotes the
sheaf whose sections over an open set U are the units in the ring r(U,@x), with
multiplication as the group operation. [Hint: For any invertible sheaf fi> on X,
cover X by open sets U; on which fi> is free, and fix isomorphisms ({J;: @u, .::::. fL>Iu,·
Then on U; n Ui, we get an isomorphism ({J;- 1 o cpi of (Du n u with itself. These
isomorphisms give an element of H1 (U,@k). Now use (E~. 4.4).]

4.6. Let (X,@ x) be a ringed space, let J be a sheaf of ideals with J 2 = 0, and let X 0
be the ringed space (X,@x/J). Show that there is an exact sequence of sheaves of
abelian groups on X,
0 --> J --> (Dk --> (Dko --> 0,
where (Dk (respectively, @k 0 ) denotes the sheaf of (multiplicative) groups of units
in the sheaf of rings (Dx (respectively, @x 0 ); the map J--> (Dk is defined by ac->
1 + a, and J has its usual (additive) group structure. Conclude there is an exact
sequence of abelian groups
•.. --> H 1 (X,J) --> Pic X --> Pic X 0 --> H 2 (X,J) --> .•.•

4.7. Let X be a subscheme of P~ defined by a single homogeneous equation

f(x 0 ,x 1 ,x 2 ) = 0 of degree d. (Do not assume f is irreducible.) Assume that (1,0,0)
is not on X. Then show that X can be covered by the two open affine subsets
U = X n {x 1 # 0} and V = X n {x 2 # 0}. Now calculate the Cech complex
r(U,@x) EB r(V,@x)--> T(U n V,@x)
explicitly, and thus show that
dim H 0 (X,@x) = 1,
1
dim H 1 (X,@x) = :2 (d - 1)(d - 2).

4.8. Cohomological Dimension (Hartshorne [3]). Let X be a noetherian separated

scheme. We define the cohomological dimension of X, denoted cd(X), to be the
least integer n such that H;(X,ff) = 0 for all quasi-coherent sheaves ff and all
i > n. Thus for example, Serre's theorem (3.7) says that cd(X) = 0 if and only
if X is affine. Grothendieck's theorem (2.7) implies that cd(X) ::::; dim X.
(a) In the definition of cd(X), show that it is sufficient to consider only coherent
sheaves on X. Use (II, Ex. 5.15) and (2.9).
(b) If X is quasi-projective over a field k, then it is even sufficient to consider only
locally free coherent sheaves on X. Use (II, 5.18).
(c) Suppose X has a covering by r + 1 open affine subsets. Use Cech cohomology
to show that cd(X) ::::; r.
*(d) If X is a quasi-projective scheme of dimension rover a field k, then X can be
covered by r + 1 open affine subsets. Conclude (independently of (2.7)) that
cd(X) ::::; dim X.
(e) Let Y be a set-theoretic complete intersection (1, Ex. 2.17) of codimension r
in X = P~. Show that cd(X - Y) ::::; r - 1.
4.9. Let X = Spec k[x~>x 2 ,x 3 ,x 4 ] be affine four-space over a field k. Let Y1 be the
plane x 1 = x 2 = 0 and let Y2 be the plane x 3 = x 4 = 0. Show that Y = Y1 u Y2
is not a set-theoretic complete intersection in X. Therefore the projective closure

224
 5 The Cohomology of Projective Space

Yin Pi is also not a set-theoretic complete intersection. [Hints: Use an affine

analogue of (Ex. 4.8e). Then show that H 2 (X - Y,@x) =I= 0, by using (Ex. 2.3)
and (Ex. 2.4). If P = Y1 n Y2 , imitate (Ex. 4.3) to show H 3 (X - P,@x) =I= 0.]
*4.10. Let X be a nonsingular variety over an algebraically closed field k, and let:#' be a
coherent sheaf on X. Show that there is a one-to-one correspondence between
the set of infinitesimal extensions of X by :#' (II, Ex. 8. 7) up to isomorphism, and
the group H 1 (X,:#' ® ff), where ff is the tangent sheaf of X (I1,§8). [Hint: Use
(II, Ex. 8.6) and (4.5).]
4.11. This exercise shows that Cech cohomology will agree with the usual cohomology
whenever the sheaf has no cohomology on any of the open sets. More precisely,
let X be a topological space, :#'a sheaf of abelian groups, and U = ( U;) an open
cover. Assume for any finite intersection V = Uio n ... n Uip of open sets of the
covering, and for any k > 0, that Hk(V,Y'iv) = 0. Then prove that for all p ;;:, 0,
the natural maps
HP(U,:#')-+ W(X,:#')
of (4.4) are isomorphisms. Show also that one can recover (4.5) as a corollary of
this more general result.

5 The Cohomology of Projective Space

In this section we make explicit calculations of the cohomology of the

sheaves (!)(n) on a projective space, by using Cech cohomology for a suitable
open affine covering. These explicit calculations form the basis for various
general results about cohomology of coherent sheaves on projective
varieties.
Let A be a noetherian ring, letS = A[x 0 , . . . ,x,], and let X = Proj S
be the projective space P~ over A. Let (!)x(1) be the twisting sheaf of Serre
(II, §5). For any sheaf of (!)x-modules :F, we denote by r *(:F) the graded
S-module E8n E z r(X,:F(n)) (see II, §5).

Theorem 5.1. Let A be a noetherian ring, and let X = P~, with r ;:;, 1. Then:
(a) the natural mapS--+ r*((!)x) = E8nEZ H 0 (X,(!)x(n)) is an isomor-
phism of graded S-modules;
(b) Hi(X,(!)x(n)) = 0 for 0 < i < rand all n E Z;
(c) H'(X,(!)x(- r - 1)) ~ A;
(d) The natural map

H 0 (X,(!)x(n)) x H'(X,(!)x(-n- r - 1))--+ H'(X,(!)x(-r- 1)) ~A

is a perfect pairing of finitely generated free A-modules, for each n E Z.
PROOF. Let :F be the quasi-coherent sheaf E8n
E z (!)x(n). Since cohomology

commutes with arbitrary direct sums on a noetherian topological space

(2.9.1), the cohomology of :F will be the direct sum of the cohomology of
the sheaves (!)(n). So we will compute the cohomology of /F, and keep track

225
III Cohomology

of the grading by n, so that we can sort out the pieces at the end. Note that
all the cohomology groups in question have a natural structure of A-module
(2.6.1).
For each i = 0, ... ,r, let U; be the open set D+(x;). Then each U; is
an open affine subset of X, and the U; cover X, so we can compute the
cohomology of:#' by using Cech cohomology for the covering U = (UJ,
by (4.5). For any set of indices i0, . .. ,iP, the open set U; 0 , . . . ,ip is just
D+(X; 0 • • • x;) so by (II, 5.11) we have

the localization of S with respect to the element X; 0 • • • X;p· Furthermore,

the grading on :#' corresponds to the natural grading of Sx, ... x, under
this isomorphism. Thus the Cech complex of:#' is given by " p

C'(U,ff): nsx,o ~ nsXioXit ~ ... ~ sxo ... Xr'

and the modules all have a natural grading compatible with the grading
on$'.
Now H 0 (X,:F) is the kernel of the first map, which is just S, as we have
seen earlier (II, 5.13). This proves (a).
Next we consider H'(X,ff). It is the cokernel of the last map in the
Cech complex, which is

We think of Sxo· .. xr as a free A-module with basis x~ · · · x~, with l; E Z.

The image of dr- 1 is the free submodule generated by those basis elements
for which at least one I; ~ 0. Thus H'(X,ff) is a free A-module with basis
consisting of the "negative" monomials
{x~ · · · x~rll; < 0 for each i}.
Furthermore the grading is given by 'f);. There is only one such monomial
of degree - r - 1, namely x 0 1 · · · x;l, so we see that H'(X,(!Jx(-r- 1))
is a free A-module of rank 1. This proves (c).
To prove (d), first note that if n < 0, then H 0 (X,(!Jx(n)) = 0 by (a), and
Hr(X,(!Jx(- n - r - 1)) = 0 by what we have just seen, since in that case
- n - r - 1 > - r - 1, and there are no negative monomials of that
degree. So the statement is trivial for n < 0. For n ~ 0, H 0 (X,(!Jx(n)) has
a basis consisting of the usual monomials of degree n, i.e., {x 0° ... x~r lm; ~ 0
and Im;
= n}. The natural pairing with H'(X,(!Jx(-n- r- 1)) into
H'(X,(!Jx( - r - 1)) is determined by

where Ll; = - n - r - 1, and the object on the right becomes 0 if any

m; + l; ~ 0. So it is clear that we have a perfect pairing, under which
x 0 m0 - 1 · · · xr-mr- 1 is the dual basis element of X 0° · · · x~r.

226
 5 The Cohomology of Projective Space

It remains to prove statement (b), which we will do by induction on r.

If r = 1 there is nothing to prove, so let r > 1. If we localize the complex
C(U,,?) with respect to x., as graded S-modules, we get the Cech complex
for the sheaf ,?lu, on the space U., with respect to the open affine covering
{U; n U,li = 0, ... ,r}. By (4.5), this complex gives the cohomology of
,?lu, on U., which is 0 for i > 0 by (3.5). Since localization is an exact
functor, we conclude that Hi(X,,?k = 0 for i > 0. In other words, every
element of H;(X,,?), for i > 0, is annihilated by some power of x,.
To complete the proof of (b), we will show that for 0 < i < r, multiplica-
tion by x, induces a bijective map of H;(X,,?) into itself. Then it will follow
that this module is 0.
Consider the exact sequence of graded S-modules
0 ~ S(-1) ~ S ~ Sj(x,) ~ 0.
This gives the exact sequence of sheaves
0 ~ (!)x( -1) ~ (!)X ~ (!)H ~ 0
on X, where His the hyperplane x, = 0. Twisting by all n E Z and taking
the direct sum, we have
0 ~ ,?( -1) ~ ,? ~ ,?H ~ 0,
where ,?H = ffin e z (!)H(n). Taking cohomology, we get a long exact
sequence
... ~ Hi(X,,?( -1)) ~ Hi(X,,?) ~ Hi(X,,? H) ~ ....
Considered as graded S-modules, Hi(X,,?( -1)) is just H;(X,,?) shifted
one place, and the map Hi( X,,?( -1)) ~ Hi(X,,?) of the exact sequence
is multiplication by x,.
Now H is isomorphic to p~-I, and Hi( X,,? H) = Hi(H,Elj(!)H(n)) by
(2.10). So we can apply our induction hypothesis to ,?H, and find that
Hi(X,,? H) = 0 for 0 < i < r - 1. Furthermore, for i = 0 we have an
exact sequence
0 ~ H 0 (X,,?( -1)) ~ H 0(X,,?) ~ H 0 (X,,? H) ~ 0
by (a), since H 0 (X,,?H) is just Sj(x,). At the other end of the exact sequence
we have
0 ~ H'- 1(X,,?H) ~ H'(X,,?( -1)) ~ H'(X,,?) ~ 0.
Indeed, we have described H'(X,,?) above as the free A-module with basis
formed by the negative monomials in x 0 , . . . ,x,. So it is clear that x, is
surjective. On the other hand, the kernel of x, is the free submodule gen-
erated by those negative monomials x~ · · · x~' with l, = -1. Since
H'- 1(X,,?H) is the free A-module with basis consisting of the negative
monomials in x 0 , ••. ,x,_ 1 , and J is division by x., the sequence is exact.
In particular, J is injective.

227
III Cohomology

Putting these results all together, the long exact sequence of cohomology
shows that the map multiplication by xr: Hi( X,$'( -1)) ~Hi( X,$') is bijec-
tive for 0 < i < r, as required. q.e.d.

Theorem 5.2 (Serre [3]). Let X be a projective scheme over a noetherian

ring A, and let CDx(l) be a very ample invertible sheaf on X over Spec A.
Let$' be a coherent sheaf on X. Then:
(a) for each i ? 0, Hi(X,ff) is a finitely generated A-module;
(b) there is an integer n0 , depending on $', such that for each i > 0
and each n ? n0 , Hi(X,ff(n)) = 0.
PROOF. Since CDx(l) is a very ample sheaf on X over Spec A, there is a closed
immersion i:X ~ PA of schemes over A, for some r, such that CDx(l) =
i*CDpr(l)-cf. (II, 5.16.1). If$' is coherent on X, then i*ff is coherent on
PA (II, Ex. 5.5), and the cohomology is the same (2.10). Thus we reduce to
the case X = PA.
For X = PA, we observe that (a) and (b) are true for any sheaf of the
form CDx(q), q E z. This follows immediately from the explicit calculations
(5.1). Hence the same is true for any finite direct sum of such sheaves.
To prove the theorem for arbitrary coherent sheaves, we use descending
induction on i. For i > r, we have Hi(X,ff) = 0, since X can be covered
by r + 1 open affines (Ex. 4.8), so the result is trivial in this case.
In general, given a coherent sheaf$' on X, we can write$' as a quotient
of a sheaf lff, which is a finite direct sum of sheaves CD(qJ, for various integers
qi (II, 5.18). Let f!Jl be the kernel,
0 ~ f!Jl ~ lff ~ $' ~ 0.
Then f!ll is also coherent. We get an exact sequence of A-modules
... ~ Hi(X,lff) ~ Hi(X,ff) ~ Hi+ 1 (X,f!ll) ~ ....
Now the module on the left is finitely generated because C is a sum of CD(qi),
as remarked above. The module on the right is finitely generated by the
induction hypothesis. Since A is a noetherian ring, we conclude that the
one in the middle is also finitely generated. This proves (a).
To prove (b), we twist and again write down a piece of the long exact
sequence
... ~ Hi(X,C(n)) ~ Hi(X,ff(n)) ~ Hi+ 1 (X,f!Jl(n)) ~ ....
Now for n » 0, the module on the left vanishes because C is a sum of CD(qJ
The module on the right also vanishes for n » 0 because of the induction
hypothesis. Hence Hi(X,ff(n)) = 0 for n » 0. Note since there are only
finitely many i involved in statement (b), namely 0 < i ~ r, it is sufficient
to determine n0 separately for each i. This proves (b).

Remark 5.2.1. As a special case of (a), we see that for any coherent sheaf
$' on X, r(X,ff) is a finitely generated A-module. This generalizes, and
gives another proof of (II, 5.19).

228
 5 The Cohomology of Projective Space

As an application of these results, we give a cohomological criterion

for an invertible sheaf to be ample (II, §7).

Proposition 5.3. Let A be a noetherian ring, and let X be a proper scheme

over Spec A. Let !l' be an invertible sheaf on X. Then the following
conditions are equivalent:
(i) !l' is ample;
(ii) For each coherent sheaf :F on X, there is an integer n0 , depending on
:F, such that for each i > 0 and each n ? n0 , H;(X,:F ® !l'") = 0.
PROOF. (i)= (ii). If !l' is ample on X, then for some m > 0, !l'm is very
ample on X over Spec A, by (II, 7.6). Since X is proper over Spec A, it is
necessarily projective (II, 5.16.1). Now applying (5.2) to each of the sheaves
:F,:F ® !l',:F ® !£' 2 , .•. ,:F ® !l'm-l gives (ii). Cf. (II, 7.5) for a similar
technique of proof.
=
(ii) (i). To show that !l' is ample, we will show that for any coherent
sheaf :F on X, there is an integer n0 such that :F ® !l'" is generated by
global sections for all n ? n0 . This is the definition of ampleness (II, §7).
Given :F, let P be a closed point of X, and let Jp be the ideal sheaf of
the closed subset {P}. Then there is an exact sequence
0 --+ .fp!F --+ :F --+ :F ® k(P) --+ 0,
where k(P) is the skyscraper sheaf (!)x/.fp. Tensoring with !l'", we get
0 --+ Jp:F ® !l'" --+ :F ® !l'" --+ :F ® !l'" ® k(P) --+ 0.
Now by our hypothesis (ii), there is an n0 such that H 1(X,.fp:F ® !l'") = 0
for all n ? n 0 . Therefore
r(X,:F ® !l'") --+ r(X,:F ® !l'" ® k(P))
is surjective for all n ? n 0 . It follows from Nakayama's lemma over the
local ring (!) p, that the stalk of :F ® !l'" at P is generated by global sections.
Since it is a coherent sheaf, we conclude that for each n ? n0 , there is an
open neighborhood U of P, depending on n, such that the global sections
of :F ® !l'" generate the sheaf at every point of U.
In particular, taking :F = (!)x, we find there is an integer n 1 > 0 and
an open neighborhood V of P such that !l'" 1 is generated by global sections
over V. On the other hand, for each r = 0,1, ... ,n 1 - 1, the above argu-
ment gives a neighborhood U, of P such that :F ® .;eno+r is generated by
global sections over U,. Now let
up = v n u 0 n 0 0 0 n unt -1·
Then over Up, all of the sheaves :F ® 2", for n ? n0 , are generated by
global sections. Indeed, any such sheaf can be written as a tensor product
(/F@ .;eno+r)@ (.!f"t)m
for suitable 0 ~ r < n 1 and m ? 0.

229
III Cohomology

Now cover X by a finite number of the open sets Up, for various closed
points P, and let the new n0 be the maximum of the n0 corresponding to
those points P. Then $' Q9 !l'" is generated by global sections over all of
X, for all n ~ n0 . q.e.d.

EXERCISES

5.1. Let X be a projective scheme over a field k, and let :F be a coherent sheaf on X.
We define the Euler characteristic of :F by
x(:F) = D -1);dimk H;(X,:F).
If
0 ---> :F' ---> :F ---> :F" ---> 0
is a short exact sequence of coherent sheaves on X, show that x(ff) = x(:F') +
x(:F").
5.2. (a) Let X be a projective scheme over a field k, let @x(1) be a very ample invertible
sheaf on X over k, and let :F be a coherent sheaf on X. Show that there is a
polynomial P(z) E Q[ z], such that x(ff(n)) = P(n) for all n E Z. We call P
the Hilbert polynomial of :F with respect to the sheaf @x(1). [Hints: Use
induction on dim Supp ff, general properties of numerical polynomials
(1, 7.3), and suitable exact sequences
0 --->f)£ ---> :F( -1) ---> :F ---> f2 ---> 0.]
(b) Now let X = P~, and let M = r *(:F), considered as a graded S = k[ x 0 , •.. ,x,]-
module. Use (5.2) to show that the Hilbert polynomial of :F just defined is
the same as the Hilbert polynomial of M defined in (1, §7).
5.3. Arithmetic Genus. Let X be a projective scheme of dimension rover a field k. We
define the arithmetic genus Pa of X by
p.(X) = ( -l)'(x(@x)-1).
Note that it depends only on X, not on any projective embedding.
(a) If X is integral, and k algebraically closed, show that H 0 (X,@x) ~ k, so that
r-1
p.(X) = L (-1); dimk H'-;(X,@x).
i=O

In particular, if X is a curve, we have

p.(X) = dimk H 1 (X,@x).
[Hint: Use (1, 3.4).]
(b) If X is a closed subvariety ofP~, show that this p.(X) coincides with the one
defined in (I, Ex. 7.2), which apparently depended on the projective embedding.
(c) If X is a nonsingular projective curve over an algebraically closed field k, show
that p.(X) is in fact a birational invariant. Conclude that a nonsingular plane
curve of degree d ~ 3 is not rational. (This gives another proof of (II, 8.20.3)
where we used the geometric genus.)
5.4. Recall from (II, Ex. 6.10) the definition of the Grothendieck group K(X) of a
noetherian scheme X.

230
 5 The Cohomology of Projective Space

(a) Let X be a projective scheme over a field k, and let {!)x(1) be a very ample
invertible sheaf on X. Show that there is a (unique) additive homomorphism
P:K(X)-+ Q[z]
such that for each coherent sheaf§' on X, P(y(ff)) is the Hilbert polynomial
of ff (Ex. 5.2).
(b) Now let X = P~. For each i = 0,1, ... ,r, let L; be a linear space of dimension
i in X. Then show that

(1) K(X) is the free abelian group generated by {y({!)L)Ii = 0, ... ,r}, and
(2) the map P:K(X)-+ Q[z] is injective.
[Hint: Show that (1) => (2). Then prove (1) and (2) simultaneously, by induc-
tion on r, using (II, Ex. 6.10c).]
5.5. Let k be a field, let X = P~, and let Y be a closed subscheme of dimension q ~ 1,
which is a complete intersection (II, Ex. 8.4). Then:
(a) for all n E Z, the natural map
H 0 (X,{!)x(n)) -+ H 0 (Y,{!)y(n))
is surjective. (This gives a generalization and another proof of (II, Ex. 8.4c),
where we assumed Y was normal.)
(b) Y is connected;
(c) H;(Y,{!)y(n)) = 0 for 0 < i < q and all n E Z;
(d) p.(Y) = dimkHq(Y,{!)y).
[Hint: Use exact sequences and induction on the codimension, starting from
the case Y = X which is (5.1).]
5.6. Curves on a Nonsingular Quadric Swface. Let Q be the nonsingular quadric sur-
face xy = zw in X = Pf over a field k. We will consider locally principal closed
subschemes Y of Q. These correspond to Cartier divisors on Q by (II, 6.17.1).
On the other hand, we know that Pic Q ~ Z ® Z, so we can talk about the
type (a,b) of Y (II, 6.16) and (II, 6.6.1). Let us denote the invertible sheaf .P(Y) by
{!)Q(a,b). Thus for any n E Z, {!)Q(n) = {!)Q(n,n).
(a) Use the special cases (q,O) and (O,q), with q > 0, when Y is a disjoint union of q
lines P 1 in Q, to show:
(1) if Ia - bl ~ 1, then H 1(Q,{!)Q(a,b)) = 0;
(2) if a,b < 0, then H 1(Q,{!)Q(a,b)) = 0;
(3) If a ~ - 2, then H 1(Q,{!)Q(a,O)) -# 0.
(b) Now use these results to show:
(1) if Y is a locally principal closed subscheme of type (a,b), with a,b > 0,
then Y is connected;
(2) now assume k is algebraically closed. Then for any a,b > 0, there exists an
irreducible nonsingular curve Y of type (a,b). Use (II, 7.6.2) and (II, 8.18).
(3) an irreducible nonsingular curve Y of type (a,b), a,b > 0 on Q is projec-
tively normal (II, Ex. 5.14) if and only if Ia - bl ~ 1. In particular, this
gives lots of examples of nonsingular, but not projectively normal curves
in P 3. The simplest is the one of type (1,3), which is just the rational
quartic curve (1, Ex. 3.18).

231
III Cohomology

(c) If Y is a locally principal subscheme of type (a,b) in Q, show that p.(Y) =

ab - a - b + 1. [Hint: Calculate Hilbert polynomials of suitable sheaves,
and again use the special case (q,O) which is a disjoint union of q copies ofP 1 .
See (V, 1.5.2) for another method.]

5.7. Let X (respectively, Y) be proper schemes over a noetherian ring A. We denote by

!£ an invertible sheaf.
(a) If!£ is ample on X, and Y is any closed subscheme of X, then i* S£' is ample on
Y, where i: Y -> X is the inclusion.
(b) !£is ample on X if and only if 2';,. = !£@ (!)x". is ample on X" •.
(c) Suppose X is reduced. Then !£ is ample on X if and only if !£@ @x, IS
ample on Xi, for each irreducible component Xi of X.
(d) Let f: X -> Y be a finite surjective morphism, and let !£ be an invertible sheaf
on Y. Then!£ is ample on Y if and only iff*!£ is ample on X. [Hints: Use
(5.3) and compare (Ex. 3.1, Ex. 3.2, Ex. 4.1, Ex. 4.2). See also Hartshorne
[5, Ch. I §4] for more details.]

5.8. Prove that every one-dimensional proper scheme X over an algebraically closed
field k is projective.
(a) If X is irreducible and nonsingular, then X is projective by (II, 6.7).
(b) If X is integral, let X be its normalization (II, Ex. 3.8). Show that X is complete
and nonsingular, hence projective by (a). Let f: X -> X be the projection. Let
!£ be a very ample invertible sheaf on X. Show there is an effective divisor
D = L,Pi on X with !£(D) ~ !£,and such that f(Pi) is a nonsingular point of
X, for each i. Conclude that there is an invertible sheaf!£ 0 on X withf* !£ 0 ~
!£. Then use (Ex. 5.7d), (II, 7.6) and (II, 5.16.1) to show that X is projective.
(c) If X is reduced, but not necessarily irreducible, let X 1 , •.. , X, be the irre-
ducible components of X. Use (Ex. 4.5) to show Pic X-> ffi Pic X, is sur-
jective. Then use (Ex. 5.7c) to show X is projective.
(d) Finally, if X is any one-dimensional proper scheme over k, use (2. 7) and (Ex. 4.6)
to show that Pic X-> Pic X". is surjective. Then use (Ex. 5.7b) to show X
is projective.
5.9. A Nonprojective Scheme. We show the result of (Ex. 5.8) is false in dimension 2.
· Let k be an algebraically closed field of characteristic 0, and let X = Pf. Let w
be the sheaf of differential 2-forms (II, §8). Define an infinitesimal extension X'
of X by w by giving the element ~ E H 1(X,w@ 3) defined as follows (Ex. 4.10).
Let x 0 ,x 1 ,x 2 be the homogeneous coordinates of X, let U 0 ,U "U 2 be the standard
open covering, and let ~ij = (x)x;)d(x;/x). This gives a Cech 1-cocycle with
values in Q}, and since dim X = 2, we have w @ 3 ~ Q 1 (II, Ex. 5.16b). Now
use the exact sequence
... -> H 1(X,w) -> Pic X' -> Pic X ~ H 2 (X,w)-> ...

of (Ex. 4.6) and show b is injective. We have w ~ (!) x(- 3) by (II, 8.20.1 ), so
H 2(X,w) ~ k. Since char k = 0, you need only show that .:5(@(1)) # 0, which can
be done by calculating in Cech cohomology. Since H 1 (X,w) = 0, we see that
Pic X' = 0. In particular, X' has no ample invertible sheaves, so it is not pro-
jective.
Note. In fact, this result can be generalized to show that for any nonsingular
projective surface X over an algebraically closed field k of characteristic 0, there
is an infinitesimal extension X' of X by w, such that X' is not projective over k.
232
 6 Ext Groups and Sheaves

Indeed, let D be an ample divisor on X. Then D determines an element c 1(D) E

H 1 (X,Q 1 ) which we use to define X', as above. Then for any divisor Eon X one
can show that b(Sf(E)) = (D.E), where (D.E) is the intersection number (Chap-
ter V), considered as an element of k. Hence if E is ample, b(Sf(E)) "# 0. There-
fore X' has no ample divisors.
On the other hand, over a field of characteristic p > 0, a proper scheme X is
projective if and only if X "d is!
5.10. Let X be a projective scheme over a noetherian ring A, and let .'F 1 --+ :F 2 --+ ... --+
.'F' be an exact sequence of coherent sheaves on X. Show that there is an integer
n0 , such that for all n ~ n0 , the sequence of global sections
T(X,:F 1(n)) --+ T(X,:F 2 (n)) --+ ... --+ F(X,:F'(n))
is exact.

6 Ext Groups and Sheaves

In this section we develop the properties of Ext groups and sheaves, which
we will need for the duality theorem. We work on a ringed space (X,@x),
and all sheaves will be sheaves of @x-modules.
If :F and '§ are @x-modules, we denote by Hom(:F,'§) the group of (!)x-
module homomorphisms, and by Yt'om(:F,'§) the sheaf Hom (II, §5). If
necessary, we put a subscript X to indicate which space we are on:
Homx(:F,'§). For fixed :F, Hom(:F, ·) is a left exact covariant functor from
9J1o)(X) to 2lb, and Yt'om(:F, ·)is a left exact covariant functor from 9J1ob(X)
to Wlob(X). Since Wlob(X) has enough injectives (2.2) we can make the follow-
ing definition.

Definition. Let (X,@x) be a ringed space, and let :F be an @x-module. We

define the functors Exti(:F, ·) as the right derived functors of Hom(:F, · ),
and !Cxti(:F, ·) as the right derived functors of Yt'om(:F, · ).

Consequently, according to the general properties of derived functors

(l.lA) we have Ext 0 = Hom, a long exact sequence for a short exact sequence
in the second variable, Exti(:F,'§) = 0 for i > 0, '§ injective in Wlob(X), and
ditto for the !Cxt sheaves.

Lemma 6.1. If J is an injective object of W1ob(X), then for any open subset
U s:::: X, Jlu is an injective object of Wlob(U).

PROOF. Letj: U ---.X be the inclusion map. Then given an inclusion :F s:::: '§
in W1ob( U), and given a map :F ---. Jlu. we get an inclusion j 1:F s:::: j,'§, and
a mapj,:F---+ j,(Jiu), wherej 1 is extension by zero (II, Ex. 1.19). Butj,(Jitr)
is a subsheaf of§, so we have a mapj,:F ---. §. Since J is injective in Wlob(X),
this extends to a map of j,'§ ,o §. Restricting to U gives the required map
of'§ to Jlu·
233
III Cohomology

Proposition 6.2. For any open subset U s; X we have

t&'xt~(g;,<§)ju ~ t&'xth(g;ju.~iu ).

PROOF. We use (1.3A). Both sides give b-functors in ~ from 9Jlob(X) to

9Jlob(U). They agree for i = 0, both sides vanish for i > 0 and~ injective,
by (6.1), so they are equaL

Proposition 6.3. For any<§ E9Jlob(X), we have:

(a) @"xt 0 (@x,':#) = ':#;
(b) t&'xti((Dx,':#) = 0 fori> 0;
(c) Exti(@x,':#) ~ Hi(X,':#) for all i?: 0.
PROOF. The functor Yf'om(@x, ·)is the identity functor, so its derived functors
are 0 fori > 0. This proves (a) and (b). The functors Hom(@x, ·)and r(X, ·)
are equal, so their derived functors (as functors from 9Jlob(X) to ~b) are the
same. Then use (2.6).

Proposition 6.4. If 0 --+ g;' --+ g; --+ g;" --+ 0 is a short exact sequence in
9Jlob(X), then for any':# we have a long exact sequence
0 --+ Hom(g;",':#) --+ Hom(g;,<§) --+ Hom(g;', '§) --+
--+ Ext 1(g;",':#) --+ Ext 1(g;,<§) --+ ... ,
and similarly for the t&'xt sheaves.
PROOF. Let 0 --+ ':# --+ f be an injective resolution of':#. For any injective
sheaf J, the functor Hom(· ,J) is exact, so we get a short exact sequence of
complexes
0--+ Hom(g;",f)--+ Hom(g;,f)--+ Hom(g;',f)--+ 0.
Taking the associated long exact sequence of cohomology groups hi gives
the sequence of Ext;.
Similarly, using (6.1) we see that Yf'om( · ,J) is an exact functor from
9Jlob(X) to 9Jlob(X). Thus the same argument gives the exact sequence
of Sxti.

Proposition 6.5. Suppose there is an exact sequence

... --+ 21--+ !l'o--+ g;--+ 0
in 9Jlob(X), where the !l'; are locally free sheaves of finite rank (in this case
we say !l'. is a locally free resolution of g;). Then for any ':# E 9Jlob(X)
we have
t&'xti(g;,<§) ~ hi(Yf' om(!l'., ':#) ).
PROOF. Both sides are b-functors in':# from 9Jlob(X) to 9Jlob(X). For i = 0
they are equal, because then Yf'om( · ,':#) is contravariant and left exact.
Both sides vanish for i > 0 and ':# injective, because then Yf'om( · ,':#) is exact.
So by (1.3A) they are equal.

234
 6 Ext Groups and Sheaves

Example 6.5.1. If X is a scheme, which is quasi-projective over Spec A, where

A is a noetherian ring, then by (II, 5.18), any coherent sheaf:#' on X is a
quotient of a locally free sheaf. Thus any coherent sheaf on X has a locally
free resolution 2. ~ :#' ~ 0. So (6.5) tells us that we can calculate rffxt
by taking locally free resolutions in the first variable.

Caution 6.5.2. The results (6.4) and (6.5) do not imply that rffxt can be con-
strued as a derived functor in its first variable. In fact, we cannot even de-
fine the right derived functors of Hom or Yt'om in the first variable because
the category Wlob(X) does not have enough projectives (Ex. 6.2). However,
see (Ex. 6.4) for a universal property.

Lemma 6.6. If 2 E Wlob(X) is locally free of finite rank, and J E Wlob(X)

is injective, then 2 0 J is also injective.

PROOF. We must show that the functor Hom(· ,2 0 J) is exact. But it is

the same as the functor Hom(· 0 2~ ,J) (II, Ex. 5.1), which is exact because
· 0 2 ~ is exact, and J is injective.

Proposition 6.7. Let 2 be a locally free sheaf of finite rank, and let 2~ =
Yt'om(2,(!Jx) be its dual. Then for any:#','§ E Wlob(X) we have

Ext;(!#' 0 2,1§) ~ Exti(:F,2~ 0 '§),

and for the sheaf rffxt we have

PROOF. The case i = 0 follows from (II, Ex. 5.1). For the general case, note
that all of them are b-functors in I§ from Wlob(X) to mb (respectively,
Wlob(X) ), since tensoring with 2 ~ is an exact functor. For i > 0 and I§
injective they all vanish, by (6.6), so by (1.3A) they are equal.

Next we will give some properties which are more particular to the case
of schemes.

Proposition 6.8. Let X be a noetherian scheme, let :#'be a coherent sheaf on X,

let I§ be any (!Jx-module, and let x EX be a point. Then we have

rffxti(:#',l§)x ~ Ext~jff x,l§ J

for any i ~ 0, where the right-hand side is Ext over the local ring (!Jx·

PROOF. Of course, Ext over a ring A is defined as the right derived functor of
HomA(M, ·) for any A-module M, considered as a functor from IDlob(A) to
Wlob(A). Or, by considering a one-point space with the ring A attached, it
becomes a special case of the Ext of a ringed space defined above.

235
III Cohomology

Our question is local, by (6.2), so we may assume that X is affine. Then

:F has a locally free (or even a free) resolution f£. --+ :F --+ 0, which on the
stalks at x gives a free resolution (fE.)x --+ ffx --+ 0. So by (6.5) we can cal-
culate both sides by these resolutions. Since Yt'om(fE,'.§)x = Hom(I?J.Px,'.§ x)
for a locally free sheaf!£, and since the stalk functor is exact, we get the
equality of Ext's.
Note that even the case i = 0 is not true without some special hypothesis
on :F, such as :F coherent.

Proposition 6.9. Let X be a projective scheme over a noetherian ring A, let

C9x(1) be a very ample invertible sheaf, and let$','.§ be coherent sheaves on X.
Then there is an integer n0 > 0, depending on :F, '.§, and i, such that for every
n ~ n0 we have

PROOF. If i = 0, this is true for any !F,'.§,n. If :F = CDx, then the left-hand
side is Hi(X,'.§(n)) by (6.3). So for n » 0 and i > 0 it is 0 by (5.2). On the
other hand, the right-hand side is always 0 fori > 0 by (6.3), so we have the
result for :F = CDx.
If :F is a locally free sheaf, we reduce to the case :F = CDx by (6.7).
Finally, if ff is an arbitrary coherent sheaf, write it as a quotient of a
locally free sheaf S (II, 5.18), and let !!A be the kernel:
0 --+ !!A --+ g --+ ff --+ 0.
Since S is locally free, by the earlier results, for n » 0, we have an exact
sequence
0 --+ Hom(ff,'.§(n)) --+ Hom(S,'.§(n)) --+ Hom(f!A,'.§(n)) --+ Ext 1 ($','.§(n)) --+ 0

and isomorphisms, for all i > 0

Exti(8i,'.§(n)) ~ Exti+ 1 (ff,'.§(n) ),

and similarly for the sheaf Yt'om and Sxt. Now by (Ex. 5.10), the sequence of
global sections of the sheaf sequence is exact after twisting a little more, so
from the case i = 0, using (6.7), we get the case i = 1 for :F. But 8i is also
coherent, so by induction we get the general result.

Remark 6.9.1. More generally, on any ringed space X, the relation between
the global Ext and the sheaf Sxt can be expressed by a spectral sequence
(see Grothendieck [1] or Godement [1, II, 7.3.3]).

Now, for future reference, we recall the notion of projective dimension of

a module over a ring. Let A be a ring, and let M be an A-module. A
projective resolution of M is a complex L. of projective A-modules, such that
. . . --+ L2 --+ L1 --+ L0 --+ M --+ 0

236
 6 Ext Groups and Sheaves

is exact. If L; = 0 for i > n, and Ln #- 0, we say it has length n. Then we

define the projective dimension of M, denoted pd(M), to be the least length
of a projective resolution of M (or + oo if there is no finite projective reso-
lution).

Proposition 6.10A. Let A be a ring, and Man A-module. Then:

(a) M is projective if and only if Ext 1 (M,N) = 0 for all A-modules N;
(b) pd(M) < n if and only if Exti(M,N) = 0 for all i > n and all A-
modules N.
PROOF. Matsumura [2, pp. 127-128].

Proposition 6.11A. If A is a regular local ring, then:

(a) for every M, pd(M) < dim A;
(b) If k = Ajm, then pd(k) = dim A.
PROOF. Matsumura [2, Th. 42, p. 131].

Proposition 6.12A. Let A be a regular local ring of dimension n, and let M be

a .finitely generated A-module. Then we have
pd M + depth M = n.

PROOF. Matsumura [2, p. 113, Ex. 4] or Serre [11, IVD, Prop. 21].

EXERCISES

6.1. Let (X,@x) be a ringed space, and let $'',$'" E Wlob(X). An extension of$'" by
$'' is a short exact sequence
0 --+ $'' --+ $' --+ $'" --+ 0
in Wlob(X). Two extensions are isomorphic if there is an isomorphism of the
short exact sequences, inducing the identity maps on $'' and $'". Given an
extension as above consider the long exact sequence arising from Hom($'",·), in
particular the map
b:Hom(ff",ff")--+ Ext 1(ff",ff'),
and let ~ E Ext 1(ff",ff') be b(l ,.-,). Show that this process gives a one-to-one
correspondence between isomorphism classes of extensions of$'" by $'', and
elements of the group Ext 1 (ff",ff'). For more details, see, e.g., Hilton and
Stammbach [1, Ch. III].
6.2. Let X = Pi, with k an infinite field.
(a) Show that there does not exist a projective object f}J E Wlob(X), together with a
surjective map f}J--+ (()x--+ 0. [Hint: Consider surjections of the form (()v--+
k(x)--+ 0, where x EX is a closed point, Vis an open neighborhood of x,
and (()v = j!((()xlv), where j: V--+ X is the inclusion.]
(b) Show that there does not exist a projective object f}J in either ,Oco(X) or <£of)(X)
together with a surjection f}J--+ (()x--+ 0. [Hint: Consider surjections of the
form !i' --+ !i' ® k(x) --+ 0, where x E X is a closed point, and !i' is an invertible
sheaf on X.]

237
III Cohomology

6.3. Let X be a noetherian scheme, and let :Y,<§ E Wlob(X).

(a) If ff,<§ are both coherent, then .fxt'(ff,<§) is coherent, for all i ~ 0.
(b) If :Y is coherent and <§ is quasi-coherent, then .fxt'(:Y,<§) is quasi-coherent,
for all i ~ 0.
6.4. Let X be a noetherian scheme, and suppose that every coherent sheaf on X is a
quotient of a locally free sheaf. In this case we say [ol)(X) has enough locally frees.
Then for any <§ E 9Jlob(X), show that the b-functor (.fxt'( · ,<§) ), from [ol)(X) to
Wlob(X ), is a contravariant universal 6-functor. [Hint: Show .fxt'( ·,<§)is coefface-
able (§I) for i > 0.]
6.5. Let X be a noetherian scheme, and assume that [ol)(X) has enough locally frees
(Ex. 6.4). Then for any coherent sheaf :Y we define the homological dimension
of ff, denoted hd(ff), to be the least length of a locally free resolution of :Y (or + oo
if there is no finite one). Show:
(a) :Y is locally free= .fxt 1(:Y,<§) = 0 for all<§ E Wlob(X);
(b) hd(ff) ~ n = bt'(:Y,<§) = 0 for all i > nand all<§ E Wlob(X);
(c) hd(.~) = SUPx pdex ffx.
6.6. Let A be a regular local ring, and let M be a finitely generated A-module. In this
case, strengthen the result (6.10A) as follows.
(a) M is projective if and only if Ext'(M,A) = 0 for all i > 0. [Hint: Use (6.11A)
and descending induction on i to show that Ext'(M,N) = 0 for all i > 0 and
all finitely generated A-modules N. Then show M is a direct summand of a
free A-module (Matsumura [2, p. 129]).]
(b) Use (a) to show that for any n, pd M ~ n if and only if Ext'(M,A) = 0 for all
i> n.
6.7. Let X = Spec A be an affine noetherian scheme. Let M, N be A-modules, with M
finitely generated. Then
Ext~(M,N) ~ Ext~(M,N)
and
.fxt~(M,iir) ~ Ext~(M.N)-.
6.8. Prove the following theorem of Kleiman (see Borelli [1]): if X is a noetherian,
integral, separated, locally factorial scheme, then every coherent sheaf on X is a
quotient of a locally free sheaf (of finite rank).
(a) First show that open sets of the form X., for various s E r(X,ll'), and various
invertible sheaves 2" on X, form a base for the topology of X. [Hint: Given a
closed point x E X and an open neighborhood U of x, to show there is an 2" ,s
such that x E Xs s; U, first reduce to the case that Z = X - U is irreducible.
Then let (be the generic point of Z. Let f E K(X) be a rational function with
f E(()x, f ¢ (!!(· Let D = (j) 00 , and let 2" = ll'(D), s E T(X,ll'(D)) correspond
to D (II, §6).]
(b) Now use (II, 5.14) to show that any coherent sheaf is a quotient of a direct sum
E±)Y?' for various invertible sheaves 2", and various integers n,.
6.9. Let X be a noetherian, integral, separated, regular scheme. (We say a scheme is
regular if all of its local rings are regular local rings.) Recall the definition of
the Grothendieck group K(X) from (II, Ex. 6.10). We define similarly another
group K 1(X) using locally free sheaves: it is the quotient of the free abelian group
generated by all locally free (coherent) sheaves, by the subgroup generated by all
expressions of the form .f - .f' - <%'", whenever 0 --> <%'' --> <%' --> <%'" --> 0 is a

238
 7 The Serre Duality Theorem

short exact sequence of locally free sheaves. Clearly there is a natural group
homomorphism e:K 1(X)-> K(X). Show that e is an isomorphism (Borel and
Serre [1, §4]) as follows.
(a) Given a coherent sheaf :F, use (Ex. 6.8) to show that it has a locally free resolu-
tion C. -> $' -> 0. Then use (6.11A) and (Ex. 6.5) to show that it has a finite
locally free resolution
0 -> c. -> ... -> c 1 -> c0 -> $' -> 0.
(b) For each $', choose a finite locally free resolution C. -> $' -> 0, and let
b($') = L( -l)iy(CJ inK 1 (X). Show that()($') is independent of the resolu-
tion chosen, that it defines a homomorphism of K(X) to K 1(X), and finally,
that it is an inverse to e.
6.10. Duality for a Finite Flat Morphism.
(a) Let f:X -> Y be a finite morphism of noetherian schemes. For any quasi-
coherent @y-module '§, Yfomy(f*@x,'§) is a quasi-coherent f*0x-module,
hence corresponds to a quasi-coherent 0x-module, which we call j!'§ (II,
Ex. 5.17e).
(b) Show that for any coherent$' on X and any quasi-coherent'§ on Y, there is a
natural isomorphism
J*Yfomx(:F,f't§) .::. Yfomy(f*$',1§).
(c) For each i ~ 0, there is a natural map
<p;: Ext~(:F,j''§) -> Ext~(f*$',1§).
[Hint: First construct a map
Ext~(:F,f''§) -> Ext~(f*:F,f*j''§).
Then compose with a suitable map from f*f' '§ to '§.]
(d) Now assume that X and Yare separated, G:oi)(X) has enough locally frees, and
assume that j*@x is locally free on Y (this is equivalent to saying f flat-see
§9). Show that <p; is an isomorphism for all i, all :F coherent on X, and all'§
quasi-coherent on Y. [Hints: First do i = 0. Then do$' = @x, using (Ex. 4.1).
Then do :F locally free. Do the general case by induction on i, writing $' as
a quotient of a locally free sheaf.]

7 The Serre Duality Theorem

In this section we prove the Serre duality theorem for the cohomology of
coherent sheaves on a projective scheme. First we do the case of projective
space itself, which follows easily from the explicit calculations of §5. Then
on an arbitrary projective scheme X, we show that there is a coherent sheaf
w~, which plays a role in duality theory similar to the canonical sheaf of a
nonsingular variety. In particular, if X is Cohen-Macaulay, it gives a
duality theorem just like the one on projective space. Finally, if X is a non-
singular variety over an algebraically closed field, we show that the dualizing
sheaf w~ coincides with the canonical sheaf wx. At the end of the section,
we mention the connection between duality and residues of differential
forms.

239
III Cohomology

Let k be a field, let X = P~ be the n-dimensional projective space over k,

and let Wx = 1\"flx;k be the canonical sheaf on X (II, §8).

Theorem 7.1 (Duality for P~). Let X = P~ over a field k. Then:

(a) H"(X,wx) ~ k. Fix one such isomorphism;
(b) for any coherent sheaf:#' on X, the natural pairing

Hom(:#',w) x H"(X,:#') --+ H"(X,w) ~ k

is a perfect pairing of finite-dimensional vector spaces over k;
(c) for every i ~ 0 there is a natural functorial isomorphism
Exti(:#',w) ~ H"-i(X,:#')',

where ' denotes the dual vector space, which for i = 0 is the one induced
by the pairing of (b).
PRooF.
(a) It follows from (II, 8.13) that Wx ~ lPx(- n - 1) (see II, 8.20.1). Thus
(a) follows from (5.1c).
(b) A homomorphism of:#' to w induces a map of cohomology groups
H"(X,:#') --+ H"(X,w). This gives the natural pairing. If:#' ~ lP(q) for some
q E Z, then Hom(:#',w) ~ H 0 (X,w(- q) ), so the result follows from (5.1d).
Hence (b) holds also for a finite direct sum of sheaves of the form lP(qJ If
:#' is an arbitrary coherent sheaf, we can write it as a cokernel {f 1 --+ {f 0 --+
:#' --+ 0 of a map of sheaves tfi, each tfi being a direct sum of sheaves lP(qJ
Now Hom(· ,w) and H"(X, · )' are both left-exact contravariant functors, so
by the 5-lemma we get an isomorphism Hom(:#',w) ~ H"(X,:#')'.
(c) Both sides are contravariant <'>-functors, for :#' E <£ol)(X), indexed by
i ~ 0. For i = 0 we have an isomorphism by (b). Thus to show they are
isomorphic, by (1.3A), it will be sufficient to show both sides are coeffaceable
for i > 0. Given :#' coherent, it follows from (II, 5.18) and its proof that we
can write :#' as a quotient of a sheaf {f = ffif= 1 lP(- q), with q » 0. Then
Exti(tf,w) = E!jHi(X,w(q)) = 0 for i > 0 by (5.1). On the other hand,
H"-i(X,tf)' = EtlH"-i(X,lP(-q))', which is 0 fori> 0, q > 0, as we see
again from (5.1) by inspection. Thus both sides are coeffaceable for i > 0,
so the <'>-functors are universal, hence isomorphic.

Remark 7.1.1. One may ask, why bother phrasing (7.1) with the sheaf wx,
rather than simply writing lPx(- n - 1), which is what we use in the proof?
One reason is that this is the form of the theorem which generalizes well.
But a more intrinsic reason is that when written this way, the isomorphism
of (a) can be made independent of the choice of basis of P", hence stable
under automorphisms of P". Thus it is truly a natural isomorphism. To do
this, consider the Cech cocycle

_
Ot;- X~
x1 ... Xn
d (X1)
Xo
/\ ••. /\ d (X")
Xo

240
 7 The Serre Duality Theorem

in C"(U,w), where U is the standard open covering. Then one can show
that rx determines a generator of Hn(X,w), which is stable under change of
variables.

To generalize (7.1) to other schemes, we take properties (a) and (b) as

our guide, and make the following definition.

Definition. Let X be a proper scheme of dimension n over a field k. A

dualizing sheaf for X is a coherent sheaf w~ on X, together with a trace
morphism t:Hn(X,w~) ~ k, such that for all coherent sheaves :F on X,
the natural pairing

followed by t gives an isomorphism

Proposition 7.2. Let X be a proper scheme over k. Then a dualizing sheaf

for X, if it exists, is unique. More precisely, if W is one, with its trace
0

map t, and if w',t' is another, then there is a unique isomorphism ({J: W -==. w'
0

such that t = t' o Hn( ({J ).

PROOF. Since w' is dualizing, we get an isomorphism Hom(w 0 ,W') ~ Hn(w 0 )'.
So there is a unique morphism ({J: W 0 ~ w' corresponding to the element
t E Hn(w 0 )', i.e., such that t' o Hn(({J) = t. Similarly, using the fact that W0 is
dualizing, there is a unique morphism tjl:w' ~ W such that to Hn(t/1) = t'.
0

It follows that to Hn(t/1 o ({J) = t. But again since W is dualizing, this implies
0

that tjJ o ({J is the identity map of W Similarly ({J o tjJ is the identity map of
0
•

w', so ({J is an isomorphism. (This proof is a special case of the uniqueness

of an object representing a functor (see Grothendieck [EGA I, new ed., Ch. 0,
§1 ]). For by definition (w ,t) represents the functor :F ~ H"(X,:F)' from
0

<£ol)(X) to ffilob(k).)

The question of existence of dualizing sheaves is more difficult. In fact

they exist for any X proper over k, but we will prove the existence here only
for projective schemes. First we need some preliminary results.

Lemma 7.3. Let X be a closed subscheme of codimension r of P = Pf. Then

!&"xt~(lDx,wp) = 0 for all i < r.

PROOF. For any i, the sheaf :#'; = @"xt~(lDx,wp) is a coherent sheaf on P

(Ex. 6.3), so after twisting by a suitably large integer q, it will be generated
by global sections (II, 5.17). Thus to show :#'; is zero, it will be sufficient to
show that F(P,:F;(q)) = 0 for all q » 0. But by (6.7) and (6.9) we have
F(P,:F;(q)) ~ Ext~(lDx,Wp(q))

241
III Cohomology

for q » 0. On the other hand, by (7.1) this last Ext group is dual to
HN-i(P,(!)x( -q)). Fori< r, N- i >dim X, so this group is 0 by (2.7) or
(Ex. 4.8d).

Lemma 7.4. With the same hypotheses as (7.3), let w~ = Sxt~((!)x,Wp). Then
for any @x-module g;, there is a functorial isomorphism
Homx(g;,w~) ~ Ext~(g;,wp).

PROOF. Let 0 ~ Wp ~ f be an injective resolution of Wp in Wlob(P). Then

we calculate Ext~(g;,wp) as the cohomology groups hi of the complex
Homp(g;,f). But since g; is an @x-module, any morphism g;-+ Ji
factors through eli = Yfomp(@x,Ji). Thus we have
Ext~(g;,wp) = hi(Homx(g;,f) ).
Now each eli is an injective {!}x-module. Indeed, for g; E Wlob(X),
Homx(g;,eli) = Homp(g;,Ji), so Homx( ·,eli) is an exact functor. Further-
more, by (7.3) we have hi(el') = 0 for i < r, so the complex e~· is exact up
to the rth step. Since the eli are injective, it is actually split exact up to the
rth step. This implies that we can write the complex as a direct sum of two
injective complexes, f = el~ EB el~, where el~ is in degrees 0::::; i ::::; rand
is exact, and el~ is in degrees i ~ r. It follows that w~ = ker(d':el2-+ el2+ 1),
and that for any @x-module g;,
Homx(g;,w~) ~ Ext~(g;,wp).

(It also follows that Ext~(g;,wp) = 0 for i < r, which we won't need.)

Proposition 7.5. Let X be a projective scheme over a field k. Then X has a

dualizing sheaf

PROOF. Embed X as a closed subscheme of P = Pf for some N, let r be its

codimension, and let wx = Sxt~((!)x,wp). Then by (7.4) we have an iso-
morphism for any @x-module g;,
Homx(g;,w~) ~ Ext~(g;,wp).

On the other hand, when g; is coherent, the duality theorem for P (7.1)
gives an isomorphism

But N - r = n, the dimension of X, and g; is a sheaf on X, so we obtain a

functorial isomorphism, for g; E (£:ol)(X),
Homx(g;,w~) ~ H"(x,g;y.
In particular, taking g; = wx, the element 1 E Hom(wx,wx) gives us a
homomorphism t:H"(X,wx)-+ k, which we take as our trace map. Then
it is clear by functoriality that (w~,t) is a dualizing sheaf for X.

242
 7 The Serre Duality Theorem

Now we can prove the duality theorem for a projective scheme X. Recall
that a scheme is Cohen-M acaulay if all of its local rings are Cohen-Macaulay
rings (II, §8).

Theorem 7.6 (Duality for a Projective Scheme). Let X be a projective scheme

of dimension n over an algebraically closed field k. Let Wx be a dualizing
sheaf on X, and let CD(l) be a very ample sheaf on X. Then:
(a) for all i ~ 0 and :F coherent on X, there are natural functorial maps
ei:Exti(:F,wx)--+ Hn-i(X,:F)',
such that eo is the map given in the definition of dualizing sheaf above;
(b) the following conditions are equivalent:
(i) X is Cohen-Macaulay and equidimensional (i.e., all irreducible com-
ponents have the same dimension);
(ii) for any :F locally free on X, we have Hi(X,:F(- q)) = 0 for i < n
and q » 0;
(iii) the maps e; of (a) are isomorphisms for all i ~ 0 and all :F coherent
on X.
PROOF.
(a) As in the proof of (7.1c), we can write any coherent sheaf :F as a
quotient of a sheaf Iff= E8f= 1 CDx(-q), with q » 0. Then Exti(tff,wx) ~
ffiHi(X,wx(q) ), which is 0 for i > 0 and q » 0 by (5.2). Thus the functor
Exti( · ,wx) is coeffaceable for i > 0, so we have a universal contravariant
15-functor by (1.3A). On the right-hand side we have a contravariant <5-
functor, indexed by i ~ 0, so there is a unique morphism of 15-functors (e;)
reducing to the given eo for i = 0.
(b) (i) => (ii). Embed X as a dosed subscheme of P = Pf. Then for any
:F locally free on X, and any closed point x E X, we have depth :F,. = n,
since X is Cohen-Macaulay and equidimensional of dimension n. Let
A = CDP.x be the local ring of x on P. Then A is a regular local ring of di-
mension N. (Since k is algebraically closed, x is rational over k, so the
fact that A is regular can be seen directly. Or it follows from the fact that P
is a nonsingular variety over k (II, §8).) Now depth :Fx is the same, whether
calculated over CDx.x or over A. Thus we conclude from (6.12A) that
pdA :Fx = N - n. Therefore by (6.8) and (6.10A) we have

tffxt~(:F, ·) = 0
fori> N- n.
On the other hand, using (7.1), we find that Hi(X,:F(- q)) is dual to
Ext~-i(:F,wp(q) ). For q » 0, this Ext is isomorphic to r(P,tffxt~-i(:F,wp(q)))
by (6.9). But this is 0 for N - i > N - n, as we have just seen. In other
words, Hi(X,:F(- q)) = 0 for i < n and q » 0.
(ii) => (i). Running the above argument backwards, using condition (ii)
with :F = CDx, we find that

243
III Cohomology

for i > N - n. This implies that over a local ring A = (!)P,x as above, we
have Ext~((!)x,x,A) = 0 for all i > N - n. Therefore by (Ex. 6.6) we have
pdA (!)x.x :( N - n, and so by (6.12A), depth (!)x.x ;;::, n. But since dim X = n,
we must have equality for every closed point of X. This shows, using
(II, 8.21Ab), that X is Cohen-Macaulay and equidimensional.
(ii) = (iii). Since we have already seen that Ext;(· ,w~)) is a universal
contravariant 6-functor, to show that the (]i are isomorphisms, it will be
sufficient to show that the 6-functor (Hn- ;(X,·)') is universal also. For this
it suffices by (1.3A) to show that Hn-i(X, · )' is coeffaceable for i > 0. So
given a coherent sheaf ff', write ff' as a quotient of Iff = EJ:j(!)( -q) with
q » 0. Then Hn-i(X,Iff)' = 0 fori > 0 by (ii), so the functor is coeffaceable.
(iii) = (ii). If (}; is an isomorphism, then for any ff' locally free, we have

But this Ext is isomorphic to w-;(X,ff'~ ® w~(q)) by (6.3) and (6.7), so

it is 0 for n - i > 0 and q » 0 by (5.2). q.e.d.

Remark 7.6.1. In particular, if X is nonsingular over k, or more generally

a local complete intersection, then X is Cohen-Macaulay (II, 8.21A) and
(II, 8.23), so the (}; are isomorphisms. In these two cases, one can show
directly (cf. proof of (7.11) below) that pdp (!)x = N - n, and thus avoid
use of the algebraic results (6.12A) and (Ex. 6.6).

Corollary 7.7. Let X he a projective Cohen-Macaulay scheme of equidimen-

sion n over k. Then for any locally free sheaf ff' on X there are natural
isomorphisms

PROOF. Use (6.3) and (6.7).

Corollary 7.8 (Lemma of Enriques-Severi-Zariski (Zariski [ 4]) ). Let X be

a normal projective scheme of dimension ;;::, 2. Then for any locally free
sheaf ff' on X,

for q » 0.

PRooF. Since X is normal of dimension ;;::, 2, we have depth ff'x ;;::, 2 for
every closed point x E X by (II, 8.22A). So the result follows by the same
method as the proof of (i) = (ii) in (7.6b).

Corollary 7.9. Let X be an integral, normal projective variety of dimension

;;::, 2 over an algebraically closed field k. Let Y be a closed subset of codimen-
sion 1 which is the support of an effective ample divisor. Then Y is con-
nected.

244
 7 The Serre Duality Theorem

PROOF. By (II, 7.6) we may assume that Y is the support of a very ample
divisor D. Let CD(1) be the corresponding very ample invertible sheaf. For
each q > 0, let ~ be the closed subscheme supported on Y corresponding
to the divisor qD (II, 6.17.1). Then we have an exact sequence (II, 6.18)

0 ~ CDx( -q) ~ CDx ~ CDy. ~ 0.

Taking cohomology and applying (7.8), we find that for q » 0,

H 0 (X,CDx) ~ H 0 (Y,CDy q ) ~ 0

is surjective. But H 0 (X,CDx) = k (1, 3.4a), and H 0 (Y,CDy.) contains k, so we

conclude that H 0 (Y,CDy q ) = k. Hence Y is connected. (If not, there would
be at least one copy of k for each connected component.)

Remark 7.9.1. This implies that the schemes H n X mentioned in Bertini's

theorem (II, 8.18) are in fact irreducible and nonsingular when dim X ~ 2.
Indeed, they are connected by (7.9). On the other hand, they are regular
by (II, 8.18). Hence the local rings are all integral domains, so we could
not have two irreducible components meeting at a point.

Now that we have proved the duality theorem (7.6), our next task is to
give more information about the dualizing sheaf w~ in some special cases.
Again we need some algebraic preliminaries.
Let A be a ring, and let f 1 , . . . ,f.. EA. We define the Koszul complex
K.(f1 , . •. ,f..) as follows: K 1 is a free A-module of rank r with basis e 1 , . . . ,e,.
For each p = 0, ... ,r, KP = NK 1 • We define the boundary map d:KP ~
Kv_ 1 by its action on the basis vectors:

Thus K.(fto ... ,f..) is a (homological) complex of A-modules. If M is any

A-module, we set K.(fto ... ,f..; M) = K.(f1 , . . . ,f..) ®AM.

Proposition 7.10A. Let A be a ring, f 1, ... ,f.. E A, and let M be an A-module.

If the J; form a regular sequence forM, then

h;(K.(f1 , ••• ,f..; M)) = 0 fori> 0

and
h 0 (K.(f1 , • .. ,f..; M)) ~ M/(/1, ... ,f..)M.

PROOF. Matsumura [2, Th. 43, p. 135] or Serre [11, IV.AJ.

Theorem 7.11. Let X be a closed subscheme of P = Pf which is a local com-

plete intersection of codimension r. Let ~ be the ideal sheaf of X. Then
w~ ~ Wp ® N(~/~ 2 (. In particular, w~ is an invertible sheaf on X.

245
III Cohomology

PROOF. We have to calculate w~ = l%'xt';,((9x,Wp). Let U be an open affine

subset over which §can be generated by r elements fb . .. ,f.. E A = T( U,(!)u)
and let x E X n U be a point corresponding to an ideal m c:; A. Because
X has codimension rand Am is Cohen-Macaulay, f 1 , . . . ,f.. form a regular
sequence for Am (II, 8.21A). Therefore the localized Koszul complex
K.(f1 , ••• ,f..; Am) gives a free resolution of Am/Ub ... ,f..)Am over A,m so
replacing U by a smaller neighborhood of x if necessary, K.(/1 , . . . ,f..) gives
a free resolution of A/(!1 , ••• ,f..) over A. Sheafifying gives a free resolution
K.(fb . .. ,f..; (!)p) of (!)x over U with which we can calculate 1%'xtp((!)x,Wp)
(6.5). We get
h'(Yfom(K.(f1 , •.• ,f..; (!)p),wp)) ;::;; wpj(f1 , ••• ,f..)wp.
In other words,
I%'Xtp({!)x,Wp) ;::;; Wp @ {!)X
over U. However, this isomorphism depends on the choice of basis f1> ... ,J,
for f. If g; = '[.cijjj, i = 1, ... ,r, is another basis, then the exterior powers
of the matrix llcijll give an isomorphism of Koszul complexes. In particular,
we have a factor of detlcijl on K" so our isomorphism of l%'xt' changes by
detlcijl·
To remedy this situation, we consider the sheaf fjf 2 on X, which is
locally free of rank r (II, 8.21A). In particular, it is free over U, with basis
f 1 , . . . ,f... Therefore 1\ '(f/f 2 ) is free of rank 1, with basis f 1 1\ . . . 1\ f...
If we change to the basis gb ... ,g" this element changes by det lcijl· There-
fore, we can obtain an intrinsic isomorphism above by tensoring with this
free sheaf of rank 1 (check variance!)
l%'xtp({!)x,Wp) ;::;; Wp@ {!)X@ 1\ '(fjf 2 f.
This isomorphism, defined over U, is independent of the choice of basis.
Therefore when we cover P with such open sets, these isomorphisms glue to-
gether, and we obtain the required isomorphism w~ ;::;; Wp@ 1\ '(f/f 2 f.
Corollary 7.12. If X is a projective nonsingular variety over an algebraically
closed field k, then the dualizing sheaf w~ is isomorphic to the canonical
sheaf Wx·
PROOF. Embed X in P == Pf. Then X is a local complete intersection in P
(II, 8.17), and Wx;::;; Wp @ 1\ '(fjf 2 f by (II, 8.20).

Remark 7.12.1. Thus for a projective nonsingular variety X, the duality

theorem (7.6) and its corollary (7.7) hold with Wx in place of w~. In particular,
we obtain an isomorphism Hn(X,wx) ;::;; k, whose existence is by no means
obvious a priori.

Remark 7.12.2. If X is a projective nonsingular curve, we find that H 1(X,(!)x)

and H 0 (X,wx) are dual vector spaces. Hence the arithmetic genus Pa =
dim H 1(X,(!)x) and the geometric genus p9 = dim r(X,wx) are equal-
cf. (Ex. 5.3a) and (II. 8.18.2).

246
 7 The Serre Duality Theorem

Remark 7.12.3. If X is a projective nonsingular surface, then H 0 (X,w) is

dual to H 2 (X,(!)x), so p9 = dim H 2 (X,(!)x). On the other hand Pa =
dim H 2 (X,(!)x) - dim H 1(X,(!)x) by (Ex. 5.3a). Thus p9 )! Pa· The difference,
p9 - Pa = dim H 1 (X,(!)x) is usually denoted by q, and is called the irregularity
of X. For example, the surface of (II, Ex. 8.3c) has irregularity 2.

Corollary 7.13. Let X be a nonsingular projective variety of dimension n. For

any p = 0,1, ... ,n, let QP = NQx k be the sheaf of d(fferential p-forms.
Then for each p,q = 0,1, ... ,n, we have a natural isomorphism
Hq(X,QP) ~ Hn-q(X,Qn-p)'.

PROOF. Indeed, for any p, Qn-p ~ (QPr ® w (II, Ex. 5.16b). Then use
(7.7).

Remark 7.13.1. The numbers hp,q = dim Hq(X,QP) are important biregular
invariants of the variety X.

Remark 7.14 (Residues of Differentials on Curves). A weakness of the duality

theorem as we have proved it is that even for a nonsingular projective variety
X, we don't have much information about the trace map t:Hn(X,w)-+ k.
We know only that it exists. In the case of curves, there is another way of
proving the duality theorem, using residues, which improves this situation.
Let X be a complete nonsingular curve over an algebraically closed field k,
and let K be the function field of X. Let Qx be the sheaf of differentials of
X over k, and for a closed point P EX, let QP be its stalk at P. Let QK be
the module of differentials of K over k. Then one first proves:

Theorem 7.14.1 (Existence of Residues). For each closed point P EX, there is
a unique k-lineur map resp: QK --+ k with the following properties:
(a) resp( r) = 0 for all r E Qp;
(b) resp(Fdf) = 0 for all f E K*, all n # -1;
(c) resp(f- 1 df) = vp(f) · 1, where Vp is the wluution associated toP.

From these properties we see immediately how to calculate the residue

of any differential. Indeed, let t E (!)P be a uniformizing parameter. Then
dt is a generator for QK as a K-vector space, so we can write any r E QK as
gdt for some g E K. Furthermore, since (!)Pis a valuation ring, we can write
g = Li<O ai + h with ai E k, hE @p, and the sum finite. Thus r =
'f.a/dt + hdt. Now from linearity and (a), (b), (c) we find
(d) res P r = a_ 1 .
Thus the uniqueness of resp is clear.
The existence is more difficult. One approach by Serre [7, Ch. II] is to
take (d) as the definition of the residue. Then one has an awkward time
proving that it is independent of the choice of the uniformizing parameter t,
especially in the case of characteristic p > 0. Another approach by Tate [2]

247
III Cohomology

gives an intrinsic construction of the residue map by a clever use of certain

k-linear transformations of K.
The basic result about residues is:

Theorem 7.14.2 (Residue Theorem). For any r E QK, we have LPex resp r = 0.
In Serre's approach this theorem is first proved on Pl, by explicit cal-
culation. Then the general case is obtained by using a finite morphism
X --+ P 1 and studying the relationship between the residues in both places.
In Tate's approach the residue theorem follows directly from the construction
of the residue map.
Once one has the theory of residues, the duality theorem for X can be
proved by a method of Weil using repartitions. We refer to the lucid ex-
positions of Serre and Tate mentioned above for the details of this classic
story.
The connection with our approach can be explained as follows. The
exact sequence
0 --+ (!)x --+ :ffx --+ :ffx/(!)x --+ 0,
where :ffx is the constant sheaf Kx, is a ftasque resolution of (!)x (cf. Ex. 2.2).
Furthermore,
:ffx/(!)x ~ · EB i*(Kxf(!)p)
PeX

where we consider Kxf(!)p as an {!}p-module, and i: fP} --+ X is the inclusion

map. Tensoring with Qx, we get a flasque resolution of Qx:
0 --+ Qx --+ Qx ® :ffx--+ EB i*(QKjQp) --+ 0.
PeX

Taking cohomology, we get an exact sequence

QK --+ EB QKjQp --+ H 1 (X,Qx) --+ 0.

We define a map

by taking the sum of all the maps resp: QKjQp --+ k. Then by (7.14.2) this map
vanishes on the image of QK, hence it passes to the quotient and gives a map
t:H 1(X,Qx) --+ k. This is the trace map of our duality theorem, which appears
now in a much more explicit form.

Remark 7.15 (The Kodaira Vanishing Theorem). Our discussion of the co-
homology of projective varieties would not be complete without mentioning
the Kodaira vanishing theorem. It says if X is a projective nonsingular
variety of dimension n over C, and if 2 is an ample invertible sheaf on X,
then:
(a) Hi(X,!£1 ® w) = 0 fori > 0;
(b) H;(X,!£- 1) = 0 fori < n.

248
 7 The Serre Duality Theorem

Of course (a) and (b) are equivalent to each other by Serre duality. The
theorem is proved using methods of complex analytic differential geometry.
At present there is no purely algebraic proof. On the other hand, Raynaud has
recently shown that this result does not hold over fields of characteristic p > 0.
The first proof was given by Kodaira [ 1]. For other proofs, including
the generalization by Nakano, see Wells [1, Ch. VI, §2], Mumford [3], and
Ramanujam [1]. For a relative version of the theorem, see Grauert and
Riemenschneider [ 1].

References for the Duality Theorem. The duality theorem was first proved
by Serre [2] (in the form of (7.7)) for locally free sheaves on a compact
complex manifold, and in the case of abstract algebraic geometry by Serre [ 1].
Our proof follows Grothendieck [5] and Grothendieck [SGA 2, exp. XII],
with some improvements suggested by Lipman. The duality 'theorem and
the theory of residues have been generalized to the case of an arbitrary
proper morphism by Grothendieck ~see Grothendieck [ 4] and Hartshorne
[2]. Deligne has given another proof of the existence of a dualizing sheaf,
and Verdier [ 1] has shown that this one agrees with the sheaf w for a non-
singular variety. Kunz [1] gives another construction, using differentials,
of the dualizing sheaf wxfor an integral projective scheme X over k.
The duality theorem has also been generalized to the case of a proper
morphism of complex analytic spaces~see Ramis and Ruget [I] and Ram is,
R uget, and Verdier [I]. For a generalization to noncom pact complex mani-
folds, see Suominen [ 1].
In the case of curves, the duality theorem is the most important ingredient
in the proof of the Riemann-Roch theorem (IV, §1). See Serre [7, Ch. II]
for the history of this approach, and also Gunning [ 1] for a proof in the
language of compact Riemann surfaces.

EXERCISES

7.1. Let X be an integral projective scheme of dimension ~ 1 over a field k, and let
!£1 be an ample invertible sheaf on X. Then H 0 (X,!£1- 1 ) = 0. (This is an easy
special case of Kodaira's vanishing theorem.)
7.2. Let f: X-+ Y be a finite morphism of projective schemes of the same dimension
over a field k, and let w~ be a dualizing sheaf for Y.
(a) Show that f' w~ is a dualizing sheaf for X, where f' is defined as in (Ex. 6.10).
(b) If X and Y are both nonsingular, and k algebraically closed, conclude that
there is a natural trace map t: f*wx -+ wy.
7.3. Let X = P~. Show that Hq(X,Q~) = 0 for p ¥- q, k for p = q, 0 ::;; p, q::;; n.
*7.4. The Cohomology Class of a Subvariety. Let X be a nonsingular projective variety
of dimension n over an algebraically closed field k. Let Y be a nonsingular sub-
variety of codimension p (hence dimension n - p). From the natural map Qx ®
(!iy-+ Qr of (II, 8.12) we deduce a map QX-p-+ a;,-p. This induces a map on
cohomology w-p(X,QX-P)-+ w-p(Y,Q'VP). Now Q~-p = Wy is a dualizing sheaf

249
III Cohomology

for Y, so we have the trace map ty:H"-P(Y,Q'Y-p)-> k. Composing, we obtain a

linear map H"-P(X,!l'Jc-P)-> k. By (7.13) this corresponds to an element IJ(Y) E
HP(X,Q'\-), which we call the cohomology class of Y.
(a) If P EX is a closed point, show that tx(IJ(P)) = 1, where IJ(P) E H"(X,Q") and
t x is the trace map.
(b) If X= P", identify W(X,QP) with k by (Ex. 7.3), and show that I](Y) = (deg Y) · 1,
where deg Y is its degree as a projective variety (1, §7). [Hint: Cut with a hyper-
plane H c::; X, and use Bertini's theorem (II, 8.18) to reduce to the case Y is a
finite set of points.]
(c) For any scheme X of finite type over k, we define a homomorphism of sheaves
of abelian groups dlog:@}-> !lx by dlog(f) = f- 1 df Here (9* is a group
under multiplication, and !lx is a group under addition. This induces a map on
cohomology Pic X = H 1 (X,@k) -> H 1 (X,Qx) which we denote by c-see
(Ex. 4.5).
(d) Returning to the hypotheses above, suppose p = 1. Show that IJ(Y) = c(£'(Y) ),
where 2'(Y) is the invertible sheaf corresponding to the divisor Y.
See Matsumura [1] for further discussion.

8 Higher Direct Images of Sheaves

For the remainder of this chapter we will be studying families of schemes.

Recall (II, §3) that a family of schemes is simply a morphism f: X --+ Y, and
the members of the family are the fibres X Y = X x y Spec k(y) for various
points y E Y. To study a family, we need some form of"relative cohomology
of X over Y," or "cohomology along the fibres of X over Y." This notion is
provided by the higher direct image functors Rj* which we define below. The
precise relationship between these functors and the cohomology of the fibres
X Y will be studied in §11, 12.

Definition. Let f: X --+ Y be a continuous map of topological spaces. Then

we define the higher direct image functors Rj*: ~b(X) --+ ~b( Y) to be the
right derived functors of the direct image functor f* (II, §1).
This makes sense because f* is obviously left exact, and ~b(X) has
enough injectives (2.3).

Proposition 8.1. For each i ~ 0 and each ~ E ~b(X), Rj*(~) is the sheaf
associated to the presheaf
v f--+ Hi(f- 1 (V),~IJ-1(V))
on Y.

PROOF. Let us denote the sheaf associated to the above presheafby Yl'i(X,~).
Then, since the operation of taking the sheaf associated to a presheaf is
exact, the functors Yl'i(X, ·) form a J-functor from ~b(X) to ~b(Y). For
i = 0 we have f* ~ = Yl' 0 (X,~) by definition off*. For an injective object
J E ~b(X) we have Rj*(J) = 0 for i > 0 because Rj* is a derived functor.

250
 8 Higher Direct Images of Sheaves

On the other hand, for each V, .J'if- 'W> is injective in m:b(f- 1(V)) by (6.1)
(think of X as a ringed space with the constant sheaf Z), so £';(X,.J') = 0
for i > 0 also. Hence there is a unique isomorphism of <5-functors Ri*( ·) ~
£';(X,·) by (1.3A).

Corollary 8.2. If V c;; Y is any open subset, then

Ri*(~)iv = R'i"~(~IJ-'(VJ)
where f' :f- 1(V) --+ Vis the restricted map.

PROOF. Obvious.

Corollary 8.3. If~ is afiasque sheaf on X, then Ri*(~) = 0 for all i > 0.
PROOF. Since the restriction of a flasque sheaf to an open subset is flasque,
this follows from (2.5).

Proposition 8.4. Let f:X--+ Y be a morphism of ringed spaces. Then the

functors R1* can be calculated on Wlob(X) as the derived functors of
f*:Wlob(X)--+ Wlob(Y).

PROOF. To calculate the derived functors off* on Wlob(X), we use resolutions

by injective objects of Wlob(X). Any injective of Wlob(X) is flasque by (2.4),
hence acyclic for f* on m:b(X) by (8.3), so they can be used to calculate
Ri* by (1.2A).

Proposition 8.5. Let X be a noetherian scheme, and let f: X --+ Y be a morphism

of X to an affine scheme Y = Spec A. Then for any quasi-coherent sheaf
~on X, we have

PROOF. By (II, 5.8), f*~ is a quasi-coherent sheaf on Y. Hence f*~ ~

r( Y,f*~)-. But r( Y,f*~) = r(X,~). So we have an isomorphism for
i = 0.
Since is an exact functor from Wlob(A) to Wlob(Y), both sides are <5-
functors from .Qco(X) to Wlob(Y). Furthermore, by (3.6), any quasi-coherent
sheaf ~ on X can be embedded in a flasque, quasi-coherent sheaf. Hence
both sides are effaceable for i > 0. We conclude from (1.3A) that there is a
unique isomorphism of <5-functors as above, reducing to the given one for
i = 0.
Note that we must work in the category .Qco(X), because already the case
i = 0 fails if~ is not quasi-coherent.

Corollary 8.6. Let f:X --+ Y be a morphism of schemes, with X noetherian.

Then for any quasi-coherent sheaf~ on X, the sheaves Ri*(~) are quasi-
coherent on Y.

251
III Cohomology

PROOF. The question is local on Y, so we may use (8.5).

Proposition 8. 7. Let f: X --+ Y be a morphism of separated noetherian schemes.

Let$' be a quasi-coherent sheaf on X, let U = (U;) be an open affine cover
of X, and let ~·(U,$') be the Cech resolution of$' given by (4.2). Then
for each p ? 0,
Wf*($') ~ hP(f* ~·(U,$') ).
PROOF. For any open affine subset V t;: Y, the open subsets Ui n f- 1(V)
of X are all affine (check !-cf. (II, Ex. 4.3) ). Hence we may reduce to the
case Y affine. The sheaves ~P(U,$') are all quasi-coherent, so we have
f* ~·(U,$') ~ C"(U,$'r
by (II, 5.8). Now the result follows from (4.5) and (8.5).

Theorem 8.8. Let f: X --+ Y be a projective morphism of noetherian schemes,

let (Ox(1) be a very ample invertible sheaf on X over Y, and let$' be a coherent
sheaf on X. Then:
(a) for all n » 0, the natural map f*f*($'(n))--+ $'(n) is surjective;
(b) for all i ? 0, Ri_f*($') is a coherent sheaf on Y;
(c) fori > 0 and n » 0, Ri_f*($'(n)) = 0.
PROOF. Since Y is quasi-compact, the question is local on Y, so we may
assume Y is affine, say Y = Spec A. Then, using (8.5), (a) says that $'(n)
is generated by global sections, which is (II, 5.17). (b) says that Hi(X,$')
is a finitely generated A-module, which is (5.2a). Finally, (c) says that
Hi(X,$'(n)) = 0, which is (5.2b).

Remark 8.8.1. Part (b) of this theorem is true more generally for a proper
morphism of noetherian schemes-see Grothendieck [EGA III, 3.2.1]. The
analogous theorem for a proper morphism of complex analytic spaces was
proved by Grauert [1].

EXERCISES

8.1. Let f:X --+ Y be a continuous map of topological spaces. Let §'be a sheaf of
abelian groups on X, and assume that Ri*(ff) = 0 for all i > 0. Show that there
are natural isomorphisms, for each i ~ 0,

(This is a degenerate case of the Leray spectral sequence-see Godement [1, II,
4.17.1 ].)
8.2. Let f: X --+ Y be an affine morphism of schemes (II, Ex. 5.17) with X noetherian,
and let§' be a quasi-coherent sheaf on X. Show that the hypotheses of (Ex. 8.1)
are satisfied, and hence that H;(X,ff) ~ H;(Y,j*§') for each i ~ 0. (This gives
another proof of (Ex. 4.1).)

252
 9 Flat Morphisms

8.3. Let f:X--+ Y be a morphism of ringed spaces, let ff be an (l)x-module, and let
rff be a locally free (l)y-module of finite rank. Prove the projection formula (cf.
(II, Ex. 5.1))
R1*(ff ® f*rff) ~ R1*(ff) ® rff.
8.4. Let Y be a noetherian scheme, and let rff be a locally free (l)y-module of rank n + 1,
n ~ 1. Let X = P(rff) (II, §7), with the invertible sheaf (l)x(1) and the projection
morphism n:X --+ Y.
(a) Thenn*((l)(0) ~ S 1(rff)forl ~ O,n*((l)(0) = Oforl < 0(11, 7.1l);R;n*((l)(0) = 0
for 0 < i < n and all I E Z; and R"n*( (1)(0) = 0 for I > - n - 1.
(b) Show there is a natural exact sequence
0 --+ QX/Y --+ (n*rff)( -1) --+ (I) --+ 0,
cf. (II, 8.13), and conclude that the relative canonical sheaf wx1r = 1\ "Qx1r is
isomorphic to (n* 1\ •+ 1 rff)(- n - 1). Show furthermore that there is a natural
isomorphism R"n*(wx1r) ~ (I)Y (cf. (7.1.1) ).
(c) Now show, for any IE Z, that

R"n*((l)(0) ~ n*((l)(-1- n- l)f ® (/\"+ 1 rfff.

(d) Show that p.(X) = (-1)"p.(Y) (use (Ex. 8.1)) and pg(X) = 0 (use (II, 8.11) ).
(e) In particular, if Y is a nonsingular projective curve of genus g, and rff a locally
free sheaf of rank 2, then X is a projective surface with Pa = - g, pg = 0, and
irregularity g (7.12.3). This kind of surface is called a geometrically ruled surface
(V, §2).

9 Flat Morphisms

In this section we introduce the notion of a flat morphism of schemes. By

taking the fibres of a flat morphism, we get the notion of a flat family of
schemes. This provides a concise formulation of the intuitive idea of a
"continuous family of schemes." We will show, through various results and
examples, why flatness is a natural as well as a convenient condition to put
on a family of schemes.
First we recall the algebraic notion of a flat module. Let A be a ring, and
let M be an A-module. We say that M is fiat over A if the functor N H
M ®A N is an exact functor for N E Wlob(A). If A ~ B is a ring homomor-
phism, we say that B is fiat over A if it is flat as a module.

Proposition 9.1A.
(a) An A-module M is fiat if and only if for every finitely generated
ideal a s:::: A, the map a ® M ~ M is injective.
(b) Base extension: If M is a fiat A-module, and A~ B is a homomor-
phism, then M ®A B is a fiat B-module.
(c) Transitivity: If B is a fiat A-algebra, and N is a fiat B-module, then
N is also fiat as an A-module.
(d) Localization: M is fiat over A if and only if Mp is fiat over Ap for all
p E Spec A.

253
III Cohomology

(e) Let 0 ~ M' ~ M ~ M" ~ 0 be an exact sequence of A-modules.

If M' and M" are both fiat then M is fiat; if M and M" are both fiat, then
M' isfiat.
(f) A finitely generated module M over a local noetherian ring A is fiat
if and only if it is free.
PROOFS. Matsumura [2, Ch. 2, §3] or Bourbaki [1, Ch. I.].

Example 9.1.1. If A is a ring and S <;; A is a multiplicative system, then the

localization s- 1 A is a flat A-algebra. If A ~ B is a ring homomorphism,
if M is a B-module which is flat over A, and if S is a multiplicative system
in B, then S- 1 M is flat over A.

Example 9.1.2. If A is a noetherian ring and a <;; A an ideal, then the a-adic
completion A is a flat A-algebra {II, 9.3A).
Example 9.1.3. Let A be a principal ideal domain. Then an A-module M
is flat if and only if it is torsion-free. Indeed, by (9.1Aa) we must check that
for every ideal a <;; A, a @ M ~ M is injective. But a is principal,. say
generated by t, so this just says that t is not a zero divisor in M, i.e., M is
torsion-free.

Definition. Let f: X ~ Y be a morphism of schemes, and let :#' be an (1} x-

module. We say that:#' is fiat over Y at a point x EX, if the stalk~ is a
flat my.r-module, where y = f(x) and we consider ffx as an my,r-module
via the natural mapf#:my,Y ~ mx,X· We say simply:#' isfiat over Yif
it is flat at every point of X. We say X is fiat over Y if (1} x is.

Proposition 9.2.
(a) An open immersion is fiat.
(b) Base change: let f: X ~ Y be a morphism, let :#' be an (1} x-module
which is fiat over Y, and let g: Y' ~ Y be any morphism. Let X' = X x r Y',
let f': X' ~ Y' be the second projection, and let :#'' = pf(ff). Then :#''
is fiat over Y'.
(c) Transitivity: let f:X ~ Y and g: Y ~ Z be morphisms. Let :#'
be an mx-module which is fiat over Y, and assume also that Y is fiat over Z.
Then:#' is fiat over Z.
(d) Let A ~ B be a ring homomorphism, and let M be a B-module. Let
f: X = Spec B ~ Y = Spec A be the corresponding morphism of affine
schemes, and let :#' = M. Then :#' is fiat over Y if and only if M is fiat
over A.
(e) Let X be a noetherian scheme, and:#' a coherent mx-module. Then:#'
is fiat over X if and only if it is locally free.
PROOF. These properties all follow from the corresponding properties of
modules, taking into account that the functor ~ is compatible with @
(II, 5.2).

254
 9 Flat Morphisms

Next, as an illustration ofthe convenience of flat morphisms, we show that

"cohomology commutes with flat base extension":

Proposition 9.3. Let f: X ~ Y be a separated morphism of finite type of noe-

therian schemes, and let ff be a quasi-coherent sheaf on X. Let u: Y' ~ Y
be a fiat morphism of noetherian schemes.

X' _ __,V:....__~ X

Y' u y

Then for all i ~ 0 there are natural isomorphisms

u* Rj*(ff) ~ Rig*(v* $').
PROOF. The question is local on Y and on Y', so we may assume they are
both affine, say Y = Spec A and Y' = Spec A'. Then by (8.5) what we have
to show is that

Since X is separated and noetherian, and ff is quasi-coherent, we can

calculate Hi(X,ff) by Cech cohomology with respect to an open affine cover
U of X (4.5). On the other hand, {v- 1( U) IU E U} forms an open affine cover
U' of X', and clearly the Cech complex C(U',v* ff) is just C"(U,ff) ®A A'.
Since A' is flat over A, the functor · ®A A' commutes with taking cohomology
groups of the Cech complex, so we get our result. Note that g is also separated
and of finite type by base extension, so X' is also noetherian and separated,
allowing us to apply (4.5) on X'.

Remark 9.3.1. Even if u is not flat, this proof shows that there is a natural
map u*Rj*(ff) ~ Rig*(v*ff).

Corollary 9.4. Let f:X ~ Y and ff be as in (9.3), and assume Y affine. For
any point y E Y, let X Y be the fibre over y, and let ffy be the induced sheaf
On the other hand, let k(y) denote the constant sheaf k(y) on the closed
subset {y}- ofY. Thenfor all i ~ 0 there are natural isomorphisms
Hi(Xy,ffy) ~ Hi(X,ff ® k(y)).
PROOF. First let Y' c:; Y be the reduced induced subscheme structure on
{ y}-, and let X' = X x r Y', which is a closed subscheme of X. Then both
sides of our desired isomorphism depend only on the sheaf ff' = :#' ® k(y)
on X'. Thus we can replace X, Y,ff by X', Y',ff', i.e., we can assume that Y is
an integral affine scheme and that y E Y is its generic point. In that case,
Spec k(y) ~ Y is a flat morphism, so we can apply (9.3) and conclude that
Hi(Xy,ffy) ~ Hi(X,ff) ® k(y).

255
III Cohomology

But after our reduction, Hi(X,g-) is already a k(y)-module, so tensoring with

k( y) has no effect, and we obtain the desired result. (This result is used in
§12.)

Flat Families
For many reasons it is important to have a good notion of an algebraic
family of varieties or schemes. The most naive definition would be just to
take the fibres of a morphism. To get a good notion, however, we should
require that certain numerical invariants remain constant in a family, such
as the dimension of the fibres. It turns out that if we are dealing with non-
singular (or even normal) varieties over a field, then the naive definition is
already a good one. Evidence for this is the theorem (9.13) that in such a
family, the arithmetic genus is constant.
On the other hand, if we deal with nonnormal varieties, or more general
schemes, the naive definition will not do. So we consider a fiat family of
schemes, which means the fibres of a fiat morphism, and this is a very good
notion. Why the algebraic condition of flatness on the structure sheaves
should give a good definition of a family is something of a mystery. But
at least we will justify this choice by showing that fiat families have many
good properties, and by giving necessary and sufficient conditions for
flatness in some special cases. In particular, we will show that a family
of closed subschemes of projective space (over an integral scheme) is fiat if
and only if the Hilbert polynomials of the fibres are the same.

Proposition 9.5. Let f:X--> Y be a flat morphism of schemes of finite type

over afield k. For any point x EX, let y = f(x). Then
dimx(X y) = dimx X - dimy Y.
Here for any scheme X and any point x EX, by dimx X we mean the di-
mension of the local ring (!} x,x.
PROOF. First we make a base change Y' --> Y where Y' = Spec (!}y,Y• and
consider the new morphism f':X'--> Y' where X' = X x y Y'. Then f'
is also fiat by (9.2), x lifts to X', and the three numbers in question are the
same. Thus we may assume that y is a closed point of Y, and dimy Y = dim Y.
Now we use induction on dim Y. If dim Y = 0, then XY is defined by a
nilpotent ideal in X, so we have dimx(Xy) = dimx X, and dimy Y = 0.
If dim Y > 0, we make a base extension to Yrect· Nothing changes, so we
may assume that Y is reduced. Then we can find an element t E my s;;; (!}y,Y
such that tis not a zero divisor. Let Y' = Spec (!}y,Y/(t), and make the base
extension Y' --> Y. Then dim Y' = dim Y - 1 by (1, 1.8A) and (1, 1.11A).
Since f is fiat, f # t E mx is also not a zero divisor. So for the same reason,
dimx X' = dimx X - 1. Of course the fibre X Y does not change under base
extension, so we have only to prove our formula for f': X' --> Y'. But this
follows from the induction hypothesis, so we are done.

256
 9 Flat Morphisms

Corollary 9.6. Let f:X --+ Y be a fiat morphism of schemes of finite type over
a field k, and assume that Y is irreducible. Then the following conditions
are equivalent:

(i) every irreducible component of X has dimension equal to dim Y + n;

(ii) for any pointy E Y (closed or not), every irreducible component of the
fibre X Y has dimension n.

PROOF.
=
(i) (ii). Given y E Y, let Z <;:: X Y be an irreducible component, and let
x E Z be a closed point, which is not in any other irreducible component
of XY. Applying (9.5) we have

dimx Z = dimx X - dimy Y.

Now dimx Z = dim Z since xis a closed point (II, Ex. 3.20). On the other
hand, since Y is irreducible and X is equidimensional, and both are of finite
type over k, we have (II, Ex. 3.20)

dimx X = dim X - dim{x}-

dimY Y = dim Y - dim { y}-.

Finally, since x is a closed point of the fibre X Y' k(x) is a finite algebraic
extension of k(y) and so
dim {x}- = dim {y}-.

Combining all these, and using (i) we find dim Z = n.

(ii) = (i). This time let Z be an irreducible component of X, and let
x E Z be a closed point which is not contained in any other irreducible
component of X. Then applying (9.5), we have

dimx(X y) = dimx X - dimy Y.

But dimx(X y) = n by (ii), dimx X = dim Z, and dimY Y = dim Y, since

y = f(x) must be a closed point of Y. Thus

dim Z = dim Y +n
as required.

Definition. A point x of a scheme X is an associated point of X if the maximal

ideal mx is an associated prime of 0 in the local ring (!) x.x. or in other
words, if every element of mx is a zero divisor.

Proposition 9.7. Let f:X --+ Y be a morphism of schemes, with Y integral

and regular of dimension 1. Then f is fiat if and only if every associated
point x E X maps to the generic point of Y. In particular, if X is reduced,
this says that every irreducible component of X dominates Y.

257
III Cohomology

PROOF. First suppose that f is flat, and let x EX be a point whose image
y = f(x) is a closed point of Y. Then @y,Y is a discrete valuation ring. Let
t E my - m; be a uniformizing parameter. Then t is not a zero divisor in
(1) y,Y· Since f is flat, f # t E mx is not a zero divisor, so x is not an associated
point of X.
Conversely, suppose that every associated point of X maps to the generic
point of Y. To show f is flat, we must show that for any x EX, letting y =
f(x), the local ring @x,x is flat over @y,Y· If y is the generic point, @y,Y is a
field, so there is nothing to prove. If y is a closed point, @y,Y is a discrete
valuation ring, so by (9.1.3) we must show that @x,x is a torsion-free module.
If it is not, then f # t must be a zero divisor in mx, where t is a uniformizing
parameter of @y,Y· Therefore f#t is contained in some associated prime
ideal :p of(O) in (1)x (Matsumura [2, Cor. 2, p. 50]). Then :p determines a point
x' E X, which is an associated point of X, and whose image by f is y, which is a
contradiction.
Finally, note that if X is reduced, its associated points are just the generic
points of its irreducible components, so our condition says that each ir-
reducible component of X dominates Y.

Example 9.7.1. Let Y be a curve with a node, and let f:X---> Y be the map
of its normalization to it. Then f is not flat. For if it were, then j*(1) x would
be a flat sheaf of @y-modules. Since it is coherent, it would be locally free by
(9.2e). And finally, since its rank is 1, it would be an invertible sheaf on Y.
But there are two points P 1 ,P 2 of X going to the node Q of Y, so (j*@x)Q
needs two generators as an <'9y-module, hence it cannot be locally free.

Example 9.7.2. The result of (9.7) also fails if Y is regular of dimension > 1.
For example, let Y = A 2 , and let X be obtained by blowing up a point. Then
X and Y are both nonsingular, and X dominates Y, but f is not flat, because
the dimension of the fibre over the blown-up point is too big (9.5).

Proposition 9.8. Let Y be a regular, integral scheme of dimension 1, let P E Y

be a closed point, and let X s; P~ _P be a closed sub scheme which is fiat
over Y - P. Then there exists a unique closed subscheme X s; P~, fiat
over Y, whose restriction to P~ _ P is X.

PROOF. Take X to be the scheme-theoretic closure of X in P~ (II, Ex. 3.1ld).

Then the associated points of X are just those of X, so by (9.7), X is flat
over Y. Furthermore, X is unique, because any other extension·of X toP~
would have some associated points mapping to P.

Remark 9.8.1. This proposition says that we can "pass to the limit," when we
have a flat family of closed subschemes of pn over a punctured curve. Hence
it implies that "the Hilbert scheme is proper." The Hilbert scheme is a

258
 9 Flat Morphisms

scheme H which parametrizes all closed subschemes ofP;:. It has the property
that to give a closed subscheme X s; P;., flat over T, for any scheme T, is
equivalent to giving a morphism cp: T--+ H. Here, naturally, for any t E T,
cp(t) is the point of H corresponding to the fibre X 1 s; P;:(tJ·
Now once one knows that the Hilbert scheme exists (see Grothendieck
[5, exp. 221]) then the question of its properness can be decided using the
valuative criterion of properness (II, 4.7). And the result just proved is the
essential point needed to show that each connected component of H is
proper over k.

Example 9.8.2. Even though the dimension of the fibres is constant in a

flat family, we cannot expect properties such as "irreducible" or "reduced"
to be preserved in a flat family. Take for example, the families given in
(II, 3.3.1) and (II, 3.3.2). In each case the total space X is integral, the base
Y is a non singular curve and the morphism f: X --+ Y is surjective, so the
family is flat. Also most fibres are integral in both families. However, the
special fibre in one is a doubled line (not reduced), and the special fibre in the
other is two lines (not irreducible).

Example 9.8.3 (Projection from a Point). We get some new insight into the
geometric process of projection from a point (I, Ex. 3.14) using (9.8). Let
P = (0,0, ... ,0,1) E pn+ 1 , and consider the projection cp:Pn+ 1 - {P}--+ pn,
which is defined by (x 0 , . . . ,xn+ 1 ) ~ (x 0 , . . . ,xn). For each a E k, a =1- 0,
consider the automorphism a a of pn+ 1 defined by (x 0 , . . . ,xn+ 1 ) ~
(x 0 , . . . ,xmaxn+ 1 ). Now let X 1 be a closed subscheme ofpn+ \not containing
P. For each a =1- 0, let Xa = aa(X 1 ). Then the Xa form a flat family param-
etrized by A 1 - {0}. It is flat, because the Xa are all isomorphic as abstract
schemes, and in fact, the whole family is isomorphic to X 1 x (A 1 - { 0}) if
we forget the embedding in pn+ 1 .
Now according to (9.8) this family extends uniquely to a flat family defined
over all of A 1 , and clearly the fibre X 0 over 0 agrees, at least set-theoretically,
with the projection cp(X 1 ) of X 1 . Thus we see that there is a flat family over
A\ whose fibres for all a =1- 0 are isomorphic to X t> and whose fibre at 0 is
some scheme with the same underlying space as cp(X 1 ).

Example 9.8.4. We will now calculate the flat family just described in the
special case where X 1 is a twisted cubic curve in P 3 , cp is a projection to P 2 ,
and cp(X 1) is a nodal cubic curve in P 2 . The remarkable result of this cal-
culation is that the special fibre X 0 of our flat family consists of the curve
cp(X 1) together with some nilpotent elements at the double point! We say
that X 0 is a scheme with an embedded point. It seems as if the scheme X 0 is
retaining the information that it is the limit of a family of space curves, by
having these nilpotent elements which point out of the plane. In particular,
X 0 is not a closed subscheme of P 2 (Fig. 11 ).

259
III Cohomology

Figure 11. A flat family of subschemes of P 3 .

Now for the calculation. We are just interested in what happens near the
double point, so we will use affine coordinates x,y in A 2 and x,y,z in A3 •
Let X 1 be given by the parametric equations
X = t2 - 1
{ y = t3 - t
z = t.

Then since t = z, t 2 = x + 1, t 3 = y + z, we recognize this as a twisted

cubic curve in A 3 (1, Ex. 1.2).
Now for any a =I= 0, the scheme X a is given by

{:z :=at.~: =:
To get the ideal I ~ k[a,x,y,z] of the total family X extended over all of A 1,
we eliminate t from the parametric equations, and make sure a is not a
zero divisor in k[a,x,y,z]/I, so that X will be flat. We find
I = (a 2 (x + 1) - z 2 , ax(x + 1) - yz, xz - ay, y 2 - x 2 (x + 1)).

From this, setting a = 0, we obtain the ideal I 0 ~ k[ x,y,z] of X 0 , which is

10 = (z 2 ,yz,xz,y2 - x 2 (x + 1)).
So we see that X 0 is a scheme with support equal to the nodal cubic curve
y2 = x 2 (x + 1). At any point where x =I= 0, we get z in the ideal, so X 0 is
reduced there. But in the local ring at the node (0,0,0), we have the element z
with z 2 = 0, a nonzero nilpotent element.
So here we have an example of a flat family of curves, whose general
member is nonsingular, but whose special member is singular, with an
embedded point. See also (9.10.1) and (IV, Ex. 3.5).

260
 9 Flat Morphisms

Example 9.8.5 (Algebraic Families of Divisors). Let X be a scheme of finite

type over an algebraically closed field k, let T be a nonsingular curve over k,
and let D be an effective Cartier divisor on X x T (II, §6). Then we can think
of D as a closed subscheme of-X x T, which is locally described on a small
open set U as the zeros of a single element f E r(U,(!)u) such that f is not a
zero divisor. For any closed point t E T, let X 1 (:~X) be the fibre of X x T
over t. We say that the intersection divisor D1 = D.X1 is defined if at every
point of X 0 the image J E r(U n X 1,(Dx,) of a local equation f of D is not a
zero divisor. In that case the covering {U n X 1 } and the elements J define a
Cartier divisor D 1 on X 1 • If D 1 is defined for all t, we say that the divisors
{D1 lt E T} form an algebraic family of divisors on X parametrized by T.
This definition, which is natural in the context of Cartier divisors, is
connected with flatness in the following way: the original Cartier divisor D,
considered as a scheme over T, is flat over T if and only if D1 = D.X 1 is
defined for each t E T. Indeed, let x E D be any point, let A = (!) x,x x r be
the local ring of x on X x T, let f E A be a local equation forD, let p 2 (x) = t,
and let u E (!)r.r be a uniformizing parameter. Then D.X 1 is defined at x if
and only if J E AjuA is not a zero divisor. Since u is automatically not a
zero divisor in A, this is equivalent to saying that (u,f) is a regular sequence
(II, §8) in A. On the other hand, D is flat over T at x if and only if (!) x,D is
flat over (!)r.r· By (9.1.3) this is equivalent to (!)x,D being torsion-free, i.e., u
not being a zero divisor in (!) x,n· But (!) x,n ~ A/fA, so this says that (f,u)
is a regular sequence in A. Since the property of being a regular sequence is
independent of the order of the sequence (Matsumura [2, Th. 28, p. 102]),
the two conditions are equivalent.

Theorem 9.9. Let T be an integral noetherian scheme. Let X £ PT be a

closed subscheme. For each point t E T, we consider the Hilbert polynomial
P 1 E Q[ z] of the fibre X 1 considered as a closed subscheme of P~<t)· Then
X is flat over T if and only if the Hilbert polynomial P 1 is independent oft.
PROOF. Recall that the Hilbert polynomial was defined in (1, §7), and com-
puted another way in (Ex. 5.2). We will use the defining property that
P 1(m) = dimk(t) H 0 (X 1,(!)x.(m))
for all m » 0.
First we generalize, replacing (!) x by any coherent sheaf §' on P~, and
using the Hilbert polynomial of ffi;. Thus we may assume X = P~. Se-
cond, the question is local on T. In fact, by comparing any point to the•
generic point, we see that it is sufficient to consider the case T = Spec A,
with A a local noetherian ring.
So now let T = Spec A with A a local noetherian domain, let X = P~,
and let §'be a coherent sheaf on X. We will show that the following con-
ditions are equivalent:
(i) §' is flat over T;
(ii) H 0 (X,§'(m)) is a free A-module of finite rank, for all m » 0;

261
III Cohomology

(iii) the Hilbert polynomial Pt of ffr on Xt = Pi:<t> is independent of t, for

any t E T.
(i) =(ii). We compute H;(X,$'(m)) by Cech cohomology using the stan-
dard open affine cover U of X. Then
H;(X,$'(m)) = h;( C(U,$'(m)) ).
Since$'" is flat, each term C(U,$'(m)) of the Cech complex is a flat A-module.
On the other hand, if m » 0, then H;(X,$'(m)) = 0 fori > 0, by (5.2). Thus
the complex C(U,$'(m)) is a resolution of the A-module H 0(X,$'(m) ): we
have an exact sequence
0 -+ H 0(X,$'(m)) -+ C 0 (U,$'(m)) -+ C 1 (U,$'(m)) -+ ... -+ C"(U,$'(m)) -+ 0.
Splitting this into short exact sequences, using (9.1Ae) and the fact that the
C; are all flat, we conclude that H 0(X,$'(m)) is a flat A-module. But it is
also finitely generated (5.2), and hence free of finite rank by (9.1Af).
(ii)= (i). Let S = A[ x 0 , . . . ,xn], and let M be the graded S-module
M = EB H 0(X,$'(m) ),
m~mo

where m0 is chosen large enough so that the H 0(X,$'(m)) are all free for
m ): m 0 . Then$'" = M by (II, 5.15). Note that M is the same as r *($'")in
degrees m ): m 0 , so M = r *($'r. Since M is a free (and hence flat) A-
module, we see that$'" is flat over A (9.1.1).
(ii) = (iii). It will be enough to show that
Pt(m) = rank A H 0(X,$'(m))
for m » 0. To prove this we will show, for any t E T, that
H 0(Xt,fft(m)) ~ H 0(X,$'(m)) 0 A k(t)
for all m » 0.
First we let T' = Spec AP, where p is the prime ideal corresponding to t,
and we make the flat base extension T' -+ T. Thus by (9.3) we reduce to
the case where t is the closed point of T. Denote the closed fibre Xt by X 0 ,
ffr by $'0 , and k(t) by k. Take a presentation of k over A,
Aq -+ A -+ k -+ 0.
Then we get an exact sequence of sheaves on X,
$'q -+ $'" -+ $'0 -+ 0.
Now by (Ex. 5.10) form » 0 we get an exact sequence
H 0(X,$'(m)q) -+ H 0(X,$'(m)) -+ H 0(X 0 ,$'0 (m)) -+ 0.
On the other hand, we can tensor the sequence Aq -+A -+ k-+ 0 with
H 0(X,$'(m) ). Comparing, we deduce that
H 0(X 0,$'0(m)) ~ H 0(X,$'(m)) @A k
for all m » 0, as required.

262
 9 Flat Morphisms

(iii) => (ii). According to (II, 8.9) we can check the freeness of H 0 (X,.?F(m))
by comparing its rank at the generic point and the closed point of T. Hence
the argument of (ii) => (iii) above is reversible.

Corollary 9.10. Let T be a connected noetherian scheme, and let X s; PT be

a closed subscheme which is flat over T. For any t E T, let X 1 be the fibre,
considered as a closed subscheme of Pi:(r)· Then the dimension of X 1 , the
degree of X 1 , and the arithmetic genus of X 1 are all independent oft.
PRooF. By base extension to the irreducible components of T with their
reduced induced structure, we reduce to the case T integral. Now the result
follows from the theorem and the facts (1, §7) and (Ex. 5.3) that
dim X 1 = deg P 1 ,
deg X 1 = (r !) · (leading coefficient of P1),
where r = dim X, and

Definition. Let k be an algebraically closed field, let f: X -+ T be a surjective

map of varieties over k, and assume that for each closed point t E T, we
have
(I) f - 1 (t) is irreducible of dimension equal to dim X - dim T, and
(2) if m1 s; (!)r,T is the maximal ideal, and if' E f- 1(t) is the generic point,
thenf#m1 generates the maximal ideal m, s; (!)I;,X·
Under these circumstances, we let X<rl be the variety f- 1(t) (with the
reduced induced structure) and we say that the X(r) form an algebraic
family of varieties, parametrized by T. The second condition is necessary
to be sure that X<rl occurs with "multiplicity one" in the family. It is
equivalent to saying that the scheme-theoretic fibre X 1 is reduced at its
generic point.

Example 9.10.1. In the flat family of (9.8.4), if we take the fibres with their
reduced induced structures, we get an algebraic family of varieties X<n
parametrized by A 1 . For t i= 0 it is a nonsingular rational curve, and for
t = 0 it is the plane nodal cubic curve. Note that the arithmetic genus is
not constant in this family: Pa(X<n) = 0 for t i= 0 and Pa(X< 0 l) = 1. This
accounts for the appearance of nilpotent elements in the scheme-theoretic
fibre X 0 , since in a flat family of schemes Pais constant by (9.10). The em-
bedded point at 0 alters the constant term of the Hilbert polynomial so
that we get Pa(X 0 ) = 0.

Theorem 9.11. Let X<rl be an algebraic family of normal varieties parame-

trized by a nonsingular curve T over an algebraically closed field k. Then
X<n is a flat family of schemes.

263
III Cohomology

PROOF. Let f:X ~ T be the defining morphism of the family. Then f is a

flat morphism by (9.7). So we have only to show that for each closed point
t E T, the scheme-theoretic fibre xt coincides with the variety x(t)• In other
words, we must show that X 1 is reduced. For any point x EX, let A = (!)x.x
be its local ring, let f(x) = t, and denote also by t a uniformizing parameter
in the local ring @1_r· Then AjtA is the local ring of x on X 1• By hypothesis
X 1 is irreducible, so t has a unique minimal prime ideal p in A. Furthermore,
t generates the maximal ideal of the local ring of the generic point of X 1
on X, which says that t generates the maximal ideal of AP. Finally, the local
ring of x on X(t) is Ajp, so our hypothesis says that Ajp is normal. Now our
result is a consequence of the following lemma, which tells us that p = tA,
so x(l) = xt.

Lemma 9.12 (Lemma ofHironaka [1]). Let A be a local noetherian domain,

which is a localization of an algebra of.finite type over a field k. Let t E A,
and assume
(1) tA has only one minimal associated prime ideal p,
(2) t generates the maximal ideal of AP,
(3) Ajp is normal.
Then p = tA and A is normal.
PROOF. Let A be the normalization of A. Then A is a finitely generated
A-module by (1, 3.9A). We will show that the maps
cp: Aft A ~ A/tA
and
ljJ: Aft A ~ Ajp
are both isomorphisms.
First we localize at p. Then ljJ is an isomorphism by hypothesis. There-
fore AP is a discrete valuation ring, hence normal. So AP = AP and cp is also
an isomorphism.
Now suppose that at least one of cp,lj; is not an isomorphism. Then, after
localizing A at a suitable prime ideal, we may assume that cp and ljJ are
isomorphisms at every localization Aq with q -=1- m, but that at least one of
cp,l/J is not an isomorphism at m. By the previous step, we have p < m, so
dim A ~ 2. Now A is normal of dimension ~ 2, so it has depth ~ 2 (II,
8.22A), so AjtA has depth ~ 1. Therefore it does not have mas an associated
prime. On the other hand, Aft A agrees with Ajp outside of m, so we conclude
that AjtA is an integral domain. Thus we have a natural map (AjtA)red ~
AjtA. But (A/tA)red ~ Ajp since ljJ is an isomorphism outside m. Thus
AjtA is a finitely generated (A/p)-module with the same quotient field. Since
Ajp is normal by hypothesis, we conclude that AjtA ~ Ajp. Therefore cp
is surjective, so we can write A = A + tA. By Nakayama's lemma, this
implies that A = A. Thus AjtA ~ Ajp and both cp and ljJ are isomorphisms.
But this is a contradiction, so we conclude that cp and ljJ were already iso-
morphisms on the original ring A before localization.

264
 9 Flat Morphisms

To conclude, we find that p = tA because 1/J is an isomorphism. Since

<pis an isomorphism, we have A = A + tA, so by Nakayama's lemma as
before, we find that A = A, so A is normal.

Corollary 9.13 (Igusa [1]). Let X<tl be an algebraic family of normal varieties
in PZ, parametrized by a variety T. Then the Hilbert polynomial of x(t)'
and hence also the arithmetic genus Pa(X<1l), are independent oft.
PROOF. Any two closed points ofT lie in the image of a morphism g: T' --+ T,
where T' is a nonsingular curve, or can be connected by a finite number of
such curves, so by base extension, we reduce to the case where T is a non-
singular curve. Then the result follows from (9.10) and (9.11).

Example 9.13.1 (Infinitesimal Deformations). Now that we have seen that

flatness is a natural condition for algebraic families of varieties, we come to
an important nonclassical example of flatness in the category of schemes.
Let X 0 be a scheme of finite type over a field k. Let D = k[t]/t 2 be the
ring of dual numbers over k. An infinitesimal deformation of X 0 is a scheme
X', flat over D, and such that X' ®v k ~ X 0 •
These arise geometrically in the following way. If f:X --+ Tis any flat
family, having a point t E T with X 1 ~ X 0 , we say that X is a (global) de-
formation of X 0 . Now given an element of the Zariski tangent space ofT
at t, we obtain a morphism Spec D --+ T (II, Ex. 2.8). Then by base extension
we obtain an X' flat over Spec D with closed fibre X 0 . Thus the study of
the infinitesimal deformations of X 0 ultimately will help in the study of
global deformations.

Example 9.13.2. Continuing the same ideas, it is often possible to classify

the infinitesimal deformations of a scheme X. In particular, if X is non-
singular over an algebraically closed field k, we will show that the set of
infinitesimal deformations of X, up to isomorphism, is in one-to-one corre-
spondence with the elements of the cohomology group H 1 (X,:Tx), where
:Tx is the tangent sheaf.
Indeed, given X' flat over D, we consider the exact sequence
0--+k..!...D--+k--+0
of D-modules. By flatness, we obtain an exact sequence
t
0 --+ (!J X --+ (!J X' --+ (!J X --+ 0
of (!Jx.-modules. Thus X' is an infinitesimal extension of the scheme X by
the sheaf (!Jx, in the sense of (II, Ex. 8.7). Conversely, such an extension
gives X' flat over D. Now these extensions are classified by H 1 (X,:Tx) by
(Ex. 4.10).

Remark 9.13.3. There is a whole subject called deformation theory devoted

to the study of deformations of a given scheme (or variety) X 0 over a field k.

265
III Cohomology

It is closely related to the moduli problem. There one attempts to classify

all varieties, and put them into algebraic families. Here we study only those
that are close to a given one X 0 •
Deformation theory is one area of algebraic geometry where the influence
of schemes has been enormous. Because even if one's primary interest is in
a variety X 0 over k, by working in the category of schemes, one can consider
flat families over arbitrary Artin rings with residue field k, whose closed
fibre is X 0 . Taking the limit of Artin rings, one can study flat families over
a complete local ring. Both of these types of families are intermediate
between X 0 itself and a global deformation f:X ~ T where Tis another
variety. Thus they form a powerful tool for studying all deformations of
X 0 . For some references on deformation theory see Schlessinger [ 1J, or
Morrow and Kodaira [1, Ch. 4].

EXERCISES

9.1. A flat morphism f:X--> Y of finite type of noetherian schemes is open, i.e, for
every open subset U <::; X,f(U) is open in Y. [Hint: Show thatf(U) is construct-
ible and stable under generization (II, Ex. 3.18) and (II, Ex. 3.19).]
9.2. Do the calculation of (9.8.4) for the curve of (I, Ex. 3.14). Show that you get an
embedded point at the cusp of the plane cubic curve.
9.3. Some examples of flatness and nonflatness.
(a) If f:X--> Y is a finite surjective morphism of nonsingular varieties over an
algebraically closed field k, then f is flat.
(b) Let X be a union of two planes meeting at a point, each of which maps iso-
morphically to a plane Y. Show that f is not flat. For example, let Y =
Spec k[x,y] and X= Spec k[x,y,z,w]j(z,w) n (x + z,y + w).
(c) Again let Y =Spec k[x,y], but take X= Spec k[x,y,z,w]/(z 2 ,zw,w 2 ,xz- yw).
Show that X"d ~ Y, X has no embedded points, but that f is not flat.
9.4. Open Nature of Flatness. Letf:X--> Y be a morphism of finite type of noetherian
schemes. Then {x E X if is flat at x} is an open subset of X (possibly empty)-see
Grothendieck [EGA IV 3 , 11.1.1].
9.5. Very Flat Families. For any closed subscheme X<::; P", we denote by C(X) <::; p•+ 1
the projective cone over X (I, Ex. 2.10). If I <::; k[ x0 , . . . ,x.] is the (largest) homo-
geneous ideal of X, then C(X) is defined by the ideal generated by I in
k[ Xo, ... ,Xn+ 1].
(a) Give an example to show that if {X,} is a flat family of closed subschemes of
P", then {C(X,)} need not be a flat family in p•+ 1 .
(b) To remedy this situation, we make the following definition. Let X <::; P';-- be a
closed subscheme, where T is a noetherian integral scheme. For each t E T,
let I, <::; S, = k(t)[ x 0 , . . . ,x.J be the homogeneous ideal of X, in Pi:(r)· We
say that the family {X,} is very fiat if for all d ;;:, 0,

dimk(r)(S,/ I,)d
is independent oft. Here ( )d means the homogeneous part of degree d.

266
 9 Flat Morphisms

(c) If {X,} is a very flat family in P", show that it is flat. Show also that {C(X,)} is
a very flat family in P" + 1 , and hence flat.
(d) If {X(r)} is an algebraic family of projectively normal varieties in P~, para-
metrized by a nonsingular curve T over an algebraically closed field k, then
{ X(r)} is a very flat family of schemes.

9.6. Let Y <;; P" be a nonsingular variety of dimension ~ 2 over an algebraically

closed field k. Suppose pn-l is a hyperplane in P" which does not contain Y,
and such that the scheme Y' = Y n pn-l is also nonsingular. Prove that Y is a
complete intersection in P" if and only if Y' is a complete intersection in pn-'.
[Hint: See (II, Ex. 8.4) and use (9.12) applied to the affine cones over Y and Y'.]
9.7. Let Y <;; X be a closed subscheme, where X is a scheme of finite type over a
field k. Let D = k[t]/t 2 be the ring of dual numbers, and define an infinitesimal
deformation of Y as a closed subscheme of X, to be a closed subscheme
Y' s; X x k D, which is flat over D, and whose closed fibre is Y. Show that these
Y' are classified by H 0 (Y,% Y/X), where
A~'y 1 x = Yfom~Jfy/.f~,@y).

*9.8. Let A be a finitely generated k-algebra. Write A as a quotient of a polynomial

ring P over k, and let J be the kernel:
0-'-- J-'-- P-'-- A-'-- 0.
Consider the exact sequence of (II, 8.4A)
J/1 2 -'-> QP!k @p A-'-> QA/k-+ 0.
Apply the functor Hom A(· ,A), and let T 1(A) be the cokernel:
HomA(QP!k ® A,A)-+ HomA(J/] 2 ,A)-'-- T 1(A)-+ 0.
Now use the construction of(Il, Ex. 8.6) to show that T 1(A) classifies infinitesimal
deformations of A, i.e., algebras A' flat over D = k[t]/t 2 , with A' ®n k ~ A. It
follows that T 1 (A) is independent of the given representation of A as a quotient
of a polynomial ring P. (For more details, see Lichtenbaum and Schlessinger [1 ].)
9.9. A k-algebra A is said to be rigid if it has no infinitesimal deformations, or equi-
valently, by (Ex. 9.8) if T 1(A) = 0. Let A= k[x,y,z,w]/(x,y) n (z,w), and show
that A is rigid. This corresponds to two planes in A4 which meet at a point.
9.10. A scheme X 0 over a field k is rigid if it has no infinitesimal deformations.
(a) Show that Pt is rigid, using (9.13.2).
(b) One might think that if X 0 is rigid over k, then every global deformation of X 0
is locally trivial. Show that this is not so, by constructing a proper, flat mor-
phism f: X -'-- A2 over k algebraically closed, such that X 0 ~ Pt, but there
is no open neighborhood U of 0 in A2 for which f- 1( U) ~ U x P 1 .
*(c) Show, however, that one can trivialize a global deformation of P 1 after a flat
base extension, in the following sense: let f: X -'-- T be a flat projective mor-
phism, where T is a nonsingular curve over k algebraically closed. Assume
there is a closed pointt E T such that X, ~ P{ Then there exists a nonsingular
curve T', and a flat morphism g: T' -'-- T, whose image contains t, such that
if X' = X x T T' is the base extension, then the new family f': X' -'-- T' is
isomorphic to P}. -+ T'.

267
III Cohomology

9.11. Let Y be a nonsingular curve of degree din P~, over an algebraically closed field k.
Show that
0 :::; p.(Y) :::; t(d - l)(d - 2).

[Hint: Compare Y to a suitable projection of Y into P 2 , as in (9.8.3) and (9.8.4).]

10 Smooth Morphisms

The notion of smooth morphism is a relative version of the notion of non-

singular variety over a field. In this section we will give some basic results
about smooth morphisms. As an application, we give Kleiman's elegant
proof of the characteristic 0 Bertini theorem. For further information about
smooth and etale morphisms, see Altman and Kleiman [1, Ch. VI, VII],
Matsumura [2, Ch. 11 ], and Grothendieck [SGA 1, exp. I, II, III].
For simplicity, we assume that all schemes in this section are of finite
type over a field k.

Definition. A morphism f: X -> Y of schemes of finite type over k is smooth

of relative dimension n if:
(1) f is flat;
(2) if X' ~ X and Y' ~ Yare irreducible components such that f(X') ~ Y',
then dim X' =dim Y' + n;
(3) for each point x EX (closed or not),
dimk(x)(Qx;Y ® k(x)) = n.

Example 10.0.1. For any Y, A~ and P~ are smooth of relative dimension n

over Y.

Example 10.0.2. If X is integral, then condition (3) is equivalent to saying

Qx;Y is locally free on X of rank n (II, 8.9).

Example 10.0.3. If Y = Spec k and k is algebraically closed, then X is smooth

over k if and only if X is regular of dimension n. In particular, if X is irre-
ducible and separated over k, then it is smooth if and only if it is a nonsingular
variety. Cf. (II, 8.8) and (II, 8.15).

Proposition 10.1.
(a) An open immersion is smooth of relative dimension 0.
(b) Base change. If f: X -> Y is smooth of relative dimension n, and
g: Y' -> Y is any morphism, then the morphism f': X' -> Y' obtained by
base extension is also smooth of relative dimension n.
(c) Composition. If f:X-> Y is smooth of relative dimension n, and
g: Y -> Z is smooth of relative dimension m, then g of: X -> Z is smooth
of relative dimension n + m.

268
 10 Smooth Morphisms

(d) Product. If X and Yare smooth over Z, of relative dimensions nand

m, respectively, then X x z Y is smooth over Z of relative dimension n + m.
PROOFS.
(a) is trivial.
(b) f' is flat by (9.2). According to (9.6), the condition (2) in the definition
of smoothness is equivalent to saying that every irreducible component of
every fibre XY off has dimension n. This condition is preserved under base
extension (II, Ex. 3.20). Finally, QXJY is stable under base extension (II, 8.1 0),
so the number dimk<x>(Qx 1y ® k(x)) is also. Hence f' is smooth.
(c) y of is flat by (9.2). If X' s X, Y' s Y, and Z' s Z are irreducible
components such that f(X') s Y' and g(Y') s Z', then clearly dim X' =
dim Z' + n + m by hypothesis. For the last condition, we use the exact
sequence of(II, 8.11)
f*QY/Z ~ QX/Z ~ QX/Y ~ 0.
Tensoring with k(x) we have
j*QY/Z @ k(x) ~ QX/Z @ k(x) ~ QX/Y @ k(x) ~ 0.
Now the first has dimension m, and the last has dimension n, by hypothesis.
So the middle one has dimension ~ n + m.
On the other hand, let z = y(f(x) ). Then
QX/Z ® k(x) = QXz/k(z) ® k(x),
since relative differentials commute with base extension. Let X' be an irre-
ducible component of X z containing x, with its reduced induced structure.
Then we have a surjective map·
QXz/k(z) ® k(x) ~ QX'/k(z) ® k(x) ~ 0
by (II, 8.12). But X' is an integral scheme of finite type over k(z), of dimension
n + m, by (9.6), so QX'Jk<z> is a coherent sheaf of rank ): n + m by (II, 8.6A).
Hence it requires at least n + m generators at every point, so
dimk(x)(QX'/k(z) ® k(x)) ): n + m.
Combining our inequalities, we find that
dimk(x)(Qx1z ® k(x)) = n + m
as required.
(d) This statement is a consequence of(b) and (c) since we can factor into
X Xz Y ~ Y ~ Z.

Theorem 10.2. Let f:X ~ Y be a morphism of schemes of finite type over k.

Then f is smooth of relative dimension 11 if and only if:
(1) f is flat; and
(2) for each pointy E Y, let XY =X)' ®k<Y> k(y)-, where k(y)- is the alge-
braic closure of k(y). Then Xy is equidimensional of dimension 11 and

269
III Cohomology

regular. (We say "the fibres off are geometrically regular of equi-
dimension n.")

PROOF. Iff is smooth of relative dimension n, so is any base extension. In

particular, X Ji is smooth of relative dimension n over k( y)-, so is regular
(10.0.3).
Conversely, suppose (1) and (2) satisfied. Then f is flat by (1 ). From (2)
we conclude that every irreducible component of X Y has dimension n, which
gives condition (2) of the definition of smoothness by (9.6). Finally, since
k(y)- is algebraically closed, regularity of XY implies that QXy/k(y)- is locally
free of rank n (10.0.3). This in turn implies that QXy/k(y) is locally free of rank n
(see e.g. Matsumura [2, (4.E), p. 29]), and so for any x EX,
dimk(x)(Qx;r @ k(x)) = dimk(x)(QXy/k(y) (8) k(x)) = n

as required.

Next we will study when a morphism of nonsingular varieties is smooth.

Recall (II, Ex. 2.8) that for a point x in a scheme X we define the Zariski
tangent space Tx to be the dual of the k(x)-vector space m)m;. Iff: X ....,. Y
is a morphism, and y = f(x), then there is a natural induced mapping on
the tangent spaces
TJ: Tx....,. Ty @k(y) k(x).
Before stating our criterion, we recall an algebraic fact.

Lemma 10.3.A. Let A ....,. B be a local homomorphism of local noetherian

rings. Let M be a finitely generated B-module, and let t E A be a nonunit
that is not a zero divisor. Then M is .fiat over A if and only if:
(1) tis not a zero divisor in M; and
(2) M/tM is .fiat over A/tA.

PROOF. This is a special case of the "Local criterion of flatness." See Bourbaki
[1, III, §5] or Altman and Kleiman [1, V, §3].

Proposition 10.4. Let f: X ....,. Y be a morphism of nonsingular varieties over

an algebraically closed field k. Let n = dim X - dim Y. Then the
following conditions are equivalent:
(i) f is smooth of relative dimension n;
(ii) Qx;r is locally free of rank n on X;
(iii) for every closed point x E X, the induced map on the Zariski tangent
spaces T f: T x ....,. T Y is surjective.
PROOF.
(i) => (ii) follows from the definition of smoothness, since X is integral
(10.0.2).

270
 10 Smooth Morphisms

(ii) ~(iii). From the exact sequence of (II, 8.11), tensoring with k(x),
we have

Now X and Yare both smooth over k, so the dimensions of these vector
spaces are equal to dim Y, dim X, and n respectively. Therefore the map
on the left is injective. But for a closed point x, k(x) ~ k, so using (II, 8. 7)
we see thatthis map is just the natural map
my/m; --+ mx/m;
induced by f Taking dual vector spaces over k, we find that T f is surjective.
(iii) ~ (i). First we show f is flat. For this, it is enough to show that
@xis flat over @Y for every closed point x EX, where y = f(x), by localiza-
tion of flatness. Since X and Y are nonsingular, these are both regular
local rings. Furthermore, since T f is surjective, we have my/m; --+ mx/m;
injective as above. So let t 1 , . . . ,tr be a regular system of parameters for
@Y. Then their images in @x form part of a regular system of parameters
of @x· Since @x/(t 1, ... ,tr) is automatically flat over @y/(t 1, ... ,tr) = k,
we can use (10.3A) to show by descending induction on i that @x/(t 1 , . . . ,t;)
is flat over @y/(t 1 , ••• ,t;) for each i. In particular, for i = 0, @x is flat over
@y· Thus f is flat.
Now we can read the argument of (ii) ~(iii) backwards to conclude that
dimk(x)(Qx;Y ® k(x)) = n
for each closed point x E X. On the other hand since f is flat, it is dominant,
so for the generic point ' E X, we have
dimk(slQx;Y ® k(()) ;;:, n

by (II, 8.6A). We conclude that Qx;Y is coherent of rank ;;:, n, so it must be

locally free of rank = n by (II, 8.9). Therefore Qx;Y ® k(x) has dimension n
at every point of X, so f is smooth of relative dimension n.

Next we will give some special results about smoothness which hold
only in characteristic zero.

Lemma 10.5. Let f:X--+ Y be a dominant morphism of integral schemes of

finite type over an algebraically closed field k of characteristic 0. Then
there is a nonempty open set U ~ X such that f: U --+ Y is smooth.

PROOF. Replacing X and Y by suitable open subsets, we may assume that

they are both nonsingular varieties over k (II, 8.16). Next, since we are in
characteristic 0, K(X) is a separably generated field extension of K(Y)
(1, 4.8A). So by (II, 8.6A), Qx;Y is free of rank n = dim X - dim Y at the
generic point of X. Therefore it is locally free of rank non some nonempty
open set U ~ X. We conclude that f: U --+ Y is smooth by (10.4).

271
III Cohomology

Example 10.5.1. Let k be an algebraically closed field of characteristic p,

let X = Y = Pf, and let f:X --+ Y be the Frobenius morphism (I, Ex. 3.2).
Then f is not smooth on any open set. Indeed, since d(tP) = 0, the natural
map f*QY/k --+ Qx;k is the zero map, and so Qx;Y ~ QX/k is locally free of
rank 1. But f has relative dimension 0, so it is nowhere smooth.

Proposition 10.6. Let f: X --+ Y be a morphism of schemes of finite type

over an algebraically closed field k of characteristic 0. For any r, let
xr = {closed points X E Xlrank Tf,x :(; r}.
Then
dim f(Xr) :(; r.

PROOF. Let Y' be any irreducible component of f(Xr), and let X' be an
irreducible component of Xr which dominates Y'. We give X' and Y'
their reduced induced structures, and consider the induced dominant mor-
phismf':X'--+ Y'. Then by (10.5) there is a nonempty open subset U' s; X'
such that f': U'--+ Y' is smooth. Now let x E U' n X, and consider the
commutative diagram of maps of Zariski tangent spaces
Tx.U' Tx,X

]Tr,x ]Tf,x

Ty,Y' Ty,Y
The horizontal arrows are injective, because U' and Y' are locally closed
subschemes of X and Y, respectively. On the other hand, rank T f,x :(; r
since x EX, and T f',x is surjective because f' is smooth (10.4). We conclude
that dim T y,Y' :(; r, and therefore dim Y' :(; r.

Corollary 10.7 (Generic Smoothness). Let f:X--+ Y be a morphism of

varieties over an algebraically closed field k of characteristic 0, and assume
that X is nonsingular. Then there is a nonempty open subset V s; Y such
that f :f- 1 V --+ V is smooth.

PROOF. We may assume Y is nonsingular by (II, 8.16). Let r = dim Y.

Let Xr- 1 s; X be the subset defined in (10.6). Then dim f(Xr_ 1 ) :(; r - 1
by (10.6), so removing it from Y, we may assume that rank T 1 ~ r for every
closed point of X. But since Y is nonsingular of dimension r, this implies
that T 1 is surjective for every closed point of X. Hence f is smooth by
(10.4).
Note that if the original f was not dominant, then V s; Y - f(X), and
f- 1 V will be empty.

For the next results, we recall the notion of a group variety (1, Ex. 3.21).
A group variety G over an algebraically closed field k is a variety G, together

272
 10 Smooth Morphisms

with morphisms 11: G x G ~ G and p: G ~ G, such that the set G(k) of

k-rational points of G (which is just the set of all closed points of G, since k
is algebraically closed) becomes a group under the operation induced by J1,
with p giving the inverses.
We say that a group variety G acts on a variety X if we have a morphism
8: G x X ~ X which induces a homomorphism G(k) ~ Aut X of groups.
A homogeneous space is a variety X, together with a group variety G
acting on it, such that the group G(k) acts transitively on the set X(k) of
k-rational points of X.

Remark 10.7.1. Any group variety is a homogeneous space if we let it act

on itself by left multiplication.

Example 10.7.2. The projective space Pi: is a homogeneous space for the
action of G = PGL(n)-cf. (II, 7.1.1).

Example 10.7.3. A homogeneous space is necessarily a nonsingular variety.

Indeed, it has an open subset which is nonsingular by (II, 8.16). But we
have a transitive group of automorphisms acting, so it is nonsingular
everywhere.

Theorem 10.8 (Kleiman [3]). Let X be a homogeneous space with group

variety G over an algebraically closed field k of characteristic 0. Let
f: Y ~ X and g: Z ~ X be morphisms of nonsingular varieties Y, Z to X.
For any CJ E G(k), let ya be Y with the morphism CJ a f to X. Then there
is a nonempty open subset V s; G such that for every CJ E V(k), ya x x Z
is nonsingular and either empty or of dimension exactly
dim Y + dim Z - dim X.
PROOF. First we consider the morphism
h:G X y ~X

defined by composing f with the group action e:G x X~ X. Now G is

nonsingular since it is a group variety (10.7.3), and Y is nonsingular by
hypothesis, so G x Y is nonsingular by (10.1). Since char k = 0, we can
apply generic smoothness (10. 7) to h, and conclude that there is a non empty
open subset Us; X such that h:h- 1(U) ~ U is smooth. Now G acts on
G x Y by left multiplication on G; G acts on X bye, and these two actions
are compatible with the morphism h, by construction. Therefore, for any
(J E G(k), h: h-l(Ua) ~ ua is also smooth. Since the ua cover X, we con-

clude that h is smooth everywhere.

Next, we consider the fibred product
w = (G X Y) X X Z,

with maps g' and h' to G x Y and Z as shown.

273
III Cohomology

h'
w ---'-'---->Z

G X y --'-'-h-~ X

Since h is smooth, h' is also smooth by base extension (1 0.1 ). Since Z is

nonsingular, it is smooth over k, so by composition (10.1), W is also smooth
over k, so W is nonsingular.
Now we consider the morphism
q = Pt og': W--> G.
Applying generic smoothness (10.7) again, we find there is a nonempty
open subset V s G such that q:q- 1 (V)--> Vis smooth. Therefore, if a E
V(k) is any closed point, the fibre W.,. will be nonsingular. But W.,. is just
Y" x x Z, so this is what we wanted to show. Note that W.,. may not be
irreducible, but our result shows that each connected component is a non-
singular variety.
To find the dimension of W.,., we first note that h is smooth of relative
dimension
dim G + dim Y - dim X.
Hence h' has the same relative dimension, and we see that
dim W = dim G + dim Y - dim X + dim Z.
If W is nonempty, then q on q- 1 (V) has relative dimension equal to
dim W - dim G, so for each a,
dim W.,. =dim Y + dimZ- dim X.

Corollary 10.9 (Bertini). Let X be a nonsingular projective variety over an

algebraically closed field k of characteristic 0. Let b be a linear system
without base points. Then almost every element of b, considered as a
closed subscheme of X, is nonsingular (but maybe reducible).

PROOF. Let f:X --> pn be the morphism to pn determined by b (II, 7.8.1).

We consider pn as a homogeneous space under the action of G = PGL(n)
(10.7.2). We apply the theorem taking g:H--> pn to be the inclusion map
of a hyperplane H ~ pn-t. We conclude that for almost all a E G(k),
X x P" H" = f - 1 (H") is nonsingular. But the divisors f- 1 (H") are just
the elements of the linear system b, by construction off Thus almost all
elements of b are non singular.

274
 10 Smooth Morphisms

Remark 10.9.1. We will see later (Ex. 11.3) that if dimf(X) ~ 2, then all
the divisors in b are connected. Hence almost all of them are irreducible
and nonsingular.

Remark 10.9.2. The hypothesis "X projective" is not necessary if we talk

about a finite-dimensional linear system b. In particular, if X was projec-
tive, and b was a linear system with base points 1:, then by considering the
base-point-free linear system b on X - 1: we obtain the more general
statement that "a general member of b can have singularities only at the
base points."

Remark 10.9.3. This result fails in characteristic p > 0. For example, in

(10.5.1) the morphism f corresponds to the one-dimensional linear system
{pPJP E P 1 }. Thus every divisor in b is a point with multiplicity p.

Remark 10.9.4. Compare this result to the earlier Bertini theorem (II, 8.18).

EXERCISES

10.1. Over a nonperfect field, smooth and regular are not equivalent. For example,
Ar
let k 0 be a field of characteristic p > 0, let k = k 0 (t), and let X <::; be the curve
defined by i = xP - t. Show that every local ring of X is a regular local ring,
but X is not smooth over k.
10.2. Let f:X--+ Y be a proper, flat morphism of varieties over k. Suppose for some
point y E Y that the fibre XY is smooth over k(y). Then show that there is an
open neighborhood U of y in Y such that f :f- 1 (U) --+ U is smooth.
10.3. A morphism f: X --+ Y of schemes of finite type over k is etale if it is smooth of
relative dimension 0. It is unramified if for every x EX, letting y = f(x), we have
my· (!)x = mx, and k(x) is a separable algebraic extension of k(y). Show that the
following conditions are equivalent:
f is etale;
(i)
f is flat, and QX/Y = 0;
(ii)
f is flat and unramified.
(iii)
10.4. Show that a morphism f: X --+ Y of schemes of finite type over k is etale if and
only if the following condition is satisfied: for each x EX, let y = f(x). Let @x and
@Y be the completions of the local rings at x andy. Choose fields of representa-
tives (II, 8.25A) k(x) c;; @x and k(y) <:; @Y so that k(y) <:; k(x) via the natural map
@Y--+ @x· Then our condition is that for every x EX, k(x) is a separable algebraic
extension of k(y), and the natural map
@Y ®k(y) k(x)--+ @x
is an isomorphism.
10.5. If x is a point of a scheme X, we define an etale neighborhood of x to be an etale
morphism f: U--+ X, together with a point x' E U such that f(x') = x. As an
example of the use of etale neighborhoods, prove the following: if ff is a coherent
sheaf on X, and if every point of X has an etale neighborhood f: U --+ X for which
f* ff is a free mu-module, then ff is locally free on X.

275
III Cohomology

10.6. Let Y be the plane nodal cubic curve y 2 = x 2 (x + 1). Show that Y has a finite
etale covering X of degree 2, where X is a union of two irreducible components,
each one isomorphic to the normalization of Y (Fig. 12).

Figure 12. A finite etale covering.

10.7. (Serre). A linear system with moving singularities. Let k be an algebraically closed
field of characteristic 2. Let P 1 , . . . ,P 7 E P~ be the seven points of the projective
plane over the prime field F 2 <::::; k. Let b be the linear system of all cubic curves
in X passing through P 1 , .•. ,P 7 •
(a) b is a linear system of dimension 2 with base points P 1 , . . . ,P 7 , which deter-
mines an inseparable morphism of degree 2 from X - { P;} to P 2 .
(b) Every curve C E b is singular. More precisely, either C consists of 3 lines all
passing through one of the P;, or C is an irreducible cuspidal cubic with
cusp P of. any P;. Furthermore, the correspondence C r-> the singular point
of Cis a 1-1 correspondence between band P 2 . Thus the singular points of
elements of b move all over.
10.8. A linear system with moving singularities contained in the base locus (any charac-
teristic). In affine 3-space with coordinates x,y,z, let C be the conic (x - 1) 2 +
y 2 = 1 in the xy-plane, and let P be the point (O,O,t) on the z-axis. Let Y, be the
closure in P 3 of the cone over C with vertex P. Show that as t varies, the surfaces
{ Y,} form a linear system of dimension 1, with a moving singularity at P. The
base locus of this linear system is the conic C plus the z-axis.
10.9. Let f: X --> Y be a morphism of varieties over k. Assume that Y is regular, X is
Cohen-Macaulay, and that every fibre off has dimension equal to dim X - dim Y.
Then f is fiat. [Hint: Imitate the proof of (10.4), using (II, 8.21A).]

11 The Theorem on Formal Functions

In this section we prove the so-called theorem on formal functions, and its
important corollaries, Zariski's Main Theorem, and the Stein factorization
theorem. The theorem itself compares the cohomology of the infinitesimal
neighborhoods of a fibre of a projective morphism to the stalk of the higher
direct image sheaves. While the corollaries use only the case i = 0 of the
theorem (which could be stated without cohomology), the proof is by
descending induction on i, and thus makes essential use of the cohomo-
logical machinery. Zariski's first proof [2] of his "Main Theorem" was
by an entirely different method which did not use cohomology.

Let f:X --+ Y be a projective morphism of noetherian schemes, let ff

be a coherent sheaf on X, and let y E Y be a point. For each n ? 1 we

276
 11 The Theorem on Formal Functions

define
xn = X X y Spec (!Jy/m~.

Then for n = 1, we get the fibre X Y' and for n > 1, we get a scheme with
nilpotent elements having the same underlying space as X y· It is a kind of
"thickened fibre" of X over the point y.

-----=-v----. X

-------+ y

Let§',. = v* ff, where v:Xn --+X is the natural map. Then by (9.3.1) we have
natural maps, for each n,
Ri*(ff) ® (!Jy/m;--+ Ri~(ffnl·
Since Spec (!Jy/m; is affine, concentrated at one point, the right-hand side is
just the group Hi( X n•ffn) by (8.5). As n varies, both sides form inverse systems
(see (II, §9) for generalities on inverse systems and inverse limits). Thus we
can take inverse limits and get a natural map
Rf*(ff); --+ li!!! Hi(X.,ff.).
Theorem 11.1 (Theorem on Formal Functions). Let f:X--+ Y be a projec-
tive morphism of noetherian schemes, let ff be a coherent sheaf on X, and
let y E Y. Then the natural map
Rf*(ff); --+ li!!! Hi( X .,ff.)
is an isomorphism, for all i ~ 0.
PROOF. As a first step, we embed X in some projective space P~, and consider
ff as a coherent sheaf on P~. Thus we reduce to the case X = P~.
Next we let A = (!JY' and make the flat base extension Spec A --+ Y. Thus,
using (9.3), we reduce to the case where Y is affine, equal to the spectrum of a
local noetherian ring A, andy is the closed point of Y. Then using (8.5) again,
we can restate our result as an isomorphism of A-modules,
Hi(X ' :#')~ ~ lim Hi(X ff)
+---- n' n ·
Now suppose ff is a sheaf of the form (!J(q) on X = P~, for some q E Z.
Then ff. is just (!J(q) on X n = P~/m". So by the explicit calculations of (5.1)
we see that
Hi(Xn,ffn) ~ Hi(X,ff) ®A A/m"
for each n. Therefore by definition of completion, we have
Hi(X :#')~ ::::: lim Hi(X ff)
' - +---- "' n
in this case. Clearly the same calculation holds for any finite direct sum of
sheaves of the form (!J(qJ

277
 III Cohomology

We will now prove the theorem for an arbitrary coherent sheaf !!i' on X,
by descending induction on i. Fori > N, both sides are 0, because X can
be covered by N + 1 open affine subsets (Ex. 4.8). So we assume the theorem
has been proved for i + 1, and for all coherent sheaves.
Given !!i' coherent on X, it follows from (II, 5.18) that we can write !!i' as
a quotient of a sheaf Iff which is a finite direct sum of sheaves (!)(q;) for suitable
qi E Z. Let f!ll be the kernel:
(1)

Now unfortunately, tensoring with (!)x" is not an exact functor-it is only

right exact, so we have an exact sequence
f!lln ~ Iff n ~ !fi'n ~ 0
of sheaves on Xn for each n. We introduce the image :!Tn and the kernel Yn
of the map &ln ~Iff", so that we have exact sequences
(2)
and
(3)
We now consider the following diagram:
Hi(X,&l)~ ------> Hi(X,rff)~ -~ H;(X,!!i')~ -----+ Hi+ 1 (X,&l)~ ~ Hi+ 1(X,f,'

(a1 (a4
ll!!! Hi( X n•f!lln) IXz IX3

1p1
lim Hi(X :!T) ~ lim H;(X Iff ) ~ lim Hi(X !!i') ~ lim Hi+ 1(X :!T) ~lim Hi+ 1 (X 0"
-+---- "' n ,....._ "' n +--- "' n +--- "' n +--- "' r,

The top row comes from the cohomology sequence of (1) by completion.
Since they are all finitely generated A-modules (5.2), completion is an exact
functor (II, 9.3A). The bottom row comes from the cohomology sequence
of (3) by taking inverse limits. These groups are all finitely generated A/m"-
modules, and so satisfy d.c.c. for submodules. Therefore the inverse systems
all satisfy the Mittag-Leffier condition (II, 9.1.2), and so the bottom row is
exact (II, 9.1). The vertical arrows a 1, . . . ,as are the maps of the theorem.
We have a 2 an isomorphism because Iff is a sum of sheaves (!)(q;), and a 4 and
as are isomorphisms by the induction hypothesis. Finally, {3 1 and {3 2 are
maps induced from the sequence (2). We will show below that {3 1 and {3 2
are isomorphisms.
Admitting this for the moment, it follows from the subtle 5-lemma that a 3
is surjective. But this is then true for any coherent sheaf on X, so a 1 must
also be surjective. This in turn implies a 3 is an isomorphism, which is what
we want.

278
 11 The Theorem on Formal Functions

It remains to prove that /3 1 and /3 2 are isomorphisms. Taking the coho-

mology sequence of (2), and passing to the inverse limit, using (II, 9.1) again,
it will be sufficient to show that
lim Hi(Xn,Y'n) = 0
+---
for all i ~ 0. To accomplish this, we will show that for any n, there is an
n' > n such that the map of sheaves Y'n· --+ Y'n is the zero map. By quasi-
compactness, the question is local on X, so we may assume that X is affine,
X= Spec B. We denote by R,E,Sn the B-modules corresponding to the
sheaves f!ll,tff,!/"' and we denote by a the ideal mB.
Recall that R is a submodule of E, and that
Sn = ker(Rja"R --+ E/a"E).
Thus
Sn = (R n a"E)/a"R.
But by Krull's theorem (3.1A), the a-adic topology on R is induced by the
a-adic topology on E. In other words, for any n, there is an n' > n such that
R n a"'E £: anR.
In that case the map Sn. --+ Sn is zero. q.e.d.

Remark 11.1.1. This theorem is proved more generally for a proper mor-
phism in Grothendieck [EGA III, §4].

Remark 11.1.2. Many applications of this theorem use only the case i = 0.
In that case the right-hand side is equal to F(X,#), where X is the formal
completion of X along XY, and # = ff ® @g (II, 9.2). In particular, if
ff = r9x, we have r(X,@g), which is the ring of formal-regular functions
(also called holomorphic functions) on X along XY. Hence the name of the
theorem.

Remark 11.1.3. One can also introduce the cohomology Hi(X,ff) of# on
the formal scheme X, and prove that it is isomorphic to the two other
quantities in the theorem [EGA III, §4].

Corollary 11.2. Let f: X --+ Y be a projective morphism of noetherian schemes,

and let r = max{dim XyiY E Y}. Then RiJ*(ff) = 0 for all i > r, and
for all coherent sheaves ff on X.
PROOF. For any y E Y, Xn is a scheme whose underlying topological space is
the same as X y· Hence
Hi(Xn,ffn) = 0
fori > r by (2.7). If follows that RiJ*(ff); = 0 for ally E Y, i > r, and there-
fore since R1*(ff) is coherent (8.8) it must be 0.

Corollary 11.3. Let f: X

--+ Y be a projective morphism of noetherian schemes,
and assume that f*@x = @y. Then f- 1(y) is connected, for every y E Y.

279
III Cohomology

PROOF. Suppose to the contrary that f- 1 (y) = X' u X", where X' and X"
are disjoint closed subsets. Then for each n, we would have
H 0 (XJDxJ = H 0 (X~,(DxJ E8 H 0 (X~/9xJ·
By the theorem, we have
&Y = (f*mx); = fu!! H 0(Xn,mxJ.
Therefore &Y = A' E8 A", where
A' =lim
+---
H 0 (X'"' mXn )
and
lim H 0 (X""'
A" = +--- mXn ).
But this is impossible, because a local ring cannot be a direct sum of two
other rings. Indeed, let e',e" be the unit elements of A' and A". Then
e' + e" = 1 in &Y. On the other hand, e'e" = 0, so e',e" are nonunits, hence
contained in the maximal ideal of &Y, so their sum cannot be 1 (cf. (II,
Ex. 2.19) ).

Corollary 11.4 (Zariski's Main Theorem). Let f: X ~ Y be a birational pro-

jective morphism of noetherian integral schemes, and assume that Y is
normal. Then for every y E Y,f- 1(y) is connected. (See also (V, 5.2).)
PROOF. By the previous result, we have only to verify that f*mx = my. The
question is local on Y, so we may assume Y is affine, equal to Spec A. Then
f*mx is a coherent sheaf of my-algebras, so B = r(Y,f*mx) is a finitely gen-
erated A-module. But A and Bare integral domains with the same quotient
field, and A is integrally closed, so we must have A =B. Thus f*mx =my.
Corollary 11.5 (Stein Factorization). Let f:X ~ Y be a projective morphism
of noetherian schemes. Then one can factor f into g of', where f': X ~ Y'
is a projective morphism with connected fibres, and g: Y' ~ Y is a finite
morphism.
PROOF. Let Y' = Spec f*mx (II, Ex. 5.17). Then since f*mx is a coherent
sheaf of my-algebras, the natural map g: Y' ~ Y is finite. On the other hand
f clearly factors through g, so we get a morphism f': X ~ Y'. Since g is
separated, we conclude that f' is projective by (II, Ex. 4.9). By construction
f~mx =mY' so f' has connected fibres by (11.3).

EXERCISES

11.1. Show that the result of (11.2) is false without the projective hypothesis. For
example, let X = A;;, let P = (0, ... ,0), let U = X - P, and let f: U --+ X be
the inclusion. Then the fibres off all have dimension 0, but R"- 1j*{!Ju i= 0.
11.2. Show that a projective morphism with finite fibres ( = quasi-finite (II, Ex. 3.5))
is a finite morphism.
11.3. Let X be a normal, projective variety over an algebraically closed field k. Let b
be a linear system (of effective Cartier divisors) without base points, and assume
that b is not composite with a pencil, which means that iff: X --+ P~ is the morphism

280
 12 The Semicontinuity Theorem

determined by b, then dim f(X) ~ 2. Then show that every divisor in b is con-
nected. This improves Bertini's theorem (10.9.1). [Hints: Use (11.5), (Ex. 5.7)
and (7.9).]
11.4. Principle of Connectedness. Let {X,} be a flat family of closed subschemes of P;:
parametrized by an irreducible curve T of finite type over k. Suppose there is a
nonempty open set U <;; T, such that for all closed points t E U, X, is connected.
Then prove that X, is connected for all t E T.
*11.5. Let Y be a hypersurface in X = Pf with N ~ 4. Let X be the formal completion
of X along Y (II, §9). Prove that the natural map Pic X --+ Pic Y is an isomorphism.
[Hint: Use (II, Ex. 9.6), and then study the maps Pic X"+ 1 --+ Pic X" for each n
using (Ex. 4.6) and (Ex. 5.5).]
11.6. Again let Y be a hypersurface in X = Pf, this time with N ~ 2.
(a) If ff is a locally free sheaf on X, show that the natural map
H 0 (X,ff) --+ H 0 (X,.#)
is an isomorphism.
(b) Show that the following conditions are equivalent:
(i) for each locally free sheaf 3 on X, there exists a coherent sheaf ff on X
such that 3 ;:;: .# (i.e., 3 is algebraizable);
(ii) for each local~y free sheaf 3 on X, there is an integer n0 such that 3(n) is
generated by global sections for all n ~ n0 •
[Hint: For (ii) => (i), show that one can find sheaves if 0 ,tf 1 on X, which are
direct sums of sheaves of the form {I)(- q;), and an exact sequence i 1 --+ i 0 --+
3 --+ 0 on X. Then apply (a) to the sheaf Jf'om(tf 1 ,S 0 ).]
(c) Show that the conditions (i) and (ii) of(b) imply that the natural map Pic X--+
Pic X is an isomorphism.
Note. In fact, (i) and (ii) always hold if N ~ 3. This fact, coupled with
(Ex. 11.5) leads to Grothendieck's proof [SGA 2] of the Lefschetz theorem
which says that if Y is a hypersurface in Pf with N ~ 4, then Pic Y ;:;: Z, and
it. is generated by {l)y(l). See Hartshorne [5, Ch. IV] for more details.
11.7. Now let Y be a curve in X = P;.
(a) Use the method of (Ex. 11.5) to show that Pic X --+ Pic Y is surjective, and its
kernel is an infinite-dimensional vector space over k.
(b) Conclude that there is an invertible sheaf i! on X which is not algebraizable.
(c) Conclude also that there is a locally free sheaf 3 on X so that no twist 3 (n) is
generated by global sections. Cf. (II, 9.9.1)
11.8. Let f: X --+ Y be a projective morphism, let ff be a coherent sheaf on X which is
flat over Y, and assume that H;(Xy,ffy) = 0 for some i and some y E Y. Then
show that Rf*(ff) is 0 in a neighborhood of y.

12 The Semicontinuity Theorem

In this section we consider a projective morphism f: X.--+ Y and a coherent
sheaf :F on X, flat over Y. We ask, how does the cohomology along the
fibre Hi( X Y':FY) vary as a function of y E Y? Our technique is to find some

281
III Cohomology

relation between these groups and the sheaves Ri*(ff). The main results
are the semicontinuity theorem (12.8), and the theorem on cohomology and
base change (12.11).
Since the question is local on Y, we will usually restrict our attention to
the case Y = Spec A is affine. Then we compare the A-modules Hi(X,ff)
and Hi(Xy,ffy). Using (9.4), the cohomology of the fibre is equal to
Hi(X,ff ® k(y) ). Grothendieck's idea is to study more generally
Hi(X,ff ®AM) for any A-module M, and consider it as a functor on A-
modules.

Definition. Let A be a noetherian ring, let Y = Spec A, let f: X ~ Y be a

projective morphism, and let ff be a coherent sheaf on X, flat over Y.
(This data will remain fixed throughout this section.) Then for each
A-module M, define

for all i ~ 0.

Proposition 12.1. Each Ti is an additive, covariant functor from A-modules to

A-modules which is exact in the middle. The collection (Ti);~o forms a
J-functor (§1).
PROOF. Clearly each Ti is an additive, covariant functor. Since ff is flat over
Y, for any exact sequence
0 ~ M' ~ M ~ M" ~ 0
of A-modules, we get an exact sequence
0 ~ ff ® M' ~ ff ®M ~ ff ® M" ~ 0
of sheaves on X. Now the long exact sequence of cohomology shows that
each Ti is exact in the middle, and that together they form a J-functor.

We reduce the calculation of the functors Ti to a process involving only

A-modules, by the following result.

Proposition 12.2. With the hypotheses above, there exists a complex L' of
finitely generated free A-modules, bounded above (i.e., U = 0 for n » 0),
such that

for any A-module M, any i ~ 0, and this gives an isomorphism of J-functors.

PROOF. For any A-module M, the sheaf ff ®AM is quasi-coherent on X,
so we can use Cech cohomology to compute Hi(X,ff ®AM). Let U = (U;)
be an open cover of X. Let C = C"(U,ff) be the Cech complex of ff (§4).
Then for any i0, ... ,iP, we have
r(U; 0 , ••• ,ip,ff @AM) = r(U; 0 , • . . ,ip,ff) @AM,

282
 12 The Semicontinuity Theorem

so
C"(U,.? @AM)= c @AM.
Hence we have

for each M.
This is a step in the right direction, since C is a bounded complex of
A-modules. However, the C will almost never be finitely generated A-
modules. But the complex e· does have the good properties that for each i,
ei is a fiat A-module (since.? is flat over Y), and for each i, hi( C) = Hi(X,ff)
is a finitely generated A-module, since ff is coherent and f is projective.
Now the result of the proposition is a consequence of the following algebraic
lemma.

Lemma 12.3. Let A be a noetherian ring, and let C be a complex of A-modules,

bounded above, such that for each i, hi( C) is a finitely generated A-module.
Then there is a complex L" of finitely generated free A-modules, also
bounded above, and a morphism of complexes g:L· -+ C, such that the
induced map hi(L") -+ hi( e·) is an isomorphism for all i. Furthermore, if
each ei is a fiat A -module, then the map
hi(L.@ M)-+ hi(C@ M)
is an isomorphism for any A-module M.
PROOF. First we fix our notation. For any complex N·, we let
zn(N") = ker(dn:Nn-+ Nn+ 1 )
and

Thus we have

Now for large n, we have en = 0, so we define U = 0 there also. Suppose

inductively that the complex L. and the morphism of complexes g: L. -+ C
has been defined in degrees i > n in such a way that
hi(L") ~hi( C) for all i > n + 1, and (1)
zn+ 1(L") -+ hn+ 1 (C) is surjective. (2)
Then we will construct L n, d: U -+ U + 1, and g: L n -+ en to propagate these
properties one step further.
Choose a set of generators x1, . . . ,x, of hn( C), which is possible since hn( C)
is finitely generated. Lift them to a set of elements xl, ... ,x, E zn(C). On
the other hand, let Yr+ 1 , . . . ,y. be a set of generators of g- 1 (Bn+ 1 (C") ), which
is a submodule of u+ 1, hence finitely generated. Let g(yJ = Yi E Bn+ 1(C"),
and lift the Yi to a set of elements x,+ 1, •.. ,x. of en.
Now take U to be a free A-module on s generators e 1, .•. ,e5 • Define
d:U -+ L n+ 1 by dei = 0 for i = 1, ... ,r, and dei = Yi for i = r + 1, ... ,s.
Define g: U -+ en by gei = xi for all i.

283
III Cohomology

Then one checks easily that g commutes with d, that h" + 1 (L") --+ h" + 1 ( C)
is an isomorphism, and that Z"(L")--+ h"(C") is surjective. So inductively, we
construct the complex L" required.
Now suppose that each Ci is a flat A-module. Then we will prove, by
descending induction on i, that
hi(L" @ M) --+ hi( C @ M)
is an isomorphism for all A-modules M. Fori » 0, both Li and Care 0, so
both sides are 0. So suppose this is true for i + 1. It is sufficient to prove
the result for finitely generated A-modules, because any A-module is a direct
limit of finitely generated ones, and both @ and hi commute with direct
limits. So given M finitely generated, write it as a quotient of a free finitely
generated A-module E, and let R be the kernel:
0 --+ R --+ E --+ M --+ 0.
Since each Ci is flat by hypothesis, and each Li is flat, because it is free, we
get an exact, commutative diagram of complexes

0 ~ L. @R ----~ L. @E L"@M ~ 0

j l j
o~C®R----~C®E---~ C@M ~ 0.

Applying hi, we get a commutative diagram of long exact sequences. Since

the result holds for i + 1 by induction, mid for E, since E is free, and
hi(L") --+ hi( c·) is an isomorphism, the result for any M follows from the
subtle 5-lemma.

Now we will study conditions under which one of the functors Ti is left
exact, right exact, or exact. For any complex N·, we define
Wi(N") = coker(di- 1 :Ni- 1 --+ Ni)

so that we have an exact sequence

0 --+ hi(N") --+ Wi(N") --+ Ni+ 1 .

Proposition 12.4. The following conditions are equivalent:

(i) Ti is left exact;
(ii) Wi = Wi(L") is a projective A-module;
(iii) there is a finitely generated A-module Q, such that
Ti(M) = HomA(Q,M)
for all M.
Furthermore the Q in (iii) is unique.

284
 12 The Semicontinuity Theorem

PROOF. Since tensor product is right exact, we have

W;(L' 0 M) = W;(L') 0 M,
for any A-module M. We will write simply W; for W;(L'). Hence
T;(M) = ker(W; 0 M ~ Li+l 0 M).
Let 0 ~ M' ~ M be an inclusion. Then we obtain an exact, commutative
diagram

j
0 T;(M') W;0M' ~ Li+l 0 M'

]~ lp j
0 T;(M) W;0M Li+10 M.

The third vertical arrow is injective, since L; + 1 is free. A simple diagram chase
shows that ~ is injective if and only if p is. Since this is true for any choice of
0 ~ M' ~ M, we see that T; is left exact if and only if W; is flat. (Recall
that in any case T; is exact in the middle (12.1).) But since W; is finitely
generated, this is equivalent to W; being projective (9.1A). This shows
(i) ¢ > (ii).
(iii) = (i) is obvious.
To prove (ii) =(iii), let fi+ 1 and W; be the dual projective modules.
Define
Q = coker(D+ 1 ~ W;).
Then for every A-module M, we have
0 ~ Hom(Q,M) ~ Hom(W;,M) ~ Hom(fi+ l,M).
But the last two groups are W; 0 M, and Li+ 1 0 M, respectively, so
Hom(Q,M) = T;(M).
To see the uniqueness of Q, let Q' be another module such that T;(M) =
Hom(Q',M) for all M. Then
Hom(Q,M) = Hom(Q',M)
for all M. In particular, the elements
1 E Hom(Q,Q) = Hom(Q',Q)
and
1' E Hom(Q',Q') = Hom(Q,Q')
give isomorphisms of Q and Q', inverse to each other, and canonically
defined.

285
III Cohomology

Remark 12.4.1. There is a general theorem to the effect that any left-exact
functor Ton A-modules, which commutes with direct sums, is of the form
Hom(Q, ·) for s0me A-module Q. But even if T takes finitely generated
modules into finitely generated modules, Q need not be finitely generated.
Thus the fact that our Q in (iii) above is finitely generated is a strong fact
about the functor Ti.
For example, let A be a noetherian ring with infinitely many maximal
ideals mi. Let Q = l:A/mi, and let T be the functor Hom(Q, · ). Then Q is
not finitely generated, but for any finitely generated A-module M, T(M) is
finitely generated, because Hom(A/mi,M) =1- 0 if and only if mi E Ass M,
which is a finite set.

Proposition 12.5. For any M, there is a natural map

q>: Ti(A)@ M ~ Ti(M).
Furthermore, the following conditions are equivalent:
(i) Ti is right exact;
(ii) q> is an isomorphism for all M;
(iii) q> is surjective for all M.
PROOF. Since Ti is a functor, we have a natural map, for any M,
M = Hom(A,M) ~ Hom(Ti(A),Ti(M)).
This gives q>, by setting
q>(Lai @ mJ = Lt/l(mJai.
Since Ti and @ commute with direct limits, it will be sufficient to consider
finitely generated A-modules M. Write
A'~ As~ M ~ 0.
Then we have a diagram
--------> 0

where the bottom row is not necessarily exact. The first two vertical arrows
are isomorphisms. Thus, if Ti is right exact, q> is an isomorphism. This
proves (i) => (ii). The implication (ii) => (iii) is obvious, so we have only to
prove (iii) => (i). We must show if
0 ~ M' ~ M ~ M" ~ 0
is an exact sequence of A-modules, then
Ti(M') ~ Ti(M) ~ Ti(M") ~ 0

286
 12 The Semicontinuity Theorem

is exact. By (12.1) it is exact in the middle, so we have only to show that

Ti(M) --+ Ti(M") is surjective. This follows from the diagram
Ti(A) ® M -------t Ti(A) ® M" 0

l<p(M) l<p(M")

Ti(M) Ti(M")
and the fact that <p(M") is surjective.

Corollary 12.6. The following conditions are equivalent:

(i) Ti is exact;
(ii) Ti is right exact, and Ti(A) is a projective A-module.
PROOF. In any case, Ti is right exact, so by (12.5) we have T;(M) ~ Ti(A) ® M
for all A-modules M. Therefore Ti is exact if and only if T;(A) is flat. But
T;(A) is a finitely generated A-module, so this is equivalent to being locally
free (9.1A), hence projective.

Now we wish to localize the above discussion. For any point y E Y =

Spec A, we denote by r;. the restriction of the functor T; to the category of
AP-modules, where p ~ A is the prime ideal corresponding to y. Then we
say "T; is left exact at y" to meanT~ is left exact, and similarly for right exact,
or exact. Note that for any Ap-module N, r;(N) = h;(L;, ® N). Also note
that Ti is left exact if and only if it is left exact at all points y E Y; similarly
for right exact, or exact. Finally, since cohomology commutes with flat base
extension (9.3), we see that r;is the functor T; associated with the morphism
f': X' --+ Y' obtained from f by the flat base extension Y' = Spec (!) Y --+ Y.
So we can apply the results (12.4), (12.5), (12.6) locally to each r;.
Proposition 12.7. If Ti is left exact (respectively, right exact, exact) at some
point y 0 E Y, then the same is true for all points y in a suitable open neigh-
borhood U of Yo·
PROOF. Ti is left exact at Yo if and only if w;o
is free, by (12.4). But since
Wi is a coherent sheaf on Y, this implies that Wi is locally free in some
neighborhood U of y 0 , and so T; is left exact at all points of U.
Ti is right exact at a pointy if and only if Ti+ 1 is left exact there, by the
long exact sequence (12.1). So the second statement follows from the first,
applied to Ti + 1 .
Ti is exact at a point if and only if it is both left exact and right exact, so
the third statement is the conjunction of the first two.

Definition. Let Y be a topological space. A function <p: Y --+ Z is upper

semicontinuous if for each y E Y, there is an open neighborhood U of y,
such that for ally' E U, <p(y') ~ <p(y). Intuitively, this means that <p may
get bigger at special points.

287
III Cohomology

Remark 12.7.1. A function cp: Y ~ Z is upper semicontinuous if and only if

for each n E Z, the set { y E Y[cp(y) ~ n} is a closed subset of Y.

Example 12.7.2. Let Y be a noetherian scheme, and let :F be a coherent

sheaf on Y. Then the function
cp(y) = dimk(yJ(:F ® k(y))
is upper semicontinuous. Indeed, by Nakayama's lemma, cp(y) is equal to
the minimal number of generators of the {9Y-module :Fy. But if s 1, ... ,s, E :FY
form a minimal set of generators, they extend to sections of :F in some
neighborhood of y, and they generate :F in some neighborhood, because :F
is coherent. So if y' is in that neighborhood, then cp(y'), which is the minimal
number of generators of :Fy, is ~ r = cp( y).

Theorem 12.8 (Semicontinuity). Let f:X ~ Y be a projective morphism of

noetherian schemes, and let :F be a coherent sheaf on X,jlat over Y. Then
for each i ~ 0, the function
hi(y,:F) = dimk(yJ Hi( X Y':FY)
is an upper semicontinuous function on Y.
PROOF. The question is local on Y, so we may assume Y = Spec A is affine,
with A noetherian. Thus we can apply the earlier results of this section.
By (9.4) we have
hi(y,:F) = dimk(yJ Ti(k(y) ).
As in the proof of (12.4), we have
T;(k(y)) = ker(W; ® k(y) ~ Li+ 1 ® k(y) ).
On the other hand, there is an exact sequence

so tensoring with k(y), we obtain a four-term exact sequence

o~ Ti(k(y)) ~ wi ® k(y) ~ Li+ 1 ® k(y) ~ wi+ 1 ® k(y) ~ o.
Therefore, counting dimensions, we have
h;(y,:F) = dimk(yJ Wi ® k(y) + dimk(yl wi+ 1 ® k(y) - dimk(yJ Li+ 1 ® k(y).
Now Wi and wi+ 1 are finitely generated A-modules, so by (12.7.2) the first
two terms of this sum are upper semicontinuous functions of y. But Li+ 1
is a free A-module, so the last term is a constant function of y. Combining,
we see that hi(y,:F) is upper semicontinuous.

Corollary 12.9 (Grauert). With the same hypotheses as the theorem, suppose
furthermore that Y is integral, and that for some i, the function hi(y,:F) is
constant on Y. Then Ri*(:F) is locally free on Y, and for every y the natural

288
 12 The Semicontinuity Theorem

map

is an isomorphism.
PROOF. As above, we may assume that Y is affine. Using the expression for
h;(y,%) in the proof of the theorem, we conclude that the functions dim
W; ® k(y) and dim wi+ 1 ® k(y) must both be constant. But this implies
(II, 8.9) that W; and J.Vi+ 1 are both locally free sheaves on Y. So by (12.4),
T; and yi+ 1 are both left exact, soT; is exact, so by (12.6), T;(A) is a projective
A-module. But Ri*(%) is just T;(Af, so it is a locally free sheaf. Finally by
(12.5) we see that
Ri*(%) ® k(y) ~ H;(Xy,%y)
is an isomorphism for all y E Y.

Example 12.9.1. Let {X,} be a flat family of integral curves in P;;, with k
algebraically closed. Then for every closed point t E T, H 0 (X,,(!Jx,) = k. On
the other hand, the arithmetic genus Pa = 1 - x((!)x,) is constant, by (9.10).
So we see that in this case the functions h0 (t,(!)x) and h 1 (t,(!)x) are both constant
on T.

Example 12.9.2. In the flat family of (9.8.4), we have h0 (X 0 (!)x,) = 1 if

t =1- 0, and 2 if t = 0, because of the nilpotent elements. On the other
hand, h 1(X,,(!)x,) = 0 for t =1- 0, since X, is rational, and h 1(X 0 ,(!)x0 ) =
h 1(X 0 ,((!)x0 )red) = 1, since (X 0 )red is a plane cubic curve. So in this case the
functions h0 ,h 1 both jump up at t = 0.

Example 12.9.3. If {X,} is an algebraic family of nonsingular projective

varieties over C, parametrized by a variety T, then the functions h;(X,,(!)x,)
are actually constant for all i. The proof of this result requires transcendental
methods, namely the degeneration of the Hodge spectral sequence-cf.
Deligne[ 4].

Now we wish to give some more precise information about when the map
T;(A) ® k(y) ~ T;(k(y))
is an isomorphism. And here we will use a new ingredient in our proof,
namely the theorem on formal functions (11.1).

Proposition 12.10. Assume that for some i,y, the map

<p: T;(A) ® k(y) ~ T;(k(y))

is surjective. Then T; is right exact at y (and conversely, by (12.5) ).

PROOF. By making a flat base extension Spec (!)Y --+ Y if necessary (9.3), we
may assume that y is a closed point of Y; A is a local ring, with maximal ideal

289
III Cohomology

m, and k(y) = k = A/m. By (12.5), it is sufficient to show that

qJ(M): Ti(A) ® M --+ Ti(M)
is surjective for all A-modules M. Since Ti and tensor product commute with
direct limits, it is sufficient to consider finitely generated M.
First, we consider A-modules M of finite length, and we show that qJ(M)
is surjective, by induction on the length of M. If the length is 1, then M = k,
and qJ(k) is surjective by hypothesis. In general, write
0 --+ M' --+ M --+ M" --+ 0,
where M' and M" have length less than length M. Then using (12.1) we have
a commutative diagram with exact rows
0

The two outside vertical arrows are surjective, by the induction hypothesis,
so the middle one is surjective also.
Now let M be any finitely generated A-module. For each n, M/mnM is a
module of finite length, so that by the previous case,
qJn: Ti(A) ® M/mnM--+ Ti(M/mnM)
is surjective. Note that ker (/Jn is an A-module of finite length, so the inverse
system (ker qJn) satisfies the Mittag-Leffler condition (II, 9.1.2). Hence by
(II, 9.1) the map

is also surjective. But by the theorem on formal functions (11.1), applied

to the sheaf$' ®A M on X, the right hand side is just Ti(Mt. So we have a
surjection
(Ti(A) ® M)~ --+ Ti(M)~.
Since completion is a faithful exact functor for finitely generated A-modules,
it follows that

is surjective, so we are done.

Combining this with our earlier results, we obtain the following theorem.

Theorem 12.11 (Cohomology and Base Change). Let f:X --+ Y be a pro-
jective morphism of noetherian schemes, and let$' be a coherent sheaf on X,
flat over Y. Let y be a point of Y. Then:
(a) if the natural map
qJi(y):Ri_{*($') ® k(y)--+ Hi(Xy,:f'y)

290
 12 The Semicontinuity Theorem

is surjective, then it is an isomorphism, and the same is true for all y' in a
suitable neighborhood of y;
(b) Assume that cpi(y) is surjective. Then the following conditions are
equivalent:
(i) cpi- 1(y) is also surjective;
(ii) RiJ*($') is locally free in a neighborhood of y.
PROOF. (a) follows from (12.10), (12.7), and (12.5). (b) follows from (12.10),
(12.6), and (12.5), using the fact that Ti is exact if and only if y i - t and Ti are
both right exact.

References for §12. The semicontinuity theorem was first proved by

Grauert [1] in the complex-analytic case. These theorems in the algebraic
case are due to Grothendieck [EGA III, 7.7]. Our proof follows the main
ideas ofGrothendieck's proof, with simplifications due to Mumford [5, II, §5].

EXERCISES

12.1. Let Y be a scheme of finite type over an algebraically closed field k. Show that the
function
cp(y) == dimk(my/m;)
is upper semicontinuous on the set of closed points of Y.
12.2. Let {X,} be a family of hypersurfaces of the same degree in PZ. Show that for
each i, the function hi(X,,(I)x,l is a constant function oft.
12.3. Let X 1 <;; P~ be the rational normal quartic curve (which is the 4-uple embedding
of P 1 in P4 ). Let X 0 <;; Pf be a nonsingular rational quartic curve, such as the
one in (1, Ex. 3.18b). Use (9.8.3) to construct a flat family {X,} of curves in P4 ,
parametrized by T = A 1 , with the given fibres X 1 and X 0 for t = 1 and t = 0.
Let .f <;;IT P' x T be the ideal sheaf of the total family X s P 4 x T. Show that
J is flat over T. Then show that
fort ¥- 0
h0 (t,J) = {~ fort= 0
and also
fort ¥- 0
h1(t,J) = {~ fort= 0.
This gives another example of cohomology groups jumping at a special point.
12.4. Let Y be an integral scheme of finite type over an algebraically closed field k.
Let f: X ---> Y be a fl~t projective morphism whose fibres are all integral schemes.
Let 2',A be invertible sheaves on X, and assume for each y E Y that 2'Y ~ AY
on the fibre XY" Then show that there is an invertible sheaf .AI on Y such that
2' ~A®!* 5. [Hint: Use the results of this section to show thatf*(2' ®A- 1 )
is locally free of rank 1 on Y.] -
12.5. Let Y be an integral scheme of finite type over an algebraically closed field k.
Let ,g be a locally free sheaf on Y, and let X = P(tff)-see (II, §7). Then show that
Pic X ~ (Pic Y) x Z. This strengthens (II, Ex. 7.9)_

291
 III Cohomology

*12.6. Let X be an integral projective scheme over an algebraically closed field k, and
assume that H 1 (X,(!)x) = 0. Let T be a connected scheme of finite type over k.
(a) If !l' is an invertible sheaf on X x T, show that the invertible sheaves !l', on
X = X x {t} are isomorphic, for all closed points t E T.
(b) Show that Pic( X x T) = Pic X x Pic T. (Do not assume that Tis reduced!)
Cf. (IV, Ex. 4.10) and (V, Ex. 1.6) for examples where Pic(X x T) -# Pic X x
Pic T. [Hint: Apply (12.11) with i = 0,1 for suitable invertible sheaves on
X x T.]

292
CHAPTER IV

Curves

In this chapter we apply the techniques we have learned earlier to study

curves. But in fact, except for the proof of the Riemann-Roch theorem (1.3),
which uses Serre duality, we use very little of the fancy methods of schemes
and cohomology. So if a reader is willing to accept the statement of the
Riemann-Roch theorem, he can read this chapter at a much earlier stage of
his study of algebraic geometry. That may not be a bad idea, pedagogically,
because in that way he will see some applications of the general theory, and
in particular will gain some respect for the significance of the Riemann-Roch
theorem. In contrast, the proof of the Riemann-Roch theorem is not very
enlightening.
After reviewing what is needed from the earlier part of the book in §1, we
study in §2, 3 various ways of representing a curve explicitly. One way is to
represent the curve as a branched covering ofP 1 . So in §2 we make a general
study of one curve as a branched covering of another. The central result
here is Hurwitz's theorem (2.4) which compares the canonical divisors on
the two curves.
In §3 we give two other ways ofrepresenting a curve. We show that any
nonsingular projective curve can be embedded in P 3 , and it can be mapped
birationally into P 2 in such a way that the image has only nodes for singu-
larities. The proof of the latter theorem has an interesting extra twist in
characteristic p > 0.
In §4 we discuss the special case of curves of genus 1, called elliptic curves.
This is a whole subject in itself, quite independent of the rest of the chapter.
We have space for only a brief glimpse of some aspects of this fascinating
theory.
In §5, 6 we discuss the canonical embedding, and some classification
questions, both for abstract curves and for curves in P 3 .

293
IV Curves

1 Riemann-Roch Theorem
In this chapter we will use the word curve to mean a complete, nonsingular
curve over an algebraically closed field k. In other words (II, §6), a curve is
an integral scheme of dimension 1, proper over k, all of whose local rings
are regular. Such a curve is necessarily projective (II, 6. 7). If we want to
consider a more general kind of curve, we will use the word "scheme,"
appropriately qualified, e.g., "an integral scheme of dimension 1 of finite
type over k." We will use the word point to mean a closed point, unless we
specify the generic point.
We begin by reviewing some of the concepts introduced earlier in the
book, which we will use in our study of curves.
The most important single invariant of a curve is its genus. There are
several ways of defining it, all equivalent. For a curve X in projective space,
we have the arithmetic genus Pa(X), defined as ~- Px(O), where Px is the
Hilbert polynomial of X (I, Ex. 7.2). On the other hand, we have the geo-
metric genus p9 (X), defined as dimk r(X,wx), where wx is the canonical
sheaf (II, 8.18.2).

Proposition 1.1. If X is a curve, then

Pa(X) = P9 (X) = dimk H 1 (X,@x),
so we call this number simply the genus of X, and denote it by g.
PROOF. The equality p0 (X) = dim H 1 (X,(!)x) has been shown in (III, Ex. 5.3).
The equality p9 = dim H 1(X,@x) is a consequence of Serre duality (III,
7.12.2).

Remark 1.1.1. From g = p9 , we see that the genus of a curve is always non-
negative. Conversely, for any g ~ 0, there exist curves of genus g. For
example, take a divisor of type (g + 1,2) on a nonsingular quadric surface.
There exist such divisors which are irreducible and nonsingular, and they
have Pa = g (III, Ex. 5.6).

A (Wei/) divisor on the curve X is an element of the free abelian group

generated by the set of points of X (II, §6). We write a divisor as D = L,niPi
with ni E Z. Its degree is I,ni. Two divisors are linearly equivalent if their
difference is the divisor of a rational function. We have seen that the degree
of a divisor depends only on its linear equivalence class (II, 6.10). Since X
is nonsingular, for every divisor D we have an associated invertible sheaf
2(D), and the correspondence D --+ 2(D) gives an isomorphism of the
group Cl(X) of divisors modulo linear equivalence with the group Pic X of
invertible sheaves modulo isomorphism (II, 6.16).
A divisor D = L,niPi on X is effective if all ni ~ 0. The set of all effective
divisors linearly equivalent to a given divisor D is called a complete linear

294
 I Riemann-Roch Theorem

system (II, §7) and is denoted by IDI. The elements of IDI are in one-to-one
correspondence with the space
(H 0 (X,f£l(D)) - {0})/k*,
so IDI carries the structure of the set of closed points of a projective space
(II, 7.7). We denote dimk H 0 (X,f£l(D)) by l(D), so that the dimension of IDI
is l(D) - 1. The number l(D) is finite by (II, 5.19) or (Ill, 5.2).
As a consequence of this correspondence we have the following elemen-
tary, but useful, result.

Lemma 1.2. Let D be a divisor on a curve X. Then if l(D) # 0, we must have

deg D ): 0. Furthermore, if l(D) # 0 and deg D = 0, we must have D ~ 0,
i.e., fil(D) ~ r9x.
PROOF. If l(D) # 0, then the complete linear system IDI is nonempty. Hence
D is linearly equivalent to some effective divisor. Since the degree depends
only on the linear equivalence class, and the degree of an effective divisor is
nonnegative, we find deg D ): 0. If deg D = 0, then D is linearly equivalent
to an effective divisor of degree 0. But there is only one such, namely the
zero divisor.

We denote by Qx1k, or simply Qx, the sheaf of relative differentials of X

over k (II, §8). Since X has dimension 1, it is an invertible sheaf on X, and
so is equal to the canonical sheaf Wx on X. We call any divisor in the cor-
responding linear equivalence class a canonical divisor, and denote it by K.
(We also occasionally use the letter K to denote the function field of X,
but it should be clear from the context which meaning is intended.)

Theorem 1.3 (Riemann-Roch). Let D be a divisor on a curve X of genus g.

Then
l(D) - l(K - D) = deg D + 1 - g.
PROOF. The divisor K - D corresponds to the invertible sheaf wx ® ffl(D) ~.
Since X is projective (II, 6.7), we can apply Serre duality (III, 7.12.1) to
conclude that the vector space H 0 (X,wx ® ffl(D) ~) is dual to H 1 (X,f£l(D) ).
Thus we have to show that for any D,
x(£(D)) = deg D + 1 - g,
where for any coherent sheaf :F on X, x(:F) is the Euler characteristic
x(ff) = dim H 0 (X,:F) - dim H 1(X,ff).
First we consider the case D = 0. Then our formula says
dim H 0 (X,r9x) - dim H 1 (X,r9x) = 0 + 1 - g.
This is true, because H 0 (X,r9x) = k for any projective variety (1, 3.4), and
dim H 1 (X,r9x) = g by (1.1).

295
IV Curves

Next, let D be any divisor, and let P be any point. We will show that the
formula is true for D if and only if it is true for D + P. Since any divisor
can be reached from 0 in a finite number of steps by adding or subtracting
a point each time, this will show the result holds for all D.
We consider Pas a closed subscheme of X. Its structure sheaf is a sky-
scraper sheaf k sitting at the point P, which we denote by k(P), and its ideal
sheaf is .P(- P) by (II, 6.18). Therefore we have an exact sequence
0--+ .P( -P)--+ (!)x--+ k(P)--+ 0.
Tensoring with .P(D + P) we get
0 --+ .P(D) --+ .P(D + P) --+ k(P) --+ 0.
(Since .P(D + P) is locally free of rank 1, tensoring by it does not affect
the sheaf k(P).) Now the Euler characteristic is additive on short exact
sequences (III, Ex. 5.1), and x(k(P)) = 1, so we have
x(.P(D + P)) = x(.P(D)) + 1.
On the other hand, deg(D + P) = deg D + 1, so our formula is true for
D if and only if it is true forD + P, as required.
Remark 1.3.1. If X is a curve of degree din P", and D is a hyperplane section
X n H, so that .P(D) = (9x(1), then the Hilbert polynomial (III, Ex. 5.2)
tells us that
x(.P(D)) = d + 1 - Pa·
This is a special case of the Riemann-Roch theorem.

Remark 1.3.2. The Riemann-Roch theorem enables us to solve the

"Riemann-Roch problem" (II, Ex. 7.6) for a divisor D on a curve X. If
deg D < 0, then dimjnDj = -1 for all n > 0. If deg D = 0, then dimjnDI
is 0 or -1 depending on whether nD "' 0 or not. If deg D > 0, then
l(K - nD) = 0 as soon as n · deg D > deg K, by (1.2), so for n » 0 we have

dimlnDI = n · deg D - g.

Example 1.3.3. On a curve X of genus g, the canonical divisor K has degree

2g - 2. Indeed, we apply (1.3) with D = K. Since l(K) = p9 = g and
1(0) = 1, we have
g - 1 = deg K + 1 - g,
hence deg K = 2g - 2.

Example 1.3.4. We say a divisor D is special if l(K - D) > 0, and that

l(K - D) is its index of speciality. Otherwise D is nonspecial. If deg D >
2g - 2, then by (1.3.3), deg(K - D) < 0, so l(K - D) = 0 (1.2). Thus D
is nonspecial.

296
 1 Riemann-Roch Theorem

Example 1.3.5. Recall that a curve is rational if it is birational to P 1 (I,

Ex. 6.1). Since curves in this chapter are complete and rionsingular by
definition, a curve X is rational if and only if X ~ P 1 (1, 6.12). Now using
(1.3) we can show that X is rational if and only if g = 0. We already know
that Pa(P 1 ) = 0 (I, Ex. 7.2), so conversely suppose given a curve X of genus
0. Let P,Q be two distinct points of X and apply Riemann-Roch to the
divisor D = P - Q. Since deg(K - D) = -2, using (1.3.3) above, we have
l(K - D) = 0, and so we find l(D) = 1. But D is a divisor of degree 0,
so by (1.2) we have D ~ 0, in other words P "' Q. But this implies X is
rational (II, 6.10.1).

Example 1.3.6. We say a curve X is elliptic if g = 1. On an elliptic curve,

the canonical divisor K has degree 0, by (1.3.3). On the other hand, l(K) =
p9 = 1, so from (1.2) we conclude that K "' 0.

Example 1.3.7. Let X be an elliptic curve, let P 0 be a point of X, and let

Pica X denote the subgroup of Pic X corresponding to divisors of degree 0.
Then the map P --+ !l'(P - P 0 ) gives a one-to-one correspondence between
the set of points of X and the elements of the group Pica X. Thus we get a
group structure (with P 0 as identity) on the set of points of X, generalizing
(II, 6.1 0.2).
To see this, it will be enough to show that if D is any divisor of degree 0,
then there exists a unique point P EX such that D "' P - P 0 . We apply
Riemann-Roch to D + P 0 , and get
l(D + P0 ) - 1 - 1.
l(K - D - P 0 ) = 1 +
Now deg K = 0, so deg(K - D - P 0 ) = -1, and hence l(K - D - P 0 ) =
0. Therefore, l(D + P 0 ) = 1. In other words, dimiD + Pol = 0. This
means there is a unique effective divisor linearly equivalent to D + P 0 .
Since the degree is 1, it must be a single point P. Thus we have shown that
there is a unique point P ~ D + P0 , i.e., D "' P - P 0 .

Remark 1.3.8. For other proofs of the Riemann-Roch theorem, see Serre
[7, Ch. II] and Fulton [1].

EXERCISES

1.1. Let X be a curve, and let P E X be a point. Then there exists a nonconstant
rational function f E K(X), which is regular everywhere except at P.
1.2. Again let X be a curve, and let P 1 ,..• ,P, EX be points. Then there is a rational

function f E K(X) having poles (of some order) at each of the P;, and regular
elsewhere.
1.3. Let X be an integral, separated, regular, one-dimensional scheme of finite type
over k, which is not proper over k. Then X is affine. [Hint: Embed X in a (proper)
curve X over k, and use (Ex. 1.2) to construct a morphism f:X-+ P 1 such that
rl(Al) =X.]

297
IV Curves

1.4. Show that a separated, one-dimensional scheme of finite type over k, none of whose
irreducible components is proper over k, is affine. [Hint: Combine (Ex. 1.3) with
(III, Ex. 3.1, Ex. 3.2, Ex. 4.2).]

1.5. For an effective divisor D on a curve X of genus g, show that dimjDj ~ deg D.
Furthermore, equality holds if and only if D = 0 or g = 0.
1.6. Let X be a curve of genus g. Show that there is a finite morphism f:X-+ P 1
of degree ~ g + 1. (Recall that the degree of a finite morphism of curves f: X -+ Y
is defined as the degree of the field extension [K(X):K(Y)] (II, §6).)
1.7. A curve X is called hypere/liptic if g ? 2 and there exists a finite morphism
f:X -+pi of degree 2.
(a) If X is a curve of genus g = 2, show that the canonical divisor defines a com-
plete linear system jK I of degree 2 and dimension 1, without base points. Use
(II, 7.8.1) to conclude that X is hyperelliptic.
(b) Show that the curves constructed in (1.1.1) all admit a morphism of degree 2
to pi_ Thus there exist hyperelliptic curves of any genus g ? 2.
Note. We will see later (Ex. 3.2) that there exist nonhyperelliptic curves. See
also (V, Ex. 2.10).
1.8. Pa of a Singular Curve. Let X be an integral projective scheme of dimension 1
over k, and let X be its normalization (II, Ex. 3.8). Then there is an exact sequence
of sheaves on X,
o-+ mx -+ f*mx-+ I @pj@p-+ o,
PEX

where i!iP is the integral closure of @p. For each P EX, let bp = length(i!ipj@p).
(a) Show that Pa(X) = Pa(X) + IPEX bp. [Hint: Use (III, Ex. 4.1) and (III,
Ex. 5.3).]
(b) If Pa(X) = 0, show that X is already nonsingular and in fact isomorphic to P 1 .
This strengthens (1.3.5).
*(c) If Pis a node or an ordinary cusp (1, Ex. 5.6, Ex. 5.14), show that bp = 1. [Hint:
Show first that bp depends only on the analytic isomorphism class of the sin-
gularity at P. Then compute bP for the node and cusp of suitable plane cubic
curves. See (V, 3.9.3) for another method.]
*1.9. Riemann-Roch for Singular Curves. Let X be an integral projective scheme of
dimension 1 over k. Let X "g be the set of regular points of X.
(a) Let D = In;P; be a divisor with support in x"g' i.e., all P; E X,,g· Then
define deg D = In;.
Let !l'(D) be the associated invertible sheaf on X, and
show that
x(!l'(D)) = deg D + 1 - Pa·
(b) Show that any Cartier divisor on X is the difference of two very ample Cartier
divisors. (Use (II, Ex. 7.5).)
(c) Conclude that every invertible sheaf !l' on X is isomorphic to !l'(D) for some
divisor D with support in X"'.
(d) Assume furthermore that X is a locally complete intersection in some pro-
jective space. Then by (III, 7.11) the dualizing sheaf wx is an invertible sheaf
on X, so we can define the canonical divisor K to be a divisor with support in
X"' corresponding to wx. Then the formula of (a) becomes
I(D) - l(K - D) = deg D + 1 - Pa·

298
 2 Hurwitz's Theorem

1.10. Let X be an integral projective scheme of dimension 1 over k, which is locally

complete intersection, and has Pa = 1. Fix a point P 0 E X,g· Imitate (1.3.7) to
show that the map P--> !l'(P - P 0 ) gives a one-to-one correspondence between
the points of X,g and the elements of the group Pico X. This generalizes (II,
6.11.4) and (II, Ex. 6.7).

2 Hurwitz's Theorem

In this section we consider a finite morphism of curves f: X ~ Y, and study

the relation between their canonical divisors. The resulting formula in-
volving the genus of X, the genus of Y, and the number of ramification points
is called Hurwitz's theorem.
Recall that the degree of a finite morphism f:X ~ Y of curves is defined
to be the degree [K(X):K(Y)] of the extension of function fields (II, §6).
For any point P EX we define the ramification index ep as follows. Let
Q = f(P), let t E (9Q be a local parameter at Q, consider t as an element of
@p via the natural map f# :(9Q ~ @p, and define
ep = vp(t),
where Vp is the valuation associated to the valuation ring @p. If ep > 1 we
say f is ramified at P, and that Q is a branch point off (Fig. 13). If ep = 1, we
say f is unramified at P. This definition is consistent with the earlier definition
of unramified (Ill, Ex. 10.3) since our groundfield k is algebraically closed,
and so k(P) = k(Q) for any point P of X. In particular, iff is unramified
everywhere, it is etale, because in any case it is flat by (III, 9.7).

X y
Figure 13. A finite morphism of curves.

If char k = 0, or if char k = p, and p does not divide ep, we say that the
ramification is tame. If p does divide ep, it is wild.
Recall that we have defined a homomorphism f*: Div Y ~ Div X of the
groups of divisors, by setting
f*(Q) = I ep · P
P--+Q

for any point Q of Y, and extending by linearity (II, §6). If D is a divisor on

Y, then f*(2(D)) ~ 2(f* D) (II, Ex. 6.8), so that this f* on divisors is com-
patible with the homomorphism .f*: Pic Y ~ Pic X on invertible sheaves.

299
IV Curves

We say the morphism f:X ~ Y is separable if K(X) is a separable field

extension of K( Y).

Proposition 2.1. Let f: X ~ Y be a finite separable morphism of curves. Then

there is an exact sequence of sheaves on X,
0 ~ f*Qy ~ QX ~ QX/Y ~ 0.

PROOF. From (II, 8.11) we have this exact sequence, but without the 0 on
the left. So we have only to show that f*Qy ~ Qx is injective. Since both
are invertible sheaves on X, it will be sufficient to show that the map is
nonzero at the generic point. But since K(X) is separable over K(Y), the
sheaf Qx;Y is zero at the generic point of X, by (II, 8.6A). Hence f*Qy ~ Qx
is surjective at the generic point.

Since QY and Qx correspond to the canonical divisors on Y and X,

respectively, we see that the sheaf of relative differentials Qx;Y measures
their difference. So we will study this sheaf. For any point P EX, let Q =
f(P), let t be a local parameter at Q, and let u be a local parameter at P.
Then dt is a generator of the free CDQ-module QY,Q, and du is a generator of
the free CDp-module Qx,P, by (II, 8.7) and (II, 8.8). In particular, there is a
unique element g E CDp such that f*dt = g · du. We denote this element by
dtjdu.

Proposition 2.2. Let f:X ~ Y be a finite, separable morphism of curves.

Then:
(a) QX/Y is a torsion sheaf on X, with support equal to the set of ramifi-
cation points off In particular, f is ramified at only finitely many points;
(b) for each P EX, the stalk (Qx;y)p is a principal CDp-module of finite
length equal to vp(dtjdu);
(c) iff is tamely ramified at P, then

length(Qx;y)p = ep - 1.

Iff is wildly ramified, then the length is > ep - 1.

PROOF.
(a) The fact that Qx;Y is a torsion sheaf follows from (2.1) since f*Qy and
Qx are both invertible sheaves on X. Now (Qx 1y)p = 0 if and only if f*dt
is a generator for Qx,P, using the above notation. But this happens if and
only if tis a local parameter for CDp, i.e., f is unramified at P.
(b) Indeed, from the exact sequence of (2.1), we see that (Qx;y}p ~
Qx,Pif*QY,Q' which is isomorphic as an CDp-module to CDpj(dtjdu).
(c) Iff has ramification index e = ep, then we can write t = aue for
some unit a E CDp. Then
dt = aeue-ldu + ueda.

300
 2 Hurwitz's Theorem

If the ramification is tame, then e is a nonzero element of k, so we have

vp(dtjdu) = e - 1. Otherwise vp(dtjdu) ~ e.

Definition. Let f: X ---+ Y be a finite, separable morphism of curves. Then

we define the ramification divisor off to be
R = L 1ength(Qx1y)p · P.
PEX

Proposition 2.3. Let f:X ---+ Y be a finite, separable morphism of curves. Let
Kx and Ky be the canonical divisors of X and Y, respectively. Then
Kx ~f*Ky + R.
PROOF. Considering the divisor Rasa closed subscheme of X, we see from
(2.2) that its structure sheaf @R is isomorphic to QXJY· Tensoring the exact
sequence of (2.1) with Qi 1, we can therefore write an exact sequence
0 ---+ f*Qy @ Qi 1 ---+ t:9x ---+ @R ---+ 0.
But by (II, 6.18), the ideal sheaf of R is isomorphic to £'(- R), so we have
f*Qy ® Qil ~ £'( -R).
Now the result follows from taking associated divisors. (One can also prove
this proposition by applying the operation det of (II, Ex. 6.11) to the exact
sequence of (2.1).)

Corollary 2.4 (Hurwitz). Let f:X---+ Y be a finite separable morphism of

curves. Let n = deg f Then
2g(X) - 2 = n · (2g(Y) - 2) + deg R.
Furthermore, iff has only tame ramification, then
deg R = L (ep - 1).

PROOF. We take the degrees of the divisors in (2.3). The canonical divisor
has degree 2g - 2 by (1.3.3); f* multiplies degrees by n (II, 6.9); and if the
ramification is tame, R has degree L(ep - 1) by (2.2).

We complete our discussion of finite morphisms by describing what

happens in the purely inseparable case. First we define the Frobenius
morphism.

Definition. Let X be a scheme, all of whose local rings are of characteristic p

(i.e., contain Z/p). We define the Frobenius morphism F:X---+ X as fol-
lows: F is the identity map on the topological space of X, and F*: t:9x ---+
t:9x is the pth power map. Since the local rings are of characteristic p,
F* induces a local homomorphism on each local ring, so F is indeed a
morphism.

301
IV Curves

Remark 2.4.1. Ifn:X ~Speck is a scheme over a field k of characteristic p,

then F:X ~X is not k-linear. On the contrary, we have a commutative
diagram
X ---=F'-------> X

Speck __F=----> Speck

with the Frobenius morphism F of Speck (which corresponds to the pth
power map of k to itself).
We define a new scheme over k, XP, to be the same scheme X, but with
structural morphism F on. Thus k acts on (!Jxp via pth powers. Then F be-
comes a k-linear morphism F':XP ~X. We call this the k-linear Frobenius
morphism.

Example 2.4.2. If X is a scheme over k, then X P may or may not be iso-

morphic to X as a scheme over k. For example, if X = Spec k[t], where k
is a perfect field, then XP is isomorphic to X, because the pth power map
k ~ k is bijective. Under this identification, the k-linear Frobenius mor-
phism F': X ~ X corresponds to the homomorphism k[ t] ~ k[ t] defined
by t ~ tP. This is the morphism given in (1, Ex. 3.2).

Example 2.4.3. If X is a curve over k, an algebraically closed field of charac-

teristic p, then F': X P ~ X is a finite morphism of degree p. It corresponds
to the field inclusion K ~ K 11 P, where K is the function field of X, and
K 11P is the field of pth roots of elements of Kin some fixed algebraic closure
of K.

Proposition 2.5. Let f: X ~ Y be a finite morphism of curves, and suppose

that K(X) is a purely inseparable field extension of K(Y). Then X andY are
isomorphic as abstract schemes, and f is a composition of k-linear Frobenius
morphisms. In particular, g(X) = g(Y).
PROOF. Let the degree off be p'. Then K(X)P' s; K(Y), or in other words,
K(X) s; K(Y) 11Pr. On the other hand, consider the k-linear Frobenius
morphisms

where for each i, YP, = (J;•-dp· The composition of these is a morphism

f': Ypr~ Y, also of degree p'. Since K(X) s; K(Y) 11Pr, and both have the
same degree over K(Y), we conclude that K(X) = K(Y) 11P'. Since a curve
is uniquely determined by its function field (1, 6.12), we have X ~ YP'' and
f = f'. Therefore X and Yare isomorphic as abstract schemes, and their
genus (which does not depend on the k-structure) is the same.

302
 2 Hurwitz's Theorem

Example 2.5.1. If X = YP, and f:X--+ Y is the k-linear Frobenius mor-

phism, then f is ramified everywhere, with ramification index p. Indeed,
f is the identity on point sets, but pth power on structure sheaves. So if
t E (!)Pis a local parameter, j#t = tP. Since d(tP) = 0, the map j*Qy--+ Qx
is the zero map, so Qx;Y ~ Qx.

Example 2.5.2. Iff: X --+ Y is separable, then the degree of the ramification
divisor R is always an even number. This follows from the formula of (2.4).

Example 2.5.3. An etale covering of a scheme Y is a scheme X, together with

a finite etale morphism f: X --+ Y. It is called trivial if X is isomorphic to
a finite disjoint union of copies of Y. Y is called simply connected if it has
no nontrivial etale coverings.
Now we show that P 1 is simply connected. Indeed, let f:X--+ P 1 be an
etale covering. We may assume X is connected. Then X is smooth over k
since f is etale (III, 10.1 ), and X is proper over k since f is finite, so X is a
curve (note connected and regular imply irreducible). Again since f is etale,
f is separable, so we can apply Hurwitz's theorem. Since f is unramified,
R = 0 so we have
2g(X) - 2 = n( -2).

Since g(X) ~ 0, the only way this can happen is for g(X) = 0 and n = 1.
Thus X= P 1 .

Example 2.5.4. Iff: X --+ Y is any finite morphism of curves, then g(X) ~
g(Y). We can factor the field extension K(Y) ~ K(X) into a separable
extension followed by a purely inseparable extension. Since the genus
doesn't change for a purely inseparable extension (2.5), we reduce to the
case f separable. If g(Y) = 0, there is nothing to prove, so we may assume
g(Y) ~ 1. Then we rewrite the formula of (2.4) as

g(X) = g(Y) + (n - 1)(g(Y} - 1) + ~ deg R.

Since n - 1 ~ 0, g(Y) - 1 ~ 0, and deg R ~ 0, we are done. By the way,
this shows also that equality occurs (for f separable) only if n = 1, or
g( Y) = 1 and f is unramified.

Example 2.5.5 (Liiroth's Theorem). This says that if L is a subfield of a pure

transcendental extension k(t) of k, containing k, then L is also pure tran-
scendental. We may assume that L -=1= k, so that L has transcendence degree
1 over k. Then Lis a function field of a curve Y, and the inclusion L ~ k(t)
corresponds to a finite morphism f: P 1 --+ Y. By (2.5.4) we conclude that
g(Y) = 0, so by (1.3.5), Y ~ P 1 . Hence L ~ k(u) for some u.

303
IV Curves

Note: This proof is only fork algebraically closed, but the theorem is true
for any field k. An analogous result over k algebraically closed is true also
in dimension 2 (V, 6.2.1). In dimension 3 the corresponding statement is
false, because of the existence of nonrational unirational3-folds-see Clemens
and Griffiths [1] and Iskovskih and Manin [1].

EXERCISES

2.1. Use (2.5.3) to show that P" is simply connected.

2.2. Classification of Curves of Genus 2. Fix an algebraically closed field k of char-
acteristic =1- 2.
(a) If X is a curve of genus 2 over k, the canonical linear system IKI determines a
finite morphism f:X---+ P 1 of degree 2 (Ex. 1.7). Show that it is ramified at
exactly 6 points, with ramification index 2 at each one. Note that f is uniquely
determined, up to an automorphism of P 1 , so X determines an (unordered)
set of6 points ofP\ up to an automorphism ofP 1 .
(b) Conversely, given six distinct elements a 1, ... ,a 6 E k, let K be the extension
of k(x) determined by the equation z 2 = (x - ad · · · (x - a 6 ). Let f: X ---+ P 1
be the corresponding morphism of curves. Show that g(X) = 2, the map f
is the same as the one determined by the canonical linear system, and f is
ramified over the six points x = ai of P 1, and nowhere else. (Cf. (II, Ex. 6.4).)
(c) Using (I, Ex. 6.6), show that if P 1 ,P 2 ,P 3 are three distinct points of P 1 , then
there exists a unique cp E Aut P 1 such that cp(P 1) = 0, cp(P 2 ) = 1, cp(P 3 ) = oo.
Thus in (a), if we order the six points ofP\ and then normalize by sending the
first three to 0,1,oo, respectively, we may assume that X is ramified over
0,1,oo,{3 1 ,{3 2 ,{3 3 , where fJ~o/3 2 ,{3 3 are three distinct elements of k, ,.CO,l.
(d) Let E 6 be the symmetric group on 6 letters. Define an action of E 6 on sets
of three distinct elements {3 1 ,{3 2 ,{3 3 of k, =F-0,1, as follows: reorder the set
O,l,oo,{3 1 ,{3 2 ,{3 3 according to a given element fiE E 6 , then renormalize as in (c)
so that the first three become 0,1,oo again. Then the last three are the new
{3'1 ,p; ,/33.
(e) Summing up, conclude that there is a one-to-one correspondence between the
set of isomorphism classes of curves of genus 2 over k, and triples of distinct
elements {3 1 ,{3 2 ,{3 3 of k, =F-0,1, modulo the action of E 6 described in (d). In
particular, there are many non-isomorphic curves of genus 2. We say that
curves of genus 2 depend on three parameters, since they correspond to the
points of an open subset of Af modulo a finite group.
2.3. Plane Curves. Let X be a curve of degree din P 2 . For each point P EX, let T p(X)
be the tangent line to X at P (1, Ex. 7.3). Considering T p(X) as a point of the dual
projective plane (P 2 )*, the map P ---+ T p(X) gives a morphism of X to its dual
curve X* in (P 2 )* (1, Ex. 7.3). Note that even though X is nonsingular, X* in
general will have singularities. We assume char k = 0 below.
(a) Fix a line L ~ P 2 which is not tangent to X. Define a morphism cp:X---+ L by
cp(P) = Tp(X) n L, for each point P EX. Show that cp is ramified at P if and
only if either (1) PEL, or (2) Pis an inflection point of X, which means that the
intersection multiplicity (1, Ex. 5.4) ofTp(X) with X at Pis ~ 3. Conclude that
X has only finitely many inflection points.

304
 2 Hurwitz's Theorem

(b) A line ofP 2 is a multiple tangent of X if it is tangent to X at more than one point.
It is a bitangent if it is tangent to X at exactly two points: If L is a multiple
tangent of X, tangent to X at the points P 1, . . . ,P., and if none of the P; is an
inflection point, show that the corresponding point of the dual curve X* is an
ordinary r-fold point, which means a point of multiplicity r with distinct tangent
directions (1, Ex. 5.3). Conclude that X has only finitely many multiple tangents.
(c) Let 0 E P 2 be a point which is not on X, nor on any inflectional or multiple
tangent of X. Let L be a line not containing 0. Let ljJ: X --+ L be the morphism
defined by projection from 0. Show that ljJ is ramified at a point P EX if and
only if the line OP is tangent to X at P, and in that case the ramification index
is 2. Use Hurwitz's theorem and (1, Ex. 7.2) to conclude that there are exactly
d(d - 1) tangents of X passing through 0. Hence the degree of the dual curve
(sometimes called the class of X) is d(d - 1).
(d) Show that for all but a finite number of points of X, a point 0 of X lies on
exactly (d + 1)(d - 2) tangents of X, not counting the tangent at 0.
(e) Show that the degree of the morphism <p of (a) is d(d - 1). Conclude that if
d ;::, 2, then X has 3d(d - 2) inflection points, properly counted. (If T p(X) has
intersection multiplicity r with X at P, then P should be counted r - 2 times as
an inflection point. If r = 3 we call it an ordinary inflection point.) Show that
an ordinary inflection point of X corresponds to an ordinary cusp of the dual
curve X*.
(f) Now let X be a plane curve of degree d ;::, 2, and assume that the dual curve
X* has only nodes and ordinary cusps as singularities (which should be true
for sufficiently general X). Then show that X has exactly td(d- 2)(d- 3)(d + 3)
bitangents. [Hint: Show that X is the normalization of X*. Then calculate
Pa(X*) two ways: once as a plane curve of degree d(d - 1), and once using
(Ex. 1.8).]
(g) For example, a plane cubic curve has exactly 9 inflection points, all ordinary.
The line joining any two of them intersects the curve in a third one.
(h) A plane quartic curve has exactly 28 bitangents. (This holds even if the curve
has a tangent with four-fold contact, in which case the dual curve X* has a
tacnode.)
2.4. A Funny Curve in Characteristic p. Let X be the plane quartic curve x 3 y + lz +
z3 x = 0 over a field of characteristic 3. Show that X is nonsingular, every point
of X is an inflection point, the dual curve X* is isomorphic to X, but the natural
map X --> X* is purely inseparable.
2.5. Automorphisms of a Curve of Genus ;::,2. Prove the theorem of Hurwitz [1] that
a curve X of genus g ;::, 2 over a field of characteristic 0 has at most 84(g - 1) auto-
morphisms. We will see later (Ex. 5.2) or (V, Ex. 1.11) that the group G =Aut X is
finite. So let G have order n. Then G acts on the function field K(X). Let L be
the fixed field. Then the field extension L <;; K(X) corresponds to a finite mor-
phism of curves f: X --> Y of degree n.
(a) If P E X is a ramification point, and ep = r, show that f- 1f(P) consists of
exactly n/r points, each having ramification index r. Let P 1 , . . . ,P, be a maxi-
mal set of ramification points of X lying over distinct points of Y, and let
ep, = r;. Then show that Hurwitz's theorem implies that

(2g - 2)/n = 2g(Y) - 2 + 2: (1 - 1/r;).

i= 1

305
IV Curves

(b) Since g ~ 2, the left hand side of the equation is >0. Show that if g(Y) ~ 0,
s ~ 0, r; ~ 2, i = 1, ... ,s are integers such that

2g(Y) - 2 + L: (1 - 1/r;) > 0,

i= 1

then the minimum value of this expression is 1/42. Conclude that n ~ 84(g- 1).
See (Ex. 5. 7) for an example where this maximum is achieved.
Note: It is known that this maximum is achieved for infinitely many values of g
(Macbeath [1]). Over a field of characteristic p > 0, the same bound holds, pro-
vided p > g + 1, with one exception, namely the hyperelliptic curve y 2 = xP - x,
which hasp= 2g + 1 and 2p(p 2 - 1) automorphisms (Roquette [1]). For other
bounds on the order of the group of automorphisms in characteristic p, see Singh
[1] and Stichtenoth [1].
2.6. f* for Divisors. Let f:X-+ Y be a finite morphism of curves of degree n. We
define a homomorphism f*: Div X -+ Div Y by f*(l:n;P;) = L:nJ(P;) for any
divisor D = L:n;P; on X.
(a) For any locally free sheaf tC on Y, of rank r, we define det tC = 1\ 'tC EPic Y
(II, Ex. 6.11). In particular, for any invertible sheaf .,It on X,f*.,H is locally free
of rank n on Y, so we can consider det f*.,H E Pic Y. Show that for any divisor
Don X,
det(f*2'(D)) ;::;: (det f*@x) ® 2'(f*D).
Note in particular that det(f*2'(D)) =/= 2'(f*D) in general! [Hint: First con-
sider an effective divisor D, apply f* to the exact sequence 0 -+ 2'(- D) -+
@x -> @v -+ 0, and use (II, Ex. 6.11).]
(b) Conclude thatf*D depends only on the linear equivalence class of D, so there is
an induced homomorphism f*: Pic X --> Pic Y. Show that f* o f*: Pic Y -->
Pic Y is just multiplication by n.
(c) Use duality for a finite flat morphism (III, Ex. 6.10) and (III, Ex. 7.2) to show that
detf*Qx;::;: (detf*@x)- 1 ® D?".
(d) Now assume that f is separable, so we have the ramification divisor R. We
define the branch divisor B to be the divisor f*R on Y. Show that
(det f*@x) 2 ;::;: 2'(- B).
2.7. Etale Covers of Degree 2. Let Y be a curve over a field k of characteristic =/= 2.
We show there is a one-to-one correspondence between finite etale morphisms
f:X-+ Y of degree 2, and 2-torsion elements of Pic Y, i.e., invertible sheaves 2'
on Y with 2' 2 ;::;: @y.
(a) Given an etale morphism f:X-> Y of degree 2, there is a natural map @y->
f*@x. Let 2' be the cokernel. Then 2' is an invertible sheaf on Y, 2' ;::;: det f*@x,
and so 2' 2 ;::;: @y by (Ex. 2.6). Thus an etale cover of degree 2 determines a
2-torsion element in Pic Y.
(b) Conversely, given a 2-torsion element 2' in Pic Y, define an @y-algebra structure
on @y EB 2' by (a,b) · (a',b') = (aa' + <p(b ® b'), ab' + a'b), where <p is an
isomorphism of 2' ® 2' -+ @y. Then take X = Spec(@y EB 2') (II, Ex. 5.17).
Show that X is an etale cover of Y.
(c) Show that these two processes are inverse to each other. [Hint: Let T:X-> X
be the involution which interchanges the points of each fibre off Use the

306
 3 Embeddings in Projective Space

trace map a c-+ a + c(a) from j*{J)x --+ @y to show that the sequence of @y-
modules in (a)

is split exact.
Note. This is a special case of the more general fact that for (n, char k) = 1, the
etale Galois covers of Y with group Z/nZ are classified by the etale cohomology
group H;,(Y, Z/nZ), which is equal to the group of n-torsion points of Pic Y. See
Serre [6].

3 Embeddings in Projective Space

In this section we study embeddings of a curve in projective space. We will

show that any curve can be embedded in P 3 . Furthermore, any curve can
be mapped birationally into P 2 in such a way that the image has at most
nodes as singularities.
Recall that an invertible sheaf ff on a curve X is very ample (II, §5) if
it is isomorphic to @x(l) for some immersion of X in a projective space.
It is ample (II, §7) if for any coherent sheaf :F on X, the sheaf :F @ !£" is
generated by global sections for n » 0. We have seen that ff is ample if
and only if ffn is very ample for some n > 0 (II, 7.6). If D is a divisor on X,
we will say D is ample or very ample if .P(D) is.
Recall that a linear system is a set b of effective divisors, which forms a
linear subspace of a complete linear system IDI. A point P is a base point
of the linear system b if P E Supp D for all D E b. We have seen that a com-
plete linear system IDI is base-point free if and only if .P(D) is generated
by global sections (II, 7.8).
Our first result is a reinterpretation in the case of curves of the criterion
of (II, §7) for when a linear system gives rise to a closed immersion into
projective space.

Proposition 3.1. L~t D be a divisor on a curve X. Then:

(a) the complete linear system IDI has no base points if and only if for
every point P EX,
dimiD - Pi = dimiDI - 1;
(b) Dis very ample if and only if for every two points P,Q EX (including
the case P = Q),
dimiD - P - Ql = dimiDI - 2.
PROOF. First we consider the exact sequence of sheaves
0 ~ ff(D - P) ~ ff(D) ~ k(P) ~ 0.
Taking global sections, we have
0 ~ r(X,ff(D - P)) ~ F(X,:t'(D)) ~ k,

307
IV Curves

so in any case, we see that dimiD - PI is equal to either dimiDI or dimiDI - 1.

Furthermore, sending a divisor E to E + P defines a linear map

which is clearly injective. Therefore, the dimensions of these two linear

systems are equal if and only if cp is surjective. On the other hand, cp is
surjective if and only if P is a base point of IDI, so this proves (a).
To prove (b), we may assume that IDI has no base points. Indeed, this
is true if D is very ample. On the other hand, if D satisfies the condition
of (b), then we must a fortiori have
dimiD - PI = dimiDI - 1
for every P EX, so IDI has no base points.
This being the case, IDI determines a morphism of X to pn (II, 7.1) and
(II, 7.8.1), so the question is whether that morphism is a closed immersion.
We use the criterion of (II, 7.3) and (II, 7.8.2), so we have to see whether IDI
separates points and separates tangent vectors. The first condition says
that for any two distinct points P,Q E X, Q is not a base point of ID - Pl.
By (a) this is equivalent to saying
dimiD- P- Ql = dimiDI- 2.
The second condition says that for any point P EX, there is a divisor D' E IDI
such that P occurs with multiplicity 1 in D', because dim Tp(X) = 1, and
dim Tp(D') = 0 if P has multiplicity 1 in D', 1 if P has higher multiplicity.
But this just says Pis not a base point of ID - Pi,
or, using (a) again,
dimiD - 2PI = dimiDI - 2.
Thus our result follows from (II, 7.3).

Corollary 3.2. Let D be a divisor on a curve X of genus g.

(a) If deg D ~ 2g, then IDI has no base points.
(b) If deg D ~ 2g + 1, then Dis very ample.
PROOF. In case (a), both D and D - Pare nonspecial (1.3.4), so by Riemann-
Roch, dimiD - PI = dimiDI - 1. In case (b), D and D - P - Q are both
nonspecial, so dimiD- P- Ql = dimiDI- 2 again by Riemann-Roch.

Corollary 3.3. A divisor D on a curve X is ample if and only if deg D > 0.

PROOF. If D is ample, some multiple is very ample (II, 7.6), so nD ~ H
where H is a hyperplane section for a projective embedding, so deg H > 0,
hence deg D > 0. Conversely, if deg D > 0, then for n » 0, deg nD ~
2g(X) + 1, so by (3.2), nD is very ample, and so D is ample (II, 7.6).

Example 3.3.1. If g = 0, then D is ample <o> very ample <o> deg D > 0.
Since X ~ P 1 (1.3.5), this is just (II, 7.6.1).

308
 3 Embeddings in Projective Space

Example 3.3.2. Let X be a curve, and let D be a very ample divisor on X,

corresponding to a closed immersion cp: X ~ P". Then the degree of cp(X),
as defined in (1, §7) for a projective variety, is just equal to deg D (II, Ex. 6.2).

Example 3.3.3. Let X be an elliptic curve, i.e., g = 1 (1.3.6). Then any

divisor D of degree 3 is very ample. Such a divisor is nonspecial, so by
Riemann-Roch, dimiDI = 2. Thus we see that any elliptic curve can be
embedded in P 2 as a cubic curve. (Conversely, of course, any nonsingular
plane cubic is elliptic, by the genus formula (1, Ex. 7.2).)
In the case g = 1 we can actually say D very ample -=- deg D ;?: 3. Be-
cause if deg D = 2, then by Riemann-Roch, dimiDI = 1, so IDI defines a
morphism of X to P 1 , which cannot be a closed immersion.

Example 3.3.4. If g = 2, then any divisor D of degree 5 is very ample. By

Riemann-Roch, dimiDI = 3, so any curve of genus 2 can be embedded in
P 3 as a curve of degree 5.

Example 3.3.5. The result of (3.2) is not the best possible in general. For
example, if X is a plane curve of degree 4, then D = X.H is a very ample
divisor of degree 4, but g = 3 so 2g + 1 = 7.

Our next objective is to show that any curve can be embedded in P 3 .

For this purpose we consider a curve X ~ P", take a point 0 ¢ X, and
project X from 0 into pn- 1 (I, Ex. 3.14). This gives a morphism of X into
pn-t, and we investigate when it is a closed immersion.
If P,Q are two distinct points of X, we define the secant line determined
by P and Q to be the line in P" joining P and Q. If P is a point of X, we
define the tangent line to X at P to be the unique line L ~ P" passing through
P, whose tangent space T p(L) is equal to T p(X) as a subspace of T p(Pn).

Proposition 3.4. Let X be a curve in P", let 0 be a point not on X, and let
cp:X ~ pn- 1 be the morphism determined by projection from 0. Then
cp is a closed immersion if and only if
(1) 0 is not on any secant line of X, and
(2) 0 is not on any tangent line of X.
PROOF. The morphism cp corresponds (II, 7.8.1) to the linear system cut
out on X by the hyperplanes H of P" passing through 0. So cp is a closed
immersion if and only if this linear system separates points and separates
tangent vectors on X (II, 7.8.2). If P,Q are two distinct points on X, then cp
separates them if and only if there is an H containing 0 and P, but not Q.
This is possible if and only if 0 is not on the line PQ. If P E X, then cp sepa-
rates tangent vectors at P if and only if there is an H containing 0 and P,
and meeting X at P with multiplicity 1. This is possible if and only if 0
is not on the tangent line at P.

309
IV Curves

Proposition 3.5. If X is a curve in pn, with n ;:;:: 4, then there is a point 0 ¢ X

such that the projection from 0 gives a closed immersion of X into pn- 1 .
PROOF. Let Sec X be the union of all secant lines of X. We call this the
secant variety of X. It is a locally closed subset of pn, of dimension ~ 3,
since (at least locally) it is the image of a morphism from (X x X - Ll) x P 1
to pn which sends (P,Q,t) to the point ton the secant line through P and Q,
suitably parametrized.
Let Tan X, the tangent variety of X, be the union of all tangent lines of
X. It is a closed subset of pn, of dimension ~ 2, because it is locally an
image of X x P 1 .
Since n ;:;:: 4, Sec X u Tan X i= pn, so we can find plenty of points 0
which do not lie on any secant or tangent of X. Then the projection from 0
gives the required closed immersion, by (3.4).

Corollary 3.6. Any curve can be embedded in P 3 .

PROOF. First embed X in any projective space pn_ For example, take a
divisor D of degree d ;:;:: 2g + 1 and use (3.2). Since D is very ample, the
complete linear system IDI determines an embedding of X in pn with n =
dimiDI. If n ~ 3, we can consider pn as a subspace ofP 3 , so there is nothing
to prove. If n ;:;:: 4, we use (3.5) repeatedly to project from points until we
have X embedded in P 3 .

Next we study the projection of a curve X in P 3 to P 2 . In general the

secant variety will fill up all of P 3 , so we cannot avoid all the secants, and
the projected curve will be singular. However, we will see that it is possible
to choose the center of projection 0 so that the resulting morphism ((J
from X to P 2 is birational onto its image, and the image ((J(X) has at most
nodes as singularities.
Recall (I, Ex. 5.6) that a node is a singular point of a plane curve of mul-
tiplicity 2, with distinct tangent directions. We define a multisecant of X
to be a line in P 3 which meets X in three or more distinct points. A secant
with coplanar tangent lines is a secant joining two points P,Q of X, whose
tangent lines Lp,LQ lie in the same plane, or equivalently, such that Lp
meets LQ.

Proposition 3.7. Let X be a curve in P~, let 0 be a point not on X, and let
qJ: X --> P 2 be the morphism determined by projection from 0. Then ((J
is birational onto its image and ((J(X) has at most nodes as singularities,
if and only if
(1) 0 lies on only finitely many secants of X,
(2) 0 is not on any tangent line of X,
(3) 0 is not on any multisecant of X, and
(4) 0 is not on any secant with coplanar tangent lines.

310
 3 Embeddings in Projective Space

PROOF. Going back to the proof of (II, 7.3), condition (1) says that <p is
one-to-one almost everywhere, hence birational. When 0 does lie on a
secant line, conditions (2), (3), (4) tell us that line meets X in exactly two
points P,Q, it is not tangent to X at either one, and the tangent lines at P,Q
are mapped to distinct lines in P 2 . Hence the image <p(X) has a node at
that point.

To show that a point 0 exists satisfying (1)-(4) of (3.7), we will count

the dimensions of the bad points, as in the proof of (3.5). The hard part is
to show that not every secant is a multisecant, and not every secant has
coplanar tangent lines. Over C, one could see this from differential geom-
etry. However, we give a different proof, valid in all characteristics, which
is achieved by an interesting application 9f Hurwitz's theorem.

Proposition 3.8. Let X be a curve in P 3 , which is not contained in any plane.

Suppose either
(a) every secant of X is a multisecant, or
(b) for any two points P,Q EX, the tangent lines Lp,LQ are coplanar.
Then there is a point A E P 3 , which lies on every tanyent line of X.

PROOF. First we show that (a) implies (b). Fix a point R in X, and consider
the morphism l/J: X - R ---> P 2 induced by projection from R. Since every
secant is a multisecant, l/J is a many-to-one map. If l/J is inseparable, then
for any P EX, the tangent line Lp at X passes through R. This gives (b)
and our conclusion immediately, so we may assume that each such ljJ is
separable. In that case, let T be a nonsingular point of l/J(X) over which l/J
is not ramified. If P,Q E l/J- 1(T), then the tangent lines Lp,LQ to X are
projected into the tangent line LT to l/J(X) at T. So Lp and LQ are both in
the plane spanned by R and LT, hence coplanar.
Thus we have shown that for any R, and for almost all P,Q such that
P,Q,R are collinear, Lp and LQ are coplanar. Therefore, there is an open
set of (P,Q) in X x X for which Lp and LQ are coplanar. But the property
of Lp and LQ being coplanar is a closed condition, so we conclude that for
all P,Q EX, Lp and LQ are coplanar. This is (b).
Now assume (b). Take any two points P,Q EX with distinct tangents,
and let A = Lp n LQ. By hypothesis, X is not contained in any plane,
so in particular, if n is the plane spanned by Lp and LQ, then X n n is a
finite set of points. For any point REX - X n n, the tangent line LR
must meet both Lp and LQ. But since LR c:j;_ n, it must pass through A.
So there is an open set of X consisting of points R such that A E LR. Since
this is a closed condition, we conclude that A E LR for all R EX.

Definition. A curve X in P" is strange if there is a point A which lies on all the
tangent lines of X.

311
IV Curves

Example 3.8.1. P 1 is strange. Indeed, the tangent line at any point is the
same P 1 , so any point A E P 1 will do.

Example 3.8.2. A conic in P 2 over a field of characteristic 2 is strange. For

example, consider the conic y = x 2 • Then dyjdx = 0, so all the tangent
lines are horizontal, so they all pass through the point at infinity on the
x-axis.

Theorem 3.9 (Samuel [2]). The only strange curves in any P" are the line
(3.8.1) and the conic in characteristic 2 (3.8.2).
PROOF. By projecting down if necessary (3.5) we may assume that X lies
in P 3 . Choose an A3 in P 3 with affine coordinates x, y,z in such a way that
(1) A is the point at infinity on the x-axis,
(2) if A EX, then its tangent line LA is not in the xz-plane,
(3) the z-axis does not meet X,
(4) X does not meet the line at infinity of the xz-plane, except possibly at
A (Fig. 14).

y
M

z
Figure 14. Proof of (3.9).

First we project from A to the yz-plane. Since A lies on every tangent

line to X, the corresponding morphism from X to P 2 is ramified everywhere.
So either the image is a point (in which case X is a line), or it is inseparable
(2.2). We conclude that the functions y and z restricted to X lie in K(X)P,
where char k = p > 0.
Next, we project from the z-axis to the line Mat infinity in the xy-plane.
In other words, for each point P E X, we define cp(P) to be the intersection
of the plane spanned by P and the z-axis with the line M. This gives a
morphism cp: X ~ M of degree d = deg X. Note that cp is ramified exactly
at the points of X which lie in the finite part of the xz-plane, but not at A.

312
 3 Embeddings in Projective Space

We will apply Hurwitz's theorem (2.4) to the morphism ((J. For any
point P EX n xz-plane, we take u = x - a as a local coordinate, where
a E k, a # 0. We take t = yjx as a local coordinate at A on M. Then by
(2.2) we have to calculate vp(dtjdu). Write x = u + a, so t = y(u + a)- 1 .
Since y E K(X)P, we have dyjdu = 0, so
dtjdu = - y(u + a)- 2 .

But u + a is a unit in the local ring @p, so

Vp(dtjdu) = Vp(J').

If we let P 1 , . . . ,Pr be all the finite points of X n xz-plane, then Hurwitz's

theorem tells us that
r

2g - 2 = -2d + I Vp,(y).
i= 1

Now we consider two cases.

Case 1. If A~ X, the xz-plane meets X only at the points P;. Since
this plane is defined by the equation y = 0, we can compute the degree of
X as the number of intersections of X with this plane, namely
r

d = L Vp,(y).
i= 1

Substituting in the above, we have

2g- 2 = -d
which is possible only if g = 0 and d = 2. Thus X ~ P 1 as an abstract
curve (1.3.5), and its embedding is by a divisor D of degree 2. We have
dimiDI = 2 by Riemann~Roch, so X is a conic in a plane P 2 . For the conic
to be strange, we must have char k = 2.
Case 2. If A E X, then by condition (2) the xz-plane meets X transversally
at A, so we see similarly
r

d = I Vp,(y) + 1.
i= 1
So
2g- 2 = -d- 1
which implies g = 0, d = 1. This is the line.

Theorem 3.10. Let X be a curve in P 3 . Then there is a point 0 ~X such

that the projection from 0 determines a birational morphism ({J from X
to its image in P 2 , and that image has at most nodes for singularities.

PROOF.If X is contained in a plane already, any 0 not in that plane will do.
So we assume X is not contained in any plane. Then in particular, X is

313
IV Curves

neither a line nor a conic, so by (3.9), X is not strange. Therefore, by (3.8),

X has a secant which is not a multisecant, and it has a secant without co-
planar tangents. Since the same must be true for nearby secants, we see
that there is an open subset of X x X consisting of pairs <P,Q) such that
the secant line through P,Q is not a multisecant and does not have coplanar
tangents. Hence the subset of X x X consisting of pairs <P,Q) where the
secant is a multisecant or has coplanar tangents is a proper subset, has
dimension !( 1, and so the union in P 3 of the corresponding secant lines
has dimension !( 2. Combining with the fact that the tangent variety to X
has dimension !( 2 (see (3.5) ), we see that there is an open subset of P 3
consisting of points 0 which satisfy (2), (3), and (4) of (3. 7).
To complete the proof, by (3.7), we must show that 0 can be chosen to
lie on only finitely many secants of X. For this we consider the morphism
(X x X - Ll) x P 1 ~ P 3 (defined at least locally) which sends <P,Q,t) to
the point t on the secant line through P and Q. If the image has dimension
< 3, then we can choose 0 lying on no secant. If the image has dimension
= 3, then since it is a morphism between two varieties of the same dimension,
we can apply (II, Ex. 3. 7), and find there is an open set of points in P 3 over
which the fibre is finite. These points lie on only finitely many secants, so
we are done.
Corollary 3.11. Any curve is birationally equivalent to a plane curve with at
most nodes as singularities.
PROOF. Combine (3.6) with (3.10).

Remark 3.11.1. In view of (3.11), one way to approach the classification

problem for all curves is to study the family of plane curves of degree d
with r nodes, for any given d and r. The family of all plane curves of degree
dis a linear system of dimension td(d + 3), so it is parametrized by a pro-
jective space of that dimension. Inside that projective space, the (irreducible)
curves with r nodes form a locally closed subset ~ ,. If X is such a curve,
then the genus g of its normalization X is given by ·
g = t(d - 1)(d - 2) - r

because of (Ex. 1.8). So in order for ~.r to be nonempty, we must have

0 !( r !( t(d - 1)(d - 2).
Furthermore, both extremes are possible. We have seen by Bertini's theorem
(II, 8.20.2) that for any d, there are irreducible nonsingular curves of degree d
in P 2 , so this gives the case r = 0. On the other hand, for any d, we can
embed P 1 in pd as a curve of degree d (Ex. 3.4), and then project it into P 2
by (3.5) and (3.10), to get a curve X of degree d in P 2 having only nodes,
and with g(X) = 0. This gives r = t(d - l)(d - 2).
But the general problem of the structure of the ~.r is very difficult. Severi
[2, Anhang F] states that for every d,r, satisfying 0 !( r !( t{d- l)(d- 2), the

314
 3 Embeddings in Projective Space

algebraic set Vd,r is irreducible and nonempty of dimension !d(d + 3) - r,

but a complete proof was given only recently by Joe Harris.

EXERCISES

3.1. If X is a curve of genus 2, show that a divisor D is very ample = deg D ~ 5.

This strengthens (3.3.4).
3.2. Let X be a plane curve of degree 4.
(a) Show that the effective canonical divisors on X are exactly the divisors X.L,
where L is a line in P 2 .
(b) If Dis any effective divisor of degree 2 on X, show that dimiDI = 0.
(c) Conclude that X is not hyperelliptic (Ex. 1.7).
3.3. If X is a curve of genus ~ 2 which is a complete intersection (II, Ex. 8.4) in some
P", show that the canonical divisor K is very ample. Conclude that a curve of
genus 2 can never be a complete intersection in any P". Cf. (Ex. 5.1).
3.4. Let X be the d-uple embedding (I, Ex. 2.12) of P 1 in Pd, for any d ~ l. We call
X the rational normal curve of degree d in pd,
(a) Show that X is projectively normal, and that its homogeneous ideal can be
generated by forms of degree 2.
(b) If X is any curve of degree din P", with d :;:;: n, which is not contained in any
pn- \ show that in fact d = n, g(X) = 0, and X differs from the rational
normal curve of degree d only by an automorphism of Pd. Cf. (II. 7.8.5).
(c) In particular, any curve of degree 2 in any P" is a conic in some P 2 .
(d) A curve of degree 3 in any P" must be either a plane cubic curve, or the twisted
cubic curve in P 3 .
3.5. Let X be a curve in P 3 , which is not contained in any plane.
(a) If 0 ¢ X is a point, such that the projection from 0 induces a birational mor-
phism rp from X to its image in P 2 , show that rp(X) must be singular. [Hint:
Calculate dim H 0 (X,0x(1)) two ways.]
(b) If X has degree d and genus g, conclude that g < !(d - 1)(d - 2). (Use
(Ex. 1.8).)
(c) Now let [X,] be the flat family of curves induced by the projection (III, 9.8.3)
whose fibre over t = 1 is X, and whose fibre X 0 over t = 0 is a scheme with
support rp(X). Show that X 0 always has nilpotent elements. Thus the example
(III, 9.8.4) is typical.
3.6. Curves of Degree 4.
(a) If X is a curve of degree 4 in some P", show that either
(1) g = 0, in which case X is either the rational normal quartic in P 4 (Ex. 3.4)
or the rational quartic curve in P 3 (II, 7.8.6), or
(2) X c::::; P 2 , in which case g = 3, or
(3) X c::::; P 3 and g = 1.
(b) In the case g = 1, show that X is a complete intersection of two irreducible
quadric surfaces in P 3 (I, Ex. 5.11). [Hint: Use the exact sequence 0-+ .fx-+
@p3 -+ CDx -+ 0 to compute dim H 0 (P 3 ,.f x(2) ), and thus conclude that X is
contained in at least two irreducible quadric surfaces.]

315
IV Curves

3.7. In view of (3.10), one might ask conversely, is every plane curve with nodes a
projection of a nonsingular curve in P 3 ? Show that the curve xy + x 4 + y 4 = 0
(assume char k =1- 2) gives a counterexample.

3.8. We say a (singular) integral curve in P" is strange if there is a point which lies
on all the tangent lines at nonsingular points of the curve.
(a) There are many singular strange curves, e.g., the curve given parametrically by
x = t, y = tP, z = t 2P over a field of characteristic p > 0.
(b) Show, however, that if char k = 0, there aren't even any singular strange
curves besides P 1.

3.9. Prove the following lemma of Bertini: if X is a curve of degree din P 3 , not con-
tained in any plane, then for almost all planes H c;:: P 3 (meaning a Zariski open
subset of the dual projective space (P 3 )*), the intersection X n H consists of
exactly d distinct points, no three of which are collinear.

3.10. Generalize the statement that "not every secant is a multisecant" as follows.
If X is a curve in P", not contained in any pn-l, and if char k = 0, show that for
almost all choices of n- 1 points P 1 , . . . ,Pn-l on X, the linear space L"- 2
spanned by the P; does not contain any further points of X.

3.11 (a) If X is a nonsingular variety of dimension r in P", and ifn > 2r + 1, show that
there is a point 0 ¢' X, such that the projection from 0 induces a closed
immersion of X into P"- 1 .
(b) If X is the Veronese surface in P 5 , which is the 2-uple embedding of P 2 (I,
Ex. 2.13), show that each point of every secant line of X lies on infinitely many
secant lines. Therefore, the secant variety of X has dimension 4, and so in this
case there is a projection which gives a closed immersion of X into P4 (II,
Ex. 7.7). (A theorem ofSeveri [1] states that the Veronese surface is the only
surface in P 5 for which there is a projection giving a closed immersion into
P 4 . Usually one obtains a finite number of double points with transversal
tangent planes.)

3.12. For each value of d = 2,3,4,5 and r satisfying 0 ::::;; r ::::;; t(d - 1)(d - 2), show
that there exists an irreducible plane curve of degree d with r nodes and no other
singularities.

4 Elliptic Curves

The theory of elliptic curves (curves of genus 1) is varied and rich, and
provides a good example of the profound connections between abstract
algebraic geometry, complex analysis, and number theory. In this section
we will discuss briefly a number of topics concerning elliptic curves, to
give some idea of this theory. First we define thej-invariant, which classifies
elliptic curves up to isomorphism. Then we discuss the group structure on
the curve, and show that the elliptic curve is its own Jacobian variety. Next
we recall without proof the main results of the theory of elliptic functions
of a complex variable, and deduce various results about elliptic curves

316
 4 Elliptic Curves

over C. Then we define the Hasse invariant of a curve over a field of char-
acteristic p, and finally we consider the group of rational points of a curve
defined over Q.
For simplicity, we will omit the case of a ground field k of characteristic 2.
Most of the results of this section remain true, but the proofs require special
care. See, e.g., Tate [3] or the "formulaire" ofDeligne and Tate in Birch and
Kuyk [1].

The j-Invariant
Our first topic is to define the j-invariant of an elliptic curve, and to show
that it classifies elliptic curves up to isomorphism. Since j can be any ele-
ment of the ground field k, this will show that the affine line Ai is a variety
of moduli for elliptic curves over k.
Let X be an elliptic curve over the algebraically closed field k. Let P 0 EX
be a point, and consider the linear system [2P 0 [ on X. The divisor 2P 0
is nonspecial, so by Riemann-Roch, this linear system has dimension 1. It
has no base points, because otherwise the curve would be rational. There-
fore, it defines a morphism f: X -+ P 1 of degree 2, and we can specify that
f(P 0 ) = oo by a change of coordinates in P 1 .
Now if we assume char k =1= 2, it follows from Hurwitz's theorem that
f is ramified at exactly four points, with P 0 being one of them. If x = a,b,c
are the three branch points in P 1 besides oo, then there is a unique auto-
morphism of P 1 leaving oo fixed and sending a to 0 and b to 1, namely
x' = (x - a)/(b - a). So after this automorphism, we may assume that
f is branched over the points 0,1,A,oo of P 1 , where A E k, A =1= 0,1. This
defines a quantity},. We define j = j(A) by the formula

. s(A2-A+1)3
J = 2 A2(A - 1)2

This is thej-invariant of the curve X. (The coefficient 28 is thrown in to make

things work in characteristic 2, despite appearances to the contrary!) Our
main result then is the following.

Theorem 4.1. Let k be an algebraically closed field of characteristic =I= 2.

Then:
(a) for any elliptic curve X over k, the quantity j defined above depends
only on X;
(b) two elliptic curves X and X' over k are isomorphic if and only if
j(X) = j(X');
(c) every element of k occurs as the j-invariant of some elliptic curve
over k.
Thus we have a one-to-one correspondence between the set of elliptic
curves over k, up to isomorphism, and the elements of k, given by X f--+ j(X).

317
IV Curves

We will prove this theorem after some other preliminary results.

Lemma 4.2. Given any two points P,Q EX (including the case P = Q), there
is an automorphism a of X such that a 2 = id, a(P) = Q, and for any
R E X, R + a(R) "' P + Q.
PROOF. The linear system IP + Ql has dimension 1 and is base-point free,
hence defines a morphism g:X--+ P 1 of degree 2. It is separable, since
X t P 1 (2.5), so K(X) is a Galois extension of K(P 1). Let a be the non-
trivial automorphism of order 2 of K(X) over K(P 1). Then a interchanges
the two points of each fibre of g. Hence a(P) = Q, and for any R EX,
R + a(R) is a fibre of g, hence R + a(R) E IP + Ql, i.e., R + a(R) "'
p + Q.

Corollary 4.3. The group Aut X of automorphisms of X is transitive.

Lemma 4.4. If f 1 : X --+ P 1 and f 2 : X --+ P 1 are any two morphisms of degree
2 from X to Pi, then there are automorphisms a E Aut X and r E Aut P 1
such that f 2 o a = r o f 1 .

----------+
(J
X X

f1j k2
pl ------------+
T pl

PROOF. Let P 1 E X be a ramification point of f 1 and let P 2 E X be a ramifica-

tion point of f 2 • Then by (4.3) there is a a E Aut X such that a(P 1) = P 2 •
On the other hand, f 1 is determined by the linear system I2P 1 and f 2 is
1

determined by I2P 2 1. Since a takes one to the other, f 1 and f 2 o a correspond

to the same linear system, so they differ only by an automorphism r of
P 1 (II, 7.8.1).

Lemma 4.5. Let the symmetric group 1: 3 act on k - {0,1} as follows: given
A E k, A =/= 0,1, permute the numbers O,l,A according to a E 1: 3 , then apply
a linear transformation of x to send the first two back to 0,1, and let a(lc)
be the image of the third. Then the orbit of A consists of

1 1 1 1 1 A A- 1
11.'3:' -A, 1- A' A- 1'_A_

PROOF. Since the linear transformation sending a,b to 0,1 is x' = (x - a)/
(b - a), we have only to evaluate (c - a)/(b - a), where {a,b,c} = {0,1,A}
in any order.

318
 4 Elliptic Curves

Proposition 4.6. Let X be an elliptic curve over k, with char k # 2, and let
P 0 EX be a given point. Then there is a closed immersion X --+ P 2 such
that the image is the curve
y 2 = x(x - 1)(x - A)
for some A E k, and the point P 0 goes to the point at infinity (0,1,0) on the
y-axis. Furthermore, this A is the same as the A defined earlier, up to an
element of 2: 3 as in (4.5).
PROOF. We embed X in P 2 by the linear system I3P 0 I, which gives a closed
immersion (3.3.3). We choose our coordinates as follows. Think of the
vector spaces H 0 ((9(nP 0 )) as contained in each other,
k = H 0 ((9) s; H 0 (@(P 0 )) s; H 0 (@(2P 0 )) s; ...
By Riemann-Roch, we have
dimH 0 (@(nP 0 )) = n
for n > 0. Choose x E H 0 ( (9(2P 0 )) so that 1,x form a basis of that space, and
choose y E H 0 ( (9(3P 0 )) so that 1,x, y form a basis for that space. Then the
seven quantities

are in H 0 ( (9( 6P 0 ) ), which has dimension 6, so there is a linear relation among

them. Furthermore, both x 3 and y 2 occur with coefficient not equal to zero,
because they are the only functions with a 6-fold pole at P 0 . So replacing x
and y by suitable scalar multiples, we may assume they have coefficient 1.
Then we have a relation
y2 + a 1 xy + a3 y = x 3 + a2 x 2 + a4 x + a6
for suitable a; E k.
Now we will make linear changes of coordinates to get the equation in the
required form. First we complete the square on the left (here we use char
k # 2), replacing y by
y = y + 1 (a 1x + a 3 ).
1
2
The new equation has y 2 equal to a cubic equation in x, so it can be written
i = (x - a)(x - b)(x - c)
for suitable a,b,c E k. Now we make a linear change of x to send a,b to 0,1,
so the equation becomes
y 2 = x(x - 1)(x - A)
as required.
Since both x andy have a pole at P 0 , that point goes to the unique point
at infinity on this curve, which is (0,1,0).

319
IV Curves

If we project from P 0 to the x-axis, we get a finite morphism of degree 2,

sending P0 to oo, and ramified at 0,1,A,oo. So the A is the same as the one
defined earlier.
PROOF OF (4.1).
(a) To show that j depends only on X, suppose we made two choices of
base point P I>p 2 EX. Let f 1 : X -+ P 1 and f 2 : X -+ P 1 be the corresponding
morphisms. Then by (4.4) we can find automorphisms a E Aut X andrE
Aut P 1 such that f 2 o a = r o f 1 . Furthermore, we could choose a such that
a(P 1 ) = P 2 , hence r( oo) = oo. So r sends the branch points 0,1) 1 of f 1 to
the branch points 0,1) 2 of f 2 in some order. Hence by (4.5), A1 and A2 differ
only by an element of 2: 3 , via the action of(4.5). So we have only to observe
that for any rx E J: 3 ,j(A) = j(rx(A)). Indeed, since 2: 3 is generated by any two
elements of order 2, it is enough to show that

j(A) =j G) and j(A) = j(l - A),

which is clear by direct computation. Thus j depends only on X.

(b) Now suppose X and X' are two elliptic curves giving rise to A and A',
such thatj(A) = j(A'). First we note thatj is a rational function of A of degree 6,
i.e.,), -+ j defines a finite morphism P 1 -+ P 1 of degree 6. Furthermore, this is
a Galois covering, with Galois group I: 3 under the action described above.
Therefore, j(A) = j(A') if and only if A and A' differ by an element of I: 3 .
Now according to (4.6), X and X' can be embedded in P 2 so as to have the
equation y2 = x(x - 1)(x - A), or same with A'. Since A and A' differ by an
element of 2: 3 as in (4.5), after a linear change of variable in x, we have A = A'.
Thus X and X' are both isomorphic to the same curve in P 2 .
(c) Given any j E k, we can solve the polynomial equation
28 (A 2 - A + 1) 3 - jA 2 (A - V = 0

for A, and find a value of A, necessarily #0,1. Then the equation y 2 =

x(x - 1)(x - A) defines a nonsingular curve of degree 3 in P 2 , which is
therefore elliptic, and has the givenj as its j-invariant.

Example 4.6.1. The curve y 2 = x 3 - x of(l, Ex. 6.2) is nonsingular over any
field k with char k # 2. It has A = -1, hence j = 26 · 33 = 1728.

Example 4.6.2. The "Fermat curve" x 3 + y 3 = z 3 is nonsingular over any

field k with char k # 3. Making a change of variables x = x' + z, and setting
x' = -1/3, the equation becomes
1
27"
From here one can reduce it to a standard form, as in the proof of(4.6), with
A = -w or -w 2 , where w 3 = 1. Therefore,j = 0.

320
 4 Elliptic Curves

Corollary 4.7. Let X be an elliptic curve over k with char k # 2. Let P 0 EX,
and let G = Aut(X,P 0 ) be the group of automorphisms of X leaving P 0
fixed. Then G is a finite group of order
2 if j # 0, 1728
4 if j = 1728 and char k # 3
6 if j = 0 and char k # 3
12 if j = 0 ( = 1728) and char k = 3.
PROOF. Let f: X --+ P 1 be a morphism of degree 2, with f(P 0 ) = oo, branched
over 0,1,A, oo as above. If a E G, then by (4.4) there is an automorphism r
ofP 1, sending oo to oo, such thatf o a = r of In particular, r sends {0,1,A}
to {0,1,A} in some order. If r = id, then either a = id or a is the automor-
phism interchanging the sheets off Thus in any case we have two elements
in G.
If r # id, then r permutes {0,1,A }, so A must be equal to one of the other
expressions of (4.5). This can happen only in the following cases:
(1) if A = -1 or t or 2, and char k # 3, then A coincides with one other
element of its orbit under I: 3 , so G has order 4. This is the case j = 1728;
(2) if A = - w or - w 2 , and char k # 3, then ), coincides with two other
elements of its orbit under I: 3 , so G has order 6. In this case j = 0;
(3) if char k = 3 and A = -1, then all six elements of the orbit are the same,
so G has order 12. In this case j = 0 = 1728.

The Group Structure

Let X be an elliptic curve, and let P 0E X be a fixed point. We have seen, as a
consequence of the Riemann-Roch theorem (1.3.7) that the map P f--+
!f(P - P 0 ) induces a bijection between the set of points of X and the group
Pic 0 X. Thus the set of points of X forms a group, with P 0 as the 0 element,
and with addition characterized by P + Q = R if and only if P + Q ~
R + P 0 as divisors on X. This is the group structure on (X,P 0 ).
If we embed X in P 2 by the linear system I3P0 1, then three points P,Q,R of
the image are collinear if and only if P + Q + R ~ 3P 0 . This in turn is
equivalent to saying P + Q + R = 0 in the group structure. This shows that
the group law can be recovered from the geometry of the embedding. It
also generalizes (II, 6.10.2), where we used the geometry to define the group
law.
Now we will show that X is a group variety in the sense of (I, Ex. 3.21).

Proposition 4.8. Let (X,P 0 ) be an elliptic curve with its group structure. Then
the maps p:X--+ X given by P c---> -P, and 11:X x X--+ X given by
<P,Q) c---> P + Q are morphisms.
PROOF. First we apply (4.2) with P = Q = P 0 . Thus there is an automor-
phism a of X such that for any R, R + a(R) ~ 2P 0 . In other words,
a(R) = - R in the group structure, so this a is just p.

321
IV Curves

Next we apply (4.2) toP and P 0 . So there is an automorphism (J of X

with R + (J(R) ~ P + P 0 , i.e., (J(R) = P - R in the group. Preceding this
(J with p, we see that R --+ P + R, i.e., translation by P, is a morphism, for
any P.
Now take two distinct points P i= Q in X. Embed X in P 2 by I3P ol·
Form the equation of the line L joining P and Q. This depends on the co-
ordinates of P and Q. Now intersect L with X. We get a cubic equation in
the parameter along L, but we already know two of the intersections, so we
obtain the coordinates of the third point of intersection R as rational func-
tions in the coordinates of P and Q. Since R = - P - Q in the group
structure, this shows that the map (X x X - Ll) --+ X defined by <P,Q) --+
- P - Q is a morphism. Composing with p, we see that f.1 is a morphism for
pairs of distinct points of X.
To show that f.1 is a morphism also at points of the form <P,P), take any
Q i= 0. Translate one variable by Q, apply f.1 to <P,P + Q), then translate
by - Q. Since translation is a morphism, we see that f.1 is also a morphism at
these points.

Example 4.8.1. By iterating f.1, we see that for any integer n, multiplication
by n gives a morphism nx:X --+X. We will see later that for any n i= 0, nx
is a finite morphism of degree n2 ; its kernel is a group isomorphic to Z/n x
Z/n if (n,p) = 1, where p = char k, and is isomorphic to Z/p or 0 if n = p,
depending on the Hasse invariant of X. See (4.10), (4.17), (Ex. 4.6), (Ex. 4.7),
(Ex. 4.15).

Example 4.8.2. If P is a point of order 2 on X, then 2P ~ 2P 0 , so P is a

ramification point of the morphism f: X--+ P 1 defined by I2P 0 1, and f is
separable since X ~ P 1 (2.5). So there are only finitely many such points, and
if char k i= 2, there are exactly 4. Thus 2x is always a finite morphism, and if
char k i= 2, we see that it has degree 4, and its kernel is Z/2 x Z/2.

Example 4.8.3. If P is a point of order 3 on X, then 3P ~ 3P 0 , so P is an

inflection point of the embedding of X in P 2 by I3P ol· If char k i= 2,3, we
see by (Ex. 2.3) that there are exactly 9 inflection points of X. Thus 3x has
degree 9, and its kernel is isomorphic to Z/3 x Z/3. By the way, this has the
amusing geometric consequence that if P,Q are inflection points of X, then
the line PQ meets X in a third inflection point R of X. Indeed, R = - P - Q,
so it is also a point of order 3.

Lemma 4.9. If X,P 0 and X',P~ are two elliptic curves, and if f:X--+ X' is
a morphism sending P 0 to P~, then f is a homomorphism of the group
structures.
PROOF. If P + Q = R on X, then P + Q ~ R + P 0 as divisors. It follows
that f(P) + f(Q) ~ f(R) + f(P 0 ) by (Ex. 2.6), and since f(P 0 ) = P~, we have
f(P) + f(Q) = f(R) in the group law on X'.

322
 4 Elliptic Curves

Definition. If f,g are two morphisms of an elliptic curve X,P 0 to itself, sending
P 0 toP 0 , we define a morphism f + g by composing f x g: X --> X x X
with f.l· In other words, (f + g)(P) = f(P) + g(P) for all P. We define
the morphism f · g to befog. Then the set of all morphisms of X to
itself sending P 0 to P 0 forms a ring R = End(X,P 0 ), which we call the
ring ofendomorphisms of X,P 0 . Its zero element 0 is the morphism sending
X to P 0 . The unit element 1 is the identity map. The inverse morphism p
is -1. The distributive law f · (g + h) = f · g + f · h is a consequence
of the fact (4.9) that f is a homomorphism.

Proposition 4.10. Assume char k # 2. The map n ~---> nx defines an injective

ring homomorphism Z --> End(X,P 0 ). In particular, for all n # 0, nx is
a finite morphism.
PROOF. We will show by induction on n that nx # 0 for n ~ 1. It follows
that nx is a finite morphism (II, 6.8). For n = 1 it is clear; for n = 2 we have
seen it above (4.8.2). So let n > 2. Ifn is odd, say n = 2r + 1, and ifnx = 0,
then (2r)x = p. But p has degree 1, and (2r)x = 2x · rx is a finite morphism
(use induction hypothesis for r) of degree ~ 4, since 2x has degree 4 (4.8.2).
So this is impossible.
If n is even, say n = 2r, then nx = 2x · rx is finite by induction.

Remark 4.10.1. The ring of endomorphisms R is an important invariant of

the elliptic curve, but it is not easy to calculate. Let us just note for the
moment that its group of units R* is the group G = Aut(X,P0 ) studied
above (4.7). In particular, if j = 0 or 1728, it is bigger than { ± 1}, so R is
definitely bigger than Z.

The Jacobian Variety

Now we will give another, perhaps more natural, proof that the group law
on the elliptic curve makes it a group variety. Our earlier proof used geo-
metric properties of the embedding in P 2 . Now instead, we will show that
the group Pico X has a structure of algebraic variety which is so natural that
it is automatically a group variety. This approach makes sense for a curve
of any genus, and leads to the Jacobian variety of a curve. The idea is to
find a universal parameter space for divisor classes of degree 0.
Let X be a curve over k. For any scheme T over k, we define Pic 0 (X x T)
to be the subgroup of Pic(X x T) consisting of invertible sheaves whose
restriction to each fibre xt for t E T has degree 0. Let p:X X T--> T be
the second projection. For any invertible sheaf JV on T, p* JV E Pico(X x T),
because it is in fact trivial on each fibre. We define Pico(X/T) =
Pico(X x T)/p* Pic T, and we regard its elements as "families of invertible
sheaves of degree 0 on X, parametrized by T." Justification for this is the
fact that if T is integral and of finite type over k, and if 2 ,A E Pic( X x T),
then 2r ~ Ar on Xr for all t E T if and only if 2 @ A- 1 E p* Pic T (III,
Ex. 12.4).

323
IV Curves

Definition. Let X be a curve (of any genus) over k. The Jacobian variety of X
is a scheme J of finite type over k, together with an element 2? E Pico(XIJ),
having the following universal property: for any scheme T of finite type
over k, and for any .A E Pic (XIT), there is a unique morphism f: T --+ J
0

such that f* 2? ~ .A in Pic (XIT). (Note that f:X x T--+ X x J in-

0

duces a homomorphism f*: Pic (XIJ) --+ Pico(XIT).)

0

Remark 4.10.2. In the language of representable functors, this definition says

that J represents the functor T--+ Pic (XIT). 0

Remark 4.10.3. Since J is defined by a universal property, it is unique if it

exists. We will prove below that if X is an elliptic curve, then J exists, and
in fact we can take J = X. For curves of genus ~2 the existence is much
more difficult. See, for example, Chow [3] or Mumford [2] or Grothen-
dieck [5].

Remark 4.10.4. Assuming J exists, its closed points are in one-to-one cor-
respondence with elements of the group Pico X. Indeed, to give a closed
point of J is the same as giving a morphism Spec k --+ J, which by the uni-
versal property is the same thing as an element of Pic (Xlk) = Pica X.
0

Definition. A scheme X with a morphism to another scheme S is a group

scheme over S if there is a section e: S --+ X (the identity) and a morphism
p:X --+X over S (the inverse) and a morphism J.l:X x X--+ X over S
(the group operation) such that
(1) the composition J.1 o (id x p):X--+ X is equal to the projection X --+ S
followed by e, and
(2) the two morphisms J.1 o(J.l x id) and J.1 o(id x J.l) from X x X x X --+ X
are the same.

Remark 4.10.5. This notion of group scheme generalizes the earlier notion
of group variety (I, Ex. 3.21 ). Indeed, if S = Spec k and X is a variety over k,
taking e to be the 0 point, the properties (1), (2) can be checked on the closed
points of X. Then (1) says that p gives the inverse of each point, and (2) says
that the group law is associative.

Remark 4.10.6. The Jacobian variety J of a curve X is automatically a group

scheme over k. Indeed, using the universal property of J, define e: Speck --+ J
by taking the element 0 E Pico(Xlk). Define p: J --+ J by taking .9?- 1 E
Pico(XIJ). Define f.1: J x J --+ J by taking pf 2? ® Pi 2? E Pic 0 (XI J x J).
The properties (1) and (2) are verified immediately by the universal property
ofJ.

Remark 4.10.7. We can determine the Zariski tangent space to J at 0 as

follows. To give an element of the Zariski tangent space is equivalent to

324
 4 Elliptic Curves

giving a morphism ofT = Speck[ s]/s 2 to J sending Spec k to 0 (II, Ex. 2.8).
By the definition of J, this is equivalent to giving .,It E Pic (X/T) whose 0

restriction to Pico(X/k) is 0. But according to (Ill, Ex. 4.6) there is an exact

sequence 0--+ H 1 (X,(!)x)--+ Pic X[s]--+ Pic X--+ 0. So we see that the
Zariski tangent space to J at 0 is just H 1 (X,(!)x).

Remark 4.10.8. J is proper over k. We apply the valuative criterion of

properness (II, 4.7). It is enough to show (II, Ex. 4.11) that if R is any discrete
valuation ring containing k, with quotient field K, then a morphism of SpecK
to J extends uniquely to a morphism of Spec R to J. In other words, we must
show that an invertible sheaf A on X x Spec K extends uniquely to an
invertible sheaf on X x Spec R. Since X x Spec R is a regular scheme, this
follows from (II, 6.5) (note that the closed fibre of X x Spec R over Spec R,
as a divisor on X x Spec R, is linearly equivalent to 0).

Remark 4.10.9. If we fix a base point P 0 EX, then for any n ~ 1 there is a
morphism cpn:Xn--+ J defined by "(P 1 , . . . ,Pn)--+ !l'(P 1 + ... + Pn- nP 0 )"
(which means cook up the appropriate sheaf on X x xn to define cpn). If g
is the genus of X, then cpn will be surjective for n ~ g, because by Riemann-
Roch, every divisor class of degree ~ g contains an effective divisor. The
fibre of cpn over a point of J consists of all n-tuples (PI> ... ,Pn) such that
the divisors p 1 + . . . + p n form a complete linear system.
Ifn = g,thenformostchoicesofPI>···,P9 ,wehavel(P 1 + ... + P 9 ) = 1.
Indeed, by Riemann-Roch,
l(P 1 + ... + P9 ) = g + 1- g + l(K - P 1 - ... - P 9 ).
But l(K) = g. Taking P 1 not a base point of K, l(K - P d = g - 1. At
each step, taking P; not a base point of K- P 1 - ••. - P;_ 1 , we get
l(K- P 1 - • • • - P 9 ) = 0. Therefore, most fibres of cp9 are finite sets of
points. We conclude that lis irreducible and dimJ =g. On the other hand,
by (4.10.7), the Zariski tangent space to J at 0 is H 1 (X,mx), which has dimen-
sion g, so J is nonsingular at 0. Since it is a group scheme, it is a homogeneous
space, hence nonsingular everywhere. Hence J is a nonsingular variety.

Theorem 4.11. Let X be an elliptic curve, and fix a point P 0 EX. Take J = X,
and take 5l' on X x J to be !l'(L1) ® pf !l'(- P 0 ), where L1 <;:::: X x X is the
diagonal. Then J,!l' is a Jacobian variety for X. Furthermore, the resulting
structure of group variety on J (4.10.6) induces the same group structure on
X,P 0 as defined earlier.
PROOF. The last statement is obvious from the definitions. So we have only
to show that if T is any scheme of finite type over k, and if A E Pic (X /T), 0

then there is a unique morphism f: T--+ J such that f* 5l' ~ A.

Let p: X x T --+ T be the projection, and let q: X x T --+ X be the other
projection. Define A' = .,It ® q* !l'(P 0 ). Then A' has degree 1 along the
fibres. Hence, for any closed point t E T, we can apply Riemann-Roch to

325
IV Curves

A; on X 1 = X, and we find
dim H 0 (X,.A;) = 1
dim H 1 (X,.A;) = 0.
Since p is a projective morphism, and .A' is flat over T, we can apply the
theorem of cohomology and base change (III, 12.11). Looking first at
R 1p*(.A'), since the cohomology along the fibres is 0, the map qJ 1(t) of
(III, 12.11) is automatically surjective, hence an isomorphism, so we con-
clude that R 1 p*(.A') is identically 0. In particular, it is locally free, so we
deduce from part (b) of the theorem that qJ 0 (t) is also surjective. Therefore,
it is an isomorphism, and since <P - 1 (t) is always surjective, we see that p*(.A')
is locally free of rank 1.
Now replacing .A by .A ® p*p*(Ar 1 in Pic 0 (X/T), we may then assume
that p*(.A') : : : : (!)T· The section 1 E r(T,(!)r) gives a sections E r(X x T,.A'),
which defines an effective Cartier divisor Z ~ X x T. By construction, Z
intersects each fibre of p in just one point, and in fact one sees easily that the
restricted morphism p: Z --+ T is an isomorphism. Thus we get a section
s: T--+ Z ~ X x T. Composing with q gives the required morphism
f:T--+X.
Indeed, since Z is the graph off, we see that Z = f* L1, where L1 ~ X x X
is the diagonal. Hence the corresponding invertible sheaves correspond:
.A':::::::: f*.P(L1). Now twisting by -P0 shows that .A:::::::: f*.P, as required.
The uniqueness off is clear for the same reasons.

Elliptic Functions
It is hard to discuss elliptic curves without bringing in the theory of elliptic
functions of a complex variable. This classical topic from complex analysis
gives an insight into the theory of elliptic curves over C which cannot be
matched by purely algebraic techniques. So we will recall some of the
definitions and results of that theory without proof (signaling those state-
ments with a Bin their number), and give some applications to elliptic curves.
We refer to the book Hurwitz-Courant [ 1] for proofs.
Fix a complex number r, r ¢ R. Let A be the lattice in the complex plane
C consisting of all n + mr, with n,m E Z (Fig. 15).

Figure 15. A lattice in C, with one period parallelogram.

326
 4 Elliptic Curves

Definition. An elliptic function (with respect to the lattice A) is a meromorphic

function f(z) of the complex variable z such that f(z + w) = f(z) for all
w E A. (Sometimes these are called doubly periodic functions, since they
are periodic with respect to the periods 1;r.)

Because of the periodicity, an elliptic function is determined if one knows

its values on a single period parallelogram, such· as the one bounded by
0,1,r,r + 1 (Fig. 15).
An example of an elliptic function is the Weierstrass t.J-function defined by

t.J(z) = 2
Z
1 + WEA'
I ((
Z -
1 )2 - -2
W
1) '
W

where A' = A - {0}. One shows (Hurwitz-Courant [1, II, 1, §6]) that this
series converges at all z ¢A, thus giving a meromorphic function having a
double pole at the points of A, and which is elliptic. Its derivative
-2
t.J'(z) = I (z- w )3
WEA

is another elliptic function.

If one adds, subtracts, multiplies, or divides two elliptic functions with
periods in A, one gets another such. Hence the elliptic functions for a given
A form a field) 1

Theorem 4.12B. The field of elliptic functions for given A is generated over
C by the Weierstrass t.J-function and its derivative t.J'. They satisfy the
algebraic relation

where
1 1
g2 = 60 I
WEA'
4
w
and g3 = 14o I
WEA'
6
W
.

PROOF. Hurwitz-Courant [1, II, 1, §8, 9].

Thus if we define a mapping cp: C -+ P~ by sending z -+ (t.J(z),t.J'(z)) in

affine coordinates, we obtain a holomorphic mapping whose image lies
inside the curve X with equation

y2 = 4x3 - g2x - g3.

In fact, cp induces a bijective mapping of C/A to X (Hurwitz-Courant [1,

II, 5, §1 ]), and X is nonsingular, hence an elliptic curve. Under this mapping
the field of elliptic functions is identified with the function field of the curve
X. Thus for any elliptic function, we can speak of its divisor In;(aJ, with
a; E C/A.

327
IV Curves

Theorem 4.13B. Given distinct points a 1, ... ,aq E C/ A, and given integers
n 1, ,nq, a necessary and sufficient condition that there exist an elliptic
•••

function with divisor In;(a;) is that In; = 0 and In;a; = 0 in the group
C/A.
PROOF. Hurwitz-Courant [1, II, 1, §5, 14].

In particular, this says that a 1 + a2 =

b (mod A) if and only if there is
an elliptic function with zeros at a 1 and a 2 , and poles at band 0. Since this
function is a rational function on the curve X, this says that qJ(a 1 ) + qJ(a 2 ) "'
<P(b) + <P(O) as divisors on X. If we let P 0 = qJ(O), which is the point at
infinity on the y-axis, and give X the group structure with origin P 0 , this
says that <P(ad + qJ(a 2 ) = <P(b) in the group structure on X. In other words,
<P gives a group isomorphism between C/ A under addition, and X with its
group law.

Theorem 4.14B. Given c 2 ,c 3 E C, with L1 =F 0, where L1 = d - 27c~, there exists

a r E C, r ¢. R, and an a E C, a =F 0, such that the lattice A= (1,r) gives
g 2 = a 4 c 2 and g 3 = a 6 c 3 by the formulas above.
PROOF. Hurwitz-Courant [1, II, 4, §4].

This shows that every elliptic curve over C arises in this way. Indeed, if X
is any elliptic curve, we can embed X in P 2 to have an equation of the form
y 2 = x(x - 1)(x - A), with A =F 0,1 (4.6). By a linear change of variable in x,
one can bring this into the form y 2 = 4x 3 - c 2 x - c 3 , with c 2 =
(.,y'4/3)(). 2 - ). + l) and c 3 = (1/27)(). + 1)(2). 2 - SA + 2). Then L1 =
). 2 (). - 1) 2 , which is =FO since A =F 0,1. Now the curve determined by the
lattice A is equivalent to this one by a change of variables y' = a 3 y, x' = a 2 x.
Next we define J(r) = g~/LI. Then the j-invariant of X which we defined
earlier is just j = 1728 · J(r). Thus J(r) classifies the curve X up to iso-
morphism.

Theorem 4.15B. Let r,r' be two complex numbers. Then J(r) = J(r') if and
only if there are integers a,b,c,d E Z with ad - be = ± 1 and

r' = ar + b_
cr +d
Furthermore, given any r', there is a unique r with J(r) = J(r') such that r
lies in the region G (Fig. 16) defined by
1 1
--::::: Re r <-
2 ---= 2
and
if Re r ~ 0
if Re r > 0.

328
 4 Elliptic Curves

1
-1 0 2
Figure 16. The region G.

PROOF. Hurwitz-Courant [1, II, 4, §3].

Now we will start drawing consequences from this theory.

Theorem 4.16. Let X be an elliptic curve over C. Then as an abstract group,

X is isomorphic to R/Z x R/Z. In particular, for any n, the subgroup of
points of order n is isomorphic to Z/n x Zjn.
PROOF. We have seen that X is isomorphic as a group to C/A, which in turn
is isomorphic to R/Z x R/Z. The points of order n are represented by
(a/n) + (b/n)r, with a,b = 0,1, ... ,n - 1. The points whose coordinates are
not rational comhinations of 1,r are of infinite order.

Corollary 4.17. The morphism multiplication by n, nx:X -+X is a .finite mor-

phism of degree n2 .
PROOF. Since it is separable, and a group homomorphism, its degree is the
order of the kernel, which is n2 •

Next we will investigate the ring of endomorphisms R = End(X,P 0 ) of

the elliptic curve X determined by the elliptic functions with periods 1,r.

Proposition 4.18. There. is a one-to-one correspondence between endomor-

phisms f E R and complex numbers a E C such that a · A ~ A. This cor-
respondence gives an injective ring homomorphism of R to C.
PROOF. Given fER, we have seen (4.9) that f is a group homomorphism of
X to X. Hence under the identification of X with C/ A it gives a group
homomorphism J of C to C, such that J(A) ~ A. On the other hand, since
f is a morphism, the induced map J: C -+ Cis holomorphic. Now expanding
J as a power series in a neighborhood of the origin, and expressing the fact
that J(z + w) = J(z) + J(w) for any z,w there, we see that J must be just
multiplication by some complex number a.

329
IV Curves

Conversely, given (I( E C, such that (I( • A ~ A, clearly multiplication by (I(

induces a group homomorphism f of Cj A to itself, hence of X to itself. But
f is also holomorphic, so in fact it is a morphism of X to itself by GAGA
(=Serre [ 4]): see (App. B, Ex. 6.6).
It is clear under this correspondence that the ring operations of R corre-
spond to addition and multiplication of the corresponding complex num-
bers (1(.

Remark 4.18.1. Note in particular that the morphism nx E R, which is multi-

plication by n in the group structure (4.8.1) corresponds to multiplication by
n in C. This gives another proof of (4.1 0) for elliptic curves over C.

Definition. If X is an elliptic curve over C, we say it has complex multiplication

if the ring of endomorphisms R is bigger than Z. This terminology is
explained by (4.18).

Theorem 4.19. If X has complex multiplication, then r E Q(~) for some

d E Z, d > 0, and in that case, R is a subring ( i= Z) of the ring of integers of
the .field Q(~). Conversely, if r = r + s~, with r,s E Q, then X has
complex multiplication, and in fact

R = {a + brla,b E Z, and 2br,b(r 2 + ds 2 ) E Z}.

PROOF. Given r, we can determineR as the set of all (I( E C such that (I( • A ~ A.
A necessary and sufficient condition for (I( • A £ A is that there exist integers
a,b,c,e such that
(I( = a + br

(I(T = c + er.

E R, then (I( E Z, so we see that R n R = Z. On the other hand, if X has

If (I(
complex multiplication, then there is an (I( ¢= R, and in this case, b i= 0.
Eliminating (I( from these equations, we see that

br 2 + (a - e)r - c = 0,

which shows that r is in a quadratic extension ofQ. Since r ¢= R, it must be an

imaginary quadratic extension, so r E Q(~) for some dE Z, d > 0.
Eliminating r from the same equations, we find that

(1( 2 - (a - e)(l( + (ae - be) = 0,

which shows that (I( is integral over Z. Therefore R must be a sub ring of the
ring of integers of the field Q( ~).
Conversely, suppose r = r + s~, with r,s E Q. Then we can deter-
mine R as the set of all (I( = a + br, with a,b E Z, such that (I(T E A. Since

330
 4 Elliptic Curves

ocr = ar + br 2 , we must have br 2 EA. Now

r 2 = r2 - ds 2 + 2rsH,
which can be written
r 2 = - (r 2 + ds 2 ) + 2rr.
So in order to have br 2 E A we must have 2br E Z and b(r 2 + ds 2 ) E Z. These
conditions are necessary and sufficient so we get the required expression for
R. In particular, R > Z, so X has complex multiplication.

Corollary 4.20. There are only countably many values of j E C for which the
corresponding elliptic curve X has complex multiplication.
PROOF. Indeed, there are only countably many elements of all quadratic
extensions of Q.

Example 4.20.1. If r = i, then R is the ringof Gaussian integers Z[i]. In

of ± 1, ± i, so R* ~ Z/4. This
this case the group of units R* of R consists
means that the group of automorphisms of X has order 4, so by (4. 7) we must
have j = 1728. So we see in a roundabout way that r = i gives J(r) = 1.
Another way to see this is as follows. Since A = Z EE> Zi, the lattice A is
stable under multiplication by i. Therefore

g3 = 140 L w- 6 = 140 ,L i- 6 w- 6 = -g 3 .
roe A' roe A'

So g 3 = 0, which implies that J(r) = 1. The equation of X can be written

y 2 = x 3 -Ax.

Example 4.20.2. If r = w, where w 3 = 1, then R = Z[ w ], which is the ring

of integers in the field Q(j=-3). In this case R* = { ±1, ±w, ±w 2 } which is
isomorphic to Z/6. So again from (4. 7) we conclude that j = 0. One can also
see this directly as in (4.20.1) by showing that g 2 = 0. The equation of X can
be written y 2 = x 3 - B.

Example 4.20.3. If r = 2i, then R = Z[2i]. In this case R is a proper sub ring
of the ring of integers in the quadratic field Q(i), with conductor 2 (Ex. 4.21).

Remark 4.20.4. Even though we have a good criterion for complex multi-
plication in terms of r, the connection between r andj is not easy to compute.
Thus if we are given a curve by its equation in P 2 , or by its j-invariant, it is
not easy to tell whether it has complex multiplication or not. See (Ex. 4.5)
and (Ex. 4.12). There is an extensive classical literature relating complex
multiplication to class field theory-see, e.g. Deuring [2] or Serre's article in
Cassels and Frohlich [1, Ch. XIII]. Here are some of the principal results:
let X be an elliptic curve with complex multiplication, let R = End(X,P 0 ),

331
IV Curves

let K = Q(FJ) be the quotient field of R (4.19), and letj be thej-invariant.

Then (1) j is an algebraic integer; (2) the field K(j) is an abelian extension of
K of degree hR = # Pic R; (3) j E Z <o> hR = 1, and there are exactly 13 such
values ofj.

The Hasse Invariant

If X is an elliptic curve over a field k of characteristic p > 0, we define an
important invariant of X as follows. Let F:X-+ X be the Frobenius mor-
phism (2.4.1). Then F induces a map

F*:H 1 (X,(!)x)-+ H 1 (X,(!)x)

on cohomology. This map is not linear, but it is p-linear, namely F*(A.a) =

A_PF*(a) for all A. E k, a E H 1 (X,(!)x). Since X is elliptic, H 1 (X,(!)x) is a one-
dimensional vector space. Thus, since k is perfect, the map F* is either 0 or
bijective.

Definition. IfF* = 0, we say that X has Hasse invariant 0 or that X is super-

singular; otherwise we say that X has Hasse invariant 1.

For other interpretations of the Hasse invariant, see (Ex. 4.15), (Ex. 4.16).

Proposition 4.21. Let the elliptic curve X be embedded as a cubic curve in P 2

with homogeneous equation f(x,y,z) = 0. Then the Hasse invariant of X
is 0 if and only if the coefficient of (xyz)p- 1 in fP- 1 is 0.
PROOF. The ideal sheaf of X is isomorphic to @p(- 3), so we have an exact
sequence
0 -+ @p(- 3) ~ {!)p -+ {!)X -+ 0.
From this, by taking cohomology, we obtain an isomorphism
H 1 (X,(!)x)-+ H 2 (@p( -3)),

since Hi(@p) = 0 fori= 1,2. Recall also (III, 5.1) that H 2 (@p(-3)) is a
one-dimensional vector space with a natural basis (xyz)- 1 .
Now we can compute the action of Frobenius using this embedding. If
F 1 is the Frobenius morphism on P 2 , then Ff takes (!)x to (!)XP• where XP is
the subscheme of P 2 defined by fP = 0. On the other hand, X is a closed
subscheme of XP, so we have a commutative diagram
0----+ (!Jp( -3p) @p {!)XP - - 0

kp-1 j j
o---. @p(- 3) (!Jp {!)X ---.o

332
 4 Elliptic Curves

Hence we have a commutative diagram

H1(X,(!)x) ~ Hz(Pz,(!)p( -3))

lFj lFj
F* H 1(XP,(!)XP) ~ H 2 (P 2 ,(!)p( -3p))

j kp-1
H1(X,(!)x) ~ Hz(Pz,(!)p( -3))

where F is the Frobenius morphism of X. Now Fj((xyz)- 1 ) = (xyz)-P,

and its image in H 2 ((!)p( -3)) will be fP- 1 · (xyz)-P. On the other hand,
H 2 ( (!)p(- 3)) has basis (xyz)- 1, and any monomial having a nonnegative
exponent on x, y, or z is 0. Thus the image is just (xyz)- 1 times the coefficient
of(xyz)p- 1 infp- 1 , and so the Hasse invariant of X is determined by whether
or not this coefficient is zero.

Corollary 4.22. Assume p =1= 2, and let X be given by the equation y 2 =

x(x - 1)(x - A.), with A =I= 0,1. Then the Hasse invariant of X is 0 if and
only if hp(A.) = 0, where

1
k = 2 (p - 1).

PROOF. Weusethecriterionof(4.21). In thiscasef = y 2 z- x(x- z)(x- A.z).

To get (xyz)p- 1 infP- 1, we must have (y 2 z)k and (x(x - z)(x - A.z) t Then,
inside ((x - z)(x - A.z) )\ we need the coefficient of xkzk. So we take the
coefficient of xizk-i in (x - z)k, and the coefficient of xk-izi in (x - A.z)k.
Summing up, the coefficient of(xyz)p- 1 infP- 1 is

Since the outer factor is = 1 (mod p), we get hp(A.) as defined above.

Corollary 4.23. For given p, there are only finitely many elliptic curves (up to
isomorphism) over k having Hasse invariant 0. In fact, there are at most
[p/12] + 2 of them.

PROOF. The polynomial hP(A.) has degree k = !(P - 1) in A., so it has at most k
distinct roots. In particular, there are only finitely many corresponding
values of j. Since the correspondence A. ~ j is 6 to 1 with two exceptions,
we can have at most k/6 + 2 values ofj, hence at most [p/12] + 2.

333
IV Curves

Nate. In fact, lgusa [2] has shown that the roots of hP(A) are always distinct.
Using this, one can easily count the exact number ofj with Hasse invariant 0:
j = 0 occurs <o> p =
2 (mod 3) (Ex. 4.14);j = 1728 occurs <o> p = 3 (mod 4)
(4.23.5); the number ofj i= 0,1728 is exactly (p/12]. There are also tables of
these j for small values of p-see Deuring [1] or Birch and Kuyk [1, Table 6].

Example 4.23.1. Let p = 3. Then hp(A) = A + 1. The only solution is

I. = -1, which corresponds to j = 0 = 1728.

=
Example 4.23.2. If p = 5, hp(A) = A2 + 4A + 1 A2 - A + 1 (mod 5). This
has roots - w,- w 2 in a quadratic extension ofF P' with w 3 = 1. So j = 0.

Example 4.23.3. If p = 7, then

hp(A) = }, 3 + 9A 2 + 9A + 1.
This has roots - 1,2,4, which correspond to j = 1728.

Remark 4.23.4. A very interesting problem arises if we "fix the curve and
vary p." To make sense of this, let X ~ Pi be a cubic curve defined by an
equation f(x,y,z) = 0 with integer coefficients, and assume that X is non-
singular as a curve over C. Then for almost all primes p, the curve X<Pl ~
N p obtained- by reducing the coefficients off (mod p) will be nonsingular
over k<Pl = F p· So it makes sense to consider the set
~ = {p primeJX<Pl is nonsingular over k<Pl' and X<Pl has Hasse invariant 0}.
What can we say about this set? The facts (which we will not prove) are
that if X, as a curve over C, has complex multiplication, then ~ has density
!. Here we define the density of a set of primes ~ to be
!~~ # {p E ~IP ~ xv# {p primeJp ~ x}.

In fact, assuming X<Pl is nonsingular, then X<Pl has Hasse invariant 0 if and
only if either pis ramified or p remains prime in the imaginary quadratic field
containing the ring of complex multiplication of X (Deuring [ 1] ). If X does
not have complex multiplication, then ~ has density 0, but Elkies has shown
that ~ is infinite (N. Elkies, The existence of infinitely many supersingular
primes for every elliptic curve over Q, Invent. Math. 89 (1987) 561-567). There
is also ample numerical evidence for the conjecture of Lang and Trotter [1],
that more precisely
#{pE~Jp ~ x},.... c·JXjlogx
as x -+ oo, for some constant c > 0.

Example 4.23.5. Let X be the curve y 2 = x 3 - x. Then j = 1728, and as

we have seen (4.20.1), X has complex multiplication by i. For any p i= 2,
X<Pl is nonsingular, and we compute its Hasse invariant by the criterion of
(4.21). With k = !(P - 1), we need the coefficient of xk in (x 2 - 1t If k is

334
 4 Elliptic Curves

odd, it is 0. If k is even, say k = 2m, it is ( -1r(!) which is nonzero. We

conclude that
if p = 1 (mod 4), then Hasse = 1
{
if p = 3 (mod 4), then Hasse = 0.
Thus ~ = {p primelp =3 (mod 4)}. According to Dirichlet's theorem on
primes in arithmetic progressions (see, e.g., Serre [14, Ch VI, §4], this is a set
of primes of density!. In particular, there are infinitely many such primes.
Note that p = 3 (mod 4) if and only ifp is prime in the ring of Gaussian integers
Z[i].

Example 4.23.6. Let X be the curve y 2 = x(x - 1)(x + 2), so A = -2, and
· r 2 · 7 3 . Then X<Pl is nonsingular for p # 2,3, but one checks by
j = 26
the criterion of (4.22), using a calculator, that the only value of p ::;:; 73 giving
Hasse = 0 is p = 23. So we can guess that~ has density 0. Indeed,j is not an
integer, so by (4.20.4), X does not have complex multiplication. See Lang and
Trotter [1] for more extensive computations.

Rational Points on an Elliptic Curve

Let X be an elliptic curve over an algebraically closed field k, let P 0 be a fixed
point, and let X be embedded in Pf by the linear system I3Pol· Suppose that X
can be defined by an equation f(x,y,z) = 0 with coefficients in a smaller
field k 0 c:; k, and that the point P 0 has coordinates in k 0 . In this case we say
(X,P 0 ) is defined over k 0 . If this happens, then it is clear from the geometric
nature of the group law on X, that the set X(k 0 ) of points of X with co-
ordinates in k 0 forms a subgroup of the group of all points of X. It is an
interesting arithmetic problem to determine the nature ofthis subgroup.
In particular, if k = C and k 0 = Q, then because x, y,z are homogeneous
coordinates in P 2 , we may assume that the equation f(x,y,z) = 0 has integer
coefficients, and we are looking for integer solutions x,y,z. So we have a
cubic Diophantine equation in three variables.
A theorem of Mordell states that the group X(Q) is a finitely generated
abelian group. We will not prove this, but just give some examples. See
Cassels [1] and Tate [3] for two excellent surveys of the subject.

Example 4.23.7. The Fermat curve x 3 + l = z3 is defined over Q. Because

Fermat's theorem is true for exponent 3, the only points of X(Q) are (1, -1,0),
(1,0,1), and (0,1,1). These are three inflection points of X. Taking any one as
base point, the group X(Q) is isomorphic to Z/3.

Example 4.23.8. The curve y 2 + y = x 3 - x is defined over Q. Take

P 0 = (0,1,0) to be the 0 element in the group law, as usual. Then (according
to Tate [3]), the group X(Q) is infinite cyclic, generated by the point P with
affine coordinates (0,0). Figure 17 shows this curve, with nP labeled as n,
for various integers n.

335
IV Curves

Figure 17. Rational points on the curve y 2 + y = x3 - x.

EXERCISES

4.1. Let X be an elliptic curve over k, with char k of. 2, let P EX be a point, and let
R be the graded ring R = ffin"o H 0 (X,C9x(nP)). Show that for suitable choice
of t,x,y,
R ~ k[t,x,y]/(y 2 - x(x - t 2 )(x - At 2 ) ),
as a graded ring, where k[t,x,yJ is graded by setting deg t = 1, deg x = 2,
deg y = 3.
4.2. If D is any divisor of degree ;::. 3 on the elliptic curve X, and if we embed X in
P" by the complete linear system JDJ, show that the image of X in P" is projec-
tively normal.
Note. It is true more generally that if D is a divisor of degree ;::. 2g + 1 on a
curve of genus g, then the embedding of X by JDJ is projectively normal (Mumford
[ 4, p. 55]).

4.3. Let the elliptic curve X be embedded in P 2 so as to have the equation y 2 =

x(x - 1)(x - A). Show that any automorphism of X leaving P0 = (0,1,0) fixed
is induced by an automorphism of P 2 coming from the automorphism of the
affine (x,y)-plane given by
x' =ax+ b
{
y' = cy.
In each of the four cases of (4.7), describe these automorphisms of P 2 explicitly,
and hence determine the structure of the group G = Aut(X,P 0 ).
4.4. Let X be an elliptic curve in P 2 given by an equation of the form

l + a 1 xy + a3 y = x3 + a2 x 2 + a4 x + a6 •
Show that the j-invariant is a rational function of the a;, with coefficients in Q.
In particular, if the a; are all in some field k 0 s; k, then j E k 0 also. Furthermore,

336
 4 Elliptic Curves

for every a E k0 , there exists an elliptic curve defined over k0 with }-invariant
equal to a.

4.5. Let X ,P 0 be an elliptic curve having an endomorphism f: X --> X of degree 2.

(a) If we represent X as a 2-1 covering of P 1 by a morphism n: X __. P 1 ramified
at P 0 , then as in (4.4), show that there is another morphism n':X--> P 1 and
a morphism g:P 1 -->Pi, also of degree 2, such that no f = go n'.
(b) For suitable choices of coordinates in the two copies of P 1 , show that g can
be taken to be the morphism x --> x 2 .
(c) Now show that g is branched over two of the branch points of n, and that g- 1
of the other two branch points of n consists of the four branch points of n'.
Deduce a relation involving the invariant A of X.
(d) Solving the above, show that there are just three values ofj corresponding to
elliptic curves with an endomorphism of degree 2, and find the corresponding
values of A andj. [Answers:}= 26 · 33 ;} = 26 · 53 ;j = -3 3 ·5 3 .]

4.6. (a) Let X be a curve of genus g embedded birationally in P 2 as a curve of degree d

with r nodes. Generalize the method of(Ex. 2.3) to show that X has 6(g - 1) +
3d inflection points. A node does not count as an inflection point. Assume
char k = 0.
(b) Now let X be a curve of genus g embedded as a curve of degree din P", n ~ 3,
not contained in any pn-'. For each point P EX, there is a hyperplane H
containing P, such that P counts at least n times in the intersection HnX.
This is called an osculating hyperplane at P. It generalizes the notion of
tangent line for curves in P 2 . If P counts at least n + 1 times in H n X, we
say H is a hyperosculating hyperplane, and that P is a hyperosculation point.
Use Hurwitz's theorem as above, and induction on n, to show that X has
n(n + 1)(g - 1) + (n + 1)d hyperosculation points.
(c) If X is an elliptic curve, for any d ~ 3, embed X as a curve of degree d in
pd- 1, and conclude that X has exactly d 2 points of order din its group law.

4.7. The Dual of a Morphism. Let X and X' be elliptic curves over k, with base points
P 0 ,P0.
(a) If f:X--> X' is any morphism, use (4.11) to show that f*:Pic X'--> Pic X
induces a homomorphism J: (X',P 0) --> (X,P 0 ). We call this the dual of f.
(b) If f:X -->X' and g:X' -->X" are two morphisms, then (go Jf =Jog.
(c) Assumef(P 0 ) = P 0,andletn = degf ShowthatifQEXisanypoint,and
f(Q) = Q', then](Q') = nx(Q). (Do the separable and purely inseparable cases
separately, then combine.) Conclude that f o j = nx· and j of= nx.
*(d) If f,g:X--> X' are two morphisms preserving the base points P 0 ,P0, then
(f + gf = J + g. [Hints: It is enough to show for any If' EPic X', that
(f + g)*ff' ~ f*!i' ® g*ff'. For any f, let rf:X--> X X X' be the graph
morphism. Then it is enough to show (for If'' = pj.!l') that

rJ+ 9 (ff'') = rjff'' ® r: If''.

Let a:X -->X x X' be the section x--> (x,P 0). Define a subgroup of
Pic(X x X') as follows:

Pic~ = { .2' EPic( X x X')lff' has degree 0 along each fibre of p 1 , and a* If' = 0
in Pic X}.

337
IV Curves

Note that this subgroup is isomorphic to the group Pic (X'/X) used in the
0

definition of the Jacobian variety. Hence there is a 1-1 correspondence

between morphisms f: X --+ X' and elements !l'1 E Pic.,. (this defines !l'1 ).
Now compute explicitly to show that r;(!l' 1 ) = rj(!l' 9 ) for any f,g.
Use the fact that !l'J+g = !l'1 (8) !l' 9 , and the fact that for any !l' on X',
p~ !l' E Pic~ to prove the result.]
(e) Using (d), show that for any n E Z, fix = nx. Conclude that deg nx = n2 .
(f) Show for any fthat deg] = degf

4.8. For any curve X, the algebraic fundamental group n 1 (X) is defined as
&!! Gal(K'/K), where K is the function field of X, and K' runs over all Galois
extensions of K such that the corresponding curve X' is etale over X (III, Ex. 10.3).
Thus, for example, n 1(P 1 ) = 1 (2.5.3). Show that for an elliptic curve X,

n 1(X) = f1 Z1 x Z1 ifchark = 0;
lprime

n 1(X) = fl Z1 x Z1 if char k = p and Hasse X = 0;

l*p

nl(X) = zp X TI Zz X Zz if char k = p and Hasse X -+ 0,

l*p

where Z1 = &!! Z/r' is the 1-adic integers.

[Hints: Any Galois etale cover X' of an elliptic curve is again an elliptic curve.
If the degree of X' over X is relatively prime top, then X' can be dominated by the
cover nx:X --+X for some integer n with (n,p) = 1. The Galois group of the
covering nx is Z/n x Z/n. Etale covers of degree divisible by p can occur only
if the Hasse invariant of X is not zero.]
Note: More generally, Grothendieck has shown [SGA 1, X, 2.6, p. 272] that
the algebraic fundamental group of any curve of genus g is isomorphic to a quo-
tient of the completion, with respect to subgroups of finite index, of the ordinary
topological fundamental group of a compact Riemann surface of genus g, i.e., a
group with 2g generators a 1 , . . . , a9 , b~o . .. , b9 and the relation (a 1 b 1 aj 1 bj 1 ) · · ·
(a 9 b9 a9- 1 b; 1 ) = 1.

4.9. We say two elliptic curves X,X' are isogenous if there is a finite morphism
f:X--+ X'.
(a) Show that isogeny is an equivalence relation.
(b) For any elliptic curve X, show that the set of elliptic curves X' isogenous to X,
up to isomorphism, is countable. [Hint: X' is uniquely determined by X and
kerf.]

4.10. If X is an elliptic curve, show that there is an exact sequence

0--+ P! Pic X EB p~ Pic X--+ Pic( X x X)--+ R --+ 0,

where R = End(X,P 0 ). In particular, we see that Pic( X x X) is bigger than the
sum of the Picard groups of the factors. Cf. (III, Ex. 12.6), (V, Ex. 1.6).

4.11. Let X be an elliptic curve over C, defined by the elliptic functions with periods
1;r. Let R be the ring of endomorphisms of X.
(a) Iff E R is a nonzero endomorphism corresponding to complex multiplication
by IJ(, as in (4.18), show that degf = 11)(1 2 •

338
 4 Elliptic Curves

(b) Iff E R corresponds to rx E C again, show that the dual J of (Ex. 4.7) corre-
sponds to the complex conjugate cz of rx.
{c) If r E Q{j=d) happens to be integral over Z, show that R = Z[r].

4.12. Again let X be an elliptic curve over C determined by the elliptic functions with
periods 1,r, and assume that r lies in the region G of(4.15B).
(a) If X has any automorphisms leaving P 0 fixed other than ± 1, show that either
r = i or r = w, as in (4.20.1) and (4.20.2). This gives another proof of the
fact (4.7) that there are only two curves, up to isomorphism, having auto-
morphisms other than ± 1.
(b) Now show that there are exactly three values of r for which X admits an
endomorphism of degree 2. Can you match these with the three values of j
determined in (Ex. 4.5)? [Answers: r = i; r = H; r = i( -1 + J=7).]

4.13. If p= 13, there is just one value of j for which the Hasse invariant of the corre-
sponding curve is 0. Find it. [Answer:j = 5 (mod 13).]

4.14. The Fermat curve X:x 3 + i = z 3 gives a nonsingular curve in characteristic p

for every p ¥ 3. Determine the set ~ = {p ¥ 3IX<Pl has Hasse invariant 0},
and observe (modulo Dirichlet's theorem) that it is a set of primes of density t.
4.15. Let X be an elliptic curve over a field k of characteristic p. Let F': X P --+ X be
the k-linear Frobenius morphism (2.4.1). Use (4.10.7) to show that the dual
morphism F': X --+ X P is separable if and only if the Hasse invariant of X is 1.
Now use (Ex. 4.7) to show that if the Hasse invariant is 1, then the subgroup of
points of order p on X is isomorphic to Z/p; if the Hasse invariant is 0, it is 0.

4.16. Again let X be an elliptic curve over k of characteristic p, and suppose X is de-
fined over the field Fq of q = p' elements, i.e., X <;; P 2 can be defined by an
equation with coefficients in Fq. Assume also that X has a rational point over
Fq. Let F':Xq--+ X be the k-linear Frobenius with respect to q.
(a) Show that Xq ~ X as schemes over k, and that under this identification,
F':X--+ X is the map obtained by the qth-power map on the coordinates
of points of X, embedded in P 2 .
(b) Show that 1x - F' is a separable morphism and its kernel is just the set
X(Fq) of points of X with coordinates in Fq.
(c) Using (Ex. 4.7), show that F' + F' = ax for some integer a, and that N =
q - a + 1, where N = #X(Fq).
(d) Use the fact that deg(m + nF') > 0 for all m,n E Z to show that ial ~ 2J(j.
This is Hasse's proof of the analogue of the Riemann hypothesis for elliptic
curves (App. C, Ex. 5.6).
(e) Now assume q = p, and show that the Hasse invariant of X is 0 if and only
=
if a 0 (mod p). Conclude for p ~ 5 that X has Hasse invariant 0 if and only
if N = p + 1.

4.17. Let X be the curve y 2 + y = x 3 - x of(4.23.8).

(a) If Q = (a,b) is a point on the curve, compute the coordinates of the point
P + Q, where P = (0,0), as a function of a,b. Use this formula to find the
coordinates ofnP, n = 1,2, ... ,10. [Check: 6P = (6,14).]
(b) This equation defines a nonsingular curve over FP for all p ¥ 37.

339
 IV Curves

4.18. Let X be the curve y 2 = x 3 - 7x + 10. This curve has at least 26 points with
integer coordinates. Find them (use a calculator), and verify that they are all
contained in the subgroup (maybe equal to all of X(Q)?) generated by P = (1,2)
and Q = (2,2).

4.19. Let X,P 0 be an elliptic curve defined over Q, represented as a curve in P 2 de-
fined by an equation with integer coefficients. Then X can be considered as the
fibre over the generic point of a scheme X over Spec Z. Let T ~ Spec Z be the
open subset consisting of all primes p =/= 2 such that the fibre X<v> of X over p
is nonsingular. For any n, show that nx:X-+ X is defined over T, and is a flat
morphism. Show that the kernel of nx is also flat over T. Conclude that for
any p E T, the natural map X(Q) -+ X<v>(F vl induced on the groups of rational
points, maps the n-torsion points of X(Q) injectively into the torsion subgroup of
X<v>(F vl, for any (n,p) = 1.
By this method one can show easily that the groups X(Q) in (Ex. 4.17) and
(Ex. 4.18) are torsion-free.

4.20. Let X be an elliptic curve over a field k of characteristic p > 0, and let R =
End(X,P 0 ) be its ring of endomorphisms.
(a) Let X P be the curve over k defined by changing the k-structure of X (2.4.1).
Showthatj(Xp) =j(X) 11 P. ThusX ~ XvoverkifandonlyifjEFP"
(b) Show that Px in R factors into a product nft of two elements of degree p if and
only if X ~ X v· In this case, the Hasse invariant of X is 0 if and only if nand
ft are associates in R (i.e., differ by a unit). (Use (2.5).)
(c) If Hasse (X)= 0 show in any casej E Fvz·
(d) For any fER, there is an induced map f* :H 1 ((9x)-+ H 1 ((9x). This must be
multiplication by an element A.1 E k. So we obtain a ring homomorphism
q;: R -+ k by sending f to A.1 . Show that any f E R commutes with the
(nonlinear) Frobenius morphism F:X-+ X, and conclude that if Hasse
(X) =/= 0, then the image of cp is Fv. Therefore, R contains a prime ideal p
with R/p ~ F v·

4.21. Let 0 be the ring of integers in a quadratic number field Q(H). Show that
any subring R ~ 0, R =/= Z, is of the form R = Z + f · 0, for a uniquely deter-
mined integer f ;;:. 1. This integer f is called the conductor of the ring R.

*4.22. If X -+ A~ is a family of elliptic curves having a section, show that the family is
trivial. [Hints: Use the section to fix the group structure on the fibres. Show
that the points of order 2 on the fibres form an etale cover of A~, which must be
trivial, since A~ is simply connected. This implies that }, can be defined on the
family, so it gives a map A~ -+ A~ - {0,1 }. Any such map is constant, so A. is
constant, so the family is trivial.]

5 The Canonical Embedding

We return now to the study of curves of arbitrary genus, and we study the
rational map to a projective space determined by the canonical linear
system. For nonhyperelliptic curves of genus g ~ 3, we will see that it is

340
 5 The Canonical Embedding

an embedding, which we call the canonical embedding. Closely related

to this discussion is Clifford's theorem about the dimension of a special
linear system, which we prove below. Using these results, we will say
something about the classification of curves.
Throughout this section, X will denote a curve of genus g over the alge-
braically closed field k. We will consider the canonical linear system IKI.
If g = 0, IKI is empty. If g = 1, IKI = 0, so it determines the constant map
of X to a point. For g ?: 2, however, IKI is an effective linear system without
base points, as we will see, so it determines a morphism to projective space
which we call the canonical morphism.

Lemma 5.1. If g ;::: 2, then the canonical linear system IKI has no base points.

PROOF. According to (3.1), we must show that for each P EX, dimiK - PI =
dimiKI - 1. Now dimiKI = dim H 0 (X,wx) - 1 = g - 1. On the other
hand, since X is not rational, for any point P, dimiPI = 0, so by Riemann-
Roch we find that dimiK - PI = g - 2, as required.

Recall that a curve X of genus g ;::: 2 is called hyperelliptic (Ex. 1.7) if

there is a finite morphism f:X--+ P 1 of degree 2. Considering the corre-
sponding linear system, we see that X is hyperelliptic if and only if it has a
linear system of dimension 1 and degree 2. It is convenient here to introduce
a classical notation. The symbol g:i will stand for "a linear system of dimen-
sion r and degree d." Thus we say X is hyperelliptic if it has a gi.
If X is a curve of genus 2, then the canonical linear system IKI is a gi
(Ex. 1.7). So X is necessarily hyperelliptic, and the canonical morphism
f: X --+ P 1 is the 2-1 map of the definition.
Proposition 5.2. Let X be a curve of genus g ;::: 2. Then IKI is very ample
if and only if X is not hyperelliptic.
PROOF. We use the criterion of (3.1). Since dimiKI = g - 1 we see that
IKI is very ample if and only if for every P,Q EX, possibly equal,
dimiK - P- Ql = g - 3. Applying Riemann-Roch to the divisor P + Q,
we have
dimiP + Ql - dimiK - P- Ql = 2 + 1 -g.
So the question is whether dimiP + Ql = 0. If X is hyperelliptic, then
for any divisor P + Q of the gi we have dimiP + Ql = 1. Conversely, if
dimiP + Ql > 0 for some P,Q, then the linear system IP + Ql contains a
gi (in fact is a gi), so X is hyperelliptic. This completes the proof.

Definition. If X is nonhyperelliptic of genus g ;::: 3, the embedding X --+

P 9 - 1 determined by the canonical linear system is the canonical em-
bedding of X (determined up to an automorphism of P 9 - 1), and its
image, which is a curve of degree 2g - 2, is a canonical curve.

341
IV Curves

Example 5.2.1. If X is a nonhyperelliptic curve of genus 3, then its canonical

embedding is a quartic curve in P 2 . Conversely, any nonsingular quartic
curve X in P 2 has wx ~ @x(1) (II, 8.20.3), so it is a canonical curve. In
particular, there exist nonhyperelliptic curves of genus 3 (see also (Ex. 3.2) ).

Example 5.2.2. If X is a nonhyperelliptic curve of genus 4, then its canonical

embedding is a curve of degree 6 in P 3 . We will show that X is contained
in a unique irreducible quadric surface Q, and that X is the complete inter-
section of Q with an irreducible cubic surface F. Conversely, if X is a non-
singular curve in P 3 which is a complete intersection of a quadric and a
cubic surface, then deg X = 6, and wx = @x(1) (II, Ex. 8.4), so X is a ca-
nonical curve of genus 4. In particular, there exist such nonsingular complete
intersections by Bertini's theorem (II, Ex. 8.4), so there exist nonhyper-
elliptic curves of genus 4.
To prove the above assertions, let X be a canonical curve of genus 4
in P 3 , and let J be its ideal sheaf. Then we have an exact sequence
0 --+ .f --+ @p --+ @X --+ 0.
Twisting by 2 and taking cohomology, we have
0 --+ H 0 (P,J(2)) --+ H 0 (P,@p(2)) --+ H 0 (X,@x(2)) --+ . . .

Now the middle vector space has dimension 10 by (III, 5.1), and the right
hand vector space has dimension 9 by Riemann-Roch on X (note that
@x(2) corresponds to the divisor 2K, which is nonspecial of degree 12).
So we conclude that
dim H 0 (P,.f(2)) ~ 1.
An element of that space is a form of degree 2, whose zero-set will be a
surface Q s; P 3 of degree 2 containing X. It must be irreducible (and
reduced), because X is not contained in any P 2 . The curve X could not
be contained in two distinct irreducible quadric surfaces Q,Q', because then
it would be contained in their intersection Q n Q' which is a curve of degree
4, and that is impossible because deg X = 6. So we see that X is contained
in a unique irreducible quadric surface Q.
Twisting the same sequence by 3 and taking cohomology, a similar
calculation shows that
dim H 0 (P,J(3)) ~ 5.
The cubic forms in here consisting of the quadratic form above times a
linear form, form a subspace of dimension 4. Hence there is an irreducible
cubic form in that space, so X is contained in an irreducible cubic surface
F. Then X must be contained in the complete intersection Q n F, and
since both have degree 6, X is equal to that complete intersection.

Proposition 5.3. Let X be a hyperelliptic curve of genus g ~ 2. Then X has

a unique g~. If f 0 :X--+ P 1 is the corresponding morphism of degree 2,
then the canonical morphism f: X --+ P9 - 1 consists of fo followed by the
342
 5 The Canonical Embedding

(g - 1)-uple embedding of P 1 in P 9 - 1 . In particular, the image X' =

f(X) is a rational normal curve of degree g - 1 (Ex. 3.4), and f is a mor-
phism of degree 2 onto X'. Finally, every effective canonical divisor on X
is a sum of g- 1 divisors in the unique gL so we write IKI = I1-
1 g~.

PROOF. We begin by considering the canonical morphism f: X --+ P 9 - \

and let X' be its image. Since X is hyperelliptic, it has a gL by definition.
We don't yet know that it is unique, so fix one for the moment. For any
divisor P + Q E gL the proof of (5.2) shows that Q is a base point of IK - PI,
so f(P) = f(Q). Since the g~ has infinitely many divisors in it, we see that
f cannot be birational. So let the degree of the map f: X --+ X' be J1 ~ 2,
and let d = deg X'. Then since deg K = 2g - 2, we have dJ1 = 2g - 2,
hence d ~ g - 1.
Next, let X' be the normalization of X', and let b be the linear system
on X' corresponding to the morphism X' --+ X' s; P 9 - 1 . Then b is a linear
system of degree d and dimension g - 1. Since d ~ g - 1, we conclude
(Ex. 1.5) that d = g - 1, the genus of X' is 0, so X' ~ P\ and the linear
system b is the unique complete linear system on P 1 of degree g - 1, namely
l(g - 1) ·Pl. Therefore, X' is the (g - 1)-uple embedding of P 1 . In par-
ticular, it is nonsingular, and it is a rational normal curve in the sense of
(Ex. 3.4).
Next, from dJ1 = 2g - 2, we conclude that J1 = 2. Since f already
collapses the pairs of the g~ we chose above, it must be equal to the com-
position of the map fo: X --+ P 1 determined by our g~ with the (g - 1)-uple
embedding of P 1 . Thus the g~ is determined by f, and so is uniquely
determined.
Finally, any effective canonical divisor K on X is f- 1 of a hyperplane
section of X'. Hence it is a sum of g - 1 divisors in the unique g~. Con-
versely, any set of g - 1 points of X' is a hyperplane section, so we can
identify the canonical linear system IK I with the set of sums of g - 1 divisors
of the g~. Hence we write
g-1
IKI = I g~.
1

Now we come to Clifford's theorem. The idea is this. For a nonspecial

divisor D on a curve X, we can compute dimiDI exactly as a function of
deg D by the Riemann-Roch theorem. However, for a special divisor,
dimiDI does not depend only on the degree. Hence it is useful to have some
bound on dimiDI, and this is provided by Clifford's theorem.

Theorem 5.4 (Clifford). Let D be an effective special divisor on the curve X.

Then

dimiDI ~ ~ deg D.
Furthermore, equality occurs if and only if either D = 0 or D = K or X
is hyperelliptic and D is a multiple of the unique g~ on X.

343
IV Curves

Lemma 5.5. Let D,E be effective divisors on a curve X. Then

dimiDI + dimiEI ~ dimiD + El.
PROOF. We define a map of sets
cp: IDI X lEI --+ ID + El
by sending (D',E') to D' + E' for any D' E IDI and E' E lEI. The map cp is
finite-to-one, because a given effective divisor can be written in only finitely
many ways as a sum of two other effective divisors. On the other hand,
since cp corresponds to the natural bilinear map of vector spaces
H 0 (X,!l'(D)) x H 0 (X,!l'(E)) --+ H 0 (X,!l'(D + E)),
we see that cp is a morphism when we endow IDI,IEI and ID + El with their
structure of projective spaces. Therefore, since cp is finite-to-one, the di-
mension of its image is exactly dimiDI + dimiEI, and from this the result
follows.
PROOF OF (5.4) (following Saint-Donat [1, §1]). If Dis effective and special,
then K - Dis also effective, so we can apply the lemma, and obtain
dimiDI + dimiK - Dl ~ dimiKI = g - 1.
On the other hand, by Riemann~Roch we have
dimiDI - dimiK - Dl = deg D + 1 - g.
Adding these two expressions, we have
2 dimiDI ~ deg D,
or in other words

dimiDI ~ ~ deg D.
This gives the first statement of the theorem. Also, it is clear that we have
equality in caseD = 0 or D = K.
For the second statement, suppose that D =f. O,K, and that dimiDI =
t deg D. Then we must show that X is hyperelliptic, and that D is a multiple
of the g~. We proceed by induction on deg D (which must be even). If
deg D = 2, then IDI itself is a gL so X is hyperelliptic and there is nothing
more to prove.
So suppose now that deg D ;;;:: 4, hence dimiDI ;;;:: 2. Fix a divisor E E
IK - Dl, and fix two points P,Q EX such that P E Supp E and Q ¢ Supp E.
Since dimiDI ;;;:: 2, we can find a divisor D E IDI such that P,Q E Supp D.
Now let D' = D n E, by which we mean the largest divisor dominated by
both D and E. This D' will accomplish our induction.
First note that since Q E Supp D but Q ¢ Supp E, we must have Q ¢
Supp D', so deg D' < deg D. On the other hand, deg D' > 0 since P E
Supp D'.

344
 5 The Canonical Embedding

Next, by construction of D', we have an exact sequence

0 --+ !E(D') --+ !E(D) (f) !E(E) --+ !E(D + E - D') --+ 0,
where we consider these as subspaces of the constant sheaf :£ on X, and
the first map is addition, the second subtraction. (Think of !E(D) =
{! E K(X)j(f) ;;::: -D}, cf. (II, 7.7).) Therefore, considering global sections
of this sequence, we have
dimjDI + dimjEj ~ dimjD'I + dimjD + E - D'j.
But E"' K - D and D +E-D' "' K - D', so the left-hand side is equal to
dimjDj + dimjK - Dj,
which must be = dimjKj = g - 1 since we have dimjDI = ! deg D by hy-
pothesis. On the other hand, the right-hand side is ~g - 1 by (5.5) applied
to D', so we must have equality everywhere. We conclude, then, that
dimjD'I =! deg D', as above. Now by the induction hypothesis, this implies
that X is hyperelliptic.
Now suppose again that D =F O,K, and dimiDI = ! deg D. Let r = dimjDj.
Consider the linear system IDI + (g - 1 - r)gi. It has degree 2g - 2, and
dimension ;;::: g - 1, by (5.5) again, so it must be equal to the canonical
system jKj. But we have already seen (5.3) that jKj = (g - 1)gi. So we
conclude that IDI = rgi, which completes the proof.

Classification of Curves
To classify curves, we first specify the genus, which as we have seen (1.1.1)
can be any nonnegative integer g ;;::: 0. If g = 0, X is isomorphic to P 1
(1.3.5), so there is nothing further to say. If g = 1, then X is classified up
to isomorphism by its j- invariant (4.1 ), so here again we have a good answer
to the classification problem. For g ;;::: 2, the problem becomes much more
difficult, and except for a few special cases (e.g., (Ex. 2.2) ), one cannot give
an explicit answer.
For g ;;::: 3 we can subdivide the set mg of all curves of genus g according
to whether the curve admits linear systems of certain degrees and dimen-
sions. For example, we have defined X to be hyperelliptic if it has a gi,
and we have seen that there are hyperelliptic curves of every genus g ;;::: 2
(Ex. 1. 7), and at least for g = 3 and 4, that there exist nonhyperelliptic curves
(5.2.1) and (5.2.2).
More generally, we can subdivide curves according to whether they have
a gJ for various d. If X has a g~ it is called trigonal.

Remark 5.5.1. The facts here are as follows. For any d ;;::: !g + 1, any
curve of genus g has a gJ; for d < !g + 1, there exist curves of genus g
having no gJ. See Kleiman and Laksov [1] for proofs and discussion. Note
in particular this implies that there exist nonhyperelliptic curves of every
genus g ;;::: 3 (V, Ex. 2.10). We give some examples of this result.

345
IV Curves

Example 5.5.2. For g = 3,4 this result states that there exist nonhyperelliptic
curves (which we have seen) and that every such curve has a g~. Of course
if X is hyperelliptic, this is trivial, by adding a point to the g~. If X is non-
hyperelliptic of genus 3, then its canonical embedding is a plane quartic
curve (5.2.1). Projecting from any point of X to Pi, we get a g~. Thus X
has infinitely many g~'s.
If X is nonhyperelliptic of genus 4, then its canonical embedding in P 3
lies on a unique irreducible quadric surface Q (5.2.2). If Q is nonsingular,
then X has type (3,3) on Q (II, 6.6.1), and each of the two families of lines
on Q cuts out a g~ on X. So in this case X has two g~'s (to see that these
are the only ones, copy the argument of (5.5.3) below). If Q is singular, it is
a quadric cone, and the one family of lines on Q cuts out a unique g~ on X.

Example 5.5.3. Let g = 5. Then (5.5.1) says that every curve of genus 5 has
a gi, and that there exist such curves with no g~. Let X be a nonhyperelliptic
curve of genus 5, in its canonical embedding as a curve of degree 8 in P 4 .
First we show that X has a g~ if and only if it has a trisecant in this em-
bedding. Let P,Q,R EX. Then by Riemann-Roch, we have
dimiP + Q + Rl = dimiK - P - Q - Rl - 1.
On the other hand, since X is in its canonical embedding, dimiK - P- Q - Rl
is the dimension of the linear system of hyperplanes in P 4 which contain
P,Q,R. Hence dimiP + Q + Rl = 1 if and only if P,Q,R are contained in
a 2-dimensional family of hyperplanes, which is equivalent to saying that
P,Q and R are collinear. Thus X has a g~ if and only if it has a trisecant
(and in that case it will have a 1-parameter family oftrisecants).
Now let X be a nonsingular complete intersection of three quadric hyper-
surfaces in P 4 . Then deg X = 8, and wx ~ CDx(1), so X is a canonical curve
of genus 5. If X had a trisecant L, then L would meet each of the quadric
hypersurfaces in three points, so it would have to be contained in these
hypersurfaces, and so L <:; X, which is impossible. So we see that there
exist curves of genus 5 containing no g~.
Now projecting this X from one of its own points P to P 3 , we obtain a
curve X' <:; P 3 of degree 7, which is nonsingular (because X had no tri-
secants). This new curve X' must have trisecants, because otherwise a
projection from one of its points would give a nonsingular curve of degree 6
in P 2 , which has the wrong genus. So let Q,R,S lie on a trisecant of X'.
Then their inverse images on X, together with P, form four points which
lie in a plane of P 4 . Then the same argument as above shows these points
give a g}_.
Coming back to the general classification question, for fixed gone would
like to endow the set 9Jl9 of all curves of genus g up to isomorphism with
an algebraic structure, in which case we call 9Jl9 the variety of moduli of
curves of genus g. Such is the case for g = 1, where the j-invariants form
an affine line.

346
 5 The Canonical Embedding

The best way to specify the algebraic structure on IDlg would be to require
it to be a universal parameter variety for families of curves of genus g, in
the following sense: we require that there be a flat family X --+ IDlg of curves
of genus g such that for any other flat family X --+ T of curves of genus g,
there is a unique morphism T--+ IDlg such that X is the pullback of X. In
this case we call IDlg a fine moduli variety. Unfortunately, there are several
reasons why such a universal family cannot exist. One is that there are
nontrivial families of curves, all of whose fibres are isomorphic to each
other (III, Ex. 9.10).
However, Mumford has shown that for g ~ 2 there is a coarse moduli
variety IDlg, which has the following properties (Mumford [1, Th. 5.11]):
(1) the set of closed points of IDlg is in one-to-one correspondence with the
set of isomorphism classes of curves of genus g;
(2) if f:X--+ Tis any flat family of curves of genus g, then there is a mor-
phism h: T--+ IDlg such that for each closed point t E T, X 1 is in the iso-
morphism class of curves determined by the point h(t) E IDlg.
In case g = 1, the affine j-line is a coarse variety of moduli for families
of elliptic curves with a section. One verifies condition (2) using the fact
that j is a rational function of the coefficients of a plane embedding of the
curve (Ex. 4.4).

Remark 5.5.4. In fact, Deligne and Mumford [1] have shown that IDlg for
g ~ 2 is an irreducible quasi-projective variety of dimension 3g - 3 over
any fixed algebraically closed field.

Example 5.5.5. Assuming that IDlg exists, we can discover some of its prop-
erties. For example, using the method of (Ex. 2.2), one can show that hy-
perelliptic curves of genus g are determined as two-fold coverings of P 1 ,
ramified at 0,1, oo, and 2g - 1 additional points, up to the action of a
certain finite group. Thus we see that the hyperelliptic curves correspond
to an irreducible subvariety of dimension 2g - 1 of IDlg. If g = 2, this is
the whole space, which confirms that 9Jl 2 is irreducible of dimension 3.

Example 5.5.6. Let g = 3. Then the hyperelliptic curves form an irreducible

subvariety of dimension 5 of 9Jl 3 . The nonhyperelliptic curves of genus 3
are the nonsingular plane quartic curves. Since the embedding is canonical,
two of them are isomorphic as abstract curves if and only if they differ by
an automorphism of P 2 . The family of all these curves is parametrized by
an open set U s:; pN with N = 14, because a form of degree 4 has 15 coeffi-
cients. So there is a morphism U --+ 9Jl 3 , whose fibres are images of the
group PGL(2) which has dimension 8. Since any individual curve has only
finitely many automorphisms (Ex. 5.2), the fibres have dimension = 8, and
so the image of U has dimension 14 - 8 = 6. So we confirm that 9Jl 3 has
dimension 6.

347
IV Curves

EXERCISES

5.1. Show that a hyperelliptic curve can never be a complete intersection in any pro-
jective space. Cf. (Ex. 3.3).
5.2. If X is a curve of genus ~ 2 over a field of characteristic 0, show that the group
Aut X ofautomorphisms of X is finite. [Hint: If X is hyperelliptic, use the unique
g1 and show that Aut X permutes the ramification points of the 2-fold covering
X ~ P 1 . If X is not hyperelliptic, show that Aut X permutes the hyperosculation
points (Ex. 4.6) of the canonical embedding. Cf. (Ex. 2.5).]
5.3. Moduli of Curves of Genus 4. The hyperelliptic curves of genus 4 form an irre-
ducible family of dimension 7. The nonhyperelliptic ones form an irreducible
family of dimension 9. The subset of those having only one g~ is an irreducible
family of dimension 8. [Hint: Use (5.2.2) to count how many complete inter-
sections Q n F 3 there are.]
5.4. Another way of distinguishing curves of genus g is to ask, what is the least degree
of a birational plane model with only nodes as singularities (3.11)? .Let X be
nonhyperelliptic of genus 4. Then:
(a) if X has two grs, it can be represented as a plane quintic with two nodes, and
conversely;
(b) if X has one gt then it can be represented as a plane quintic with a tacnode
(1, Ex. 5.14d), but the least degree of a plane representation with only nodes is 6.

5.5. Curves of Genus 5. Assume X is not hyperelliptic.

(a) The curves of genus 5 whose canonical model in P4 is a complete intersection
F 2 .F 2 .F 2 form a family of dimension 12.
(b) X has a g~ if and only if it can be represented as a plane quintic with one node.
These form an irreducible family of dimension 11. [Hint: If D E g ~, use K - D
to map X ~ P 2 .]
*(c) In that case, the conics through the node cut out the canonical system (not
counting the fixed points at the node). Mapping P 2 -+ P4 by this linear system
of conics, show that the canonical curve X is contained in a cubic surface
V £ P4 , with Visomorphic to P 2 with one point blown up (II, Ex. 7.7). Further-
more, Vis the union of all the trisecants of X corresponding to the g~ (5.5.3),
so V is contained in the intersection of all the quadric hypersurfaces containing
X. Thus Vand the g~ are unique.
Note. Conversely, if X does not have a gL then its canonical embedding is a
complete intersection, as in (a). More generally, a classical theorem of Enriques
and Petri shows that for any nonhyperelliptic curve of genus g ~ 3, the canonical
model is projectively normal, and it is an intersection of quadric hypersurfaces
unless X has a g ~ or g = 6 and X has a g;. See Saint-Donat [ l].
5.6. Show that a nonsingular plane curve of degree 5 has no g~. Show thatthere are
nonhyperelliptic curves of genus 6 which cannot be represented as a nonsingular
plane quintic curve.
5.7. (a) Any automorphism of a curve of genus 3 is induced by an automorphism ofP 2
via the canonical embedding.
*(b) Assume char k -# 3. If X is the curve given by

x3y + y3z + z3 x = 0,
348
 6 Classification of Curves in P 3

the group Aut X is the simple group of order 168, whose order is the maximum
84(g - 1) allowed by (Ex. 2.5). See Burnside [1, §232] or Klein [1].
*(c) Most curves of genus 3 have no automorphisms except the identity. [Hint:
For each n, count the dimension of the family of curves with an automorphism
T of order n. For example, if n = 2, then for suitable choice of coordinates,
T can be written as x--> -x, y--> y, z--> z. Then there is an 8-dimensional
family of curves fixed by T; changing coordinates there is a 4-dimensional
family of such T, so the curves having an automorphism of degree 2 form a
family of dimensional 12 inside the 14-dimensional family of all plane curves
of degree 4.]
Note: More generally it is true (at least over C) that for any g ;;:, 3, a "sufficiently
general" curve of genus g has no automorphisms except the identity-see Baily [ 1].

6 Classification of Curves in P 3

In 1882, a cash prize (the Steiner prize) was offered for the best work on
the classification of space curves. It was shared by Max Noether and
G. Halphen, each of whom wrote a 200-page treatise on the subject (Noether
[ 1], Halphen [ 1] ). They each proved a number of general results, and then
to illustrate their theory, constructed exhaustive tables of curves of low
degree (up to about degree 20).
Nowadays the theoretical aspect of this problem is well understood.
Using either the Chow variety or the Hilbert scheme, one can show that
the nonsingular curves of given degree d and genus g in P 3 are parametrized
by a finite union of quasi-projective varieties, in a very natural way. How-
ever, the more specific task of determining the number and dimensions of
these parameter varieties for each d,g is not solved. It is not even clear
exactly for which pairs of integers d,g there exists a curve of degree d and
genus g in P 3 . Halphen stated the result, but a correct proof was given only
recently by Gruson and Peskine.
In this section we will give a few basic results concerning curves in P 3 ,
and then illustrate them by classifying all curves of degree ~ 7 in P 3 .
We begin by investigating when a curve has a nonspecial very ample
divisor of a given degree. In the case of g = 0,1, this is answered by (3.3.1)
and (3.3.3), so we will consider the case g ? 2.

Proposition 6.1 (Halphen). A curve X of genus g ? 2 has a nonspecial very

ample divisor D of degree d if and only if d ? g + 3.
PROOF. First we show the necessity of the condition. If D is nonspecial and
very ample of degree d, then by Riemann-Roch, we have dimiDI = d - g
and IDI gives an embedding of X in pd- 9 . Since X ;f. P\ we must have
d - g ? 2, i.e., d ? g + 2. But if d = g + 2, then X is a plane curve of
degree d. In this case wx ~ @x(d - 3), so in order for D to be nonspecial
we must have d ~ 3. But then g = 0 or 1, contrary to hypothesis. So we
conclude that d ? g + 3.
349
IV Curves

So now we fix d ~ g + 3, and we search for a nonspecial very ample

divisor D of degree d. In order forD to be very ample, by (3.1) it is necessary
and sufficient that for all P,Q E X, we have
dimiD - P - Ql = dimiDI - 2.
Since D is nonspecial, this is equivalent, by Riemann-Roch, to saying that
D - P - Q is also nonspecial. Replacing D by a linearly equivalent divisor
D', we may always assume that D' - P - Q is effective.
Now consider Xd, the product of X with itself d times. We associate an
element (P 1, . . . ,Pd) E Xd and all its permutations, with the effective divisor
D = pl + 0 0 + Pd, and then by abuse of notation write DE xd. We will
0

show that the set S of divisors D E Xd such that there exists D' ~ D and
there exist points P,Q EX with E = D' - P - Q an effective special divisor,
has dimension ~ g + 2. Since d ~ g + 3, this shows that S i= Xd. Then
any D ¢ S will be a nonspecial very ample divisor of degree d.
Let E be an effective special divisor of degree d - 2. Since dimiKI =
g - 1, and since an effective special divisor is a subset of an effective ca-
nonical divisor, we see that the set of all such E, as a subset of xd-z, has
dimension ~ g - 1. Thus the set of divisors of the form E + P + Q in Xd
has dimension ~ g + 1. Since the special divisors in Xd form a subset of
dimension ~ g - 1, for the same reason, we may ignore them, so we may
assume E + P + Q is nonspecial.
Since E is special, we have dimiEI ~ d - 1 - g, by Riemann-Roch. On
the other hand, since E + P + Q is nonspecial, we have dimiE + P + Ql =
d - g. The difference between these two is 1, so we see that the set of
D E Xd which are linearly equivalent to some divisor of the form E + P + Q
has dimension ~g + 2, as required.

Corollary 6.2. There exists a curve X of degree d and genus g in P 3 , whose

hyperplane section D is nonspecial, if and only if either
(1) g = 0 and d ~ 1,
(2) g = 1 and d ~ 3, or
(3) g ~ 2 and d ~ g + 3.
PROOF. This follows immediately from (3.3.1), (3.3.3), and the proposition.
Given D very ample on X, the complete linear system IDI gives an em-
bedding of X into pn for some n, and if n > 3 we project down to P 3 , using
(3.5).

Proposition 6.3. If X is a curve in P 3 , not lying in any plane, for which the
hyperplane section Dis special, then d ~ 6 and g ~ td + 1. Furthermore,
the only such curve with d = 6 is the canonical curve of genus 4 (5.2.2).
PROOF. If D is special, then by Clifford's theorem (5.4) we have dimiDI ~
td. Since X is not in any plane, dimiDI ~ 3, so d ~ 6. And since D is

350
 6 Classification of Curves in P 3

special, d :(; 2g - 2, hence g ?: !d + 1. Now if d = 6, then we have

equality in Clifford's theorem, so either D = 0 (which is absurd) or D = K,
in which case X is the canonical curve of genus 4, or X is hyperelliptic and
IDI is a multiple of the unique g~. But this last case is impossible, because
then IDI would not separate points, so could not be very ample.

Next, we have a result which bounds the genus of a space curve of given
degree.

!
Theorem 6.4 (Castelnuovo [1]). Let X be a curve of degree d and genus gin
P 3 , which is not contained in any plane. Then d ?: 3, and
1 2
-d -d+1 if dis even
4
g=(;
1 2
4(d -1)-d+1 if dis odd.

Furthermore, the equality is attained for every d ?: 3, and any curve for
which equality holds lies on a quadric surface.
PROOF. Given X, let D be its hyperplane section. The idea of the proof is to
estimate dimlnDI - diml(n - 1)DI for any n, and then add. First of all, we
choose the hyperplane section D = P 1 + ... + Pd in such a way that no
three of the points Pi are collinear. This is possible because not every secant
of X is a multisecant (3.8), (3.9), (Ex. 3.9).
Now I claim for each i = 1,2, ... ,min(d,2n + 1), that Pi is not a base
point of the linear system lnD - P 1 - . . . . - Pi_ 1 1. To show this, it is suffi-
cient to find a surface of degree n in P 3 containing P 1 , .•. ,Pi_ 1 , but not Pi. In
fact, a union of n planes will do. We take the first plane to contain P 1 and P 2 ,
but no other Pi, which is possible, since no three Pi are collinear. We take
the second plane to contain P 3 and P 4 , and so on, until our planes contain
P 1 , . . . ,Pi_ 1 , and take the remaining planes to miss all the Pi. This is possible
for any i such that i - 1 :(; 2n, and of course i :(; d since there are only d
points.
It follows that for any n ?: 1, we have
dimlnDI - diml(n - 1)DI ?: min(d,2n + 1),
because (n - 1)D = nD - P 1 - . . . - Pd, and each time we remove a non-
base point from a linear system, the dimension drops by 1.
Now we take n » 0, and add these expressions together, starting with
n = 1, up to the given n, using the fact that dimiO · Dl = 0. If we let r =
[!(d - 1)], then we can write the answer as
dimlnDI ?: 3 + 5 + ... + (2r + 1) + (n - r)d
or
dimlnDI ?: r(r + 2) + (n - r)d.

351
IV Curves

On the other hand, for large n, the divisor nD will be nonspecial, so by

Riemann-Roch we have
dimlnDI = nd - g.
Combining, we find that
g ::::; rd - r(r + 2).
To interpret this, we consider two cases. If dis even, then r = !d - 1, and
we get
1 2
g::::;4d -d+l.

If dis odd, then r = !(d - 1), and we get

1 2
g:::;;4(d -1)-d+1,

which is the bound of the theorem.

If X is a curve for which equality holds, then we must have had equality at
every step of the way. In particular, we see that dimi2DI = 8 (or even less if
d < 5), from which it follows that X is contained in a quadric surface. Indeed,
from the exact sequence

we obtain
0 --+ H 0 (.J" x(2)) --+ H 0 (mp(2)) --+ H 0 (mx(2)) --+ ...

Since dim H 0 ((9p(2)) = 10 and dim H 0 ((9x(2)) = 9 (or less, by the above),
we conclude that H 0 (.J"x(2)) i= 0, hence X is contained in a quadric surface.
Finally, to show that equality is achieved, we look at certain curves on a
nonsingular quadric surface Q. If d is even, d = 2s, we take a curve of type
(s,s), which has degree d and genus s2 - 2s + 1 = ±d2 - d + 1, by (III,
Ex. 5.6). This curve is a complete intersection of Q with a surface of
degree s. If d is odd, d = 2s + 1, we take a curve of type (s,s + 1) on Q,
which has degreed and genus s 2 - s = t(d 2 - 1) - d + 1.

Remark 6.4.1. Let us gather together everything we know about curves in P 3 .

First, we recall various classes of curves which we know to exist.
(a) For every d ~ 1, there are nonsingular plane curves of degree d, and
they have g = !(d - 1)(d - 2). See (II, 8.20.2) and (II, Ex. 8.4).
(b) For every a,b ~ 1, there are complete intersections of surfaces of
degrees a,b in P 3 which are nonsingular curves. They have degree d = ab
and genus g = !ab(a + b - 4) + 1 (II, Ex. 8.4).
(c) For every a,b ~ 1, there are nonsingular curves of type (a,b) on a non-
singular quadric surface. They have degree d = a + b and genus g =
ab - a - b + 1 (III, Ex. 5.6).
(d) We will see later (V, Ex. 2.9) that if X is a curve on a quadric cone Q,
there are two cases. If d is even, d = 2a, then X is a complete intersection

352
 6 Classification of Curves in P 3

of Q with a surface of degree a, so g = a2 - 2a + 1. If dis odd, d = 2a + 1,

then X has genus g = a 2 - a. Comparing with (c) above, we note that these
values of d and g are among those possible on a nonsingular quadric surface.

Example 6.4.2. Now we can classify curves of degree d ~ 7 in P 3 .

d = 1. The only curve with d = 1 is P 1 (1, Ex. 7.6).
d = 2. The only curve of degree 2 is the conic in P 2 (1, Ex. 7.8).
d = 3. Here we have the plane cubic with g = 1, and the twisted cubic curve
in P 3 with g = 0. That is all, by (Ex. 3.4).
d = 4. The plane quartic has g = 3; in P 3 there are rational quartic curves,
and elliptic quartic curves, the latter being the complete intersection
of two quadric surfaces (Ex. 3.6).
d = 5. The plane quintic has g = 6. In P 3 , there are curves with (1)(1) non-
special of genus 0,1,2, by (6.2), and these are all by (6.3).
d = 6. The plane sextic has g = 10. In P 3 there are curves with (1)(1)non-
special of genus 0,1,2,3 by (6.2), and a curve of genus 4 which is the
canonical curve of genus 4, a complete intersection of a quadric and
a cubic surface (6.3).
d = 7. The plane septic has g = 15. There are curves in P 3 with (1)(1) non-
special of genus 0,1,2,3,4, and any curve of g ~ 4 must be nonspecial.
On the other hand, there is a curve of type (3,4) on a nonsingular
quadric surface, which has g = 6. This is the maximum genus for
this degree, by (6.4), so any curve of g = 6 must lie on a quadric.
We are left with the question, does there exist a curve of degree 7 with
g = 5? By (6.4.1) there is no such curve on a quadric surface. So we approach
the question from a different angle. Given an abstract curve X of genus 5,
can we embed it as a curve of degree 7 in P 3 ? We need a very ample divisor D
of degree 7, with dimiDI ~ 3. By Riemann-Roch, such a divisor must be
special. Since deg K = 8, we can write D = K - P, and so we see dimiDI = 3.
In order for D to be very ample, we must have
dimiK - P - Q - Rl = dimiK - PI - 2
for all Q,R in X. Using Riemann-Roch again, this says that
dimiP + Q + Rl = 0
for all Q,R. But this is possible if and only if X does not have a g~. Indeed,
if X has no gL then dimiP + Q + Rl = 0 for all P,Q,R. On the other hand,
if X does have a gL then for any given P, there exist Q,R such that
dimiP + Q + Ri = 1.
Summing up, we see that the abstract curve X of genus 5 admits an em-
bedding of degree 7 in P 3 if and only if X has no g~. Now from (5.5.3) we
know there are such curves, so we see that curves of degree 7 and genus 5 do
exist in P 3 . This example should give some idea of the complexities which
compound themselves when trying to classify curves of higher degree and
genus in P 3 .

353
IV Curves

Figure 18 summarizes what we know about the existence of curves of

degree d and genus gin P 3 , ford :( 10 and g :( 12, using the results of this
section. See (V, 4.13.1) and (V, Ex. 4.14) for further information.

12 I ~I
I
I
11 -e~lsts 1 ~I I

10 r----- !---=does not ~-~

9
exist
1

=don't ~r
r:~I
I
.. /
8
b"
/
know yet 11 "'1 1 ...... ~
g 7 I -o)/ "''' /
/

6 0~.----- 'b
~I /tlo./ /
/

5 s-~- of't /
~

0 I /

4
.__

/ /
/ /
(:-0 /
3 ~0I/ ///
/

2 ,;,.Q" /'/
. . :::1.-,...v d
0 1 2 3 4 5 6 7 8 9 10
Figure 18. Curves of degree d and genus gin P 3 .

Example 6.4.3. As another example, we consider curves X of degree 9 and

genus 10 in P 3 . This is the first case where there are two distinct families of
curves of the same degree and genus, neither being a special case of the other.
Type 1 is a complete intersection of two cubic surfaces. In this case,
Wx ~ mx(2) (II, Ex. 8.4), so (?)x(2) is special, and dim H 0 ((1)x(2)) = 10.
Furthermore, X is projectively normal (II, Ex. 8.4) so the natural map
H 0 ((1)p(2)) --+ H 0 ((1)x(2)) is surjective. Since dim H 0 ((1)p(2)) = 10, we con-
clude that H 0 (f x(2)) = 0, so X is not contained in any quadric surface.
Type 2 is a curve of type (3,6) on a nonsingular quadric surface Q. In
this case, using the calculations of cohomology of (III, Ex. 5.6), from the
exact sequence
0 --+ (I)Q(- 3,- 6) --+ (I)Q --+ (I) X --+ 0,
twisting by 2, and taking cohomology, we find that dim H 0 ((1)x(2)) = 9.
Thus (1)(2) is nonspecial. On the other hand, since X cannot be contained
in two distinct quadric surfaces, we have dim H 0 (f x(2)) = 1.
Since the dimension of cohomology groups can only increase under
specialization, by semicontinuity (III, 12.8), we see that neither of these
types can be a specialization of the other. Indeed, dim H 0 (f x(2)) increases
from type 1 to type 2, whereas dim H 0 ((1)x(2)) decreases.
To complete the picture, we show that any curve of degree 9 and genus 10
is one of the two types above. If (1)(2) is nonspecial, then dim H 0 ((1)(2)) = 9,
so X must be contained in a quadric surface Q. Checking the possibilities
ford and g (6.4.1), we see that Q must be nonsingular, and X must be of type
(3,6) on Q. On the other hand, if (1)(2) is special, then X cannot lie on a quadric
surface (because if it did, then it would have to be type 2, in which case (1)(2)
is nonspecial). Since (1)(3) is nonspecial (its degree is > 2g - 2), we have

354
 6 Classification of Curves in P 3

dim H 0 (0x(3)) = 18, so we find dim H 0 (J x(3)) ~ 2. The corresponding

cubic surfaces must be irreducible, so X is contained in the intersection of
two cubic surfaces; then by reason of its degree, X is equal to the complete
intersection of those two cubic surfaces, so X is type 1.

EXERCISES

6.1. A rational curve of degree 4 in P 3 is contained in a unique quadric surface Q, and

Q is necessarily nonsingular.
6.2. A rational curve of degree 5 in P 3 is always contained in a cubic surface, but there
are such curves which are not contained in any quadric surface.
6.3. A curve of degree 5 and genus 2 in P 3 is contained in a unique quadric surface Q.
Show that for any abstract curve X of genus 2, there exist em beddings of degree 5
in P 3 for which Q is nonsingular, and there exist other em beddings of degree 5 for
which Q is singular.
6.4. There is no curve of degree 9 and genus 11 in P 3 . [Hint: Show that it would have
to lie on a quadric surface, then use (6.4.1).]
6.5. If X is a complete intersection of surfaces of degrees a,b in P 3 , then X does not lie
on any surface of degree <min(a,b).
6.6. Let X be a projectively normal curve in P 3 , not contained in any plane. If d = 6,
then g = 3 or 4. If d = 7, then g = 5 or 6. Cf. (II, Ex. 8.4) and (III, Ex. 5.6).
6.7. The line, the conic, the twisted cubic curve and the elliptic quartic curve in P 3
have no multisecants. Every other curve in P 3 has infinitely many multisecants.
[Hint: Consider a projection from a point of the curve to P 2 .]
6.8. A curve X of genus g has a nonspecial divisor D of degree d such that JDJ has no
base points if and only if d ::;:: g + 1.
*6.9. Let X be an irreducible nonsingular curve in P'. Then for each m » 0, there is a
nonsingular surface F of degree m containing X. [Hint: Let 7T : P~ P3 be the
blowing-up of X and let Y = n- 1(X). Apply Bertini's theorem to the projective
embedding ofP corresponding to Jy ® n*ep3(m).]

355
CHAPTER V

Surfaces

In this chapter we give an introduction to the study of algebraic surfaces.

This includes the basic facts about the geometry on a surface, and about
birational transformations of surfaces. Also we treat two special classes of
surfaces, the ruled surfaces, and the nonsingular cubic surfaces in P 3 , both
to illustrate the general theory, and as a first step in the more detailed study
of various types of surfaces.
This chapter should be adequate preparation for reading some more
advanced works, such as Mumford [2], Zariski [5], Shafarevich [1], Bom-
bieri and Husemoller [1]. We have mentioned the classification of surfaces
only very briefly in §6, since it is adequately treated elsewhere.
Sections 1, 3 and 5 are general. Here we develop intersection theory on a
surface and prove the Riemann-Roch theorem. As applications we give the
Hodge index theorem and the Nakai-Moishezon criterion for an ample
divisor. In §3 we study the behavior of a surface and the curves on it under a
single monoidal transformation, which is blowing up a point. Then in §5
we prove the theorem of factorization of a birational morphism into monoidal
transformations, and prove Castelnuovo's criterion for contracting an
exceptional curve of the first kind.
In §2 we discuss ruled surfaces. Here the theory of curves gives a good
handle on the ruled surfaces, because many properties of the surface are
closely related to the study of certain linear systems on the base curve. Also
there is a close connection between ruled surfaces over a curve C and locally
free sheaves of rank 2 on C, so as a byproduct, we get some information about
the classification of these locally free sheaves on a curve.
In §4, we study the nonsingular cubic surfaces in P 3 , and the famous 27lines
which lie on those surfaces. By representing the surface as a P 2 with 6
points blown up, the study of linear systems on the cubic surface is reduced

356
 1 Geometry on a Surface

to the study of certain linear system of plane curves with assigned base
points. This is a very classical subject, about which whole books have been
written, and which we rewrite here in modern language.

1 Geometry on a Surface

We begin our study of surfaces with the internal geometry of a surface. A

divisor on a surface is a sum of curves, so (in the absence of a projective
embedding) it does not make sense to talk about the degree of a divisor, as
in the case of curves. However, we can talk about the intersection of two
divisors on a surface, and this gives rise to intersection theory. The Riemann-
Roch theorem for surfaces gives a connection between the dimension of a
complete linear system jDj, which is essentially a cohomological invariant,
and certain intersection numbers on the surface. As in the case of curves,
the Riemann-Roch theorem is basic to all further work with surfaces, espe-
cially questions of classification.
Throughout this chapter, a surface will mean a nonsingular projective
surface over an algebraically closed field k. It is true that any complete
nonsingular surface is projective (cf. II, 4.10.2), but since we will not prove
that, we assume that our surfaces are projective. A curve on a surface will
mean any effective divisor on the surface. In particular, it may be singular,
reducible or even have multiple components. A point will mean a closed
point, unless otherwise specified.

Let X be a surface. We wish to define the intersection number C.D for

any two divisors C,D on X in such a way as to generalize the intersection
multiplicity defined in (1, Ex. 5.4) and (1, §7). If C and D are curves on X, and
if P E C n D is a point of intersection of C and D, we say that C and D meet
transversally at P if the local equations f,g of C,D at P generate the maximal
ideal mp of (!)P.x· This implies, by the way, that C and Dare each nonsingular
at P, because f will generate the maximal ideal of P in (!)P,D = (!)P,x/(g),
and vice versa.
If C and D are two nonsingular curves, which meet transversally at a
finite number of points P 1 , . . . ,P" then it is clear that the intersection number
C.D should be r. So we take this as our starting point, together with some
natural properties the intersection pairing should have, to define our inter-
section theory. We denote by Div X the group of all divisors on X, and by
Pic X the group of invertible sheaves up to isomorphism, which is isomorphic
to the group of divisors modulo linear equivalence (II, §6).

Theorem 1.1. There is a unique pairing Div X x Div X--+ Z, denoted by

C.D for any two divisors C,D, such that
(1) if C
and D are nonsingular curves meeting transversally, then C.D =
#(C n D), the number of points of C n D,

357
V Surfaces

(2) it is symmetric: C.D = D.C,

(3) it is additive: (C 1 + C 2 ).D = C 1 .D + C 2 .D, and
(4) it depends only on the linear equivalence classes: if cl ~ c2 then
C1.D = C 2 .D.

Before giving the proof, we need some auxiliary results. Our main tool is
Bertini's theorem, which we will use to express any divisor as a difference of
nonsingular curves, up to linear equivalence.

Lemma 1.2. Let C 1 , . . . ,C, be irreducible curves on the surface X, and let D
be a very ample divisor. Then almost all curves D' in the complete linear
system IDI
are irreducible, nonsingular, and meet each of the C; transversally.

PROOF. We embed X in a projective space P" using the very ample divisor D.
Then we apply Bertini's theorem (II, 8.18) and (III, 7.9.1) simultaneously to X
and to the curves Cto ... ,C,. We conclude that most D' E IDI
are irreducible
nonsingular curves in X, and that the intersections C; n D are nonsin-
gular, i.e., points with multiplicity one, which means that the C; and D' meet
transversally. Since we did not assume the C; were nonsingular, we need to
use (II, 8.18.1).

Lemma 1.3. Let C be an irreducible nonsingular curve on X, and let D be any

curve meeting C transversally. Then

#(C n D) = degc(2'(D) ® @c).

PROOF. Here, of course, £>(D) is the invertible sheaf on X corresponding to D
(II, §7), and degc denotes the degree of the invertible sheaf £>(D) ® (!Jc
on C (IV, §1). We use the fact (II, 6.18) that 2'(- D) is the ideal sheaf of Don X.
Therefore, tensoring with (!Jc, we have an exact sequence

where now C n D denotes the scheme-theoretic intersection. Thus £>(D) ®

(!Jc is the invertible sheaf on C corresponding to the divisor C n D. Since the
intersection is transversal, the degree of the divisor C n D is just the number
ofpoints #(C n D).

PROOF OF (1.1). First we show the uniqueness. Fix an ample divisor H on X.

Given any two divisors C,D on X, we can find an integer n > 0 such that
C + nH, D + nH, and nH are all very ample. Indeed, we first choose k > 0
such that 2'( C + kH), 2'(D + kH) and 2(kH) are all generated by global
sections. This is possible by definition of ampleness (II, §7). Then we choose
l > 0 so that lH is very ample (II, 7.6). Taking n = k + l, it follows that
C + nH, D + nH, and nH are all very ample (II, Ex. 7.5).

358
 1 Geometry on a Surface

Now using (1.2), choose nonsingular curves

C' E IC + nH!
D' E ID + nH!, transversal to C'
E' E !nH!, transversal to D'
F' E !nH!, transversal to C' and E'.
Then C ~ C' - E' and D ~ D' - F', so by the properties (1)-(4) of the
theorem, we have
C.D = #(C' n D') - #(C' n F') - #(E' n D') + #(E' n F').
This shows that the intersection number of any two divisors is determined by
(1)-(4), so the intersection pairing is unique.
For the existence, we use the same method, and check that everything is
well-defined. To simplify matters, we proceed in two steps. Let ~ s; Div X
be the set of very ample divisors. Then ~ is a cone, in the sense that the sum
of two very ample divisors is again very ample. For C,D E ~'we define the
intersection number C.D as follows: by (1.2) choose C' E ICI nonsingular, and
choose D' E IDI nonsingular and transversal to C'. Define C.D = # (C' n D').
To show this is well-defined, first fix C', and let D" E IDI be another non-
singular curve, transversal to C'. Then by (1.3), we have
#(C' n D') = deg .P(D')@ (!)c.,
and ditto forD". But D' ~ D", so .P(D') ~ .P(D"), so these two numbers
are the same. Thus our definition is independent of D'. Now suppose C" E ICI
is another nonsingular curve. By the previous step, we may assume D' is
transversal to both C' and C". Then by the same argument, restricting to the
curveD', we see that #(C' n D') = #(C" n D').
So now we have a well-defined pairing ~ x ~ ---+ Z, which is clearly
symmetric, and by definition it depends only on the linear equivalence
classes of the divisors. It also follows from (1.3) that it is additive, since
.P(D 1 + D 2 ) ~ .P(D 1 ) @ .P(D 2 ), and the degree is additive on a curve.
Finally, this pairing on ~ x ~satisfies condition (1) by construction.
To define the intersection pairing on all of Div X, let C and D be any two
divisors. Then, as above, we can write C ~ C' - E' and D ~ D' - F' where
C',D',E',F' are all in~. So we define
C.D = C'.D' - C'.F' - E'.D' + E'.F'.
If, for example, we used another expression C ~ C" - E" with C",E" also
very ample, then
C' + E" ~ C" + E',
so by what we have shown for the pairing in~' we have
C'.D' + E".D' = C".D' + E'.D'

359
V Surfaces

and ditto for F' in place of D'. Thus the resulting two expressions for C.D are
the same. This shows that the intersection pairing C.D is well-defined on all
ofDiv X.
It satisfies (2), (3), (4) by construction and by the corresponding properties
on ~. The condition (1) follows using (1.3) once more. q.e.d.

Now that we have defined the intersection pairing, it is useful to have a way
of calculating it without having to move the curves. If C and D are curves
with no common irreducible component, and if P E C n D, then we define
the intersection multiplicity (C.D)p of C and D at P to be the length of {!)P,xl(f,g),
where f,g are local equations of C,D at P (1, Ex 5.4). Here length is the same
as the dimension of a k-vector space.

Proposition 1.4. If C and D are curves on X having no common irreducible

component, then
C.D = L (C.D)p.
PeCnD
PROOF. As in the proof of (1.3), let !l'(D) be the invertible sheaf corresponding
to D. Then we have an exact sequence
0--+ 2( -D)@ {!)c--+ {!)c--+ {!)enD--+ 0

where we consider C n D as a scheme. Now the scheme C n D has support

at the points of C n D, and for any such P, its structure sheaf is the k-algebra
@p x/(f,g). Therefore

dimk H 0 (X,{!)cnv) = L (C.D)p.

PeCnD
On the other hand, we can calculate this H 0 from the cohomology sequence
of the exact sequence above. We obtain

where, as usual, for any coherent sheaf /F,

x(:F) = D -1)i dimk Hi(X,:F)

is the Euler characteristic (III, Ex. 5.1).
This shows that the expression L( C.D)p depends only on the linear equiva-
lence class of D. By symmetry, it also depends only on the linear equivalence
class of C. Replacing C and D by differences of nonsingular curves, all
transversal to each other as in the proof of (1.1), we see that this quantity is
equal to the intersection number C.D defined in (1.1).

Example 1.4.1. If D is any divisor on the surface X, we can define the self-
intersection number D.D, usually denoted by D 2 • Even if Cis a nonsingular
curve on X, the self-intersection C 2 cannot be calculated by the direct method

360
 1 Geometry on a Surface

of (1.4). We must use linear equivalence. However, by (1.3), we see that

C 2 = degc(.P( C) ® me). To reinterpret this, note that since the ideal sheaf .J'
of Con X is 2'( -C) (II, 6.18), we have .1'/.1' 2 ~ 2'( -C)® me. Therefore
its dual.P(C) ®me is isomorphic to the normal sheaf .Aie1x, which is defined
as Yfom(.J'j.J' 2 ,mc) (II, §8). So we have C 2 = dege .Aic;x·

Example 1.4.2. Let X = P 2 . Then Pic X ~ Z, and we can take the class h
of a line as generator. Since any two lines are linearly equivalent, and since
two distinct lines meet in one point, we have h2 = 1. This determines the
intersection pairing on P 2 , by linearity. Thus if C,D are curves of degrees
n,m respectively, we have C "" nh, D "" mh and so C.D = nm. If C and D
have no component in common this can be interpreted in terms of the local
intersection multiplicities of(l.4), and we get a new proof ofBezout's theorem
(1, 7.8).

Example 1.4.3. Let X be the nonsingular quadric surface in P 3 . Then Pic

X ~ Z Et> Z (II, 6.6.1) and we can take as generators lines l of type (1,0) and m
of type (0,1), one from each family. Then F = 0, m 2 = 0, l.m = 1, because
two lines in the same family are skew, and two lines of opposite families meet
in a point. This determines the intersection pairing on X. So for example
if C has type (a,b) and D has type (a',b'), then C.D = ab' + a' b.

Example 1.4.4. Using the self-intersection, we can define a new numerical

invariant of a surface. Let QX/k be the sheaf of differentials of Xjk, and let
wx = f\ 2Qx;k be the canonical sheaf, as defined in (II, §8). Any divisor K in
the linear equivalence class corresponding to wx is called a canonical divisor.
Then K 2 , the self-intersection of the canonical divisor, is a number depending
only on X. For example, if X = P 2 , K = - 3h, so K 2 = 9. If X is the
quadric surface (1.4.3), then K has type (- 2,- 2) (II, Ex. 8.4), so K 2 = 8.

Proposition 1.5 (Adjunction Formula). If C is a nonsingular curve of genus g

on the surface X, and if K is the canonical divisor on X, then
2g - 2 = C.( C + K).
PROOF. According to (II, 8.20) we have We ~ Wx@ .P(C)@ me. The degree
of We is 2g - 2 (IV, 1.3.3). On the other hand, by (1.3) we have

degc(wx @ .P(C) @me) = C.(C + K).

Example 1.5.1. This gives a quick method of computing the genus of a curve
on a surface. For example, if Cis a curve of degree din P 2 , then
2g - 2 = d(d - 3)

so g = ~(d - 1)(d - 2). Cf. (II, Ex. 8.4).

361
V Surfaces

Example 1.5.2. If Cis a curve of type (a,b) on the quadric surface, then C +K
has type (a - 2, b - 2), so
2g - 2 = a(b - 2) + (a - 2)b,
so g = ab - a - b + 1. Cf. (III, Ex. 5.6).

Now we come to the Riemann-Roch theorem. For any divisor Don the
surface X, we let l(D) = dimk H 0 (X,.!l'(D) ). Thus l(D) = dim IDI + 1, where
IDI is the complete linear system of D. We define the superabundance s(D) to
be dim H 1 (X,.!l'(D) ). The reason for this terminology is that before the in-
vention of cohomology, the Riemann-Roch formula was written only with
l(D) and l(K - D), and the superabundance was the amount by which it
failed to hold. Recall also that the arithmetic genus Pa of X is defined by Pa =
x(@x)- 1 (III, Ex. 5.3).

Theorem 1.6 (Riemann-Roch). If Dis any divisor on the surface X, then

1
l(D) - s(D) + l(K - D) = 2 D.(D - K) + 1 + Pa·

PROOF. By Serre duality (III, 7.7) we have

l(K - D) = dim H 0 (X,.!l'(Dr @ Wx) = dim H 2 (X,.!l'(D) ).
Thus the left-hand side is just the Euler characteristic, so we have to show
for any D that
1
x(.!l'(D)) = 2 D.(D - K) + 1 + Pa·

Since both sides depend only on the linear equivalence class of D, as in (1.1)
we can write D as the difference C - E of two nonsingular curves. Now let
us calculate. Since the ideal sheaves of C,E are .!l'(- C), .!l'(- E) respectively,
we obtain exact sequences, tensoring with .!l'(C),
0--+ .!l'(C - E) --+ .!l'(C)--+ .!l'(C)@ (i')E--+ 0
and

Since xis additive on short exact sequences (III, Ex. 5.1), we have

Now X((i')x) = 1 + Pa by definition of Pa· Using the Riemann-Roch theorem

for the curves C and E (IV, 1.3), and using (1.3) to find the degree, we have

x(.!l'(C) ® @c) = C2 + 1 - 9c
and

362
 I Geometry on a Surface

Finally, we use (1.5) to compute the genus of C and£:

1
gc = 2 C.(C + K) + 1
and
1
gE = 2 E.(E + K) + 1.

Combining all these, we obtain

1
x(!l'(C- £)) = 2(C- E).(C- E- K) + 1 + Pa

as required.

Remark 1.6.1. There is another formula, which is sometimes considered to

be part of the Riemann-Roch theorem, namely
12(1 + Pal = K2 + C2,

where c2 is the second Chern class of the tangent sheaf of X. This is a con-
sequence ofthe generalized Grothendieck-Hirzebruch Riemann-Roch theo-
rem (App. A, 4.1.2).

As applications of the Riemann-Roch theorem, we will prove the Hodge

index theorem and Nakai's criterion for an ample divisor.

Remark 1.6.2. In the following, note that if we fix a very ample divisor H
on a surface X, then for any curve Con X, the intersection number C.H is
just equal to the degree of C in the projective embedding determined by H
(Ex. 1.2). In particular, it is positive. More generally, having fixed an ample
divisor H on X, the number C.H plays a role similar to the degree of a
divisor on a curve.

Lemma 1.7. Let H be an ample divisor on the surface X. Then there is an

integer n0 such that for any divisor D, if D.H > n0 , then H 2 (X,!l'(D)) = 0.
PROOF. By Serre duality on X, for any divisor D we have dim H 2 (X,!l'(D)) =
l(K -D). If l(K -D)> 0, then the divisor K -Dis effective, so (K - D).H >
0. In other words, D.H < K.H. So we have only to take n0 = K.H to get
the result.

Remark 1.7.1. This result can be regarded as the analogue for surfaces of
the result that says on a curve X, there is an integer n0 (namely 2gx - 2)
such that if deg D > n0 , then H 1 (X,!l'(D)) = 0 (IV, 1.3.4).

Corollary 1.8. Let H be an ample divisor on X, and let D be a divisor such

that D.H > 0 and D 2 > 0. Then for all n » 0, nD is linearly equivalent
to an effective divisor.

363
V Surfaces

PROOF. We apply the Riemann-Roch theorem to nD. Since D.H > 0, for
n » 0 we will have nD.H > n0 , so by (1.7), l(K - nD) = 0. Since s(nD) ~ 0,
the Riemann-Roch theorem gives
1 1
l(nD) ~ 2 n2 D2 - 2 nD.K + 1 + Pa·
Now since D 2 > 0, the right-hand side becomes large for n » 0, so we see
that l(nD) -> oo as n -> oo. In particular, nD is effective for all n » 0.

Definition. A divisor D on a surface X is numerically equivalent to zero,

written D = 0, if D.E = 0 for all divisors E. We say D and E are nu-
merically equivalent, written D =
E, if D - E 0. =
Theorem 1.9 (Hodge Index Theorem). Let H be an ample divisor on the sur-
face X, and suppose that D is a divisor, D ¢ 0, with D.H = 0. Then
D 2 < 0.

PROOF. Suppose to the contrary that D 2 ~ 0. We consider two cases. If

D2 > 0, let H' = D + nH. For n » 0, H' is ample, as in the proof of (1.1).
Furthermore, D.H' = D2 > 0, so by (1.8), we have mD is effective for all
m » 0. But then mD.H > 0 (think of the projective embedding defined by
a multiple of H), hence D.H > 0, which is a contradiction.
If D 2 = 0, we use the hypothesis D ¢ 0 to conclude that there is a divisor
E with D.E =f. 0. Replacing E by E' = (H 2 )E - (E.H)H, we may assume
furthermore that E.H = 0. Now let D' = nD + E. Then D'.H = 0, and
D' 2 = 2nD.E + E 2 • Since D.E =f. 0, by suitable choice of n E Z we can make
D' 2 > 0. But then the previous argument applies to D', and again we have
a contradiction.

Remark 1.9.1. We explain the title of this theorem as follows. Let Picn X
be the subgroup of Pic X of divisor classes numerically equivalent to zero,
and let Num X = Pic X/Picn X. Then clearly the intersection pairing in-
duces a nondegenerate bilinear pairing Num X x Num X-> Z. It is a
consequence of the Neron-Severi theorem (Ex. 1.7) that Num X is a free
finitely generated abelian group (see also (Ex. 1.8) ). So we can consider the
vector space Num X ®z R over R, and the induced bilinear form. A theo-
rem of Sylvester (Lang [2, XIV, §7, p. 365]) shows that such a bilinear form
can be diagonalized with ± 1's on the diagonal, and that the number of
+ 1's and the number of -1's are invariant. The difference of these two
numbers is the signature or index of the bilinear form. In this context, (1.9)
says that the diagonalized intersection pairing has one + 1, corresponding
to a (real) multiple of H, and all the rest -1's.

Example 1.9.2. On the quadric surface X (1.4.3) we can take H of type (1,1)
and D of type (1,-1). Then H 2 = 2, H.D = 0, D2 = -2, and D,H form a
basis of Pic X. In this case the only divisor numerically equivalent to 0 is

364
 1 Geometry on a Surface

0, so Pic X = Num X. The pairing on Num X ®z R is diagonalized by

taking the basis (1/J'i)H, (1/J'i)D. .

Theorem 1.10 (Nakai-Moishezon Criterion). A divisor D on the surface X is

ample if and only if D 2 > 0 and D.C > 0 for all irreducible curves C in X.
PROOF. The condition is clearly necessary, because if D is ample, then mD
is very ample for some m > 0, in which case m 2 D 2 is the degree of X in the
corresponding embedding, and mD.C is the degree of C, both of which must
be positive (Ex. 1.2).
Conversely, suppose D 2 > 0 and D.C > 0 for all irreducible curves C.
If H is a very ample divisor on X, then H is represented by an irreducible
curve, so D.H > 0 by hypothesis. Therefore by (1.8) some multiple mD for
m > 0 is effective. Replacing D by mD, we may assume that Dis effective. So
we think of D as a curve in X, possibly singular, reducible, and nonreduced.
Next, let !l' = !l'(D). We will show that the sheaf !l' ® (!Jv is ample on
the scheme D. For this it is sufficient to show that !l' ® (!JD,.d is ample on
the reduced scheme Drect (III, Ex. 5.7). And if D,ect is a union of irreducible
curves Ct. ... ,C" it is enough to show that !l' ® (!Jc is ample on each C;
(foe. cit.). Finally, iff: C; ---+ C; is the normalization ~f C;, it is enough to
show that f*(!l'@ (!)c) is ample on C;, since f is a finite surjective morphism
(foe. cit.). But deg f*(!l' @ (!)c) is just D.C; > 0, because we can represent
!l' as a difference of nonsingular curves meeting C; transversally, so this
degree is preserved by f*. Since the degree is positive, this sheaf is ample
on the nonsingular curve C; (IV, 3.3). Therefore !l' ® (!JD is ample on D.
Next we will show that !l'n is generated by global sections for n » 0.
We use the exact sequence
0---+ !l'- 1 ---+ (!Jx---+ (!JD---+ 0
tensored with !l'n, and the resulting cohomology sequence
0---+ Ho(X,!l'n-1)---+ Ho(X,!l'")---+ Ho(D,!l'n@ (!Jv)---+
---+ H1(X,!l'"-1)---+ H1(X,!l'n)---+ H1(D,!l'n@ (!Jv)---+ ...
Since !l' @ (!JD is ample on D, we have H 1(D,!l'n @ (!Jv) = 0 for n » 0
(III, 5.3). So we see that for each n,
dim H 1(X,!l'n) ~ dim H 1(X,!l'n- 1).
Since these are finite-dimensional vector spaces, these dimensions must
eventually be all equal. Therefore the map
Ho(X,!l'n) ---+ Ho(D,!l'n ® (!Jv)
is surjective for all n » 0. Again since !l' ® (!JD is ample on D, the sheaf
!£'" @ (!JD will be generated by global sections for all n » 0. These sections
lift to global sections of !l'n on X, as we have just seen, so by Nakayama's
lemma, the global sections of !l'n generate the stalks at every point of D.
But since !l' = !l'(D), it has a section vanishing only along D, so in fact !l'n
is generated by global sections everywhere.

365
V Surfaces

Fixing an n such that !lm is generated by global sections, we obtain a

morphism cp:X--+ pN defined by fr (II, 7.1). Next we show that the mor-
phism cp has finite fibres. If not, there would be an irreducible curve C in
X with cp( C) = a point. In this case, taking a hyperplane in pN which misses
that point, we would have an effective divisor E ~ nD with En C = 0.
Therefore E.C = 0, which contradicts the hypothesis D.C > 0 for all C.
So we see that cp has finite fibres.
Then it is a consequence of the Stein factorization theorem (III, 11.5)
that cp is actually a finite morphism (III, Ex. 11.2). So cp*(C9(1)) = ;r is
ample on X by (III, Ex. 5.7) and we conclude that Dis ample. q.e.d.

Example 1.10.1. On the quadric surface X (1.4.3), the effective divisors are
those of type (a,b) with a,b ~ 0. So a divisor D of type (a,b) is ample if and
only if a = D.(1,0) > 0 and b = D.(0,1) > 0 (II, 7.6.2). In this case the con-
dition D.C > 0 for all irreducible curves C implies D 2 > 0. However, there
is an example of Mumford of a divisor Don a surface X, with D.C > 0 for
every irreducible curve, but D 2 = 0, hence D not ample. See Hartshorne
[5, I, 10.6].

References for§ 1. For another approach to intersection theory on a sur-

face, see Mumford [2]. The proof of the Riemann~Roch theorem follows
Serre [7, Ch. IV no. 8]. The proof of the Hodge index theorem is due to
Grothendieck [2]. The criterion for an ample divisor is due to Nakai [1]
and independently Moishezon [1]. See Appendix A for intersection theory
and the Riemann~Roch theorem in higher dimensions.

EXERCISES

1.1. Let C,D be any two divisors on a surface X, and let the corresponding invertible
sheaves be .P,A. Show that

1.2. Let H be a very ample divisor on the surface X, corresponding to a projective

embedding X <;; PN. If we write the Hilbert polynomial of X (III, Ex. 5.2) as
1
F(z) = 2 azz + bz + c,
show that a = Hz, b = !Hz + 1 - n, where n is the genus of a nonsingular
curve representing H, and c = 1 + Pa· Thus the degree of X in PN, as defined
in (I, §7), is just Hz. Show also that if Cis any curve in X, then the degree of C
in pN is just C. H.
1.3. Recall that the arithmetic genus of a projective scheme D of dimension 1 is defined
as Pa = 1 ~ x(lPv) (III, Ex. 5.3).
(a) If Dis an effective divisor on the surface X, use (1.6) to show that 2pa - 2 =
D.(D + K).
(b) p.(D) depends only on the linear equivalence class of Don X.

366
 Geometry on a Surface

(c) More generally, for any divisor Don X, we define the virtual arithmetic genus
(which is equal to the ordinary arithmetic genus if Dis effective) by the same
formula: 2p. - 2 = D.(D + K). Show that for any two divisors C,D we have
p.( -D)= D2 - p.(D) +2
and
p.(C + D) = p.(C) + p.(D) + C.D - 1.
1.4. (a) If a surface X of degree d in P 3 contains a straight line C = P 1 , show that
C2 = 2- d.
(b) Assume char k = 0, and show for every d ~ 1, there exists a nonsingular
surface X of degree d in P 3 containing the line x = y = 0.
1.5. (a) IfXisasurfaceofdegreedinP 3 ,thenK 2 = d(d- 4) 2 •
(b) If X is a product of two nonsingular curves C,C', of genus g,g' respectively,
then K 2 = 8(g - 1)(g' - 1). Cf. (II, Ex. 8.3).
1.6. (a) If Cis a curve of genus g, show that the diagonal Ll <;: C x C has self-inter-
section Ll 2 = 2 - 2g. (Use the definition of QCfk in (II, §8).)
(b) Let l = C x pt and m = pt x C. If g ~ 1, show that l,m, and Ll are linearly
independent in Num( C x C). Thus Num( C x C) has rank ~ 3, and in parti-
cular, Pic(C x C) i= Pi Pic C EB P! Pic C. Cf. (III, Ex. 12.6), (IV, Ex. 4.10).
1.7. Algebraic Equivalence of Divisors. Let X be a· surface. Recall that we have
defined an algebraic family of effective divisors on X, parametrized by a non-
singular curve T, to be an effective Cartier divisor D on X x T, flat over T
(III, 9.8.5). In this case, for any two closed points 0,1 E T, we say the corresponding
divisors D 0 ,D 1 on X are prealgebraically equivalent. Two arbitrary divisors
are prealgebraically equivalent if they are differences of prealgebraically equiva-
lent effective divisors. Two divisors D,D' are algebraically equivalent if there is
a finite sequence D = D0 ,D 1 , . .. ,Dn = D' with D; and D;+ 1 prealgebraically
equivalent for each i.
(a) Show that the divisors algebraically equivalent to 0 form a subgroup ofDiv X.
(b) Show that linearly equivalent divisors are algebraically equivalent. [Hint: If
(f) is a principal divisor on X, consider the principal divisor (if- u) on X x P\
where t,u are the homogeneous coordinates on P 1.]
(c) Show that algebraically equivalent divisors are numerically equivalent. [Hint:
Use (III, 9.9) to show that for any very ample H, if D and D' are algebraically
equivalent, then D.H = D'.H.]
Note. The theorem of Neron and Severi states that the group of divisors
modulo algebraic equivalence, called the Nhon-Seueri group, is a finitely gen-
erated abelian group. Over C this can be proved easily by transcendental methods
(App. B, §5) or as in (Ex. 1.8) below. Over a field of arbitrary characteristic, see
Lang and Neron [1 J for a proof, and Hartshorne [ 6] for further discussion. Since
Num X is a quotient of the Neron-Severi group, it is also finitely generated, and
hence free, since it is torsion-free by construction.
1.8. Cohomology Class of a Divisor. For any divisor D on the surface X, we define
its cohomology class c(D) E H 1(X,Qx) by using the isomorphism Pic X ~
H 1 (X,@k) of (III, Ex. 4.5) and the sheaf homomorphism dlog:@*--. Qx (III,
Ex. 7.4c). Thus we obtain a group homomorphism c:Pic X--> H 1(X,Qx). On
the other hand, H 1(X,Q) is dual to itself by Serre duality (III, 7.13), so we have a

367
V Surfaces

nondegenerate bilinear map

(a) Prove that this is compatible with the intersection pairing, in the following
sense: for any two divisors D,E on X, we have
(c(D),c(E)) = (D.E) · 1
in k. [Hint: Reduce to the case where D and E are nonsingular curves meeting
transversally. Then consider the analogous map c: Pic D ...... H 1(D,QD), and
the fact (III, Ex. 7.4) that c(point) goes to 1 under the natural isomorphism of
H 1(D,QD) with k.]
(b) If char k = 0, use the fact that H 1(X,Qx) is a finite-dimensional vector space
to show that Num X is a finitely generated free abelian group.
1.9. (a) If H is an ample divisor on the surface X, and if D is any divisor, show that
(D 2 )(H 2 ) ~ (D.H) 2 •
(b) Now let X be a product of two curves X = C x C'. Let l = C x pt, and
m = pt x C'. For any divisor Don X, let a = D.l, b = D.m. Then we say D
has type (a,b). If D has type (a,b), with a,b E Z, show that
D 2 ~ 2ab,
=
and equality holds if and only if D bl +am. [Hint: Show that H = l + m
is ample, let E = l - m, let D' = (H 2 )(E 2 )D - (E 2 )(D.H)H - (H 2 )(D.E)E, and
apply (1.9). This inequality is due to Castelnuovo and Severi. See
Grothendieck [2].]
1.10. Wei/'s Proof [2] of the Analogue of the Riemann Hypothesis for Curves. Let C
be a curve of genus g defined over the finite field Fq, and let N be the number of
points of C rational over Fq. Then N = 1 - a + q, with lal
~ 2g~q. To prove
this, we consider C as a curve over the algebraic closure k of Fq. Let f: C ...... C
be the k-linear Frobenius morphism obtained by taking qth powers, which
makes sense since Cis defined over Fq, so Xq ~ X (IV, 2.4.1). Let r <;; C x C
be the graph off, and let Ll <;; c X c be the diagonal. Show that r 2 = q(2 - 2g),
and r.LI = N. Then apply (Ex. 1.9) to D = rr + sLI for all r and s to obtain
the result. See (App. C, Ex. 5.7) for another interpretation of this result.
1.11. In this problem, we assume that X is a surface for which Num X is finitely gen-
erated (i.e., any surface, if you accept the Neron~Severi theorem (Ex. 1. 7) ).
(a) If His an ample divisor on X, and dE Z, show that the set of effective divisors
D with D.H = d, modulo numerical equivalence, is a finite set. [Hint: Use
the adjunction formula, the fact that Pa of an irreducible curve is ;;, 0, and the
fact that the intersection pairing is negative definite on Hj_ in Num X.]
(b) Now let C be a curve of genus g ;;, 2, and use (a) to show that the group of
automorphisms of C is finite, as follows. Given an automorphism a of C, let
r <;; X = c X c be its graph. First show that if r = Ll, then r = Ll, using
the fact that Ll 2 < 0, since g ;;, 2 (Ex. 1.6). Then use (a). Cf. (IV, Ex. 2.5).
1.12. If D is an ample divisor on the surface X, and D' = D, then D' is also ample.
Give an example to show, however, that if D is very ample, D' need not be very
ample.

368
 2 Ruled Surfaces

2 Ruled Surfaces

In this section we will illustrate some of the general concepts discussed in

§1 by studying a particular class of surfaces, the ruled surfaces. By using
some results from the theory of curves, we get a good hold on these surfaces,
and can describe them and the curves lying on them quite explicitly.
We begin by establishing some general properties of ruled surfaces. Then
we will define an invariant e, and give some examples. After that we give a
classification of elliptic ruled surfaces, a detailed description of the rational
ruled surfaces, and we determine the ample divisors on a ruled surface of
any genus.

Definition. A geometrically ruled surface, or simply ruled surface, is a surface

X, together with a surjective morphism n:X-+ C to a (nonsingular) curve
C, such that the fibre XY is isomorphic to P 1 for every point y E C, and
such that n admits a section (i.e., a morphism a: C-+ X such that no a =
ide).

Nate: In fact, one can show using Tsen's theorem that the existence of a
section is a consequence of the other provisions of the definition-see, e.g.,
Shafarevich [1, p. 24].

Example 2.0.1. If C is a curve, then C x P 1 with its first projection is a

ruled surface. In particular, the quadric surface in P 3 is a ruled surface in
two different ways. We consider the data n,C as given when we speak of a
ruled surface.

Lemma 2.1. Let n:X -+ C be a ruled surface, let D be a divisor on X, and

suppose that D.f = n ;, 0, where f is a fibre of n. Then n*:i'(D) is a
locally free sheaf of rank n + 1 on C. In particular, n*@x = @c.

PROOF. First note that any two fibres of n are algebraically equivalent
divisors on X, since they all are parametrized by the curve C. Therefore
they are numerically equivalent (Ex. 1.7), so that D.f is independent of the
choice of the fibre.
Now for any y E C, we consider the sheaf :i'(D)y on the fibre XY. This
is an invertible sheaf of degree n on X Y ~ P 1 , so H 0 (:£(D)y) has dimension
n + 1. This is independent of y, so by Grauert's theorem (III, 12.9), n*:E(D)
is locally free of rank n + 1.
In caseD = 0, n*@x is locally free of rank 1. But (III, 12.9) tells us fur-
thermore that the natural map

369
V Surfaces

is an isomorphism for each y. The right-hand side is canonically isomorphic

to k. Therefore the image of the global section 1 of C9c via the structural
map C9c -+ n*C9x generates the stalk at every point, showing that n*C9x ~
C9c.

Proposition 2.2. If n: X -+ Cis a ruled surface, then there exists a locally free
sheaf G of rank 2 on C such that X ~ P(G) over C. (See (II, §7) for the
definition of P(G).) Conversely, every such P(G) is a ruled surface over C.
If iff and iff' are two locally free sheaves of rank 2 on C, then P(G) and
P(G') are isomorphic as ruled surfaces over C if and only if there is an
invertible sheaf ff on C such that G' ~ @" @ ff.
PROOF. Given a ruled surface n:X -+ C, then by definition n has a section
u. Let D = u(C). Then Dis a divisor on X, and D.f = 1 for any fibre. By
the lemma, @" = n*ff(D) is a locally free sheaf of rank 2 on C. Furthermore,
there is a natural map n*S = n*n*ff(D) -+ ff(D) on X. This map is sur-
jective. Indeed, by Nakayama's lemma, it is enough to check this on any
fibre X y· But XY ~ P 1 , and ff(D)y is an invertible sheaf of degree 1, which
is generated by its global sections, and@" @ k(y) -+ H 0 (ff(D)y) is surjective
by (III, 12.9).
Now we apply (II, 7.12) which shows that the surjection n*G-+ ff(D)-+ 0
determines a morphism g: X-+ P(G) over C, with the property that ff(D) ~
g*@P(c)(1). Since ff(D) is very ample on each fibre, g is an isomorphism on
each fibre, and so g is an isomorphism.
Conversely, let @" be a locally free sheaf of rank 2 on C, let X = P(G)
and let n: X -+ C be the projection. Then X is a nonsingular projective
surface over k, and each fibre of n is isomorphic to P 1 . To show the exis-
tence of a section, let U s;;; C be an open subset on which @" is free. Then
n- 1 (V) ~ U x Pl, so we can define a section u: U-+ n- 1 (U) by y r--+ y x
pt. Then, since X is a projective variety, by (1, 6.8) there is a unique exten-
sion of u to a map of C to X, which is necessarily a section.
For the last statement, see (II, Ex. 7.9).

Remark 2.2.1. A surface X is called a birationally ruled surface if it is hi-

rationally equivalent to C x P 1 for some curve C. (This includes the
rational surfaces, because P 2 is birational to P 1 x P 1 .) We see from (2.2)
that every ruled surface is birationally ruled.

Proposition 2.3. Let n:X -+ C be a ruled surface, let C0 s;;; X be a section,

and let f be a fibre. Then
Pic X ~ Z EB n* Pic C,
where Z is generated by C 0 . Also
NumX ~ Z EB Z,
generated by C 0 ,f, and satisfying C 0 .f = 1, j2 = 0.

370
 2 Ruled Surfaces

PROOF. Clearly C 0 .f = 1, because C 0 and f meet at only one point, and

are transversal there. We have f 2 = 0 because two distinct fibres don't
meet.
Now if DE Pic X, let n = D.f, and let D' = D - nC 0 . Then D'.f = 0.
Therefore by (2.1), n*(2'(D')) is an invertible sheaf on C, and clearly 2'(D') ~
n*n*(2'(D') ). Since n*: Pic C --+ Pic X is clearly injective, we see that
Pic X ~ Z EEl n* Pic C. Then, since any two fibres are numerically equiva-
lent, Num X ~ Z EEl Z, generated by C0 and f See also (II, Ex. 7.9) and
(III, Ex. 12.5).

Lemma 2.4. Let D be a divisor on the ruled surface X, and assume that
D.f ~ 0. Then Rin*2'(D) = 0 fori > 0; and for all i,
Hi(X,2'(D)) ~ Hi( C,n*2'(D) ).
PROOF. Since 2'(D)y is an invertible sheaf of degree D.f ~ 0 on XY ~ P\
we have Hi(Xy,2'(D)y) = 0 for all i > 0. Therefore Rin*2'(D) = 0 for
i > 0 (III, Ex. 11.8) or (III, 12.9). The second statement follows from (III,
Ex. 8.1).

Corollary 2.5. If the genus of Cis g, then Pa(X) = -g, p9 (X) = 0, q(X) =g.
PROOF. The arithmetic genus Pais defined by 1+pa=x(@x). Since n*@x=(r)c
by (2.1), we have dim H 0 (X,(r)x) = 1, dim H 1(X,(r)x) = g, dim H 2 (X,(r)x) =
0 using (2.4). So Pa = -g. By (III, 7.12.3), the geometric genus p9 =
dim H 2 (X,(r)x) = 0. The irregularity q = dim H 1 (X,(r)x) = g. See also (III,
Ex. 8.4).

Proposition 2.6. Let Iff be a locally free sheaf of rank 2 on the curve C, and
let X be the ruled surface P(!C). Let (9x(1) be the invertible sheaf (r)P(&)(1)
(II, §7). Then there is a one-to-one correspondence between sections
0': C--+ X and surjections Iff --+ 2' --+ 0, where 2' is an invertible sheaf on C,
given by 2' = O"*@x(1). Under this correspondence, if JV = ker(lff --+ 2'),
then JV is an invertible sheaf on C, and JV ~ n*((r)x(l) ® 2'(- D)), where
D = O'(C), and n*JV ~ (r)x(1) ® 2'( -D).
PROOF. The correspondence between sections 0' and surjections Iff --+ 2' --+ 0
is given by (II, 7.12). (See also (II, Ex. 7.8).) Given 0', with O"(C) = D, we
consider the exact sequence

Taking n*' we have

0--+ n*((r)x(1) ® 2'( -D))--+ Iff--+ 2'--+ 0,
with 0 on the right because R 1 n*((r)x(1) ® 2'( -D)) = 0 by (2.4). The middle
term is Iff by (II, 7.11), and the right-hand term is 2', because (r)x(l) ® (r)D
is a sheaf on D ~ C, so O"* and n* have the same effect. We conclude that

371
V Surfaces

JV ~ n*(lDx(l) ® 2( -D)). Since the sheaf lDx(l) ® 2( -D) has degree 0

along the fibres, we see that it is isomorphic to n* JV by (2.3) and JV is
invertible (2.1 ).

Corollary 2.7. Any locally free sheaf$ of rank 2 on a curve Cis an extension
of invertible sheaves.
PROOF. Since P($) has a section (2.2), we get an exact sequence 0 -+ JV -+
$ -+ 2 -+ 0 where JV and 2 are invertible sheaves. This also follows from
(II, Ex. 8.2).

Remark 2.7.1. The same result holds for locally free sheaves of arbitrary
rank (Ex. 2.3).

Proposition 2.8. If 1t: X -+ C is a ruled surface, it is possible to write X ~

P($) where$ is a locally free sheaf on C with the property that H 0 ($) =I 0
but for all invertible sheaves 2 on C with deg 2 < 0, we have H 0 ($ ® 2) =
0. In this case the integer e = -deg $is an invariant of X. Furthermore
in this case there is a section u 0 :C-+ X with image C 0 , such that 2(C0 ) ~
lD x( 1).
PROOF. First write X ~ P($') for some locally free sheaf$' on C (2.2). Then
we will replace $' by $ = $' ® .A for a suitable invertible sheaf .A on C
so as to have H 0 ($) =1 0 but H 0 ($ ® 2) = 0 for all 2 with deg 2 < 0.
An invertible sheaf of positive degree on Cis ample (IV, 3.3), so it is possible
to make H 0 ($) =1 0 by taking deg .A large enough. On the other hand,
since $' is an extension of invertible sheaves (2. 7), and since an invertible
sheaf of negative degree can have no global sections, we see that H 0 ($) = 0
for deg .A sufficiently negative. So we achieve our result by taking an .A
of least degree such that H 0 ($' ® .A) =1 0.
Since all possible representations of X as a P($) are given by the sheaves
$ = $' ® .A (2.2) we see that the integer e = - deg $ depends only on X.
(The degree of$ is defined as the degree of the invertible sheaf (\ 2 $ (II,
Ex. 6.12).)
Finally, let s E H 0($) be a nonzero section. It determines an injective
map 0-+ lDc-+ $. I claim the quotient 2 = cff/lDc is an invertible sheaf on
C. Since C is a nonsingular curve, and 2 has rank 1 in any case, it is enough
to show that 2 is torsion-free. If not, let g; <;; $ be the inverse image of
the torsion subsheaf of 2 by the map $ -+ 2 -+ 0. In that case g; is torsion-
#
free of rank 1 on C, hence invertible. Furthermore, lD c g;, so deg g; > 0.
But then, since g; <;; $, we have H 0 ($ ® g;~) =1 0, and deg g;~ < 0, so
this contradicts the choice of$.
Now, since 2 is invertible, it gives a section u0 : C -+ X by (2.6). Let C0
be its image. Then JV = lDc in the notation of (2.6), so lDx(l) ® 2(- C0 ) ~
lDx, which shows that 2(C 0 ) ~ lDx(l).

372
 2 Ruled Surfaces

Figure 19. A ruled surface.

Notation 2.8.1. For the rest of this section, we fix the following notation
(Fig. 19). Let C be a curve of genus g, and let n:X--+ C be a ruled surface
over C. We write X ~ P(t&"), where !&" satisfies the conditions of (2.8), in
which case we say !&" is normalized. This does not necessarily determine
!&" uniquely, but it does determine deg !&". We let e be the divisor on C
corresponding to the invertible sheaf f\ 2 !&", so that e = -deg e. (This
sign is put in for historical reasons.) We fix a section C 0 of X with
2(C0 ) ~ (!)P(.c)(1). If b is any divisor on C, then we denote the divisor
n*b on X by bf, by abuse of notation. Thus any element of Pic X can be
written aC 0 + bf with a E Z and b E Pic C. Any element of Num X can
be written aC 0 + bf with a,b E Z.

Proposition 2.9. If D is any section of X, corresponding to a surjection !&" --+

2 --+ 0, and if 2 = 2(b) for some divisor b on C, then deg b = C 0 .D,
and
D "' C 0 + (b - e)f

In particular, we have C6 = deg e = -e.

PROOF. Since 2 = a*(2(C 0 ) ® (!)v), we have deg 2 = C 0 .D by (1.1) and

(1.3). Writing

we have 2(C 0 -D)~ n*JV by (2.6) and the choice of C 0 (2.8.1). But
JV = 2(e - b), so we have D "' C 0 + (b - e)f in Pic X. Finally, in the
caseD = C 0 , JV = (!)c, sob = e and we have C6 = deg e = -e.

Lemma 2.10. The canonical divisor K on X is given by

K "' -2C 0 + (f + e)f

where f is the canonical divisor on C.

373
V Surfaces

PROOF. Let K - aC 0 + bf Using the adjunction formula (1.5) for a fibre

f, we have
- 2 = f.(f + K) = a.

Now we use the adjunction formula for C0 in its invertible sheaf form
(II, 8.20), which says that
Wc 0 :;:;: Wx ® Y(Co) ® 0c 0 :;:;: 2"(- Co + bf) ® 0ca·
Identifying C0 with C via n, the corresponding statement for divisors on C
is f = - e + b, so b = e + f. This result also follows from {III, Ex. 8.4).

Corollary 2.11. For numerical equivalence, we have

K = - 2C 0 + (2g - 2 - e)f
and therefore
K 2 = 8(1 - g).

PROOF. We have deg f = 2g - 2 (IV, 1.3.3) and deg e = -e. Then we

compute K 2 using (2.3) and (2.9).

Example 2.11.1. For any curve C, the ruled surface X = C x P 1 corre-

sponds to the (normalized) locally free sheaf iff = 0c EB 0c on C. In this
case e = 0, and C0 is any fibre of the second projection.

Example 2.11.2. If C is a curve of genus ~ 1, and iff = 0c EEl Y where

deg Y = 0 but Y ;f. 0c, then there are two choices of normalized iff,
namely iff and iff ® Y- 1 . We have e = 0, deg e = 0, but e is determined
only up to sign. There are exactly two choices of C0 , both with C6 = 0.

Example 2.11.3. On any curve C, let iff = 0c EB Y with deg Y < 0. Then
the normalized iff is unique, Y = Y(e) and e is unique. The section C0 is
unique, with C6 = -e < 0. In this case e = -deg Y > 0.

Example 2.11.4. Let C be any curve embedded in P", of degree d. Let X 0

be the cone over C in P" + 1 , with vertex P 0 (I, Ex. 2.10). If we blow up the
point P0 , we will show that we obtain a ruled surface X over C, of the kind
(2.11.3) above, with Y :;:;: (!Jc( -1). In particular, e = d, and the inverse
image of P 0 in X is the section C 0 with C5 = -d.
First of all, we show that P"+ 1 with one point blown up is isomorphic
to P((!J EB 0(1)) over P". Indeed, let P"+ 1 have coordinates x 0 , . . . ,xn+ 1 .
If we blow up the point P 0 = (1,0, ... ,0), then we get the variety V s; P" x
pn+ 1 defined by the equations X;Yi = xiyi for i,j = 1,2, ... ,n + 1, where
Yb . .. ,Yn+ 1 are the coordinates for P" (II, 7.12.1). On the other hand, if
iff = (!J EB 0(1) on P", then P(iff) is defined as Proj S(iff), where S(iff) is the
symmetric algebra of iff (II, §7). Now iff is generated by the global sections 1
of (!J andY!, ... ,Yn+ 1 of 0(1). Therefore S(iff) is a quotient ofthe polynomial

374
 2 Ruled Surfaces

algebra m[ Xo, 0 0,xn+ 1] by the mapping Xo f-> 1, xi f-> Yi fori = 1,

0 ,n + 1.
0 0 0

The kernel of this map is the ideal generated by all xiyi - xiyi, i =
1, ... ,n + 1. Therefore P(lff) is isomorphic to the subscheme of pn x pn+ 1
defined by these equations, which is the same as the variety V s; pn x pn+ 1
defined above. The first projection makes V look like P(lff), the second
projection makes V look like blowing up a point.
Now let Y be any subvariety ofPn, and X 0 its cone in pn+ 1 , with vertex P 0 .
If we blow up P 0 on X 0 , we get a variety X which is the strict transform of X 0
in V (II, 7.15.1). On the other hand, this variety X is clearly the inverse
image of Y under the projection n: V ~ P(lff)--+ pn_ So we see that X ~
P((Dy EB (Dy(1) ). Twisting by (Dy( -1), we still have the same variety, so
X ~ P(my EB my( -1) ).
In particular, if Y is a nonsingular curve C of degree d in pn, then
fi' = (D c( - 1) has degree - d.

Example 2.11.5. As a special case of (2.11.4), we see that P 2 with one point
blown up is isomorphic to the rational ruled surface over P 1 defined by
Iff = (D EB (()( -1), having e = 1.

Example 2.11.6. For an example of a ruled surface withe < 0, let C be an

elliptic curve, let P E C be a point, and construct a locally free sheaf Iff of
rank 2 as an extension
0 --+ (D --+ Iff --+ fi'(P) --+ 0
defined by a nonzero element ~ E Ext 1(fi'(P),m) (III, Ex. 6.1). In this case
Ext 1 (fi'(P},(D) ~ H 1 ( C,fi'(- P)) (III, 6.3) and (III, 6. 7). This is dual to
H 0 ( C,fi'(P)) which has dimension 1. Thus~ is unique up to a scalar multiple,
and so Iff is uniquely determined up to isomorphism.
I claim this Iff is normalized. Clearly H 0 (0") #- 0 by construction. If .A
is any invertible sheaf, then we have an exact sequence
0 --+ .A --+ Iff 0 .A --+ fi'(P) 0 .A --+ 0.
If deg .A < 0, then we have H 0 (.A) = 0, and H 0 (fi'(P) 0 .A) = 0, and
therefore also H 0 (0" 0 .A) = 0, except for the case .A = fi'(- P). In that
case we look at the cohomology sequence
0 --+ H 0 (.A) --+ H 0 (0" 0 .A) --+ H 0 (fi'(P) 0 .A) ~ H 1 (.A) --+ ....
The image of 1 E H 0 (fi'(P) 0 .A) = H 0 ((Dc) by <5 is just the element ~ de-
fining Iff (III, Ex. 6.1), which is nonzero. Therefore <5 is injective, and again
H 0 (0" 0 .A) = 0. Thus Iff is normalized.
Now taking X = P(lff), we have an elliptic ruled surface with e = -1.

Now that we have established some general properties of ruled surfaces

and have given some examples, we can look more closely at some special
cases. We begin by discussing the possible values of the invariant e.

375
V Surfaces

Theorem 2.12. Let X be a ruled surface over the curve C of genus g, determined
by a normalized locally free sheaf <C.
(a) If <C is decomposable (i.e., a direct sum of two invertible sheaves)
then <C ~ 0c ffi 2 for some 2 with deg 2 ~ 0. Therefore e ~ 0. All
values of e ~ 0 are possible.
(b) If <C is indecomposable, then - 2g ~ e ~ 2g - 2. (In fact, there
are even stronger restrictions on e (Ex. 2.5).)
PROOF. If <Cis decomposable, then <C ~ 2 1 ffi 2 2 for two invertible sheaves
2 1 and 2 2 on C. We must have deg 2i ~ 0 because of the normalization
(2.8) and furthermore H 0 (2i) # 0 for at least one of them. Thus one of
them is 0c, so we have <C ~ 0c ffi 2 with deg 2 ~ 0. From (2.11.1),
(2.11.2), and (2.11.3), we see that all values of e ~ 0 are possible.
Now suppose <C is indecomposable. Then, corresponding to the section
C 0 , we have an exact sequence
0 --+ 0c --+ <C --+ 2 --+ 0
for some 2 (2.8). This must be a nontrivial extension, so it corresponds to
a nonzero element~ E Ext 1 (2,0c) ~ H 1 (C,2~) (III, Ex. 6.1). In particular,
H 1 (2~) # 0, so we must have deg 2~ ~ 2g - 2 (IV, 1.3.4). Since e =
-deg 2, we have e ~ 2g - 2.
On the other hand, we have H 0 (<C ®A) = 0 for all deg A < 0 by the
normalization. In particular, taking deg A = -1, we have
0 = H 0 (<C ® A) --+ H 0 (2 ® A) --+ H 1 (A) --+ ... ,

so we must have
dim H 0 (2 ® A) ~ dim H 1(A).
Since deg A < 0, H 0 (A) = 0, so by Riemann-Roch, we have dim H 1 (A) = g.
On the other hand, also by Riemann-Roch, we have
dim H 0 (2 ® A) ~ deg 2 - 1 + 1 - g.
Combining, we get deg 2 ~ 2g, hence e ~ - 2g.

Corollary 2.13. If g = 0, then e ~ 0, and for each e ~ 0 there is exactly one

rational ruled surface with invariant e, given by <C = 0 ffi 0(- e) over
c= p1_
PROOF. If g = 0, case (b) of (2.12) cannot occur. Hence <C ~ 0c ffi 2. But
the only invertible sheaves on P 1 are 0(n) for n E Z (II, 6.4). So for each
e ~ 0 there is just one possibility.

Corollary 2.14. Every locally free sheaf <C of rank 2 on P 1 is decomposable.

PROOF. After tensoring with a suitable invertible sheaf, it becomes normal-
ized, in which case it is isomorphic to 0 ffi 0( -e) by (2.13). See (Ex. 2.6)
for a generalization.

376
 2 Ruled Surfaces

Theorem 2.15. If X is a ruled surface over an elliptic curve C, corresponding

to an indecomposable iff, then e = 0 or -1, and there is exactly one such
ruled surface over C for each of these two values of e.
PROOF. According to (2.12) we must have e = 0,-1,-2. If e = 0, then we
have an exact sequence
0 ~ (!)c ~ iff ~ !l' ~ 0
with deg !l' = 0. This extension corresponds to a nonzero element
~ E H 1 (!l'~ ). In particular, H 1 (!l'~) #- 0. It is dual to H 0 (!l'), so we must
have !l' ~ (!)c· Conversely, taking !l' = (!)c, we have dim H 1 ((!)c) = 1, so
there is just one choice of nonzero ~ E H 1 ((!)c), up to isomorphism, which is
a nontrivial extension
0 ~ (!)c ~ iff ~ (!)c ~ 0.
Clearly this iff is normalized. Furthermore, this iff is indecomposable,
because if iff were decomposable, being normalized, it would be isomorphic
to (!)c EB !l' for some !l', by (2.12). But f\ 2 iff ~ (!)c, so !l' ~ (!)c, so in fact
this extension would have to split, which it doesn't. Thus we get exactly
one elliptic ruled surface X with e = 0 and iff indecomposable.
If e = -1, then we have an exact sequence
0 ~ (!)c ~ iff ~ !l'(P) ~ 0
for some point P E C, because every invertible sheaf of degree 1 on C is of
the form !l'(P) (IV, 1.3.7). Furthermore, for each P there exists such a nor-
malized bundle, unique up to isomorphism, by (2.11.6). To show that there
is just one elliptic ruled surface with e = -1, it will be sufficient to show
that if iff is defined by P as above, and iff' is similarly defined by Q #- P, then
there exists an invertible sheaf A on C such that iff' ~ iff ® A.
Take a point R E C such that 2R ~ P + Q. This is possible, because
the linear system IP + Ql defines a two-to-one map of C to PI, ramified
at four points (assume char k #- 2), and we can take R to be one of them
(IV, §4). We will show that iff' ~ iff ® !l'(R - P). In any case, we have an
exact sequence
0 ~ !l'(R - P) ~ iff ® !l'(R - P) ~ !l'(R) ~ 0.
Since H 0 (!l'(R)) #-0 and H 1(!l'(R- P) )=0, we see that H 0 (iff®!l'(R- P)) #-0.
So we get an exact sequence
0 ~ (!)c ~ iff ® !l'(R - P) ~ % ~ 0,
and the quotient % must be invertible, as in the proof of (2.8). So we have
% ~ f\ 2 (iff ® !l'(R- P)) ~ (J\2 iff) ® !l'(2R- 2P).
Since f\ 2 iff ~ !l'(P), we have % ~ !l'(2R - P) ~ !l'(Q). Therefore iff®
!l'(R - P) ~ iff' as required. This proves the uniqueness of the elliptic
ruled surface with e = -1.

377
V Surfaces

Finally, we will show that the case e = -2 does not occur. If it did, we
would have a normalized bundle Iff with an exact sequence
0 --+ {!)c --+ Iff --+ st'(P + Q) --+ 0
for some P,Q E C, since every invertible sheaf of degree 2 is of the form
st'(P + Q). Now take any pair of points R,S E C with R + S "' P + Q, and
let A = Sf(- R). Then, since Iff is normalized, H 0 (tff ® A) = 0, so the map
y:H 0 (st'(P + Q- R))--+ H 1 (st'( -R))mustbeinjective. On the other hand,
let ~ E H 1(st'(- P - Q)) be the element defining the extension C. Then we
have a commutative diagram, writing st'(P + Q - R) as st'(S),

where c:5(1) = ~. ()((1) = t, a nonzero section defining the divisor S, and fJ is

induced from the map {!)c --+ st'(S) corresponding to t. Now fJ is dual to
the map
{J': H 0 (st'(R)) --+ H 0 (st'(P + Q))
also induced by t. The image of any nonzero element of H 0 (st'(R)) by {J' is
a section of H 0 (st'(P + Q)) corresponding to the effective divisor
R + s E IP + Ql.
By varying R and S, we get every divisor in the linear system IP + Ql.
Therefore the image of {J' as R varies fills up the whole 2-dimensional vector
space H 0 (st'(P + Q) ). In particular, we can chooseR so that the image of {J'
lands in the kernel of~. considered as a linear functional on H 0 (st'(P + Q) ).
In that case, {J(~) = 0, which contradicts the injectivity of y. Thus the case
e = -2 is impossible.

Caution 2.15.1. One point which came up in the first part of this proof should
be noted. It is possible for a locally free sheaf of rank 2 to be a nontrivial
·extension of two invertible sheaves, and yet be decomposable. For example,
the sheaf of differentials on P 1 is isomorphic to {!)(- 2), so we have an exact
sequence (II, 8.13)
0 --+ @(- 2) --+ @( -1) EB @( -1) --+ {!) --+ 0.
The sequence cannot be split (because for example H 0 ({!)( -1) EB {!)( -1)) = 0),
but the sheaf in the middle is decomposable.

Corollary 2.16 (Atiyah). For each integer n, there is a natural one-to-one cor-
respondence (described explicitly in the proof below) between the set of
isomorphism classes of indecomposable locally free sheaves of rank 2 and
degree non the elliptic curve C, and the set of points of C.

378
 2 Ruled Surfaces

PROOF. Fix a point P0 E C. Let <%'' be an indecomposable locally free sheaf

of rank 2 and degree n on C. Tensoring with !f'(mP 0 ) for some m, we may
assume n = 0 or 1. If n = 0, then by (2.2), (2.8), and (2.15) there is a (unique)
invertible sheaf !I' of degree 0 on C such that <%'' ® !I' is isomorphic to the
unique nontrivial extension of (!Jc by (!)c· Since the invertible sheaves of
degree 0 are in one-to-one correspondence with the closed points of C (IV, §4),
we have the result. If n = 1, then as in the proof of (2.15) we find that <%'' is
an extension of !f'(P) by (!)c for some uniquely determined point P E C,
whence the result.
Remark 2.16.1. More generally, for any curve C, of genus g, one can consider
the problem of classifying all locally free sheaves<%' on Cup to isomorphism.
The rank rand the degree d (which is deg /\'C) are numerical invariants. For
fixed r and d, one expects some kind of continuous family. For g = 0, all
locally free sheaves are direct sums of invertible sheaves (Ex. 2.6). For g = 1
the general classification, which is similar to the rank 2 case we have just
done, has been accomplished by Atiyah [ 1]. For g )!: 2, the situation be-
comes more complicated. Among the indecomposable locally free sheaves,
one has to distinguish between the stable ones (Ex. 2.8) in the sense of Mum-
ford [1 ], and the rest. The stable ones form nice algebraic families, whereas
the others do not. See for example, Narasimhan and Seshadri [1]. Similarly,
for the ruled surfaces themselves, the ones with e < 0 are stable, and form
nice algebraic families, but the others do not.
Next we will study the rational ruled surfaces, which were classified
in (2.13).
Theorem 2.17. Let Xe, for any e )!: 0, be the rational ruled surface defined by
<%' = (!) EB (!)(-e) on C = P 1 (2.13). Then:
(a) there is a section D ~ C0 + nf if and only if n = 0 or n )!: e. In
particular, there is a section C 1 ~ C0 + ef with C 0 n C 1 = 0 and
q = e;
(b) the linear system ICo + iiflis base-point-free if and only if n ?: e;
(c) the linear sys_tem ICo + iiflis very ample if and only if n > e.
PROOF.
(a) According to (2.6) and (2.9), giving a section D ~ C 0 + iif is equivalent
to giving a surjective map<%' --+ !I' --+ 0 with deg !I' = C 0 .D = n - e. Since
we are on P 1 , this means a surjective map
(!) EB (!)(-e)--+ (!)(n - e)--+ 0.
If n < e, there are no nonzero maps of(!) to (!)(n - e), so the map {!}(-e) --+
(!)(n - e) must be an isomorphism, and therefore n = 0. This corresponds
to the section C 0 , which is unique if e > 0. Otherwise we have n )!: e, and
any such n is possible. We have only to take maps (S --+ (!(n - e) and
(!)(-e) --+ (!)(n - e) corresponding to effective divisors of degrees n - e and
n on C which do not meet. Then the corresponding map (!' EB (!)(-e) --+
(!)(n - e) will be surjective.

379
V Surfaces

In particular, if we take n = e, there is a section C 1 "" C 0 + ef Then

Cf = e, and C 0 .C 1 = 0, so C 0 n C 1 = 0.
(b) If ICo + J?JI is base-point-free, then C 0 .(C 0 + nf) ?: 0 so n ?: e.
Conversely, ifn?: e, then C 0 + nf"" C 1 + (n- e)f, and since C0 n C 1 = 0,
and any f is linearly equivalent to any other, we can find a divisor of the form
C 0 + nf or C 1 + (n - e)f which misses any given point.
(c) If D = C 0 + nf is very ample, then we must have D.C 0 > 0, son > e.
Conversely, suppose n > e. Then we will show that D is very ample by
showing that the linear system IDI separates points and tangent vectors
(II, 7.8.2).
Case 1. Let P #- Q be two points not both in C 0 , and not both in any
fibre. Then a divisor of the form C 0 + nf for suitable f will separate them.
Case 2. Let P be a point and t a tangent vector at P, such that P,t are not
both in C 0 and not both in any fibre. Then a divisor of the form C 0 + Ii'=
1 };,
for suitable fibres};, will contain P but not t.
Case 3. Suppose P,Q or P,t are both in C 0 . Then a divisor of the form
cl + 2::?:1}; will separate them.
Case 4. Suppose P,Q, or P,t are both in the same fibre f Since D.f = 1,
the invertible sheaf 2(D) ® (!) 1 is very ample on f ~ P 1 . Thus to separate
P,Q or P,t, it will be sufficient to show that the natural map H 0 (X,2(D)) ---+
H 0 (f,2(D) ® (!) 1 ) is surjective. The co kernel of this map lands in
H 1 (X,2(D - f)), which by (2.4) is isomorphic to H 1 ( C,n*2(D - f)). On
the other hand, D - f "" C 0 + (n - 1)f, so
n*(2(D- f))~ n*(2(C 0 )) ® (!)c(n- 1)
by the projection formula (II, Ex. 5.1). Now n*(2(C 0 )) ~ ~ by (2.8) and
(II, 7.11), so we have
n*(2(D - f)) ~ (!)(n - 1) EB (!)(n - e - 1).
Since n > e ;:: 0, both n - 1 ;:: 0 and n - e - 1 ;:: 0, so H 1 = 0 and the
above map is surjective. q.e.d.

Corollary 2.18. Let D be the divisor aC 0 + bf on the rational ruled surface Xe,
e ;:: 0. Then:
(a) Dis very ample=- Dis ample=- a > 0 and b > ae;
(b) the linear system IDI contains an irreducible nonsingular curve=- it
contains an irreducible curve=- a = 0, b = 1 (namely f); or a = 1, b = 0
(namely C 0 ); or a > 0, b > ae; ore > 0, a > 0, b = ae.
PROOF.
(a) If D is very ample, it is certainly ample (II, 7.4.3). If D is ample,
then D.f > 0, so a > 0, and D.C 0 > 0, so b > ae (1.6.2). Now suppose
that a > 0 and b > ae. Then we can write D = (a - 1)(C0 + ~cf) +
(C 0 + (b - ae + e)f). Since ICo + efl has no base points, and C0 +
(b - ae + e)f is very ample (2.17), we conclude that D is also very ample
(II, Ex. 7.5).

380
 2 Ruled Surfaces

(b) If IDI contains an irreducible nonsingular curve, then in particular it

contains an irreducible curve. If D is an irreducible curve, then D could be
f (in which case a = 0, b = 1) or C 0 (in which case a = 1, b = 0). Other-
wise, n maps D surjectively to C, so D.f = a > 0, and D.C 0 :;:, 0, so b :;:, ae.
If e = 0 and b = ae, then b = 0 soD = aC 0 . But in this case X 0 is P 1 x P 1 ,
and C 0 is one of the rulings, so forD to be irreducible, we must have a = 1.
Thus the restrictions on a,b are necessary. To complete the proof, we must
show that if a > 0, b > ae, or e > 0, a > 0, b = ae, then IDI contains an
irreducible nonsingular curve. In the first case, Dis very ample by (a), so the
result follows from Bertini's theorem (II, 8.18) applied to X. In the second
case, we use the fact (2.11.4) that X e can be obtained from the cone Y over a
nonsingular rational curve C of degree e > 0 in some P", by blowing up the
vertex. In this case, the curve C 1 on X e is the strict transform of the hyper-
plane section H of Y. By Bertini's theorem applied to the very ample divisor
aH on Y (II, 8.18.1), we can find an irreducible nonsingular curve in the
linear system iaHI, notcontaining the vertex of Y. Its strict transform on xe
is then an irreducible nonsingular curve in the linear system laC 1 1 = IDI.

Remark 2.18.1. In case e = 0, we get some new proofs of earlier results about
curves on the nonsingular quadric surface, which is isomorphic to X 0
(II, 7.6.2), (III, Ex. 5.6), (1.10.1).

Corollary 2.19. For every n > e :;:, 0, there is an embedding of the rational
ruled surface X e as a rational scroll of degree d = 2n - e in pd+ 1 . (A
scroll is a ruled surface embedded in pN in such a way that all the fibres
f have degree 1.)
PROOF. Use the very ample divisor D = C0 + nf. Then D.f = 1, so the
image of X e in pN is a scroll, and D2 = 2n - e, so the image has degree
d = 2n - e. To find N, we compute H 0 (X,!f(D) ). As in the proof of (2.17),
we find that
H 0 (X,!f(D)) = H 0 (C,n*!f(D)) = H 0 (C,tff ® @(n)) = H 0 (@(n) E8 @(n- e)).
This has dimension 2n + 2 - e, so N = 2n + 1- e = d + 1.

Example 2.19.1. For e = 0, n = 1, we recover the nonsingular quadric sur-

face in P 3 .
For e = 1, n = 2, we get a rational scroll of degree 3 in P 4 , which is
isomorphic to P 2 with one point blown up (II, Ex. 7. 7).
In P 5 , there are two different kinds of rational scrolls of degree 4, cor-
responding to e = 0, n = 2, and e = 2, n = 3.

Remark 2.19.2. In fact, it is known that every nonsingular surface of degree d

in pd+ 1, not contained in any hyperplane, is either one of these rational
scrolls (2.19), or P 2 c;:; P 2 (if d = 1), or the Veronese surface in P 5 (I, Ex. 2.13).
See, for example, Nagata [5, I, Theorem 7, p. 365].

381
V Surfaces

Now we will try to determine the ample divisors on a ruled surface over a
curve of any genus, as an application ofNakai's criterion (1.10). In order to
apply Nakai's criterion, we need to know which numerical equivalence
classes of divisors on the surface contain an irreducible curve. On a general
ruled surface, we cannot expect to get nearly as precise an answer to this
question as in the case of the rational ruled surfaces (2.18), but at least we
can get some estimates which allow us to apply Nakai's criterion successfully.

Proposition 2.20. Let X be a ruled surface over a curve C, with invariant e ~ 0.

(a) If Y = aC 0 + bf is an irreducible curve # C 0 ,f, then a > 0, b ~ ae.
(b) A divisor D =
aC 0 + bf is ample if and only if a > 0, b > ae.

PROOF.
(a) Since Y # f, n: Y ~ C is surjective, so Y.f = a > 0. Also since
Y # C 0 , Y.C 0 = b - ae ~ 0.
(b) If D is ample, then D.f = a > 0, and D.C 0 = b - ae > 0. Con-
versely,ifa > O,b- ae > O,thenD.f > O,D.C 0 > O,D 2 = 2ab- a 2 e > 0
and if Y = a'C 0 + b'f is any irreducible curve # C 0 ,f, then

D.Y = ab' + a'b- aa'e > aa'e + aa'e- aa'e = aa'e ~ 0.

Therefore by (1.10) Dis ample.

Proposition 2.21. Let X be a ruled surface over a curve C of genus g, with in-
variant e < 0, and assume furthermore either char k = 0 or g ::::;; 1.
(a) If Y =
aC 0 + bf is an irreducible curve =I= C 0 ,f, then either a = 1,
b ~ 0 or a ~ 2, b ~ !ae.
(b) A divisor D = aC 0 + bf is ample if and only if a > 0, b > !ae.

PROOF.
(a) We will use Hurwitz's theorem (IV, 2.4) to get some information about
Y. Let Y be the normalization of Y, and consider the composition of the
natural map Y ~ Y with the projection n: Y ~ C. If char k = 0, this map
is a finite, separable map of degree a, so by (IV, 2.4) we have

2g(Y) - 2 = a(2g - 2) + deg R,

where R is the(effective) ramification divisor. On the other hand,pa(Y) ~g(Y)
by (IV, Ex. 1.8), so we find that

2pa(Y) - 2 ~ a(2g - 2).

Furthermore, this last inequality is true in any characteristic if g = 0,1,

since in any case pa(Y) ~ g (IV, 2.5.4).
By the adjunction formula (1.5), we have
2pa(Y) - 2 = Y.(Y + K).

382
 2 Ruled Surfaces

Substituting Y =
aC 0 + bf and K =
-2C 0 + (2g - 2 - e)f from (2.11),
and combining with the inequality above, we find that
b(a - 1) ;:?: iae(a - 1).
Therefore if a ;:?: 2, we have b ;:?: iae as required. Now Y.f = a > 0 in any
case, so it remains to show that if a = 1, then b ;:::: 0. In the case a = 1, Y is
a section, corresponding to a surjective map rff --+ ff! --+ 0. Because of the
normalization of rff, we must have deg ff! ;:?: deg rff. But deg ff! = C 0 . Y
(2.9), so we have b - e ;:::: - e, hence b ;:::: 0.
(b) If D is ample, then D.f = a > 0, and D 2 = 2ab - a 2 e > 0, so
b > iae. Conversely, if a > 0, b > iae, then D.f > 0, D 2 > 0, D.C 0 =
b - ae > -iae > 0, and if Y =
a'C 0 + b'f is any irreducible curve
=f. C 0 ,f, then
D.Y = ab' + a'b - aa'e.
Now if a' = 1, then b' ;:?: 0, so D. Y > iae - ae = -iae > 0. If a' ;:?: 2,
then b' ;:?: !a' e, so D. Y > iaa' e + iaa' e - aa' e = 0. Therefore by (1.10),
Dis ample.

Remark 2.21.1. In the remaining case e < 0, char k = p > 0, g ;:?: 2, we can-
not get necessary and sufficient conditions forD to be ample, but it is possible
to get some partial results (Ex. 2.14) and (Ex. 2.15).

Remark 2.22.2. The determination of the very ample divisors on a ruled sur-
face with g ;:?: 1 is more subtle than in the rational case (2.18), because it
does not depend only on the numerical equivalence class of the divisor
(Ex. 2.11) and (Ex. 2.12).

References for §2. Since the theory of ruled surfaces is very old, I cannot
trace the origins of the results given here. Instead, let me simply list a few
recent references: Atiyah [1], Hartshorne [4], Maruyama [1], Nagata [5],
Shafarevich [1, Ch. IV, V], Tjurin [1], [2].

EXERCISES
2.1. If X is a birationally ruled surface, show that the curve C, such that X is birationally
equivalent to C x P 1 , is unique (up to isomorphism).

2.2. Let X be the ruled surface P(6') over a curve C. Show that @' is decomposable if
and only if there exist two sections C',C" of X such that C' n C" = 0.

2.3. (a) If@' is a locally free sheaf of rank ron a (nonsingular) curve C, then there is a
sequence
Q = @' 0 s; @'I s; · · · s; @', = @'

of subsheaves such that 6';/6';_ 1 is an invertible sheaf for each i = 1, ... ,r. We
say that@' is a successive extension of invertible sheaves. [Hint: Use (II, Ex. 8.2).]

383
V Surfaces

(b) Show that this is false for varieties of dimension ;::, 2. In particular, the sheaf of
differentials Q on P 2 is not an extension of invertible sheaves.
2.4. Let C be a curve of genus g, and let X be the ruled surface C x P 1 . We consider
the question, for what integers s E Z does there exist a section D of X with D 2 = s?
First show that s is always an even integer, say s = 2r.
(a) Show that r = 0 and any r ;::, g + 1 are always possible. Cf. (IV, Ex. 6.8).
(b) If g = 3, show that r = 1 is not possible, and just one of the two values r = 2,3
is possible, depending on whether C is hyperelliptic or not.
2.5. Values of e. Let C be a curve of genus g ;::, 1.
(a) Show that for each 0 ~ e ~ 2g - 2 there is a ruled surface X over C with
invariant e, corresponding to an indecomposable t£. Cf. (2.12).
(b) Let e < 0, let D be any divisor of degree d = - e, and let~ E H 1(£'(- D)) be a
nonzero element defining an extension
0 --> (!)c --> t£ --> !l'(D) --+ 0.
Let H <;::: ID + Kl be the sub linear system of codimension 1 defined by ker ~.
where ~ is considered as a linear functional on H 0 (!l'(D + K) ). For any
effective divisor E of degree d - 1, let LE <;::: ID + Kl be the sublinear system
ID + K - El + E. Show that t£ is normalized if and only if for each E as
above, LE 'j. H. Cf. proof of (2.15).
(c) Now show that if -g ~ e < 0, there exists a ruled surface X over C with
invariant e. [Hint: For any given Din (b), show that a suitable~ exists, using
an argument similar to the proof of (II, 8.18).]
(d) For g = 2, show that e;::, -2 is also necessary for the existence of X.
Note. It has been shown that e ;::, -g for any ruled surface (Nagata [8]).
2.6. Show that every locally free sheaf of finite rank on P 1 is isomorphic to a direct
sum of invertible sheaves. [Hint: Choose a subinveitible sheaf of maximal degree,
and use induction on the rank.]
2.7. On the elliptic ruled surface X of(2.11.6), show that the sections C 0 with C6 = 1
form a one-dimensional algebraic family, parametrized by the points of the base
curve C, and that no two are linearly equivalent. ·
2.8. A locally free sheaf t£ on a curve Cis said to be stable if for every quotient locally
free sheaf t£ --+ ff -> 0, ff ¥ t£, ff ¥ 0, we have
(deg ff)/rank ff > (deg tff')/rank t£.
Replacing > by ;::, defines semistable.
(a) A decomposable t£ is never stable.
(b) If t£ has rank 2 and is normalized, then t£ is stable (respectively, semistable) if
and only if deg t£ > 0 (respectively, ;::,0).
(c) Show that the indecomposable locally free sheaves t£ of rank 2 that are not
semistable are classified, up to isomorphism, by giving (1) an integer 0 < e ~
2g - 2, (2) an element£' EPic C of degree - e, and (3) a nonzero~ E H 1 (£'~),
determined up to a nonzero scalar multiple.
2.9. Let Y be a nonsingular curve on a quadric cone X 0 in P 3 . Show that either Y is a
complete intersection of X 0 with a surface of degree a ;::, 1, in which case deg Y =
2a, g(Y) = (a - 1) 2 , or, deg Y is odd, say 2a + 1, and g(Y) = a 2 - a. Cf.
(IV, 6.4.1). [Hint: Use (2.11.4).j

384
 2 Ruled Surfaces

2.10. For any n > e ~ 0, let X be the rational scroll of degree d = 2n - e in pd+ 1
given by (2.19). If n ~ 2e - 2, show that X contains a nonsingular curve Y of
genus g = d + 2 which is a canonical curve in this embedding. Conclude that for
every g ~ 4, there exists a nonhyperelliptic curve of genus g which has a g~.
Cf. (IV, §5).
2.11. Let X be a ruled surface over the curve C, defined by a normalized bundle Iff,
and let e be the divisor on C for which 2'(e) ~ f\2 tC (2.8.1). Let b be any divisor on
c.
(a) If lbl and lb + el have no base points, and if b is nonspecial, then there is a
section D - C 0 + bf, and IDI has no base points.
(b) lfb and b + e are very ample on C, and for every point P E C, we have b - P
and b + e - P nonspecial, then C 0 + bf is very ample.
2.12. Let X be a ruled surface with invariant e over an elliptic curve C, and let b be a
divisor on C.
(a) If deg b ~ e + 2, then there is a section D- C 0 + bf such that IDI has no
base points.
(b) The linear system !Co + bfl is very ample if and only if deg b ~ e + 3.
Note. The case e = -1 will require special attention.

2.13. For every e p -1 and n ~ e + 3, there is an elliptic scroll of degree d = 2n - e

in pd- 1. In particular, there is an elliptic scroll of degree 5 in P4 .
2.14. Let X be a ruled surface over a curve C of genus g, with invariant e < 0, and assume
that char k = p > 0 and g ~ 2.
(a) If Y = aC 0 + bf is an irreducible curve i= C 0 ,f, then either a = 1, b ~ 0, or
2 ~ a ~ p - 1, b ~ !ae, or a ~ p, b ~ !ae + 1 - g.
(b) If a > 0 and b > a(!e + (1/p)(g - 1) ), then any divisor D = aC 0 + bf is
ample. On the other hand, if Dis ample, then a > 0 and b > !ae.
2.15. Funny behavior in characteristic p. Let C be the plane curve x 3 y + y 3 z + z3 x = 0
over a field k of characteristic 3 (IV, Ex. 2.4).
(a) Show that the action of the k-linear Frobenius morphism f on H 1 (C,f9c) is
identically 0 (Cf. (IV, 4.21) ).
(b) Fix a point P E C, and 8how that there is a nonzero~ E H 1(2'(- P)) such that
f*~ = 0 in H 1(2'(- 3P) ).
(c) Now let Iff be defined by~ as an extension

0 -> f9c -> Iff -> 2'(P) -> 0,

and let X be the corresponding ruled surface over C. Show that X contains a
nonsingular curve Y =
3C 0 - 3/, such that n: Y-> Cis purely inseparable.
Show that the divisor D = 2C 0 satisfies the hypotheses of(2.21b), but is not
ample.
2.16. Let C be a nonsingular affine curve. Show that two locally free sheaves tff,tff' of
the same rank are isomorphic if and only if their classes in the Grothendieck group
K(X) (II, Ex. 6.10) and (II, Ex. 6.11) are the same. This is false for a projective curve.
*2.17. (a) Let tp:Pl -> Pt be the 3-uple embedding (1, Ex. 2.12). Let f be the sheaf of
ideals of the twisted cubic curve C which is the image of qJ. Then f / f 2 is a
locally free sheaf of rank 2 on C, so tp*(f /f 2 ) is a locally free sheaf of rank 2 on

385
V Surfaces

~ (r)(l) EB (r)(m)for some l,m E Z. Determine

P 1 . By(2.14), therefore, cp*(J/J 2 )
land m.
(b) Repeat part (a) for the embedding cp:P 1 -> P 3 given by x 0 = t 4 , x 1 = t 3 u,
x 2 = tu 3 , x3 = u4 , whose image is a nonsingular rational quartic curve.
[Answer: If char k =1= 2, then l = m = -7; if char k = 2, then l,m = -6, -8.]

3 Monoidal Transformations

We define a monoidal transformation of a surface X to be the operation of

blowing up a single point P. This new terminology is to distinguish it from
the more general process of blowing up an arbitrary closed subscheme
(II, §7). It also goes by many other names in the literature: locally quadratic
transformation, dilatation, a-process, Hopf map, to mention a few.
We will see later (5.5) that any birational transformation of surfaces can be
factored into monoidal transformations and their inverses. Thus the mono-
ida! transformation is basic to the birational study of surfaces.
In this section we will study what happens under a single monoidal
transformation. As an application, we will show how to resolve the singulari-
ties of a curve on a surface by monoidal transformations, and begin a study
of the different types of curve singularities.

First we fix our notation. Let X be a surface, and let P be a point of X.

We denote the monoidal transformation with center P by n:X--+ X. Then
we know (1, §4) or (II, §7) that n induces an isomorphism of X - n - 1 (P) onto
X - P. The inverse image of P is a curve E, which we call the exceptional
curve (1, 4.9.1).

Proposition 3.1. The new variety X is a nonsingular projective swface. The

curve E is isomorphic to P 1 . The self-intersection of E on X is E 2 = - 1.

PROOF. Since a single point is nonsingular, we can apply (II, 8.24). This tells
us that X is nonsingular, and we know already from (II, 7.16) that X is
projective, of dimension 2, and birational to X. We also conclude from
(II, 8.24) that E ;;;: P 1 , since it is the projective space bundle over the point P
corresponding to the two-dimensional vector space mpjm~. Finally, the
normal sheaf JV E/X is just (DE( -1), so by (1.4.1) we have E 2 = -1.

Remark 3.1.1. There is a converse to this result, which we will prove later
(5.7), namely, any curve E ;;;: P 1 in a surface X', with E 2 = -1, is obtained as
the exceptional curve by a monoidal transformation from some other
surface X.

Proposition 3.2. The natural maps n*: Pic X --+ Pic X and Z --+ Pic X defined
by 1 ~ 1 · E give rise to an isomorphism Pic X ;;;: Pic X E8 Z. The inter-

386
 3 Monoidal Transformations

section theory on X is determined by the rules:

(a) if C,D EPic X, then (n*C).(n*D) = C.D;
(b) if C EPic X, then (n*C).E = 0;
(c) E2 = -1.
Finally, if n*: Pic X --> Pic X denotes the projection on the first factor,
then:
(d) ifC EPic X and DE Pic X, then (n*C).D = C.(n*D).
PROOF. (See also (II, Ex. 8.5).) From (II, 6.5) we see that Pic X ~ Pic( X - P).
But X - P ~ X - E, so also from (II, 6.5) we have an exact sequence
Z --> Pic X --> Pic X --> 0,
where the first map sends 1 to 1· E. Since for any n # 0 we have
(nE) 2 = - n 2 # 0, this map is injective. On the other hand, n* splits this
sequence, so we have Pic X ~ Pic X EB Z.
We have already seen that E 2 = -1. To prove (a) and (b), we use the
fact (§1) that C and D are linearly equivalent to differences of nonsingular
curves, meeting everywhere transversally, and not containing P. For in the
proof of(1.2) we can require also that D' misses any given finite set of points.
Then n* does not affect their intersection, which proves (a). Also clearly
n*C does not meet E, So (n*C).E = 0. The same argument also proves (d),
because we may assume that Cis a difference of curves not containing P.

Proposition 3.3. The canonical divisor of X is given by Kx = n* Kx + E.

Therefore Kj = KI - 1.
PROOF. (See also (II, Ex. 8.5).) Since the canonical sheaf on X - E and X - P
is the same, clearly Kx = n* Kx + nE for some n E Z. To determine n, we
use the adjunction formula (1.5) for E . .It says -2 = E.(E + Kg), so using
(3.2) we find n = 1. The formula for K 2 follows directly from (3.2).

Remark 3.3.1. Thus the invariant K 2 of a surface is not a birational invariant.

For a specific example, we have K 2 ofP 2 is 9 (1.4.4), K 2 of the rational ruled
surface X 1 is 8 (2.11), and X 1 is isomorphic to a monoidal transformation
of P 2 (2.11.5).

Next we want to show that the arithmetic genus Pa is preserved by a

monoidal transformation. For that, we must compare the cohomology of
the structure sheaves on X and X. We will use the theorem on formal
functions (III, 11.1) to compute R;n/9x.

Proposition 3.4. We have n*(l)x = (!)x, and R;n*(!)x = 0 fori > 0. Therefore
H;(X,(!)x) ~ H;(X,(!)x) for all i ;::: 0.
PROOF. Since n is an isomorphism of X - E onto X - P, it is clear that the
natural map (!)x --> n*(!)x is an isomorphism except possibly at P, and that
the sheaves :F; = R;n*(!)x fori > 0 have support at P. We use the theorem on

387
V Surfaces

formal functions (III, 11.1) to compute these !Fi. It says (taking completions
of the stalks at P) that
/#i :-: : : lim
<E---
H;(E n' (!) En )
where En is the closed subscheme of X defined by ,r, where f is the ideal
of E. There are natural exact sequences
0 ----> fn / fn + 1 ----> {!)En+ 1 ----> {!)En ----> 0
for each n. Furthermore, by (II, 8.24) we have f// 2 = (!)E(1), and by
(II,8.21Ae),fn/fn+l;::; S\f// 2 );::; (!)E(n). NowE;::; P\soHi(E,(!)E(n)) =
0 for i > 0 and all n > 0. Since E 1 = E, we conclude from the long exact
sequence of cohomology, using induction on n, that Hi((!)EJ = 0 for all i > 0,
all n ~ 1. It follows that /#i = 0 fori > 0. Since !Fi is a coherent sheaf with
support at P, !Fi = /#i, so !Fi = 0.
The fact that (!)x ;::; n*(!)x follows simply from the fact that X is normal
and n is birational. Cf. proof of (III, 11.4).
Now from (III, Ex. 8.1) we conclude that Hi(X,(!)x) ;::; Hi(X,(!)g) for all
i ~ 0.

Corollary 3.5. Let n:X ---->X be a monoidal transformation. Then pa(X) =

Pa(X).
PROOF. From (III, Ex. 5.3) we have Pa(X) = dim H 2 (X,(!)x) - dim H 1 (X,(!)x)
and similarly for Pa(X).

Remark 3.5.1. It follows also from (3.4) that X and the same ir- X have
regularity q(X) = dim H (X,(!)x) and the same geometric
1 genus p9 (X) =
dim H (X,(!)x) (III, 7.12.3). The invariance
2 of p9 is also of course a con-
sequence of the fact that p9 is a birational invariant in general (II, 8.19).

Next we will investigate what happens to a curve under a monoidal

transformation. Let C be an effective divisor on X, and let n:X----> X be
the monoidal transformation with center P. Recall that the strict transform
C of C is defined as the closed subscheme of X obtained by blowing up P
on C (II, 7.15). It is also the closure in X ofn- 1(C n (X- P)) (II, 7.15.1).
So it is clear that C can be obtained from n*C by throwing away E (with
whatever multiplicity it has in n*C).
The multiplicity of E in n*C will depend on the behavior of Cat the point
P. So we make the following definition of the multiplicity, which generalizes
the definition for plane curves given in (I, Ex. 5.3).

Definition. Let C be an effective Cartier divisor on the surface X, and let f

be a local equation for C at the point P. Then we define the multiplicity
of Cat P, denoted by JiAC), to be the largest integer r such that f Em~,
where mp <;: (!)P,x is the maximal ideal.

388
 3 Monoidal Transformations

Remark 3.5.2. We always have Jl-p{C) ~ 0, since f E r9p,x· Furthermore

f.lp( C) ~ 1 if and only if P E C, and equality holds if and only if C is non-
.
singular at P, because in that case mp c will be a principal ideal, so f ¢ m~ x· .
Proposition 3.6. Let C be an effective divisor on X, let P be a point of multi-
plicity r on C, and let n: X ---+ X be the monoidal transformation with
center P. Then
n*C = C + rE.

PROOF. We will go back to the definition of blowing up, and compute ex-
plicitly what happens in a neighborhood of P, so that we can trace the local
equation of Con X and n*C on X.
Let m be the sheaf of ideals of P on X. Then X is defined as Proj !/',
where !/' is the graded sheaf of algebras !/' = ffid ~ 0md (II, §7). Let x, y be
local parameters at P. Then x,y generate min some neighborhood U of P,
which we may assume to be affine, say U = Spec A. The Koszul complex
(III, 7.10A) gives a resolution ofm over U:

where we denote the two generators of@~ by t,u, and send t to x, u to y. Then
the kernel is generated by ty - ux. Therefore!/' over U is the sheaf associated
to the A-algebra A[t,u ]/(ty - ux), so X is the closed subscheme ofPb defined
by ty - ux, where t,u are the homogeneous coordinates of P 1 . (Note how
this construction generalizes the example (1, 4.9.1).)
Now let f be a local equation for C on U (shrinking U if necessary).
Then by definition of the multiplicity, we can write

f = .f..(x,y) + g
where f.. is a nonzero homogeneous polynomial of degree r with coefficients
ink, and gEm~+ 1 . Indeed, f Em', f ¢ m'+ 1 , and m'/m'+ 1 is the k-vector
space with basis x',x'- 1 y, ... ,y'.
Consider the open affine subset V of Pb defined by t = 1. Then on
X n V we have y = ux, so we can write
n*f = x'(.f..(1,u) + xh)

for some hE A[ u]. Indeed, m'+ 1 A[ u] is generated by x'+ 1 ,x'+ 1 u, ... ,

x'+ 1 u'+ 1 , so n*g is divisible by x'+ 1 .
Now x is a local equation forE, and .f..(1,u) is zero at only finitely many
points of E, so we see that E occurs with multiplicity exactly r in n*C, which
is locally defined by n*f

Corollary 3.7. With the same hypotheses, we have C.E = r, and Pa(C) =
Pa(C) - tr(r - 1).

389
V Surfaces

PROOF. Since C = n*e - rE, we have C.E = r by (3.2). We compute Pa(C)

by the adjunction formula (1.5) and (Ex. 1.3)
2pa(C) - 2 = C.(C + Kg)
= (n*e - rE)(n*e - rE + n* Kx + E)
= 2pa(e) - 2 - r(r - 1),
so
-
Pa(e) = Pa(e) - 21 r(r - 1).

Proposition 3.8. Let e be an irreducible curve in the surface X. Then there

exists a finite sequence of monoidal transformations (with suitable centers)
Xn ~ Xn_ 1 ~ . . . ~X 1 ~X 0 = X such that the strict transform en of e
on X n is nonsingular.
PROOF. If e is already nonsingular, take n = 0. Otherwise, let P E e be a
singular point, and let r ): 2 be its multiplicity. Let X 1 ~ X be the monoidal
transformation with center P, and let e 1 be the strict transform of C. Then
from (3.7) we see that pa(e 1 ) < Pa(e). If e 1 is nonsingular, stop. Otherwise
choose a singular point of e 1 and continue. In this way we obtain a sequence
of monoidal transformations

such that the strict transform ei of e on Xi satisfies Pa(eJ < pa(ei- d for
each i. Since the arithmetic genus of any irreducible curve is nonnegative
(Pa(eJ =dim H 1 (@c) (III, Ex. 5.3)), this process must terminate. Thus for
some n, en is nonsingular.

Remark 3.8.1. The general problem of resolution of singularities is, given a

variety V, to find a proper birational morphism f: V' ~ V with V' non-
singular. If V is a curve, we know this is possible, because each birational
equivalence class of curves contains a unique nonsingular projective curve
(I, 6.11). In fact, in this case it is sufficient to take V' to be the normalization
of V. But in higher dimensions, this method does not work.
So we approach the general problem as follows. Whenever Vis singular,
blow up some subvariety contained in the singular locus, to get a morphism
f 1 : V1 ~ V. Then (and this is the hard part) find some quantitative way of
showing that the singularities of V1 are less severe than those of V, so that as
we repeat this process, we must eventually obtain a nonsingular variety.
Two things become clear quite soon: first, to maintain reasonable control
of the singularities, one should only blow up subvarieties which are them-
selves nonsingular (such as a point); second, to set up an induction on the
dimension of V, one should also consider the problem of embedded resolution.
This problem is, given a variety V, contained in a nonsingular variety W, to
find a proper birational morphism g: W' ~ W with W' nonsingular, such

390
 3 Monoidal Transformations

that not only is the strict transform V of V in W' nonsingular, but the entire
inverse image g- 1 (V) is a divisor with normal crossings, which means that
each irreducible component of g- 1 (V) is nonsingular, and whenever r irre-
ducible components Y1 , . . . , Y,. of g- 1 (V) meet at a point P, then the local
equationsf1 , . . . ,/,.of the~ form part of a regular system of parameters at P
(i.e., f 1 , . . . ,/,.are linearly independent (mod m;) ).
The result just proved (3.8) shows that if C is a curve contained in a non-
singular surface, then one can resolve the singularities of C by successive
monoidal transformations. We will prove the stronger theorem of embedded
resolution for curves in surfaces below (3.9).
The status of the general resolution problem is as follows. The resolution
of curves was known in the late 19th century. The resolution of surfaces
(over C) was known to the Italians, but the first "rigorous" proof was given
by Walker in 1935. Zariski gave the first purely algebraic proof of resolution
for surfaces (char k = 0) in 1939. Then in 1944 he proved embedded reso-
lution for surfaces and resolution for threefolds (char k = 0). Abhyankar
proved resolution for surfaces in characteristic p > 0 in 1956, and in 1966
he proved resolution for threefolds in characteristic p > 5. Meanwhile in
1964 Hironaka proved resolution and embedded resolution in all dimensions
in characteristic 0. For more details and precise references on the resolution
problem, see Lipman [1], Hironaka [ 4] and Hironaka's introduction to
Zariski's collected papers on resolution in Zariski [8].

Theorem 3.9 (Embedded Resolution of Curves in Surfaces). Let Y be any

curve in the surface X. Then there exists a finite sequence of monoidal
transformations X'= X"--+ Xn-t--+ . . . --+ X 0 =X, such that if f:X'--+ X
is their composition, then the total inverse image f- 1(¥) is a divisor with
normal crossings (3.8.1).
PROOF. Clearly we may assume that Y is connected. Furthermore, since the
multiplicities of the irreducible components do not enter into the definition
of normal crossings, we may assume that Y is reduced, i.e., each irreducible
component has multiplicity 1. Now for any birational morphism f: X' --+ X,
let us denote by f- 1 (Y) the reduced inverse image divisor f*(Y)red· In other
words, f- 1(Y) is the sum of all the irreducible components of f*(Y), with
multiplicity 1. If f is a composition of monoidal transformations, then
f- 1(¥) will also be reduced and connected, so H 0((!!1 _,(YJ) = k, and
Pa(f- 1(¥)) = dimH 1(@f_'(Yl) ~ 0.
Let n:X --+X be the monoidal transformation at a point P, and let
Jlp(Y) = r. Then the divisor n- 1(¥) is just Y + E = n*(Y) - (r- l)E, by
(3.6), so we can easily compute the arithmetic genus, using the adjunction
formula as in (3.7). We find
1
Pa(n- 1 (Y)) = Pa(Y) - 2 (r - l)(r - 2).

391
V Surfaces

To prove our result, we proceed as follows. First we apply (3.8) to each

irreducible component of Y. Thus we reduce to the case where each irre-
ducible component of Y is nonsingular, because all the new exceptional
curves we add are already nonsingular. Then, if the total curve Y has a
singular point P other than a node, we blow it up.
If f!p(Y) ?: 3, then Pa(n- 1(Y)) < pu(Y), so there can be only finitely
many steps of this kind. If ,Up(Y) = 2, then Pa(n- 1 (Y)) = Pa(Y), and we
must look more closely. In that case we have Y.E = 2, by (3.7). There are
three possibilities. One is that Y meets E transversally in two distinct points,
in which case we can stop. The second is that Y meets E in one point Q, Y
is nonsingular there, but Y and E have intersection multiplicity 2 at Q. In
this case, blowing up Q produces a triple point (check!), so one further
blowing-up makes Pa(Y) drop again. The third possibility is that Y has a
singular point Q of multiplicity 2 where it meets E. In this case Y + E has
multiplicity 3 at Q, so blowing up Q makes Pa drop again.
So we see that any kind of singularity except a node gives rise to a
monoidal transformation, or a finite sequence of such, which forces Pa to
drop. Therefore the process must terminate. When it does, f- 1 (Y) will be
a divisor with normal crossings, because each irreducible component is non-
singular, and the only singularities of the total curve f- 1 (Y) are nodes.

Example 3.9.1. If Y is the plane cuspidal curve y 2 = x 3 , then the singularity

of Y is resolved by one monoidal transformation. However, to get f- 1(Y)
to have normal crossings, we need three monoidal transformations (Fig. 20).

Figure 20. Embedded resolution of a cusp.

In the context of successive monoidal transformations, it is convenient

to introduce the language of infinitely near points.

Definition. Let X be a surface. Then any point on any surface X', obtained
from X by a finite succession of monoidal transformations, is called an
infinitely near point of X. If g: X" -+ X' is a further succession of monoidal
transformations, and if Q" E X" is a point in the open set where g is an
isomorphism, then we identify Q" with g(Q") as infinitely near points
of X. In particular, all the ordinary points of X are included among
the infinitely near points. We say "Q is infinitely near P" if P lies on
some X' and Q lies on the exceptional curve E obtained by blowing up
P. If C is a curve in X, and Q' E X' is an infinitely near point of X, we

392
 3 Monoidal Transformations

say Q' is an infinitely near point of C if Q' lies on the strict transform of
Con X'.

Example 3.9.2. Let C be an irreducible curve on a surface X, with nor-

malization C. Then we have
- 1
g(C) = Pa(C)- ~lrp(rp- 1),

where rp is the multiplicity, and the sum is taken over all singular points
P of C, including infinitely near singular points. Indeed, by (3.8) we pass
from C to C by blowing up the singular points in succession, until there
are none left. Each time, by (3.7), the arithmetic genus drops by !r(r - 1).

Example 3.9.3. In particular, working with the infinitesimal neighborhood

of one point at a time, we see that the integer bp of (IV, Ex. 1.8) can be com-
puted as I!rQ(rQ - 1) taken over all infinitely near singular points Q
lying over P, including P.

Remark 3.9.4 (Classification of Curve Singularities). With the ideas of this

section we can begin a new classification of the possible singularities of a
(reduced) curve lying on a surface, which is weaker than the classification
by analytic isomorphism introduced in (I, 5.6.1) and (I, Ex. 5.14)-see
(Ex. 3.6). For references, see Walker [1, Ch. III, §7] and Zariski [10, Ch. I].
As a first invariant of a singular point P on a curve C (lying always on a
surface X) we have its multiplicity. Next we have the multiplicities of the
infinitely near singular points of C, and their configuration around P.
This data already suffices to determine bp (3.9.3).
We define a slightly more complex, but still discrete, invariant of a
singular point (or set of singular points), to be its equivalence class for the
following equivalence relation. A (reduced) curve C in an open set U of a
surface X is equivalent to another C' ~ U' ~ X' if there is a sequence of
monoidal transformations un ~ un- 1 ~ . . . ~ u 0 = u and another u~ ~
U~_ 1 ~ ••. ~ U 0 = U', which give embedded resolutions for C and C'
respectively, and if there is a one-to-one correspondence between the ir-
reducible components of the reduced total transforms and their singular
points at each step, preserving multiplicities, incidence, and compatible
with the maps U; ~ U;_ 1 and u; ~ u;_ 1 for each i. One checks easily
that this is in fact an equivalence relation. To define equivalence of a single
singular point P E C, just take U so small that C n U has no other singular
points.

Example 3.9.5. To illustrate this concept, let us classify all double points
up to equivalence. Let P E C be a double point, and let n:X ~X be the
monoidal transformation with center P. Then as in the proof of (3.9), there
are three possibilities: (a) C meets E transversally in two points, in which

393
V Surfaces

case P is a node; (b) C meets E in one point, and is tangent to it there, in

which case P is a cusp; (c) C is singular, with a double point Q. In this
case E must pass through Q in a direction not equal to any tangent direction
of Q, since C.E = 2. Therefore the equivalence class of Q determines the
equivalence class of P.
Thus we can classify double points according to the number of times n
we must blow up to get C nonsingular, and the behavior (a) or (b) at the
last step.
In this case it happens that the classification for equivalence coincides
with the classification for analytic isomorphism, although that is not true
in general (Ex. 3.6). Indeed, up to analytic isomorphism, any double point
is given by y 2 = x' for some r ;:::;: 2 (I, Ex. 5.14d). To blow up, set y = ux.
Then we get u2 = x'- 2 • So we see inductively that the equivalence class
is given by n = [r/2] and the type is (a) if r is even, (b) if r is odd.

ExERCISES

3.1. Let X be a nonsingular projective variety of any dimension, let Y be a nonsingular

subvariety, and let n:X-+ X be obtained by blowing up Y. Show that PaC¥) =
Pa(X).
3.2. Let C and D be curves on a surface X, meeting at a point P. Let n:X ...... X be
the monoidal transformation with center P. Show that C.D = C.D - Jlp( C)· Jlp(D).
Conclude that C.D = LJlp(C) · Jlp(D), where the sum is taken over all intersection
points of C and D, including infinitely near intersection points.
3.3. Let n:X -+X be a monoidal transformation, and let D be a very ample divisor
on X. Show that 2n*D - E is ample on X. [Hint: Use a suitable generalization
of (1, Ex. 7.5) to curves in P".]
3.4. Multiplicity of a Local Ring. (See Nagata [7, Ch III, §23] or Zariski-Samuel
[1, vol 2, Ch VIII, §10].) Let A be a noetherian local ring with maximal ideal m.
For any l > 0, let 1/J(I) = length(A/m1). We callljl the Hilbert-Samuel function of A.
(a) Show that there is a polynomial P A(z) E Q [ z] such that P A(l) = 1/J(l) for all
l » 0. This is the Hilbert-Samuel polynomial of A. [Hint: Consider the graded
ring grm A = EB D 0 md /md+ 1, and apply (1, 7.5).]
(b) Show that deg P A = dim A.
(c) Let n = dim A. Then we define the multiplicity of A, denoted Jl(A), to be (n!) ·
(leading coefficient of P A)· If P is a point on a noetherian scheme X, we define
the multiplicity of P on X, Jlp{X), to be Jl((!)P.x).
(d) Show that for a point P on a curve Con a surface X, this definition of Jlp(C)
coincides with the one in the text just before (3.5.2).
(e) If Y is a variety of degree din P", show that the vertex of the cone over Y is a
point of multiplicity d.
3.5. Let a 1 , . . . ,a, r ~ 5, be distinct elements of k, and let C be the curve in P 2 given
by the (affine) equation y 2 = fli= 1 (x - aJ Show that the point P at infinity
on they-axis is a singular point. Compute bp and g(Y), where Y is the normaliza-
tion of Y. Show in this way that one obtains hyperelliptic curves of every genus
g ~ 2.

394
 4 The Cubic Surface in P 3

3.6. Show that analytically isomorphic curve singularities (1, 5.6.1) are equivalent in
the sense of (3.9.4), but not conversely.
3.7. For each of the following singularities at (0,0) in the plane, give an embedded
resolution, compute Op, and decide which ones are equivalent.
(a) x 3 + y 5 = 0.
(b) x3 + x 4 + y 5 = 0.
(c) x3 + Y4 + ys = 0.
(d) x3 + ys + y6 = 0.
(e) x3 + xy 3 + y 5 = 0.
3.8. Show that the. following two singularities have the same multiplicity, and the
same configuration of infinitely near singular points with the same multiplicities,
hence the same Op, but are not equivalent.
(a) x 4 - xy4 = 0.
(b) x4 - xzl - xzys + YB = 0.

4 The Cubic Surface in P 3

In this section, as in §2, we consider a very special class of surfaces, to illus-

trate some general principles. Our main result is that the projective plane
with six points blown up is isomorphic to a nonsingular cubic surface in P 3 .
We use this isomorphism to study the geometry of curves on the cubic
surface. The isomorphism is accomplished using the linear system of plane
cubic curves with six base points, so we begin with some general remarks
about linear systems with base points.
Let X be a surface, let IDI be a complete linear system of curves on X, and
let P b . . . ,P, be points of X. Then we will consider the sublinear system b
consisting of divisors D E IDI which pass through the points P 1, ... ,P" and
we denote it by ID - P 1 - . . . - P,l. We say that P 1 , . . . ,P, are the assigned
base points of b.
Let n::X'-+ X be the morphism obtained by blowing up P 1 , . . . ,P"
and let E 1 , . . . ,E, be the exceptional curves: Then there is a natural one-to-
one correspondence between the elements of b on X and the elements of the
complete linear system b' = In* D - E 1 - . . . - E,l on X' given by D f---*
n:* D - E 1 - . . . - E" because the latter divisor is effective on X' if and
only if D passes through P 1 , . . . ,P,.
The new linear system b' on X' may or may not have base points. We call
any base point ofb', considered as an infinitely near point of X, an unassigned
base point of b.
These definitions also make sense if some of the P; themselves are infinitely
near points of X, or if they are given with multiplicities greater than 1. For
example, if P 2 is infinitely near P 1, then for D E b we require that D contain
P 1 , and that nf D - E 1 contain P 2 , where n 1 is the blowing-up of P 1 . On the
other hand, if P 1 is given with multiplicity r ~ 1, then we require that D have
at least an r-fold point at P 1 , and in the definition of b', we taken* D - rE 1 .

395
V Surfaces

(Note that if we assign base points Qt. ... ,Qs infinitely near a point P,
then every divisor containing Q 1 , . . . ,Qs will automatically have at least an
s-fold point at P. So we make the convention that every point P must be
assigned with a multiplicity at least equal to the sum of the multiplicities of
the assigned base points infinitely near P.)
The usefulness of this language is that it gives us a way of talking about
linear systems on various blown-up models of X, in terms of suitable linear
systems with assigned base points on X.

Remark 4.0.1. Using this language, we can rephrase the condition (II, 7.8.2)
for a complete linear system IDI to be very ample as follows: IDI is very ample
if and only if(a) IDI has no base points, and (b) for every P EX, ID - PI has no
unassigned base points. Indeed, IDI separates the points P and Q if and only
if Q is not a base point of ID - PI, and IDI separates tangent vectors at P if
and only if ID - PI has no unassigned base points infinitely near to P.

Remark 4.0.2. If we observe that the dimension drops by exactly one when
we assign a base point which was not already an unassigned base point of a
linear system, then we can rephrase this condition in a form reminiscent
of (IV, 3.1) as follows: IDI is very ample if and only if for any two points
P,Q EX, including the case Q infinitely near P,

dimiD- P- Ql = dimiDI - 2.

Remark 4.0.3. Applying (4.0.1) to a blown-up model of X, we see that ifb =

ID - P 1 - . . . - P,l is a linear system with assigned base points on X,
then the associated linear system b' on X' is very ample on X' if and only
if (a) b has no unassigned base points, and (b) for every P EX, including
infinitely near points on X', b - P has no unassigned base points.

Now we turn our attention to the particular situation ofthis section, which
is linear systems of plane curves of fixed degree with assigned base points.
We ask whether they have unassigned base points, and if not, we study the
corresponding morphism of the blown-up model to a projective space. To
get the cubic surface in P 3 we will use the linear system of plane cubic curves
with six base points. But first we need to consider linear systems of conics
with base points. Here we use the word conic (respectively, cubic) to mean any
effective divisor in the plane of degree 2 (respectively, 3).

Proposition 4.1. Let b be the linear system of conics in P 2 with assigned base
points P 1 , . . . ,P, and assume that no three of the P; are collinear. If r :::::; 4,
then b has no unassigned base points. This result remains true if P 2 is
infinitely near P 1 .

396
 4 The Cubic Surface in P 3

PROOF. Clearly it is sufficient to consider the case r = 4. First suppose

P 1 ,P 2 ,P 3 ,P4 are all ordinary points. Let Lii denote the line containing Pi
and Pi. Then b contains L 12 + L 34 and L 13 + L 24 . Since no three of the Pi
are collinear, the intersection of these two divisors consists of the points
P 1 ,P 2 ,P 3 ,P4 with multiplicity 1 each, so there are no unassigned base points.
Now suppose P 2 is infinitely near P 1 . In this case b contains L 12 + L 34
and L 13 + L 14 . (Here, of course, L 12 denotes the line through P 1 with the
tangent direction given by P 2 .) This intersection again is just {P 1 ,P 2 ,P 3 ,P4 },
so there are no further base points.

Corollary 4.2. With the same hypotheses, we have:

(a) if r :( 5, then dim b = 5 - r;
(b) if r = 5, then there exists a unique conic containing P 1 , . . . ,P 5 , which
is necessarily irreducible.
Furthermore, these results remain true if P 5 is infinitely near any one of
P1, ... ,P4.
PROOF.
(a) Every time we prescribe a new base point on a linear system without
unassigned base points, the dimension drops by one. Since the linear system
of all conics in P 2 has dimension 5, this follows from (4.1).
(b) For r = 5, dim b = 0, so there is a unique conic containing P to ••• ,P 5 .
It must be irreducible since no three of the Pi are collinear.

Remark 4.2.1. This last statement is the classical result that a conic is
uniquely determined by giving 5 points, or 4 points and a tangent direction
at one of them, or 3 points with tangent directions at two of them, or even 3
points with a tangent direction and a second order tangent direction at one
of them (when P 5 is infinitely near P 2 which is infinitely near P 1 ).

Example 4.2.2. If r = 1, then b has no unassigned base points, and for any
point P, b - P has no unassigned base points, so by (4.0.3), b' is very ample
on X'. Since dim b' = 4, it gives an embedding of X' in P 4 , as a surface of
degree 3, which is the number of unassigned intersection points oftwo divisors
in b. In fact, X' is just the rational ruled surface with e = 1, and this em-
bedding is the rational cubic scroll (2.19.1).

Example 4.2.3. If r = 3, then X' is P 2 with three points blown up, and
dim b = 2. Since b' has no base points, it determines a morphism lf; of X' to
P 2 . We may take the three points to be P 1 = (1,0,0), P 2 = (0,1,0) and P 3 =
(0,0,1). Then the vector space V £:: H 0 (0pz(2)) corresponding to b is spanned
by x 1 x 2 , x 0 x 2 , and x 0 x1o so lf; can be defined by y 0 = x 1 x 2 , y 1 = x 0 x 2 ,
y 2 = x 0 x 1 , where Yi are the homogeneous coordinates of the new P 2 . Con-
sidered as a rational map from P 2 to P 2 , this is none other than the quadratic
transformation qJ of (I, Ex. 4.6).

397
V Surfaces

Now we will show that t/1 identifies X' with the second P 2 , blown up at
the points Q 1 = (1,0,0), Q 2 = (0,1,0), and Q 3 = (0,0,1), in such a way that
the exceptional curve t/J- 1 (QJ is the strict transform of the line Ljk joining
Pj,Pk on the first P 2 , for each (i,j,k) = (1,2,3) in some order. Furthermore
t/J(E;) is the line Mjk joining Qj and Qk, for each (i,j,k) = (1,2,3). Thus we
can say that the quadratic transformation cp is just "blowing up the points
P 1,P 2 ,P 3 and blowing down the lines L12 ,L 13 ,l 23 " (Fig. 21).

r
--~----

Figure 21. The quadratic transformation of P 2 •

To prove this, we consider the variety V in P 2 x P 2 defined by the

bihomogeneous equations x 0 y 0 = x 1 y 1 = x 2 Y2. I claim the first projection
p 1 : V --+ P 2 identifies V with X'. This is a local question, since blowing up
a point depends only on a neighborhood of the point, so we consider the
open set U ~ P 2 defined by x 0 = 1. Then U =Spec A with A = k[x 1 ,x 2 ],
and p1 1 (U) can be written as
p1 1(U) = Proj A[y 0 ,y 1 ,Y2]/(Y0 - X1Y1, X1Y1 - X2Y2).

We can eliminate y 0 from the graded ring, so

P1 1 (U) ~ Proj A[y 1 ,Y2]/(x 1 y 1 - X2Y2).
But this shows that p1 1 (U) is isomorphic to U with the point (x 1 ,x 2 ) = (0,0)
blown up, as in the proof of (3.6).
Doing the same with the open sets x 1 = 1 and x 2 = 1 of P 2 , we see that
V, via the first projection, is just P 2 with the points P 1 ,P 2 ,P 3 blown up, so
V ~ X'. By symmetry, V with the second projection is the second P 2 with
the points Q 1 ,Q 2 ,Q 3 blown up. So p 2 a p1 1 gives a birational transforma-
tion ofP 2 to itself. Solving the equations x 0 y 0 = x 1 y 1 = x 2 Y2 in the case
x 0 ,x 1 ,x 2 =f. 0, we get Yo = x 1 x 2 , y 1 = x 0 x 2 , y 2 = x 0 x 1, so that this trans-
formation is again the quadratic transformation cp above.
We conclude that t/f:X' --+ P 2 is the same as p 2 : V--+ P 2 , so that cp is just
blowing up three points and blowing down three lines. Finally, it is clear
from the equations that pz(Lij) = Qk and pz(E;) = Mjk for each i,j,k.

398
 4 The Cubic Surface in P 3

Proposition 4.3. Let b be the linear system of plane cubic curves with assigned
base points P 1 , . . . ,P" and assume that no 4 of the P; are collinear, and
no 7 of them lie on a conic. If r :::::; 7, then b has no unassigned base points.
This result remains true if P 2 is infinitely near P 1 .

PROOF. It is sufficient to consider the case r = 7. We will show first that if

P 1, ... ,P 7 are all ordinary points, then b has no unassigned ordinary base
points. For this it is sufficient to exhibit, for each point Q not equal to any
P;, a cubic curve containing P 1 , . . . ,P 7 but not Q.
Case 1. Suppose there exist some three points P 1 ,P 2 ,P 3 lying on a line L *
with Q. The points P 4 ,P 5 ,P6 ,P 7 are not all collinear, so we may assume that
P 4 ,P 5 ,P 6 are not collinear. Then the conic r 12456 through P 1 ,P 2,P4 ,P 5 ,P 6 ,
together with the line L 37 through P 3 and P 7 forms a cubic curve containing
P 1 , . . . ,P 7 but not Q. Indeed, if Q E F 12456 , then this conic contains the
line L*, so it is reducible, in which case P 4 ,P 5 ,P 6 must be collinear, which
is a contradiction. If Q E L 37 , then P 7 E L*, so P 1 ,P 2,P 3,P 7 are collinear,
which is a contradiction.
Case 2. Suppose that Q is not collinear with any set of three of the points
P;, but that Q lies on a conic r* (necessarily irreducible) containing 6 of
them, say P 1 , . . . ,P 6 . Then F 12347 + L 56 is a cubic not containing Q.
Indeed, if Q E r12347, then P1,P2,P3,P4,Q are in this conic and also in r*,
so by (4.2), F 12347 = r*. But then P 1 , . . . ,P 7 are all in r*, a contradiction.
If Q E L56, then r* is reducible, a contradiction.
Case 3. Q is not collinear with any 3 of the P;, and not on a conic with
any 6 of them. Then consider the three cubic curves C; = r 1234 ; + Ljk,
where (i,j,k) = (5,6,7) in some order. We will show that one of these does
not contain Q. If Q E L 56 , then Q ¢ L 57 and Q ¢ L 67 , because in either case
P 5 ,P6 ,P 7 ,Q would be collinear. So, ruling out C 7 , we may assume Q ¢ L 57
and Q ¢ L 67 . Then, if Q E C 5 and Q E C 6 , we have Q E r 12345 and Q E
r 12346 . Consider the conic r' = r 1234Q. If r' is irreducible, then from
(4.2) we have all three conics equal, so Q E r 123456 , which is a contradic-
tion. If r' is reducible, then for a suitable relabeling, we have either (a) r' =
L 123 + L 4 Q or (b) r' = L 12 Q + L 34 . In case (a), F 12345 = L 123 + L 45 and
r 12346 = L 123 + L 46 , soP 4 ,P 5 ,P 6 and Q are collinear, a contradiction. In
case (b), r 12345 = L 12 + L 345 and r 12346 = L 12 + L 346 , so P 3 ,P4 ,P 5 ,P6
are collinear, a contradiction.
THis completes the proof in the case P 1 , . . . ,P 7 and Q are all ordinary
points. The same proof also works in case P 2 is infinitely near P 1 , or Q is
infinitely near one of P 1 , . . . ,P 7 , or both. One has to relabel the P; occa-
sionally so that the constructions make sense, and one has to use (4.2) in
the case of infinitely near points also. (Details left to reader.)

Corollary 4.4. With the same hypotheses, we have

(a) if r :::::; 8, then dim b = 9 - r, and
(b) if r = 8, dim b = 1 and almost every curve in b is irreducible.

399
V Surfaces

PROOF. For r ~ 7 there are no unassigned base points, so at each step, the
dimension drops by one. The cubics with no base points form a linear
system of dimension 9. This proves (a). To prove (b), we observe that with
no 4 points collinear and no 7 on a conic, there are only finitely many ways
of passing three lines, or one line and one irreducible conic through the
8 points.

Corollary 4.5. Given 8 points P 1 , . . . ,P 8 in the plane, no 4 collinear, and no 7

lying on a conic, there is a uniquely determined point P 9 (possibly an
infinitely near point) such that every cubic through P 1 , . . . ,P 8 also passes
through P 9 . This is still true if P 2 is infinitely near P 1 , and P 8 is infinitely
near any one of P 1 , . . . ,P 7 .
PROOF. By (4.4) the linear system b of all cubics through P 1 , . . . ,P 8 has
dimension one, and we can choose two distinct irreducible ones C,C' E b.
Then by Bezout's theorem (1.4.2) (cf. Ex. 3.2), C and C' meet in 9 points,
8 of which are P 1, . . . ,P 8 . So this determines a ninth point P 9 , possibly an
infinitely near point. Now since dim b = 1, any other curve C" E b, irre-
ducible or not, is a linear combination of C and C', so it must also pass
through P 9 . Thus P 9 is an unassigned base point of b.

Remark 4.5.1. This classical result has a number of interesting geometrical

consequences. See (Ex. 4.4), (Ex. 4.5).

Theorem 4.6. Let b be the linear system of plane cubic curves with assigned
(ordinary) base points P 1 , . . . ,P_, and assume that no 3 of the P; are col-
linear, and no 6 of them lie on a conic. If r ~ 6, then the corresponding
linear system b' on the surface X' obtained from P 2 by blowing up P to . . . ,
P_, is very ample.
PROOF. According to (4.0.3) we must verify that b has no unassigned base
points, and that for every point P, possibly infinitely near, b - P has no
unassigned base points. The first statement is an immediate consequence of
(4.3). For the second, we note that since no 3 of the Pi are collinear, and
no 6 of them lie on a conic, the r + 1 points P to ... ,P" P satisfy the hy-
potheses of (4.3). So this case also follows from (4.3).

Corollary 4.7. With the same hypotheses, for each r = 0,1, ... ,6, we obtain
an embedding of X' in P 9 -' as a surface of degree 9 - r, wbose canonical
sheaf Wx· is isomorphic to lDx.( -1). In particular, for r = 6, we obtain a
nonsingular cubic surface in P 3 .
PROOF. We embed X' in pN via the very ample linear system b'. Since
dim b = dim b' = 9 - r by (4.4), we have N = 9 - r. If L is a line in P 2 ,
then b' = ln*3L - E 1 - ... - E,l, so for any D' E b', we have D' 2 = 9 - r.
Therefore the degree of X' in pN is 9 - r. Finally, since the canonical
divisor on P 2 is -3L, we see from (3.3) that Kx· = -n*3L + E 1 + ... +

400
 4 The Cubic Surface in P 3

E., which is just - D'. Therefore wX' ~ (l)x,( -1) in the given projective
embedding.

Remark 4.7.1. A Del Pezza surface is defined to be a surface X of degree d

in pd such that Wx ~ (l)x( -1). So (4.7) gives a construction of Del Pezza
surfaces of degrees d = 3,4, ... ,9. A classical result states that every Del
Pezza surface is either one given by (4.7) for a suitable choice of points
P; E P 2 , or the 2-uple embedding of a quadric surface in P\ which is a Del
Pezza surface of degree 8 in P 8 . In particular, every nonsingular cubic
surface in P 3 can be obtained by blowing up 6 points in the plane. Indeed,
for a cubic surface in P 3 , the condition Wx ~ (l)x( -1) is automatic (II, Ex.
8.4). For proofs see, e.g., Manin [3, §24] or Nagata [5, I, Thm. 8, p. 366].

Remark 4.7.2. In the case of cubic surfaces in P 3 , we can prove a slightly

weaker result by counting constants. The choice of 6 points in the plane
requires 12 parameters. Subtract off the automorphisms of P 2 (8 param-
eters) and add automorphisms of P 3 (15 parameters). Thus we see that the
cubic surfaces in P 3 given by (4.7) form a 19-dimensional family. But the
family of all cubic surfaces in P 3 has dimension equal to dim H 0 ( (l)p,(3)) - 1,
which is also 19. Thus we see at least that almost all nonsingular cubic
surfaces arise by (4. 7).

Notation 4.7.3. For the rest of this section, we specialize to the case of the
cubic surface in P\ and fix our notation. Let P 1 , . . . ,P 6 be six points of
the plane, no three collinear, and not all six lying on a conic. Let b be
the linear system of plane cubic curves through P 1 , . . . ,P 6 , and let X be
the nonsingular cubic surface in P 3 obtained by (4.7). Thus X is iso-
morphic to P 2 with the six points P 1 , . . . ,P 6 blown up. Let n: X ~ P 2
be the projection. Let E 1 , . . . ,E 6 ~ X be the exceptional curves, and
let e 1 , . . . ,e6 EPic X be their linear equivalence classes. Let l EPic X be
the class of n* of a line in P 2 .

Proposition 4.8. Let X be the cubic surface in P 3 (4.7.3). Then:

(a) Pic X ~ Z 7 , generated by l,e 1 , . . . ,e6 ;
(b) the intersection pairing on X is given by 12 = 1, e? = -1, I.e; = 0,
e;.ei = 0 fori =1- j;
(c) the hyperplane section h is 31 - Ie;;
(d) the canonical class is K = - h == - 31 + Ie;;
(e) if D is any effective divisor on X,D ,...., al - Lb;e;, then the degree
of D, as a curve in P 3 , is
d = 3a- ~b-·
L."
(f) the self-intersection of D is D 2 = a2 - Ib?;
(g) the arithmetic genus of D is
1 1 1
pu(D) = 2 (D 2 - d) + 1 = 2 (a - 1)(a - 2) - 2 l:b;(b; - 1).

401
V Surfaces

PROOF. All of this follows from earlier results. (a) and (b) follow from (3.2).
(c) comes from the definition of the embedding in P 3 . (d) comes from (3.3).
For (e), we note that the degree of D is just D.h. (f) is immediate from (b),
and (g) follows from the adjunction formula 2pa(D) - 2 = D.(D + K)
(Ex. 1.3) and the fact that D.K = -D.h = -d by (d).

Remark 4.8.1. If Cis any irreducible curve in X, other than E 1, . . . ,E 6 , then

n(C) is an irreducible plane curve C 0 , and C in turn is the strict transform
of C 0 . Let C 0 have degree a, and suppose that C 0 has a point of multiplicity
b; at each P;. Then n*C 0 = C + 'j);E;, by (3.6). Since C 0 ~ a · line, we
conclude that C ~ al - 'j);e;. Thus for any a,b 1 , . .. ,b 6 ~ 0, we can in-
terpret an irreducible curve C on X in the class al - 'L.b;e; as the strict
transform of a plane curve of degree a with a b;-fold point at each P;. So
the study of curves on X is reduced to the study of certain plane curves.

Theorem 4.9 (Twenty-Seven Lines). The cubic surface X contains exactly 27

lines. Each one has self-intersection -1, and they are the only irreducible
curves with negative self-intersection on X. They are
(a) the exceptional curves E;, i = 1, ... ,6 (six of these),
(b) the strict transform Fii of the line in P 2 containing P; and Pi, 1 :::::;
i < j :::::; 6 (fifteen of these), and
(c) the strict transform Gi of the conic in P 2 containing the five P; for
i #- j,j = 1, ... ,6 (six of these).
PROOF. First of all, if L is any line in X, then deg L = 1 and Pa(L) = 0, so
by (4.8) we have L 2 = -1. (See also (Ex. 1.4).) Conversely, if C is an irre-
ducible curve on X with C 2 < 0, then since Pa(C) ~ 0, we must have C 2 =
-1, Pa(C) = 0, deg C = 1 again by (4.8), soC is a line.
Next, from (4.8.1) we see that E; ~ e;, Fii ~ 1 - e; - ei, and Gi ~ 21 -
Li>'i e;, and we see immediately from (4.8) that each of these has degree 1,
i.e., is a line.
It remains to show that if C is any irreducible curve on X with deg C = 1
and C 2 = -1, then C is one of those 27 lines listed. Assuming Cis not one
of theE;, we can write C ~ al - 'L.b;e;, and by (4.8.1) we must have a > 0,
b; ~ 0. Furthermore,
deg C = 3a - 'j); = 1
cz = az - 'L.bf = - 1.
We will show that the only integers a,b 1 , • .. ,b 6 satisfying all these condi-
tions are those corresponding to the Fii and Gi above.
Recall Schwarz's inequality, which says that if x 1 ,x 2 , . . . ,y 1 ,y2 , • •• are
two sequences of real numbers, then
I'Ix;Y;I 2 :::::; I'Ixfl· I'L.Yfl.
Taking X; = 1, Y; = b;, i = 1, ... ,6, we find

402
 4 The Cubic Surface in P 3

Substituting 'f.bi = 3a - 1 and "[}? = a 2 + 1 from above, we obtain

3a 2 - 6a - 5 :(; 0.
Solving the quadratic equation, this implies a :(; 1 + (2/3)ft < 3. There-
fore a = 1 or 2. Now one quickly finds all possible values of the bi by trial:
if a = 1, then bi = bj = 1 for some i,j, the rest 0. This gives Fij. If a = 2,
then all bi = 1 except for one bj = 0. This gives Gj.

Remark 4.9.1. There is lots of classical projective geometry associated with

the 27lines. For example, the configuration of 12lines £ 1 , . . . ,E 6 , G 1 , ... ,G 6
in P 3 with the property that the Ei are mutually skew, the Gj are mutually
skew, and E; meets Gj if and only if i =I j, is called Sch/afli's double-six.
One can show that given a line £ 1 , and five lines G2 , . . . ,G 6 meeting it, but
otherwise in sufficiently general position, then other lines £ 2 , •.• ,E 6 and
G1 are uniquely determined so as to form a double-six. Furthermore, each
double-six is contained in a unique nonsingular cubic surface, and thus
forms part of a set of 27 lines on a cubic surface. See Hilbert and Cohn-
Vossen [1, §25]. The 27 lines have a high degree of symmetry, as we see in
the next result.

Proposition 4.10. Let X be a cubic surface as above, and let E'1 , • •• ,E~ he
any subset of six mutually skew lines chosen from among the 27 lines on
X. Then·there is another morphism n':X--> P 2 , making X isomorphic to
that P 2 with six points P'b ... ,P~ blown up (no 3 collinear and not all 6
on a conic), such that E'1 , ••• ,E6 are the exceptional curves for n'.
PROOF. We proceed stepwise, working with one line at a time. We will show
first that it is possible to find n' such that £'1 is the inverse image of P'1 .
Case I. If E~ is one of the E;, we take n' = n, but relabel the Pi so that
Pi becomes P~.
Case 2. If £'1 is one of the Fij, say £'1 = F 12 , then we apply the quadratic
transformation with centers P 1 ,P 2 ,P 3 (4.2.3) as follows. Let X 0 be P 2 with
P 1 ,P 2 ,P 3 blown up, let n 0 :X 0 --> P 2 be the projection, and let tjJ:X 0 --> P 2
be the other map to P 2 of(4.2.3), so that X 0 via tjJ is P 2 with Q 1 ,Q 2 ,Q 3 blown
up. Since n: X --> P 2 expresses X as P 2 with P 1 , . . . ,P 6 blown up, n factors
through n 0 , say n = n 0 o 8, where 8:X--> X 0 • Now we define n' as tjJ o 8.

~p2"\\
X X0 I([J

~p2/
Then, using the notation of (4.2.3), 8(F 12 ) = L 12 , so n'(Fu) = Q3 . Fur-
thermore, n' expresses X as P 2 with Q 1 ,Q 2 ,Q 3 ,P~,P~,P6 blown up, where
P~,P~,P6 are the images of P 4 ,P 5 ,P 6 under tjJ a n 0 1 . Now taking P'1 = Q3 ,
and P~,P~ to be Q 1 ,Q 2 , we have £'1 = n'- 1(P'1 ).

403
V Surfaces

We still have to verify that no 3 of Q 1 ,Q 2 ,Q 3 ,P~,P~,P~ lie on a line, and

no 6 on a conic. Q1 ,Q 2 ,Q 3 are noncollinear by construction. If QI>Q 2 ,P~
were collinear, then I/J- 1 (P~) E E 3 , so P 4 would be infinitely near P 3 • If
Q 1 ,P~,P~ were collinear, let L' be the line containing them. Then the strict
transform of L' by cp - l will be a line L containing P 1 ,P4 ,P 5 . Indeed, cp- 1
is the rational map determined by the linear system of conics through
Q1 ,Q 2 ,Q 3 . Such a conic has one free intersection with L', so the strict trans-
form of L' is a line L. Furthermore, L' meets M 23 , soL passes through P 1 .
Finally, suppose P~,P~,P~ were collinear. Since cp is determined by the
conics through P 1 ,P 2 ,P 3 , the strict transform of the line L' containing
P~,P~,P~ would be a conic r containing P 1 , . . . ,P 6 , which is impossible.
For the same reason, if Q 1 ,Q 2 ,Q 3 ,P~,P~,P~ lay on a conic, then P 4 ,P 5 ,P 6
would be collinear. This completes Case 2.
Case 3. If E'1 is one of the Gi' say E'1 = G6 , we again apply the quadratic
transformation of (4.2.3) with centers P 1 ,P 2 ,P 3 . Since n(G 6 ) is the conic
through P 1 , . . . ,P 5 , we see that n'(G 6 ) is the line through P~,P~. Thus E'1
is the curve F~ 5 for n', which reduces us to Case 2.
Now that we have moved E'1 to the position of E 1 , we may assume E'1 =
E 1 , and we consider E~. Since E~ does not meet E 1 , the possible values of
E~ are E 2 , •.• ,E 6 , Fii with 1 < i,j, or G1 . We apply the same method as in
Cases 1, 2, and 3 above, and find that we can move E'z to the role of E 2 without
touching P 1 . That is to say, we allow ourselves only to relabel P 2 , . . . ,P 5 ,
or use quadratic transformations based at three points among P 2 , . . . ,P 5 .
Continuing in this manner, we eventually have E'1, . . . ,E~ in the position
of E 1 , . . . ,E 6 , which proves the proposition. For example, the last step is
this. Assuming that Ei = Ei for i = 1,2,3,4, there are only 3 lines left which
do not meet EI> ... ,E4 . They are E 5 ,E 6 ,F 56 . Since F 56 meets E 5 and E 6 ,
the lines E~ and E~ must be E 5 and E 6 in some order. So for the last step
we have only to permute 5 and 6 if necessary.

Remark 4.10.1. The proposition says that any six mutually skew lines among
the 27 lines play the role of E 1 , . . . ,E6 . Another way of expressing this is
to consider the corifiguration of the 27 lines (forgetting the surface X). That
means we consider simply the set of 27 elements named Ei,Fii,Gi (the lines)
together with the incidence relations they satisfy. These incidence relations
are easily deduced from (4.8) and (4.9), and say (explicity) Ei does not meet
Ei for i =1- j; Ei meets Fik if and only if i = j or i = k; Ei meets Gi if and
only if i =1- j; Fii meets Fk 1 if and only if i,j,k,l are all distinct; Fii meets Gk
if and only if i = k or j = k; Gi does not meet Gk for j =1- k.
Now to say that E'1 , . . . ,E6 play the same role as E 1 , . . . ,E 6 means that
there is another way of labeling all 27 lines, starting with E'1 , . . . ,E6, so as
to satisfy the same incidence relations. In other words, there is an automor-
phism of the configuration (meaning a permutation of the set of 27 elements,
preserving the incidence relations) which sends E 1, . . . ,E 6 to E'1, . . . ,E6.
Notice furthermore that naming E 1, . . . ,E 6 uniquely determines the names

404
 4 The Cubic Surface in P 3

of the remaining 21 lines: Fii is the unique line which meets Ei,Ei but not
other Ek; Gi is the unique line which meets all Ei except Ei.
So (4.10) tells us that for every (ordered) set of six mutually skew lines
among the 27 lines, there is a unique automorphism of the configuration
taking Et. ... ,E6 to those six. Since any automorphism must send skew
lines to skew lines, we get all elements of the group G of automorphisms of
the configuration this way. From the incidence relations it is easy to count
the ways of choosing six mutually skew lines: there are 27 choices for E 1 ,
16 for E 2 , 10 for E 3 , 6 for £ 4 , 2 for E 5 and 1 for E 6 • So the order of the
group G is 27 · 16 · 10 · 6 · 2 = 51,840.
One can show that G is isomorphic to the Weyl group E 6 , and that it
contains a normal subgroup of index 2 which is a simple group of order
25,920. See (Ex. 4.11) and Manin [3, §25, 26].

We will use this symmetry of the 27 lines to determine the ample and
very ample divisor classes on the cubic surface.

Theorem 4.11. The following conditions are equivalent, for a divisor D on the
cubic surface X:
(i) D is very ample;
(ii) D is ample;
(iii) D 2 > 0, and for every line L ~ X, D.L > 0;
(iv) for every line L ~ X, D.L > 0.
PROOF. Of course (i) => (ii) => (iii) => (iv), using the easy direction of Nakai's
criterion (1.10). For (iv) => (i) we will first prove a lemma.

Lemma 4.12. Let D ~ al - 'i);e; be a divisor class on the cubic surface X,

and suppose that b 1 ~ b 2 ~ ••. ~ b 6 > 0 and a ~ b 1 + b 2 + b 5 • Then
D is veryample.
PROOF. We use the general fact that a very ample divisor plus a divisor
moving in a linear system without base points is very ample (II, Ex. 7.5).
Let us consider the divisor classes
D 0 =I
D1 = I - e1
D2 = 21 - e 1 - e2
D 3 = 21 - e 1 - e2 - e3
D 4 = 21 - e 1 - e2 - e3 - e4
D 5 = 31 - e 1 - e2 - e3 - e4 - e5
D 6 = 31 - e 1 - e2 - e3 - e4 - e5 - e6 •

Then 0 ID I,ID
1 1 correspond to the linear systems of lines in P with 0 or 1
2

assigned base points, which have no unassigned base points. 2 3 41 ID I,ID I,ID
405
V Surfaces

have no base points by (4.1), IDsl has no base points by (4.3), and D 6 is very
ample by (4.6). Therefore any linear combination of these, D = L,c;D;, with
c; ? 0 and c6 > 0, will be very ample.
Clearly D 0 , . .. ,D 6 form a free basis for Pic X ~ Z 7 . Writing D "' al -
'L,b;e;, we have b6 = c6 , b 5 = c 5 + c6 , .•. , b 1 = c 1 + ... + c6 , a = c 1 +
2(c 2 + c 3 + c4 ) + 3(c 5 + c6 ). Then one checks easily that the conditions
c; ? 0, c6 > 0 are equivalent to the conditions b 1 ? ... ? b6 > 0 and a ?
b 1 + b 2 + b 5 , so all divisors satisfying these conditions are very ample.
PROOF OF (4.11 ), CONTINUED. Suppose D is a divisor satisfying D.L > 0 for
every line L ~ X. Choose six mutually skew lines E'1 , . .. ,E~ as follows:
choose E~ so that D.E~ is equal to the minimum value of D.L for any line L;
choose E~ so that D.E~ is equal to the minimum value of D.L among those
lines L which do not meet E~; and choose E~,E~ similarly. There will be
just three remaining lines which do not meet E~,E~,E~,E~, one of them
meeting the other two. Choose E't>E~ so that D.E'1 ? D.E~.
Now according to (4.10), we may assume that E; = E; for each i. Writing
D "' al - L,b;e;, we have D.E; = b;, so by construction we have b 1 ?
b2 ? ... ? b6 > 0. On the other hand, F 12 was available as a candidate
at the time we chose E 3 , so we have D.F 12 ? D.E 3 . This translates as
a - b 1 - b2 ? b 3 , i.e., a ? b 1 + b 2 + b 3 . Since b3 ? b 5 , these conditions
imply the conditions of the lemma, so D is very ample. q.e.d.

Corollary 4.13. Let D "' al - L,b;e; be a divisor class on X. Then:

(a) D is ample¢;> very ample¢;> b; > 0 for each i, and a > b; + bi for
each i,j, and 2a > Li,.,
i b; for each j;
(b) in any divisor class satisfying the conditions of (a), there is an irre-
ducible nonsingular curve.
PROOF. (a) is just a translation of(4.11), using the enumeration ofthe 27lines
in (4.9), and (b) is a consequence of Bertini's theorem (II, 8.18) and
(III, 7.9.1).

Example 4.13.1. Taking a = 7, b 1 = b 2 = 3, b 3 = b4 = b 5 = b6 = 2 we

obtain an irreducible nonsingular curve C "' al - L,b;e;, which according
to (4.8) has degree 7 and genus 5. This gives another proof of the existence
of a curve of degree 7 and genus 5 in P 3 (IV, 6.4.2).

ExERCISEs
4.1. The linear system of conics in P 2 with two assigned base points P 1 and P 2 (4.1)
determines a morphism l/1 of X' (which is P 2 with P 1 and P 2 blown up) to a
nonsingular quadric surface Y in P\ and furthermore X' via l/J is isomorphic
to Y with one point blown up.
4.2. Let cp be the quadratic transformation of (4.2.3), centered at P 1 ,P 2 ,P 3 • If Cis
an irreducible curve of degree d in P 2 , with points of multiplicity r 1 .rzh at
P 1 ,P 2 ,P 3 , then the strict transform C' of C by cp has degree d' = 2d- r 1 - r2 - r3 ,

406
 4 The Cubic Surface in P 3

and has points of multiplicity d - r2 - r3 at Q 1 , d - r 1 - r3 at Q2 and d -

r 1 - r2 at Q3 . The curve C may have arbitrary singularities. [Hint: Use(Ex. 3.2).J
4.3. Let C be an irreducible curve in P 2 . Then there exists a finite sequence of qua-
dratic transformations, centered at suitable triples of points, so that the strict
transform C' of C has only ordinary singularities, i.e., multiple points with all
distinct tangent directions (1, Ex. 5.14). Use (3.8).
4.4. (a) Use (4.5) to prove the following lemma on cubics: If C is an irreducible plane
cubic curve, if Lis a line meeting C in points P,Q,R, and L' is a line meeting C
in points P',Q',R', let P" be the third intersection of the line PP' with C, and
define Q",R" similarly. Then P",Q",R" are collinear.
(b) Let P0 be an inflection point of C, and define the group operation on the set
of regular points of C by the geometric recipe "let the line PQ meet Cat R, and
let P 0 R meet Cat T, then P + Q = T" as in (II, 6.10.2) and (II, 6.11.4). Use
(a) to show that this operation is associative.
4.5. Prove Pascal's theorem: if A,B,C,A',B',C' are any six points on a conic, then the
points P = AB'.A'B, Q = AC'.A'C, and R = BC'.B'C are collinear (Fig. 22).

B'
Figure 22. Pascal's theorem.

4.6. Generalize (4.5) as follows: given 13 points P 1 , ••. ,P 13 in the plane, there are
three additional determined points P 14 ,P 15 ,P 16 , such that all quartic curves
through P 1 , ••. ,P 13 also pass through P 14 ,P 15 ,P 16 . What hypotheses are
necessary on P 1o ••• ,P 13 for this to be true?
4.7. If Dis any divisor of degree don the cubic surface (4.7.3), show that

=1,2 (mod 3)
p.(D) ~ {
~ (d -
1
1)(d - 2)

2
if d

6(d - 1)(d - 2) +J if d = 0 (mod 3).

Show furthermore that for every d > 0, this maximum is achieved by some
irreducible nonsingular curve.
*4.8. Show that a divisor class Don the cubic surface contains an irreducible curve=
it contains an irreducible nonsingular curve= it is either (a) one of the 27 lines,
or (b) a conic (meaning a curve of degree 2) with D 2 = 0, or (c) D.L;;:. 0 for every

407
V Surfaces

line L, and D2 > 0. [Hint: Generalize (4.11) to the surfaces obtained by blowing
up 2, 3, 4, or 5 points of P 2 , and combine with our earlier results about curves
on P 1 x P 1 and the rational ruled surface X 1o (2.18).]
4.9. If C is an irreducible non-singular curve of degree d on the cubic surface, and
if the genus g > 0, then

g~ { ~(d- 6) if d is even, d ~ 8,

~ (d - 5) if d is odd, d ~ 13,
2
and this minimum value of g > 0 is achieved for each din the range given.
4.10. A curious consequence of the implication (iv) =(iii) of (4.11) is the following
numerical fact: Given integers a,b 1 , • •. ,b 6 such that bi > 0 for each i, a - bi -
bj > 0 for each i,j and 2a - Li,;j bi > 0 for each j, we must necessarily have
a 2 - Ib? > 0. Prove this directly (for a,b 1 , . •• ,b 6 E R) using methods of
freshman calculus.
4.11. The Weyl Groups. Given any diagram consisting of points and line segments
joining some of them, we define an abstract group, given by generators and
relations, as follows: each point represents a generator xi. The relations are
x? = 1 for each i; (xix) 2 = 1 if i and j are not joined by a line segment, and
(xixj) 3 = 1 ifi andj are joined by a line segment.
(a) The Weyl group A. is defined using the diagram
o-o-o . . . --()
of n - 1 points, each joined to the next. Show that it is isomorphic to the
symmetric group L:. as follows: map the generators of A. to the elements
(12),(23), ... ,(n - 1,n) of L:., to get a surjective homomorphism A. -+ 1:..
Then estimate the number of elements of A. to show in fact it is an isomorphism.
(b) The Weyl group E 6 is defined using the diagram

Call the generators x 1 , •.. ,x 5 andy. Show that one obtains a surjective homo-
morphism E 6 -+ G, the group of automorphisms of the configuration of 27
lines (4.10.1), by sending x 1 , •.. ,x 5 to the permutations (12),(23), ... ,(56) of
the Ei, respectively, and y to the element associated with the quadratic trans-
formation based at P 1 ,P 2 ,P 3 .
*(c) Estimate the number of elements in E 6 , and thus conclude that E 6 ~ G.
Note: See Manin [3, §25,26] for more about Weyl groups, root systems, and
exceptional curves.
4.12. Use (4.11) to show that if D is any ample divisor on the cubic surface X, then
H 1 (X,@x(- D)) = 0. This is Kodaira's vanishing theorem for the cubic surface
(III, 7.15).

4.13. Let X be the Del Pezzo surface of degree 4 in P 4 obtained by blowing up 5.points
of P 2 (4.7).
(a) Show that X contains 161ines.

408
 5 Birational Transformations

(b) Show that X is a complete intersection of two quadric hypersurfaces in P 4

(the converse follows from (4.7.1) ).
4.14. Using the method of (4.13.1), verify that there are nonsingular curves in P 3
with d = 8, g = 6,7; d = 9, g = 7,8,9; d = 10, g = 8,9,10,11. Combining with
(IV, §6), this completes the determination of all posible g for curves of degree
d~10inP 3 •

4.15. Let P 1 , .•• ,P, be a finite set of(ordinary) points ofP 2 , no 3 collinear. We define
an admissible transformation to be a quadratic transformation (4.2.3) centered
at some three of the P; (call them P 1 ,P2 ,P 3 ). This gives a new P 2 , and a new set
ofrpoints, namely Q 1 ,Q 2 ,Q 3 , and the images of P 4 , . . . ,P,. We say that P 1 , . . . ,P,
are in general position if no three are collinear, and furthermore after any finite
sequence of admissible transformations, the new set of r points also has no three
collinear.
(a) A set of 6 points is in general position if and only if no three are collinear and
not all six lie on a conic.
(b) If P 1 , •.. ,P, are in general position, then the r points obtained by any finite
sequence of admissible transformations are also in general position.
(c) Assume the ground field k is uncountable. Then given P 1 , .•• ,P, in general
position, there is a dense subset V ~ P 2 such that for any P,+ 1 E V, P 1 , ••• ,P,+ 1
will be in general position. [Hint: Prove a lemma that when k is uncountable,
a variety cannot be equal to the union of a countable family of proper closed
subsets.]
(d) Now take Pt. . .. ,P, E P 2 in general position, and let X be the surface obtained
by blowing up P 1 , •.. ,P,. If r = 7, show that X has exactly 56 irreducible
nonsingular curves C with g = 0, C 2 = -1, and that these are the only
irreducible curves with negative self-intersection. Ditto for r = 8, the number
being 240.
*(e) For r = 9, show that the surface X defined in (d) has infinitely many irreducible
nonsingular curves C with g = 0 and C 2 = -1. [Hint: Let L be the line
joining P 1 and P 2 • Show that there exist finite sequences of admissible trans-
formations such that the strict transform of L becomes a plane curve of
arbitrarily high degree.] This example is apparently due to Kodaira-see
Nagata [5, II, p. 283].
4.16. For the Fermat cubic surface x6 + xi + x~ + x~ = 0, find the equations of
the 27 lines explicitly, and verify their incidence relations. What is the group of
automorphisms of this surface?

5 Birational Transformations

Up to now we have dealt with one surface at a time, or a surface and its
monoidal transforms. Now we will show that in fact any birational trans-
formation of (nonsingular projective) surfaces can be factored into a finite
sequence of monoidal transformations and their. inverses. This confirms
the central role played by the monoidal transformations in the study of
surfaces. A consequence of this result is the fact that the arithmetic genus
of a surface is a birational invariant.
409
V Surfaces

In this section we will also prove Castelnuovo's criterion for contracting

exceptional curves of the first kind, and deduce the existence of relatively
minimal models for surfaces. See Shafarevich [1, Ch. I, II], Zariski [5], and
Zariski [10, Ch. IV] as general references.
We begin by recalling some general facts about birational maps between
varieties of any dimension, including Zariski's Main Theorem.

Let X and Y be projective varieties of any dimension. Recall (1, §4)

that to give a birational transformation T from X to Y is to give an open
subset U s; X and a morphism cp: U --+ Y which induces an isomorphism
of function fields K(Y) ~ K(X). If we have another open set V s; X and
another morphism ljJ: V --+ Y representing T, then cp and ljJ agree where
both are defined, so we can glue them to obtain a morphism defined on
U u V (1, Ex. 4.2). So there is a largest open set U s; X on which T is
represented by a morphism cp: U --+ Y. We say that Tis defined at the points
of U, and we call the points of X - U fundamental points ofT.
Now let T:X --+ Y be a birational transformation, represented by the
morphism cp: u --+ Y. Let r 0 s; u X y be the graph of cp, and let r s;
X X y be the closure of r o· We call r the graph of T. For any subset
Z s; X, we define T(Z) to be p 2 (p1 1(Z) ), where p 1 and p 2 are the projec-
tions of r to X and Y. We call T(Z) the total transform of Z. If Tis defined
at a point P, then T(P) will be the point cp(P). However, if Pis a fundamental
point of T, then in general T(P) will consist of more than one point.

Lemma 5.1. If T:X --+ Y is a birational transformation of projective va-

rieties, and if X is normal, then the fundamental points of T form a closed
subset of codimension ~ 2.
PROOF. If p EX is a point of codimension 1, then mP,X is a discrete valuation
ring. Since Tis defined at the generic point of X, and Y is projective, hence
proper, it follows from the valuative criterion of properness (II, 4.7) that
T is also defined at P. (We have already used this argument in the proof
of (II, 8.19), to show that the geometric genus is a birational invariant.)

Example 5.1.1. Let X be a surface, and let n:X --+X be the monoidal
transformation with center P. Then n is defined everywhere. Its inverse
n- 1 : X --+ X is a birational transformation having P as a fundamental point.

Theorem 5.2 (Zariski's Main Theorem). Let T: X --+ Y be a birational trans-

formation of projective varieties, and assume that X is normal. If P is a
fundamental point of T, then the total transform T(P) is connected and
of dimension ~ 1.
PRooF. This is just a variant of the earlier form of Zariski's Main Theorem
(III, 11.4). Let r be the graph ofT, and consider the morphism p 1 :r--+ X.
This is a birational projective morphism, so by (III, 11.4), p1 1 (P) is connected.

410
 5 Birational Transformations

If it has dimension 0, then the same is true in a neighborhood V of P (II,

Ex. 3.22). In that case p1 1 (V) --+ V is a projective, birational morphism,
with finite fibres, so it is a finite morphism (III, Ex. 11.2). But V is normal,
so it must be an isomorphism. This says that T is defined at P, which is a
contradiction. We conclude that p! 1 (P) is connected of dimension ~ 1.
Since p 2 maps this set isomorphically onto T(P), we have the result.

Now we come to the key result which enables us to factor birational

transformations of surfaces.

Proposition 5.3. Let f:X' --+X be a birational morphism of (nonsingular,

projective) surfaces. Let P be a fundamental point of f- 1 . Then f factors
through the monoidal transformation n:X--+ X with center P.

PROOF. Let T be the birational transformation of X' to X defined as n- 1 of

Our object is to show that Tis a morphism. If not, then it has a fundamental
(closed) point P'. Clearly f(P') = P. Furthermore T(P') has dimension ~ 1
in X, by (5.2). Thus T(P') must be the exceptional curve E of n.
On the other hand, by (5.1), T- 1 is defined at all except finitely many
points of X, so we can find a closed point Q E E where T- 1 is defined, and
hence T- 1 (Q) = P'. We will show that this situation leads to a contradiction.
Choose local coordinates x,y at P on X. Then as in the proof of (3.6),
there is an open neighborhood V of P such that n - 1 (V) is defined by the
equation ty - ux in P~. By a linear change of variables in x,y and t,u, we
may assume that Q is the point t = 0, u = 1 in E. Then t,y form local
coordinates at Q on X; the local equation of E is y = 0, and x = ty.
Since Pis a fundamental point off- 1 , by (5.2) we have f- 1 (P) connected
of dimension ~ 1, so there is an irreducible curve C in f- 1 (P) containing
P'. Let z = 0 be a local equation for C at P'.
Since f- 1(P) is defined by x = y = 0, the images of x,y in (!JP, are in
the ideal generated by z, so we can write x = az, y = bz, a,b E (!Jr· On
the other hand, (!JQ dominates (!Jr· (We consider (!Jp,(!Jp.,(!JQ all as subrings
of the common function field K of X,X',X.) Since t,y are local coordinates
at Q, y ¢ m~, so we conclude that y ¢ m~. in (!Jr· Therefore b is a unit in
(!Jp., and so t = xjy = a/b is in the local ring (!Jr· Since t E mQ, we must
have t E mr.
Now we use the fact that T(P') = E. This implies that for any wE mp·,
the image of w in (!) Q must be contained in the ideal generated by y, since y
is the local equation for E. In particular, taking w = t, we find t E (y),
which is a contradiction, since t and y are local coordinates at Q.

Corollary 5.4. Let f: X' --+ X be a birational morphism of surfaces. Let

n(f) be the number of irreducible curves C' s; X' such that f( C') is a
point. Then n(f) is finite and f can be factored into a composition of
exactly n(f) monoidal transformations.

411
V Surfaces

PROOF. If f(C') is a point P, then Pis a fundamental point of f- 1 . By (5.1)

the fundamental points off- 1 form a finite set, and for each one, its inverse
image f - 1(P) is a closed subset of X', having only finitely many irreducible
components, so the set of curves C' which are mapped to a point is finite.
Now let P be a fundamental point off- 1 . Then by (5.3) f factors through
the monoidal transformation n:X -X with center P, i.e., f = no f 1 for
some morphism f 1 :X' -X. We will show that n(f1 ) = n(f) - 1. Indeed,
if f 1 ( C') is a point, then certainly f( C') is a point. Conversely, iff( C') is a
point, then either f 1( C') is a point, or f 1( C') = E, the exceptional curve of n.
Furthermore, since f 1 1 is a morphism except at finitely many points,
there is a unique irreducible curve E' in X' with f 1(E') = E. Thus n(f1 ) =
n(f) - 1.
Continuing in this fashion, after factoring through n(f) monoidal trans-
formations, we reduce to a morphism with n(f) = 0. But by (5.2), such a
morphism has no fundamental points, so it is an isomorphism. Thus f is
factored into n(f) monoidal transformations.

Remark 5.4.1. It is interesting to compare the factorization of (5.3) with the

universal property of blowing up proved in (II, 7.14). While the new result
implies the old one in the special case ofblowing up a point (sincef- 1 mp ·(!)X'
being invertible implies f- 1 (P) has dimension 1, so P is a fundamental
point), it is actually stronger, since it uses Zariski's Main Theorem. We
cannot deduce (5.3) from (II, 7.14), because the hypothesis that f- 1 mp ·(!)X'
is invertible is impossible to verify in our case.

Remark 5.4.2. Comparing (5.4) to the earlier theorem (II, 7.17), we see that
the new result is more precise, because it uses only monoidal transforma-
tions, rather than the more general concept of blowing up an arbitrary sheaf
of ideals.

Remark 5.4.3. It is easy to see that (5.3) is false for nonsingular projective
varieties of dimension ~ 3. For example, let f: X' - X be the blowing-up
of a nonsingular curve C in a nonsingular projective 3-fold X. Then any
point P E C is a fundamental point off- 1, but f cannot factor through the
monoidal transformation n:X -X with center P, because f- 1(P) has
dimension 1, while n- 1 (P) has dimension 2.

Remark 5.4.4. The example (5.4.3) suggests posing the following modified
problem: given a birational morphism f: X' - X of nonsingular projective
varieties, is it possible to factor f into a finite succession of monoidal
transformations along nonsingular subvarieties? This is also false in
dimension ~ 3: see Sally [ 1J and Shannon [ 1].

Theorem 5.5. Let T:X -X' be a birational transformation of surfaces. Then

it is possible to factor T into a finite sequence of monoidal transformations
and their inverses.

412
 5 Birational Transformations

PROOF. Using (5.4), it will be sufficient to show that there is a surface X",
and birational morphisms f: X" ~ X and g: X" ~ X' such that T = g of- 1 .
To construct X", we proceed as follows.
Let H' be a very ample divisor on X', and let C' be an irreducible non-
singular curve in the linear system I2H'I, which does not pass through any
of the fundamental points of r- 1 . In other words, C' is entirely contained
in the largest open set U' s X' on which r- 1 is represented by a morphism
cp: U' ~X. Let C = cp(C') be the image of C' in X. We define an integer m
by m = Pa(C) - pa(C'). Since we have a finite birational morphism of C'
to C, we see that m ~ 0, and m = 0 if and only if C' is isomorphic to C
(IV, Ex. 1.8). Note also that if we replace C' by a linearly equivalent curve C'1 ,
also missing the fundamental points of r-1, then C 1 = cp( C'1) is linearly
equivalent to C. In fact, if C' - C'1 = (f) for some rational function f on X',
then C - C 1 = (f) on X. Since the arithmetic genus of a curve depends
only on its linear equivalence class (Ex. 1.3), we see that the integer m depends
only on T and H', and not on the particular curve C' E I2H'I chosen.
Now fix C' temporarily. If m > 0, then C must be singular. Let P be a
singular point of C, let n:X ~X be the monoidal transformation with
center P, and let C be the strict transform of C. Then by (3.7), Pa(C) < Pa(C).
Thus if T = Ton, then m('T} < m(T).
Continuing in this fashion, as in the proof of (3.8), we see that there is a
morphism f:X" ~X, obtained by a finite number of monoidal trans-
formations, such that if T' = To f, then m(T') = 0.
We will show that in fact T' is a morphism. If not, then T' will have a
fundamental point P. By (5.2), T'(P) contains an irreducible curve E' s X'.
Since H' is very ample, E'.H' > 0, and so C'.E' ~ 2for any C' E I2H'I· Let
us choose C', not containing any fundamental point of T'-1, such that C'
meets E' transversally (1.2). Then C' meets E' in at least two distinct points,
so the corresponding curve C in X" has at least a double point at P. But this
contradicts m(T') = 0.
We conclude that T' is a morphism of X" to X', so that applying (5.4)
completes the proof, as mentioned above.

Corollary 5.6. The arithmetic genus of a nonsingular projective surface is a

birational invariant.
PROOF. Indeed, Pa is unchanged by a monoidal transformation (3.5), so this
follows directly from the theorem.

Remark 5.6.1. Even though the factorization theorem (5.5) in a form analo-
gous to (5.4.4) is false in dimension ~ 3, Hironaka [3] is able to deduce
the birational invariance of Pa, for nonsingular projective varieties over a
field of characteristic 0, from the following statement, which is a consequence
of his resolution of singularities: If T: X ~ X' is any birational transforma-
tion of nonsingular projective varieties over a field of characteristic 0, then
there is a morphismf:X" ~X, obtained by a finite succession ofmonoidal

413
V Surfaces

transformations along nonsingular subvarieties, such that the birational map

T' = To f is a morphism. There is another proof of the birational in variance
of Pa, for varieties over C, by Kodaira and Spencer [1 ], using the equalities
h0 q = hqo from Hodge theory, and the birational invariance of hq 0 (II,
Ex. 8.8).

Now we come to Castelnuovo's criterion for contracting a curve on a

surface. We have seen that if E is the exceptional curve of a monoidal
transformation, then E ~ P 1, and E 2 = -1 (3.1). In general, any curve Y
on a surface X, withY ~ P 1 and Y 2 = -1 is classically called an exceptional
curve of the first kind. The following theorem tells us that any exceptional
curve of the first kind is the exceptional curve of some monoidal trans-
formation.

Theorem 5.7 (Castelnuovo). If Y is a curve on a surface X, with Y ~ P 1 and

Y 2 = -1, then there exists a morphism f: X ---> X 0 to a (nonsingular
projective) surface X 0 , and a point P E X 0 , such that X is isomorphic via f
to the monoidal transformation of X 0 with center P, and Y is the exceptional
curve.

PROOF. We will construct X 0 using the image of X under a suitable mor-

phism to a projective space. Choose a very ample divisor H on X such that
H 1 (X,!f>(H)) = 0: for example, a sufficiently high multiple of any given very
ample divisor (III, 5.2). Let k = H. Y, and let us assume k ~ 2. Then we
will use the invertible sheaf A = !fl(H + kY) to define a morphism of X
to pN_
Step 1. First we prove that H 1 (X,!f>(H + (k - 1) Y)) = 0. In fact, we
will prove more generally that for every i = 0,1, ... ,k, we have
H 1 (X,!fl(H + iY)) = 0. For i = 0, this is true by hypothesis, so we pro-
ceed by induction on i. Suppose it is true for i - 1. We consider the exact
sequence of sheaves
0---> !fl(H + (i- 1)Y)---> !fl(H + iY)---> c'9y@ !fl(H + iY)---> 0.
Now Y ~ P 1, and (H + iY). Y = k - i, so
c'9y @ !fl(H + iY) ~ @p1(k - i).
We get an exact cohomology sequence
... ---> H 1(X,!fl(H + (i - 1) Y)) ---> H 1(X,!fl(H + iY)) --->
---> H 1 (Pi,c'9pl(k - i)) ---> ....
So from the induction hypothesis, and the known cohomology of P 1 , we
conclude that H 1 (X,!f>(H + iY)) = 0 for any i :( k.
Step 2. Next we show that A is generated by global sections. Since H is
very ample, the corresponding linear system IH + k Yl has no base points
away from Y, so A is generated by global sections off Y. On the other hand,

414
 5 Birational Transformations

the natural map

H 0 (X,A)--+ H 0 (Y,A ® @y)
is surjective, because A® ~r ~ Sf(H + (k - 1)Y), and
H (X,Sf(H + (k - 1) Y)) = 0
1

by Step 1. Next observe that (H + k Y). Y = 0, so A ® @y :;-=: (I)P,, which

is generated by the global section 1. Lifting this section to H 0 (X,A), and
using Nakayama's lemma, we see that A is generated by global sections
also at every point of Y.
Step 3. Therefore A determines a morphismf1 :X--+ pN (II, 7.1). Let X 1
be its image. Since ff(l)(1) ~ A, and since the degree of A ® (i)r is 0,
f 1 must map Y to a point P 1 . On the other hand, since H is very ample,
the linear system IH + k Yl separates points and tangent vectors away from
Y, and also separates points of Y from points not on Y, so f 1 is an isomorphism
of X - Y onto X 1 - P 1 (II, 7.8.2).
Step 4. Let X 0 be the normalization of X 1 (II, Ex. 3.8). Since X is non-
singular, hence normal, the map f 1 factors to give a morphism f:X--+ X 0 •
Since Y is irreducible, f( Y) is a point P, and since X 1 - P 1 was nonsingular,
we still have f: X - Y --+ X 0 - P an isomorphism.
Step 5. Now we will show that X 0 is nonsingular at the point P. Since in
any case X 0 is normal, and f is birational, we have f*(i)x ~ (i)xo (see proof of
(III, 11.4) ). So we can apply the theorem on formal functions (Ill, 11.1) to
conclude that
f!JP ~ lli!! Ho(Y,,(I)yJ,
where Y, is the closed subscheme of X defined by m~ · (i)x· But since
f- 1(P) = Y, the sequence of ideals m~(i)x is cofinal with the sequence of ideals
~~'so we may use these instead in the definition of Y, (II, 9.3.1).
We will show for each n that H 0 (Y,(I)y ) is isomorphic to a truncated power
seriesringAn = k[[x,y]]/(x,y)n. Itwillfollowthatf!Jp ~ ll!!!An ~ k[[x,y]J,
which is a regular local ring. This in turn implies that (I) P is regular (1, 5.4A),
hence P is a nonsingular point.
For n = 1, we have H 0 ( Y,(l) r) = k. For n > 1, we use the exact sequences
0 --+ ~~~~~+ 1 --+ (i}Yn+ 1 --+ (i}Yn --+ 0.
Since Y ~Pi, and Y 2 = -1, we have ~/~ 2 ~ @p,(1) by (1.4.1), and
~n ;~n+ 1 ~ (l)p,(n) for each n, as in the proof of (3.4). Taking cohomology,
we have
0 --+ H 0 ((1)p,(n)) --+ H 0 ((1)yn+) --+ H 0 ((i)yJ --+ 0.
For n = 1, H 0 ((1)p,(1)) is a 2-dimensional vector space. Take a basis x,y.
Then H 0 (@y,), which in any case contains k, is seen to be isomorphic to A 2 •
Now inductively, if H 0 ((1)rJ is isomorphic to An, lift the elements x,y to
H ((i)rn+J Since H 0 ((1)p,(n)) is the vector space with basis xn,xn- 1 y, ... ,yn,
0

we see that H 0 (@yn+',) ~ An+ 1 . Now as above we see that Pis a nonsingular
point.

415
V Surfaces

Step 6. We complete the proof using the factorization theorem (5.4). Since
X 0 is nonsingular, we can apply (5.4) to f:X-+ X 0 . We have n(f) = 1 by
construction, so f must be the monoidal transformation with center P.
Step 7. As an addendum, we show in fact that X 0 = X 1 , so the normaliza-
tion was unnecessary. The natural map
H 0 (X,.A@ Jy)-+ H 0 (Y,.A@ JyjJ~)
is surjective, because the next term of the cohomology sequence is
H 1 (X,!E(H + (k - 2) Y) ), which is 0 by Step 1. Since A @ CDy ~ CDy, this
shows that there are global sections s,t E H 0 (X,.A @ J Y) <;: H 0 (X,.A),
which map to the parameters x,y E H 0{CDy 2 ) ~ A 2 • On the other hand, these
sections s,t become sections of CD(l) on PN, defining hyperplanes containing
P 1. So they give elementss;t E mp,, whose images in (D P generate the maximal
ideal mp. Since in any case CDp is a finitely generated CDp,-module, we conclude
that CDp ~ CDp, (II, 7.4), and so X 0 ~ X 1 .

Example 5.7.1. Let n:X-+ C be a geometrically ruled surface (§2), let P be a

point of X, and let L be the fibre of n containing P. Let f:X -+X be the
monoidal transformation with center P. Then the strict transform L of L
on X is isomorphic to P 1 , and hasP = -1. Indeed, L 2 = 0 by (2.3), and P
is a nonsingular point of L, so L ~ f* L - E by (3.6). It follows that
L2 = -1. Therefore by the theorem, we can blow down L. In other words,
there is a morphism g:X -+X' sending L to a point Q, and such that g is
the monoidal transformation with center Q. If M = g(E), then M ~ P 1
and M 2 = 0, for a similar reason to the above. Note also that the rational
mapn':X'-+ CobtainedfromnonX- L ~X'- Misinfactamorphism.
Therefore n': X' -+ C is another geometrically ruled surface. Indeed, the
fibres of n' are all isomorphic to P 1 , and since n has a section, its strict
transform on X' will be a section of n'. This new ruled surface is called the
elementary transform of X with center P, and is denoted by elmp X (Fig. 23).
See (Ex. 5.5) for some applications.

~,.-............-~~c
Figure 23. An elementary transformation of a ruled surface.

416
 5 Birational Transformations

Remark 5.7.2 (The General Contraction Problem). With the theorem of

Castelnuovo (5.7) in mind, one can pose the following general problem:
Given a variety X, and a closed subset Y £ X, find necessary and sufficient
conditions for the existence of a birational morphism f: X ~ X 0 such that
f( Y) is a single point P, and f: X - Y ~ X 0 - P is an isomorphism. If
such a morphism exists, we say that Y is contractible. A number of special
cases of this problem have been treated, but a general solution is unknown.
See Artin [3], [ 4], Grauert [2], and Mumford [ 6].
In case Y is an irreducible curve on a surface X, here is what is known.
If we require X 0 to be nonsingular, then by (5.7) the necessary and sufficient
conditions are that Y ~ P 1 and Y 2 = -1. If we allow X 0 to be singular,
then a necessary condition is that Y 2 < 0 (Ex. 5.7). If Y ~ P\ this condition
is also sufficient (Ex. 5.2). If Y is arbitrary, with Y 2 < 0, and if the base field
is C, then a theorem of Grauert [2] shows that X 0 exists as a complex
analytic space. However, Y may not be contractible to an algebraic variety,
as we show in the following example.

Example 5.7.3 (Hironaka). Let Y0 be a nonsingular cubic curve in P 2 over an

uncountable algebraically closed field k (e.g., k = C). Fix an inflection point
P 0 E Y0 to be the origin of the group law on Y0 . Since the abelian group Y0
is uncountable, and since the torsion points are countable (IV, 4.8.1), the
torsion-free part must have infinite rank. Therefore we can choose 10 points
P 1 , . . . ,P 10 E Y0 which are linearly independent over Z in the group law.
Now blow up P 1 , . . . ,P 10 in P 2 , let X be the resulting surface, and let Y
be the strict transform of Y0 . Since Y~ = 9, and we have blown up 10 points
on Y, we have Y 2 = -1, using (3.6). So by Grauert's theorem (5.7.2), if
k = C, then Yin X would be contractible to a complex analytic space. We
will show, however, that Y is not contractible to a point P in an algebraic
variety X 0 . If it were, let P E U £ X 0 be an open affine neighborhood of P.
Let C~ £ U be a curve not containing P. Let C 0 £ X 0 be its closure, which
still does not contain P. Then its inverse image in X will be a curve C £ X
which does not meet Y. The image of C in P 2 will be a curve C* which does
not meet Y0 except at the points Pt. ... ,P 10 .
But this is impossible. Let d = deg C*. Then by Bezout's theorem
(1.4.2), C*. Y0 = 3d > 0. So we can write
10
C*. Y0 = L n;P;
i= 1

on Y0 , with n; ?: 0, In; = 3d. But C* "' dL, where Lis a line in P 2 , and
L. Y0 "' 3P 0 , so we have
10
L n;P; = 0
i= 1

in the group law on Y0 (IV, 1.3.7). This contradicts the fact that P 1 , ... ,P 10
were chosen to be linearly independent over Z.

417
V Surfaces

To conclude this section, we will prove the existence of relatively minimal

models of surfaces. The idea is to find, within each birational equivalence
class of surfaces, one which is as canonical as possible. Since one can always
blow up a point, there is never a unique nonsingular projective model of a
function field, as in the case of curves. However, we can look for one which
is minimal for the relation of domination. So we say that a (nonsingular
projective) surface X is a relatively minimal model of its function field, if
every birational morphism f:X--+ X' to another (nonsingular projective)
surface X' is necessarily an isomorphism. If X is the unique relatively
minimal model in its birational equivalence class, then we say that X is a
minimal model. (This somewhat irregular use of the word "minimal" is re-
tained for historical reasons.)

Theorem 5.8. Every surface admits a birational morphism to a relatively mini-

mal model.
PROOF. Combining (5.4) and (5.7), it is clear that a surface is a relatively min-
imal model if and only if it contains no exceptional curves of the first kind.
So given a surface X, if it is already a relatively minimal model, stop. If not,
let Y be an exceptional curve of the first kind. By (5.7) there is a morphism
X --+ X 1 contracting Y.
We continue in this manner, contracting exceptional curves of the first
kind whenever one exists, and so we obtain a sequence of birational mor-
phisms X --+ X 1 --+ X 2 --+ .... We must show that this process eventually
stops.
The following proof is due to Matsumura [1]. Suppose that we have a
sequence of n contractions
X = X 0 --+ X 1 . . . --+ Xn
as above. For each i = 1, ... ,n, let E; ~ X;_ 1 be the exceptional curve of
the contraction X;_ 1 --+ X;, and let E; be its total transform on X. Then by
(3.2) we have Ef = -1 for each i, and E;.Ei = 0 for i #- j.
Now for each i, let e; = c(E;) be the cohomology class of E; in H 1 (X,Q)
(Ex. 1.8). Then we have (e;,e;) = -1 and (e;,ei) = 0 in the intersection
pairing on H 1 (X,Q), by (Ex. 1.8). It follows that e 1 , . . . ,en are linearly in-
dependent elements of the vector space H 1 (X,Q) over k.
We conclude that n ~ dimk H 1(X,Q). Since this is a finite-dimensional
vector space, n is bounded, so the contraction process must terminate.
Note: One can give another proof of this result by showing that the rank
of the Neron-Severi group drops by 1 with each contraction, so that n ~
rank NS(X), which is finite-Cf. (Ex. 1. 7).

Remark 5.8.1. In spite of this result, it is not true that a surface necessarily
has only finitely many exceptional curves of the first kind. For example, if
we blow up r points in general position in P 2 , with r ~ 9, then the resulting
surface has infinitely many exceptional curves of the first kind (Ex. 4.15).

418
 5 Birational Transformations

Example 5.8.2. In the birational equivalence class of rational surfaces, P 2 is

a relatively minimal model, and so is the rational ruled surface X e for each
e ~ 0, e #- 1. This follows easily from the determination of all irreducible
curves on Xe (2.18). On the contrary, X 1 is not relatively minimal (2.11.5).

Example 5.8.3. In the class of surfaces birational to P 1 x C, where C is a

curve of genus g > 0, every geometrically ruled surface n: X ---+ Cis relatively
minimal. Indeed, if Y is any rational curve in X, then n( Y) is a point because
of (IV, 2.5.4). Thus Y is a fibre of n, Y 2 = 0, and so we see that X has no
exceptional curves of the first kind.

Remark 5.8.4. A classical theorem, proved in all characteristics b)

Zariski [ 5], [ 6], [9] states that except for the rational and ruled surfaces, every
surface is birational to a (unique) minimal model. One can also show that in
the case of rational and ruled surfaces, every relatively minimal model is one
of those listed in (5.8.2) and (5.8.3). See Nagata [ 5] or Hartshorne [ 4].

EXERCISES

5.1. Let f be a rational function on the surface X. Show that it is possible to "resolve
the singularities of f" in the following sense: there is a birational morphism g:
X'--> X so that f induces a morphism of X' to P 1 . [Hints: Write the divisor off
as (f) = l;n;C;. Then apply embedded resolution (3.9) to the curve Y = UC;.
Then blow up further as necessary whenever a curve of zeros meets a curve of
poles until the zeros and poles off are disjoint.]
5.2. Let Y ~ P 1 be a curve in a surface X, with Y 2 < 0. Show that Y is contractible
(5.7.2) to a point on a projective variety X 0 (in general singular).
5.3. If n:X--> X is a monoidal transformation with center P, show that H 1(X,Qx) ~
H 1(X,Qx) EB k. This gives another proof of (5.8). [Hints: Use the projection
formula (III, Ex. 8.3) and (III, Ex. 8.1) to show that Hi(X,Qx) ~ Hi(X,n*Qx) for each
i. Next use the exact sequence
0 --> n*Qx --> !2x --> !2x;x --> 0
and a local calculation with coordinates to show that there is a natural isomorphism
!2x;x ~ QE, where E is the exceptional curve. Now use the cohomology sequence
of the above sequence (you will need every term) and Serre duality to get the result.]
5.4. Let f: X --> X' be a birational morphism of nonsingular surfaces.
(a) If Y ,;: X is an irreducible curve such that f(Y) is a point, then Y ~ P 1 and
yz < 0.
(b) (Mumford [6].) Let P' EX' be a fundamental point off-\ and let Y1 , . . . ,Y,. be
the irreducible components ofj- 1(P'). Show that the matrix p;.lJII is negative
definite.
5.5. Let C be a curve, and let n: X --> C and n': X' --> C be two geometrically ruled
surfaces over C. Show that there is a finite sequence of elementary transformations
(5.7.1) which transform X into X'. [Hints: First show if D ,;: X is a section of n
containing a point P, and if i5 is the strict transform of D by elmp, then 15 2 = D 2 - 1

419
V Surfaces

(Fig. 23). Next show that X can be transformed into a geometrically ruled surface
X" with invariant e » 0. Then use (2.12), and study how the ruled surface P(6')
with @' decomposable behaves under elmp.]
5.6. Let X be a surface with function field K. Show that every valuation ring R of K/k
is one of the three kinds described in (II, Ex. 4.12). [Hint: In case (3), let fER. Use
(Ex. 5.1) to show that for all i » 0, f E {f)x,, so in fact f E R 0 .]
5.7. Let Y be an irreducible curve on a surface X, and suppose there is a morphism
f: X -+ X 0 to a projective variety X 0 of dimension 2, such that f( Y) is a point P
andr 1 (P) = Y. Then show that Y 2 < 0. [Hint: Let JHJ be a very ample (Cartier)
divisor class on X 0 , let H 0 E JHJ be a divisor containing P, and let H 1 E Jl;lJ be a
divisor not containing P. Then consider f* H 0 ,/* H 1 and fi 0 = f*(H 0 - P)- .]
5.8. A surface singularity. Let k be an algebraically closed field, and let X be the surface
in M defined by the equation x 2 + y 3 + z 5 = 0. It has an isolated singularity at
the origin P = (0,0,0).
(a) Show that the affine ring A = k[ x,y,z ]/(x 2 + y 3 + z5 ) of X is a unique
factorization domain, as follows. Let t = z- 1 ; u = t 3 x, and v = t 2 y. Show
that z is irreducible in A; t E k[ u,v], and A[ z- 1 J = k[u,v,t- 1 ]. Conclude that
A is a UFD.
(b) Show that the singularity at P can be resolved by eight successive blowings-up.
If X is the resulting nonsingular surface, then the inverse image of P is a union
of eight projective lines, which intersect each other according to the Dynkin
diagram E 8 :

Here each circle denotes a line, and two circles are joined by a line segment
whenever the corresponding lines intersect.
Nate. This singularity has interesting connections with local algebra, invariant
theory, and topology.
In case k = C, Mumford [6] showed that the completion A of the ring A at the
maximal ideal m = (x,y,z) is also a UFD. This is remarkable, because in general the
completion of a local UFD need not be UFD, although the converse is true (theorem
of Mori)-see Samuel [3]. Brieskorn [2] showed that the corresponding analytic
local ring C{x,y,z}/(x 2 + y 3 + z5 ) is the only nonregular normal2-dimensional ana-
lytic local ring which is a UFD. Lipman [2] generalized this as follows: over any
algebraically closed field k of characteristic ¥- 2,3,5, the only nonregular normal
complete 2-dimensional local ring which is a UFD is k[[x,y,z]]/(x 2 + y 3 + z5 ).
See also Lipman [3] for a report on recent work connected with UFD's.
This singularity arose classically out of Klein's work on the icosahedron. The
group I of rotations of the icosahedron, which is isomorphic to the simple group
of order 60, acts naturally on the 2-sphere. Identifying the 2-sphere with P~ by
stereographic projection, the group I appears as a finite subgroup of Aut P~. This
action lifts to give an action of the binary icosahedral group I on C 2 by linear
transformations of the complex variables t 1 and t 2 • Klein [2, I, 2, §13, p.62] found
three invariant polynomials x,y, z in t 1 and t 2 , related by the equation x 2 + y 3 +
z 5 = 0. Thus the surface X appears as the quotient of A~ by the action of the
group I. In particular, the local fundamental group of X at P is just I.

420
 6 Classification of Surfaces

With regard to the topology of algebraic varieties over C, Mumford [ 6] showed

that a normal algebraic surface over C, whose underlying topological space (in its
"usual" topology) is a topological manifold, must be nonsingular. Brieskom
showed that this is not so in higher dimensions. For example, the underlying
topological space of the hypersurface in C4 defined by xi + X~ + X~ + xl = 0 is
a manifold. Later Brieskorn [ 1] showed that if one intersects such a singularity
with a small sphere around the singular point, then one may get a topological sphere
whose differentiable structure is not the standard one. Thus for example, by
intersecting the singularity
xi + x~ + x~ + xl + x~k-l = 0
inC5 with a small sphere around the origin, fork = 1,2, ... ,28, one obtains all28
possible differentiable structures on the 7-sphere. See Hirzebruch and Mayer [1 J
for an account of this work.

6 Classification of Surfaces

In the case of curves, we could achieve a classification as follows. Each hi-

rational equivalence class has a unique nonsingular projective model. There
is a numerical invariant, the genus g, which can take on every value g ~ 0.
For fixed g, the curves of genus g are parametrized by the points of the variety
of moduli 9Jl 9 (IV, §5).
For surfaces, the situation is much more complicated. First of all, the non-
singular projective model is not unique. However, we can standardize by
always considering a relatively minimal model. For rational and ruled
surfaces, these are known, and for other birational classes there is a unique
minimal model (5.8.4).
Next, we have the birational invariants Pa (5.6) and p9 (II, 8.19), and K 2 ,
which is well-defined if we specify the minimal model. However, it is not
known exactly which triples of integers can occur as Pa,p9 ,K 2 of a surface.
As to the existence of varieties of moduli, this question is wide open except in
some special cases. So we must settle for less complete information than in
the case of curves.
In this section we will mention very briefly a few basic results, and refer
to Bombieri and Husemoller [1] and Shafarevich [1] for more details and
further references.

To begin with, for any projective variety X over k, we define the Kodaira
dimension K(X) to be the transcendence degree over k of the ring
R = EB H 0 (X,.!£(nK) ),
n?!:O

minus 1, where K is the canonical divisor. One sees, as in the proof of

(II, 8.19), that R and hence K are birational invariants. Another way of
expressing this is that K is the largest dimension of the image of X in pN by
the rational map determined by the linear system lnKI, for some n ~ 1, or

421
V Surfaces

K = -1 if lnKI = 0 for all n ?:; 1. It is known, for varieties of dimension n,

that K can take on every value from -1 to n. For example, for curves, we
have K = - 1 <=> g = 0; K = 0 <=> g = 1 ; K = 1 <=> g ?:; 2.

We will classify surfaces according to K = -1,0,1,2. Some more specific

information about each group is provided by the following theorems.

Theorem 6.1. K = -1 <=> I12KI = 0 <=>X is either rational or ruled.

Theorem 6.2 (Castelnuovo). X is rational<=> Pa = P 2 = 0, where P 2 = dim

H 0 (X,2(2K)) is the second plurigenus.
PROOF. A modern proof over C, due to Kodaira, is given in Serre [13]. In
characteristic p > 0, the proof is due to Zariski [5], [6], [9].

Remark 6.2.1. As a consequence of (6.2), one can prove the analogue of

Liiroth's theorem (IV, 2.5.5) in dimension 2: let k be an algebraically closed
field, let L be a subfield of a pure transcendental extension k(t,u) of k, con-
taining k, such that k(t,u) is a finite separable extension of L. Then L is also
a pure transcendental extension of k. This is Castelnuovo's theorem "on
the rationality of plane involutions."
For the proof, let X' be a nonsingular projective model of L, and let X
be a nonsingular projective model of k(t,u). Then as in (II, 8.19) or (II, Ex. 8.8),
using separability, one shows that p9 (X') ~ p9 (X) and P 2 (X') ~ P 2 (X), hence
p9 (X') = P 2 (X') = 0 since the same is true of X. One must also show that
q(X') ~ q(X) to conclude that Pa(X') = p9 (X') - q(X') = 0. Then the
rationality of X follows from (6.2). See Serre [13] and Zariski [9].
This result is false if one does not assume k(t,u) separable over L-see
Zariski [9] or Shioda [1].

Theorem 6.3. K = 0 <=> 12K = 0. A surface in this class must be one of the
following (assume char k #- 2,3):
(1) a K3 surface, which is defined as a surface with K = 0 and irregularity
q = 0. These have Pa = p9 = 1;
(2) an Enriques surface, which has Pa = p9 = 0 and 2K = 0;
(3) a two-dimensional abelian variety, which has Pa = -1, p9 = 1; or
(4) a hyperelliptic surface, which is a surface .fibred over P 1 by a pencil of
elliptic curves.

Theorem 6.4. A surface with K = 1 is an elliptic surface, which is a surface X

with a morphism n:X --t C to a curve C, such that almost all .fibres of n
are nonsingular elliptic curves (assume char k #- 2,3).

Theorem 6.5. K = 2 if and only if for some n > 0, InK I determines a birational
morphism of X onto its image in pN_ These are called surfaces of general
type.

422
 6 Classification of Surfaces

EXERCISES

6.1. Let X be a surface in P", n ;::, 3, defined as the complete intersection ofhypersurfaces
of degrees d 1 , . . . ,d._ 2 , with each d; ;::, 2. Show that for all but finitely many choices
of (n,d 1 , • •• ,d._ 2 ), the surface X is of general type. List the exceptional cases, and
where they fit into the classification picture.
6.2. Prove the following theorem of Chern and Griffiths. Let X be a nonsingular surface
of degree d in p~+ 1 , which is not contained in any hyperplane. If d < 2n, then
p9 (X) = 0. If d = 2n, then either p9(X) = 0, or pg(X) = 1 and X is a K3 surface.
[Hint: Cut X with a hyperplane and use Clifford's theorem (IV, 5.4). For the last
statement, use the Riemann-Roch theorem on X and the Kodaira vanishing
theorem (III, 7.15).]

423
APPENDIX A

Intersection Theory

In this appendix we will outline the generalization of intersection theory and

the Riemann-Roch theorem to nonsingular projective varieties of any
dimension. To motivate the discussion, let us look at the case of curves and
surfaces, and then see what needs to be generalized. For a divisor D on a
curve X, leaving out the contribution of Serre duality, we can write the
Riemann-Roch theorem (IV, 1.3) as
x(.!Z'(D)) = deg D +1- g,

where xis the Euler characteristic (III, Ex. 5.1). On a surface, we can write
the Riemann-Roch theorem (V, 1.6) as
1
x(!l'(D)) = 2 D.(D - K) + 1 + Pa·
In each case, on the left-hand side we have something involving cohomol-
ogy groups of the sheaf !l'(D), while on the right-hand side we have some
numerical data involving the divisor D, the canonical divisor K, and some
invariants of the variety X. Of course the ultimate aim of a Riemann-Roch
type theorem is to compute the dimension of the linear system IDI or of lnDI
for large n (II, Ex. 7.6). This is achieved by combining a formula for x(!l'(D))
with some vanishing theorems for Hi(X,!l'(D)) fori > 0, such as the theorems
of Serre (III, 5.2) or Kodaira (III, 7.15).
We will now generalize these results so as to give an expression for x(!l'(D))
on a nonsingular projective variety X of any dimension. And while we are
at it, with no extra effort we get a formula for x(t&"), where @" is any coherent
locally free sheaf.
To generalize the right-hand side, we need an intersection theory on X.
The intersection of two divisors, for example, will not be a number, but a

424
 1 Intersection Theory

cycle of codimension 2, which is a linear combination of subvarieties of

codimension 2. So we will introduce the language of cycles and rational
equivalence (which generalizes the linear equivalence of divisors), in order
to set up our intersection theory.
We also need to generalize the correspondence between the invertible
sheaf !l'(D) and the divisor D. This is accomplished by the theory of Chern
classes: to each locally free sheaf. ~ of rank r, we associate Chern classes
c 1 (~), . . . ,c.(~), where ci(~) is a cycle of codimension i, defined up to rational
equivalence.
As for invariants of the variety X, the canonical class K and the arithmetic
genus Pa are not enough in general, so we use all the Chern classes of the
tangent sheaf of X as well.
Then the generalized Riemann-Roch theorem will give a formula for x(~)
in terms of certain intersection numbers of the Chern classes of~ and of the
tangent sheaf of X.

1 Intersection Theory

The intersection theory on a surface (V, 1.1) can be summarized by saying

that there is a unique symmetric bilinear pairing Pic X x Pic X --+ Z, which
is normalized by requiring that for any two irreducible nonsingular curves C,D
meeting transversally, C.D is just the number of intersection points of C and D.
Our main tool in proving this theorem was Bertini's theorem, which allowed
us to move any two divisors in their linear equivalence class, so that they
became differences of irreducible nonsingular curves meeting transversally.

In higher dimensions, the situation is considerably more complicated.

The corresponding moving lemma is weaker, so we need a stronger normal-
ization requirement. It turns out that the most convenient way to develop
intersection theory is to do it for all varieties at once, and include some
functorial mappings f* and f* associated to a morphism f: X --+ X' as part
of the structure.
Let X be any variety over k. A cycle of codimension r on X is an element
of the free abelian group generated by the closed irreducible subvarieties of X
of codimension r. So we write a cycle as Y = L:ni }j where the }j are sub-
varieties, and ni E Z. Sometimes it is useful to speak of the cycle associated
to a closed subscheme. If Z is a closed subscheme of codimension r, let
Y1, . . . , r; be those irreducible components of Z which have codimension r,
and define the cycle associated to Z to be L:ni }j, where ni is the length of
the local ring (1J y;,z of the generic point Yi of }j on Z.
Let f: X --+ X' be a morphism of varieties, and let Y be a subvariety of X.
If dim f(Y) < dim Y, we set f*(Y) = 0. If dim f(Y) = dim Y, then the
function field K(Y) is a finite extension field of K(f(Y) ), and we set
f*(Y) = [K(Y):K(f(Y))] · f(Y).
425
Appendix A Intersection Theory

Extending by linearity defines a homomorphism f* of the group of cycles on X

to the group of cycles on X'.

Now we COf!le to the definition of rational equivalence~ For any subvariety

V of X, let f: V ---+ V be the normalization of V. Then V satisfies the condi-
tion(*) of(II, §6), so we can talk about Weil divisors and linear equivalence
on V. Whenever D and D' are linearly equivalent Weil divisors on V, we say
that f*D and f*D' are rationally equivalent as cycles on X. Then we define
rational equivalence of cycles on X in general by dividing out by the group
generated by all such f*D ""' f*D' for all subvarieties V, and all linearly
equivalent Weil divisors D,D' on V. In particular, if X itself is normal, then
rational equivalence for cycles of codimension 1 coincides with linear
equivalence ofWeil divisors.
For each r we let A'(X) be the group of cycles of codimension r on X
modulo rational equivalence. We denote by A( X) the graded group c:B~= 0
A'(X), where n = dim X. Note that A 0 (X) = Z, and that A'(X) = 0 for
r > dim X. Note also that if X is complete there is a natural group homo-
morphism, the degree, from A"(X) to Z, defined by degQ)iPJ = Ini, where
the Pi are points. This is well-defined on rational equivalence classes because
of (II, 6.1 0).

An intersection theory on a given class of varieties mconsists of giving a

pairing A'( X) x As(X) ---+ Ar+s(X) for each r,s, and for each X E m, satisfying
the axioms listed below. If YEA'( X) and Z E As( X) we denote the inter-
section cycle class by Y.Z.
Before stating the axioms, for any morphism f: X ---+ X' of varieties in m,
we assume that X x X' is also in m, and we define a homomorphism
f*: A(X') ---+ A(X) as follows. For a subvariety Y' <;; X' we define
f*( Y') = P1 *(rJ·P1 1 ( Y') ),
where p 1 and p 2 are the projections of X x X' to X and X', and r J is the
graph off, considered as a cycle on X x X'.
This data is now subject to the following requirements.
Al. The intersection pairing makes A(X) into a commutative associative
graded ring with identity, for every X E m. It is called the Chow ring of X.
A2. F 9r any morphism f: X ---+ X' of varieties in m, f*: A(X') ---+ A(X) is a
ring homomorphism. If g: X' ---+ X" is another morphism, then f* o g* =
(g 0 f)*.
A3. For any proper morphism f:X---+ X' of varieties in m, f*:A(X)---+
A(X') is a homomorphism of graded groups (which shifts degrees). If
g:X' ---+X" is another morphism, then g* of* = (g a f)*.
A4. Projection formula. Iff: X ---+ X' is a proper morphism, if x E A(X)
andy E A(X'), then

426
 I Intersection Theory

A5. Reduction to the diagonal. If Y and Z are cycles on X, and if L1: X ---+
X x X is the diagonal morphism, then

Y.Z = L1*(Y X Z).

A6. Local nature. If Y and Z are subvarieties of X which intersect properly

(meaning that every irreducible component of Y n Z has codimension equal
to codim Y + codim Z), then we can write
Y.Z = Li(Y,Z; ltj)ltj,
where the sum runs over the irreducible components ltj of Y n Z, and where
the integer i(Y,Z; ltj) depends only on a neighborhood of the generic point
of ltj on X. We call i(Y,Z; ltj) the local intersection multiplicity of Y and Z
along ltj.
A7. Normalization. If Y is a subvariety of X, and Z is an effective Cartier
divisor meeting Y properly, then Y.Z is just the cycle associated to the Cartier
divisor Y n Z on Y, which is defined by restricting the local equation of
Z to Y. (This implies in particular that transversal intersections of non-
singular subvarieties have multiplicity 1.)

Theorem 1.1. Let lD be the class of nonsingular quasi-projective varieties over

a fixed algebraically closed field k. Then there is a unique intersection
theory for cycles modulo rational equivalence on the varieties X E m
which
satisfies the axioms Al-A 7 above.

There are two main ingredients in the proof of this theorem. One is the
correct definition of the local intersection multiplicities; the other is Chow's
moving lemma. There are several ways of defining intersection multiplicity.
We just mention Serre's definition, which is historically most recent, but
has the advantage of being compact. If Y and Z intersect properly, and if W
is an irreducible component of Y n Z, we define
i(Y,Z; W) = D -1); length Tor!(A/a,A/b)
where A is the local ring (l)w.x of the generic point of Won X, and a and bare
the ideals of Y and Z in A. Serre [ 11] shows that this is a nonnegative integer,
and that it has the required properties. Note in particular that the naive
definition, taking the length of A/(a + b) = A/a ® A/b, modeled after the
case of curves on a surface (V, 1.4) does not work (1.1.1).
The other ingredient is Chow's moL·ing lemma, which says that if Y,Z are
cycles on a nonsingular quasi-projective variety X, then there is a cycle Z',
rationally equivalent to Z, such that Y and Z' intersect properly. Further-
more, if Z" is another such, then Y.Z' and Y.Z" are rationally equivalent.
There are proofs of this moving lemma by Chevalley [2] and Roberts [1].
The uniqueness of the intersection theory is proved as follows: given cycles
Y,Z on X, by the moving lemma we may assume they intersect properly.

427
Appendix A Intersection Theory

Then using the reduction to the diagonal (AS) we reduce to the case of
computing Ll.( Y x Z) on X x X. This has the advantage that L1 is a local
complete intersection. Since the intersection multiplicity is local (A6) we
reduce to the case where one of the cycles is a complete intersection of
Cartier divisors, and then repeated application of the normalization (A 7)
gives the uniqueness.
Some general references for intersection theory are Wei! [1 ], Chevalley
[2], Samuel [1 ], and Serre [11]. For discussion of some other equivalence
relations on cycles, and attempts to calculate the groups Ai(X), see Harts-
horne [6].

Example 1.1.1. To see why the higher Tor's are necessary, let Y be the union
of two planes in A4 meeting at a point, so the ideal of Y is (x,y) n (z,w) =
(xz,xw,)".?.yw). Let Z be the plane (x - z, y - w). Since Z meets each
component of Y in one point P, we have i( Y,Z; P) = 2 by linearity. How-
ever, if we naively take A/(a + b) where a,b are the ideals of Y and Z, we get
k[x,y,z,w]/(xz,xw,yz,yw,x- z,y- w);::;:: k[x,y]/(x 2 ,xy,y 2 ),
which has length 3.

Example 1.1.2. We cannot expect to have an intersection theory like the

one of the theorem on singular varieties. For example, suppose there was
an intersection theory on the quadric cone Q given by xy = z 2 in P 3 . Let L
be the ruling x = z = 0, and M the ruling y = z = 0. Then 2M is linearly
equivalent to a hyperplane section, which could be taken to be a conic Con Q
which meets L,M each transversally in one point. So
1 = L.C = L.(2M).
By linearity we would have to have L.M = !, which is not an integer.

2 Properties of the Chow Ring

For any nonsingular quasi-projective variety X we now consider the Chow

ring A( X), and list some of its properties. See Chevalley [2] for proofs.
A8. Since the cycles in codimension 1 are just Wei! divisors, and rational
equivalence is the same as linear equivalence for them, and X is nonsingular,
we have A 1 (X) ;::;:: Pic X.
Thus, for example, if X is a nonsingular projective surface, we recover the
intersection theory of (V, 1.1), using the pairing A 1 (X) x A 1 (X) ---+ A 2 (X)
followed by the degree map.
A9. For any affine space Am, the projection p:X x Am---+ X induces an
isomorphism p*: A(X) ---+ A( X x Am).

428
 3 Chern Classes

A10 Exactness. If Y is a nonsingular closed subvariety of X, and U =

X - Y, there is an exact sequence
i* j*
A(Y) ~ A(X) ~ A(U) ~ 0,

where i: Y ~X is one inclusion, andj: U ~X is the other.

The proofs of these two results are similar to the corresponding results
for divisors (II, 6.5), (II, 6.6).

Example 2.0.1. A(P") ~ Z[h ]/h"+ 1, where h in degree 1 is the class of a

hyperplane. One can prove this inductively from (A9) and (AlO), or directly,
by showing that any subvariety of degree d in P" is rationally equivalent to
d times a linear space of the same dimension (Ex. 6.3).

The next property is important for the definition of the Chern classes in
the next section.
All. Let tff be a locally free sheaf of rank ron X, let P(tff) be the associated
projective space bundle (II, §7), and let ~ E A 1(P(tff)) be the class of the
divisor corresponding to @P(c)(1). Let n:P(tff) ~ X be the projection. Then
n* makes A(P(tff)) into a free A(X)-module generated by 1,~,~ 2 , . . . ,~r- 1 .

3 Chern Classes

Here we follow the treatment of Grothendieck [3].

Definition. Let C be a locally free sheaf of rank r on a nonsingular quasi-

projective variety X. For each i = 0,1, ... ,r, we define the ith Chern
class c;(C) E A;(X) by the requirement c0 (C) = 1 and
r
L (-1)in*c;(C).~r-i = 0
i=O

in Ar(P(C) ), using the notation of (All).

This makes sense, because by (All), we can express ~rasa unique linear
combination of 1,~, ... ,~r-t, with coefficients in A(X), via n*. Here are
some properties of the Chern classes. For convenience we define the total
Chern class

and the Chern polynomial

429
Appendix A Intersection Theory

Cl. If tff ~ 2(D) for a divisor D, ct(tff) = 1 + Dt. Indeed, in this case
P(tff) = X, (9P(c)(1) = 2(D), so ~ = D, so by definition 1.~ - c 1 (tff).1 = 0,
so c 1(tff) =D.
C2. If f:X' ~X is a morphism, and tff is a locally free sheaf on X, then
for each i
c;(f*tff) = f*c;(tff).
This follows immediately from the functoriality properties of the P(tff) con-
struction and f*.
C3. If 0 ~ tff' ~ tff ~ tff" ~ 0 is an exact sequence of locally free sheaves
on X, then

In fact, forgetting the definition for a moment, one can show that there
is a unique theory of Chern classes, which for each locally free sheaf tff on
X assigns c;(tff) E A;(X), satisfying (C1), (C2), and (C3). For the proof of this
uniqueness and for the proof of (C3) and the other properties below, one
uses the splitting principle, which says that given tff on X, there exists a
morphism f: X' ~ X such that f*: A(X) ~ A(X') is injective, and Iff' = j*tff
splits, i.e., it has a filtration Iff' = tff~ ;2 1!'1 ;2 . . . ;2 tff~ = 0 whose successive
quotients are all invertible sheaves. Then one uses the following property.
C4. If tff splits, and the filtration has the invertible sheaves 2 1, ... ,2,
as quotients then
ct(tff) = n ct(2;).
i= 1
r

(And of course we know each ct(2;) from (C1).)

Using the splitting principle, we can also calculate the Chern classes of
tensor products, exterior products, and dual locally free sheaves. Let tff
have rank r, and let fF have rank s. Write

ct(tff) = nr

i= 1
(1 + a;t)
and

ct(ff) = ns

i= 1
(1 + b;t),
where a 1 , ... ,a" b 1, ... ,bs are just formal symbols. Then we have
cs.
ct(tff 0 ff) = n (1 + (a; + b)t)
i,j

cr( 1\ Ptff) =
l~i1<
n
... <ip~r
(1 + (a;, + ... + a;)t)

430
 4 The Riemann-Roch Theorem

These expressions make sense, because when multiplied out, the coeffi-
cients of each power oft are symmetric functions in the a; and the bj. Hence
by a well-known theorem on symmetric functions, they can be expressed as
polynomials in the elementary symmetric functions of the a; and bj, which
are none other than the Chern classes of C and :F. For a further reference
on this formalism, see Hirzebruch [1, Ch. I, §4.4].
C6. Let s E r(X,C) be a global section of a locally free sheaf C of rank r
on X. Then s defines a homomorphism ((jx--+ C by sending 1 to s. We
define the scheme of zeros of s to be the closed subscheme Y of X defined
by the exact sequence
g~ :::. ((jx --+ @y --+ 0

where s~ is the dual of the map s. Let Y also denote the associated cycle
of Y. Then if Y has codimension r, we have cr(C) = Yin Ar(X).
This generalizes the fact that a section of an invertible sheaf gives the
corresponding divisor (II, 7.7).
C7. Self-intersection formula. Let Y be a nonsingular subvariety of X of
codimension r, and let JV be the normal sheaf (II, §8). Let i: Y--+ X be the
inclusion map. Then
i*i*(1y) = cr(JV).
Therefore, applying the projection formula (A4) we have
i*(cr(JV)) = Y. Y
on X.
This result, due to Mumford (see Lascu, Mumford, and Scott [1]), gen-
eralizes the self-intersection formula (V, 1.4.1) for a curve on a surface.

4 The Riemann-Roch Theorem

Let C be a locally free sheaf of rank r on a nonsingular projective variety X

of dimension n, and let fT = fix be the tangent sheaf of X (II, §8). We
want to give an expression for x(C) in terms of the Chern classes of C and
fT. For this purpose we introduce two elements of A(X) ® Q, which are
defined as certain universal polynomials in the Chern classes of a sheaf C.
Let
r

C1(C) = fl (1 + a;t)
i= 1

as above, where the a; are formal symbols. Then we define the exponential
Chern character
r
ch(C) = L ea;,
i= 1
where

431
Appendix A Intersection Theory

and the Todd class of Iff,

r ai
td($) = I l l - e a,
where
X 1 1 2 1 4
1- e x = l + 2X + 12 X - 720 X + ...
As before, these are symmetric expressions in the a;, so can be expressed as
polynomials in the c;(S), with rational coefficients. By elementary but
tedious calculation, one can show using these definitions that

ch($) = r + c 1 + 21 (c 21 - 2c 2 ) + 61 (c 31 - 3c 1 c 2 + 3c 3 )

and

1
- 720 4
(cl - 4c 21 c 2 - 3c 22 - c 1c3 + c4 ) + ...
where we set c; = c;($), c; = 0 if i > r.

Theorem 4.1 (Hirzebruch-Riemann-Roch). For a locally free sheaf @" of

rank r on a nonsingular projective variety X of dimension n,
x(S) = deg(ch(S).td(§") )",
where ( )n denotes the component of degree n in A(X) ® Q.

This theorem was proved by Hirzebruch [1 J over C, and by Grothendieck

in a generalized form (5.3) over any algebraically closed field k (see Borel and
Serre [1 ]).

Example 4.1.1. If X is a curve, and@" = !l'(D), we have ch($) = 1 + D. The

tangent sheaf 5""x is the dual of f2x. Therefore 5""x ~ !l'(- K), where K is
the canonical divisor, and so td(5""x) = 1 - !K. Thus (4.1) tells us that

x(!l'(D)) = deg((l + D)(l- ~K)) 1

= deg (D- ~K}
ForD = 0, this says that 1 - g = -t deg K, so we can write the theorem as
x(!l'(D)) = deg D + 1 - g,
which is the Riemann-Roch theorem for curves proved earlier (IV, 1.3).

432
 4 The Riemann-Roch Theorem

Example 4.1.2. Now let X be a surface, and again let $ = !l'(D). Then
ch($) = 1 + D + !D 2 . We denote by c 1 and c 2 the Chern classes of the
tangent sheaf ffx. These depend only on X, so they are sometimes called
the Chern classes of X. Since ffx is the dual of Qx, and since c 1 (Qx) =
c 1 (/\ 2 Qx)by(C5),andsince J\ 2 Qx = wxisjust!l'(K),whereKisthe canoni-
cal divisor, we see that c 1C'1x) = -K. But c2 (or rather its degree) is a new
numerical invariant of a surface which we have not met before.
Using c 1 = -K and c 2 , we have

td(ffx) = 1 - ~ K + 112 (K 2 + c2 ).
Multiplying, and taking degrees (by abuse of notation we let D 2 denote both
the class in A 2 ( X), and its degree), we can wr.ite (4.1) as
1 1
x(!l'(D)) = 2 D.(D - K) + 12 (K 2 + c2).

In particular, forD = 0, we find that

1 2
X(@x) = 12 (K + C2).

By definition of the arithmetic genus (III, Ex. 5.3), this says

1 2
1 + Pa = 12 (K + c2).
So the new Riemann-Roch theorem for surfaces gives us the earlier one
(V, 1.6), together with the additional information that c 2 can be expressed
in terms of the invariants p0 ,K 2 by this last formula.

Example 4.1.3. As an application, we derive a formula relating the numerical

invariants of a surface in P 4 . For perspective, note that if X is a surface of
degree din P 3 , then the numerical invariants p0 ,K 2 , and hence c 2 , are uniquely
determined by d (1, Ex. 7.2) and (V, Ex. 1.5). On the other hand, any projective
surface can be embedded in P 5 (IV, Ex. 3.11), so for surfaces in P 5 we do not
expect any particular connection between these invariants. However, a
surface cannot in general be embedded in P 4 , so for those which can, we
expect some condition to be satisfied.
So let X be a nonsingular surface of degree din P4 . In the Chow ring of
P , X is equivalent to d times a plane, so X.X = d 2 . On the other hand, we
4

can compute X.X, using the self-intersection formula (C7), as deg cz(JV)
where JV is the normal bundle of X in P 4 . There is an exact sequence

where i:X -+ P 4 is the inclusion. We use this sequence to compute c2 (JV),

and thus get our condition.

433
Appendix A Intersection Theory

First we use the exact sequence

0 --+ C9p4 --+ C9p.(1 )5 --+ ffp4 --+ 0.
Letting h E A 1(P 4 ) be the class of a hyperplane, we see that
C1 (ffp.) = (1 + ht) 5 = 1 + 5ht + 10h 2 t 2 + ...
On the other hand,
C1 (ffx) = 1 - Kt + c2 t 2
as in (4.1.2). Therefore, denoting by HE A \X) the class of a hyperplane
section of X, we have from the exact sequence above, using (C3), that
(1 - Kt + c 2 t 2 )(1 + clA')t + c 2 (%)t 2 ) = 1 + 5Ht + 10H 2 t 2 •
Comparing coefficients of t and t 2 , we find that
c 1(%) = 5H +K
c 2 (%) = 10H 2 - c2 + 5H.K + K 2•
Now take degrees and combine with deg c2 (%) = d 2 • Also note that
deg H 2 = d, and use the expression for c2 in (4.1.2). The final result is
d2 - 10d - 5H.K - 2K 2 + 12 + 12pa = 0.
This holds for any nonsingular surface of degree d in P 4 . See (Ex. 6.9) for
some applications.

5 Complements and Generalizations

Having developed an intersection theory for n-dimensional varieties, we can

ask whether some of the other theorems we proved for surfaces in Ch. V
also extend. They do.

Theorem 5.1 (Nakai-Moishezon Criterion). Let D be a Cartier divisor on a

scheme X which is proper over an algebraically closed field k. Then D is
ample on X if and only if for every closed integral subscheme Y ~ X
(including the case Y = X if X is integral), we have D'. Y > 0 where
r = dim Y.

This theorem was proved by Nakai [1] for X projective over k, and in-
dependently by Moishezon [ 1J for X an abstract complete variety. The
proof was clarified and simplified by Kleiman [ 1]. Strictly speaking, this
theorem uses a slightly different intersection theory than the one we have
developed. We do not assume X nonsingular projective, so we do not have
Chow's moving lemma. On the other hand, the only intersections we need
to consider are those of a number of Cartier divisors with a single closed
subscheme. And this intersection theory is in fact more elementary to develop

434
 5 Complements and Generalizations

than the one we outlined in §1. See Kleiman [1] for details. Notice that this
theorem extends the one given earlier for a surface (V, 1.10), because taking
Y = X gives D 2 > 0, and for Y a curve we have D. Y > 0.

We can also generalize the Hodge index theorem (V, 1.9) to a nonsingular
projective variety X over C. We consider the associated complex manifold
Xh (App. B) and its complex cohomology Hi(Xh,C). For any cycle Y of
codimension ron X, one can define its cohomology class 17(Y) E H 2 '(Xh,C).
We say that Y is homologically equivalent to zero, written Y ~hom 0, if
17(Y) = 0.

Theorem 5.2 (Hodge Index Theorem). Let X be a nonsingular projective

variety over C, of even dimension n = 2k. Let H be an ample divisor on
X, let Y be a cycle of codimension k, and assume that Y.H ~ham 0, and
Y '7'-ham 0. Then ( -1)kY 2 > 0.

This theorem is proved using Hodge's theory of harmonic integrals-

see Weil [5, Th. 8, p. 78]. It generalizes the earlier result for surfaces (V, 1.9),
because for divisors, one can show that homological and numerical equiva-
lence coincide. It is conjectured by Grothendieck [9] to be true over an
arbitrary algebraically closed field k, using 1-adic cohomology for the
definition of homological equivalence. He suggests that it might also be
true using numerical equivalence of cycles, but that isn't known even over
C. See Kleiman [2] for a discussion of these and Grothendieck's other
"standard conjectures."

Now let us turn to further generalizations of the Riemann-Roch theorem

following Borel and Serre [1]. The first step is to extend the definition of
the Chern classes to the Grothendieck group K(X) (II, Ex. 6.10). For X
nonsingular, we can compute K(X) using only locally free sheaves (Ill,
Ex. 6.9). Then, because of the additivity property (C3) of Chern classes, it
is clear that the Chern polynomial c1 extends to give a map
c1 :K(X)-+ A(X)[t].
Thus we have Chern classes defined on K(X). The exponential Chern
character ch extends to give a mapping
ch: K(X) -+ A( X) ® Q.
One shows that K(X) has a natural ring structure (defined by Iff ® :#' for
locally free sheaves Iff and :#'), and that ch is a ring homomorphism. If
f: X' -+ X is a morphism of nonsingular varieties, then there is a ring
homomorphism
f':K(X)-+ K(X')
defined by Iff H f*!% for locally free Iff. The exponential Chern character
ch commutes with f'.

435
Appendix A Intersection Theory

lff:X --+ Yis a proper morphism, one defines an additive mapf,:K(X)--+

K(Y) by

for :#' coherent.

This map f, does not commute with ch. The extent to which it fails to
commute is the generalized Riemann-Roch theorem of Grothendieck.

Theorem 5.3 (Grothendieck-Riemann-Roch). Let f:X--+ Y be a smooth

projective morphism of nonsingular quasi-projective varieties. Then for
any x E K(X) we have
ch(f,(x)) = f*(ch(x).td(§"f))
in A(Y) ® Q, where §"f is the relative tangent sheaf off

If Y is a point, this reduces to the earlier form (4.1 ).

After wrestling with formidable technical obstacles, this theorem has been
further generalized in the Paris seminar of Grothendieck [SGA 6], to the
case where Y is a noetherian scheme admitting an ample invertible sheaf,
and f is a projective locally complete intersection morphism. See Manin
[1] for a readable account of this work.
We should also mention another Riemann-Roch formula, for the case
of a closed immersion f: X c. Y of nonsingular varieties, due to Jouanolou
[1]. In the case of a closed immersion, the formula of (5.3) gives a way of
computing the Chern classes c;(f*:F), for any coherent sheaf:#' on X, in
terms of f*(C;(:F)) and f*(C;(%) }, where % is the normal sheaf. It turns
out this can be done using polynomials with integer coefficients, but the
proofof(5.3) gives the result only in A(Y) ® Q, i.e., mod torsion. Jouanolou's
result is that the result actually holds in A(Y) itself.
Recently another kind of generalization of the Riemann-Roch theorem
to singular varieties has been developed by Baum, Fulton, and MacPherson.
See Fulton [2].

EXERCISES

6.1. Show that the definition of rational equivalence in §1 is equivalent to the equiva-
lence relation generated by the following relation: two cycles Y,Z of codimension
r on X are equivalent if there exists a cycle W of codimension r on X x A 1 ,
which intersects X x { 0} and X x { 1} properly, and such that Y = W.(X x { 0} ),
Z=W.(Xx {1}).

6.2. Prqve the following result about Wei! divisors, which generalizes (IV, Ex. 2.6),
and which is needed to show that f* is well-defined modulo rational equivalence
(A3): Let f:X--> X' be a proper, generically finite map of normal varieties, and
let D 1 and D2 be linearly equivalent Weil divisors on X. Then f*D 1 and f*D 2
are linearly equivalent Wei! divisors on X'. [Hint: Remove a subset ofcodimen-
sion ~ 2 from X' so that f becomes a finite flat morphism, then generalize
(IV, Ex. 2.6).]

436
 5 Complements and Generalizations

6.3. Show directly that any subvariety of degree d in P" is rationally equivalent to d
times a linear space of the same dimension, by using a projection argument
similar to (III, 9.8.3).
6.4. Let n:X-+ C be a ruled surface (V, §2) over a nonsingular curve C. Show that
the group A 2 (X) of zero-cycles modulo rational equivalence is isomorphic to
Pic C.
6.5. Let X be a surface, let P E X be a point, and let n: X -+ X be the monoidal trans-
formation with center P (V, §3). Show that A( X) ~ n* A( X) EB Z, where Z is
generated by the exceptional curve E E A 1(X), and the intersection theory is
determined by E 2 = -n*(P).
6.6. Let X be a nonsingular projective variety of dimension n, and let Ll .-:; X x X
be the diagonal. Show that c.(ffx) = Ll 2 in A"(X), under the natural isomor-
phism of X with Ll.
6.7. Let X be a nonsingular projective 3-fold, with Chern classes c 1 ,c 2 ,c 3 . Show that
1
1 - Pa = 24 c 1 c2,
and for any divisor D,
1 1
x(.'l'(D)) = 12 D.(D - K).(2D - K) + 12 D.c 2 + 1 - Pa·

6.8. Let Iff be a locally free sheaf of rank 2 on P 3 , with Chern classes c 1 ,c 2 • Since
A(P 3 ) = Z[h]/h 4 , we can think of c 1 and c2 as integers. Show that c 1 c 2 = 0
(mod 2). [Hint: In the Riemann-Roch theorem for Iff, the left-hand side is au-
tomatically an integer, while the right-hand side is a priori only a rational number.]
6.9. Surfaces in P 4 .
(a) Verify the formula of (4.1.3) for the rational cubic scroll in P 4 (V, 2.19.1).
(b) If X is a K3 surface in P4 , show that its degree must be 4 or 6. (Examples of
such are the quartic surface in P 3 , and the complete intersection of a quadric
and a cubic hypersurface in P 4 .)
(c) If X is an abelian surface in P 4 , show that its degree must be 10. (Horrocks
and Mumford [1] have shown that such abelian surfaces exist.)
*(d) Determine which of the rational ruled surfaces X., e ~ 0 (V, §2), admit an
embedding in P4 .
6.10. Use the fact that the tangent sheaf on an abelian variety is free, to show that it
is impossible to embed an abelian 3-fold in P 5 .

437
APPENDIX B

Transcendental Methods

If X is a nonsingular variety over C, then we can also consider X as a complex

manifold. All the methods of complex analysis and differential geometry
can be used to study this complex manifold. And, given an adequate dic-
tionary between the language of abstract algebraic geometry and complex
manifolds, these results can be translated back into results about the original
variety X.
This is an extremely powerful method, which has produced and is still
producing many important results, proved by these so-called "transcen-
dental methods," for which no purely algebraic proofs are known.
On the other hand, one can ask where do the algebraic varieties fit into
the general theory of complex manifolds, and what special properties
characterize them among all complex manifolds?
In this appendix we will give a very brief report of this vast and important
area of research.

1 The Associated Complex Analytic Space

A complex analytic space (in the sense of Grauert) is a topological space X,

together with a sheaf of rings (!Jx, which can be covered by open sets, each
of which is isomorphic, as a locally ringed space, to one of the following
kind Y: let U ~ C" be the polydisc {Jzd < 1Ji = 1, ... ,n}, let f 1, . . . ,fq
be holomorphic functions on U, let Y ~ U be the closed subset (for the
"usual" topology) consisting of the common zeros of f 1 , . . . ,fq, and take
(!JY to be the sheaf (!Ju!Ut. ... ,fq), where (!Ju is the sheaf of germs of holo-
morphic functions on U. Note that the structure sheaf may have nilpotent
elements. See Gunning and Rossi [1] for a development of the general theory

438
 1 The Associated Complex Analytic Space

of complex analytic spaces, coherent analytic sheaves, and cohomology.

See also Banica and Stana~ila [ 1] for a survey of recent techniques in the
cohomology of complex analytic spaces, parallel to the algebraic techniques
in Chapter III.
Now if X is a scheme of finite type over C, we define the associated com-
plex analytic space X h as follows. Cover X with open affine subsets Y; =
Spec A;. Each A; is an algebra of finite type over C, so we can write it as
A;~ C[x 1 , . . . ,xn]/(!1 , . . . ,Jq). Here/1 , . . . ,fqarepolynomialsinx 1 , . . . ,xn.
We can regard them as holomorphic functions on C", so that their set of
common zeros is a complex analytic subspace (Y;)h ~ C". The scheme X is
obtained by glueing the open sets Y;, so we can use the same glueing data
to glue the analytic spaces (Y;h into an analytic space Xh. This is the asso-
ciated complex analytic space of X.
The construction is clearly functorial, so we obtain a functor h from the
category of schemes of finite type over C to the category of complex analytic
spaces. In a similar way, if :F is a coherent sheaf on X, one can define the
associated coherent analytic sheaf :Fh as follows. The sheaf :F is locally
(for the Zariski topology) a cokernel
mv ~ m~ --+ :F --+ o
of a morphism cp of free sheaves. Since the usual topology is finer than the
Zariski topology, Uh is open in Xh. Furthermore, since cp is defined by a
matrix of local sections of mu, these give local sections of muh' so we can
define :Fh as the co kernel of the corresponding map cph of free coherent
analytic sheaves locally.
One can prove easily some basic facts about the relationship between a
scheme X and its associated analytic space Xh (see Serre [4]). For example,
X is separated over C if and only if X h is Hausdorff. X is connected in the
Zariski topology if and only if X h is connected in the usual topology. X is
reduced if and only if X h is reduced. X is smooth over C if and only if X h
is a complex manifold. A morphism f: X --+ Y is proper if and only if
J,: X h --+ Y, is proper in the usual sense, i.e., the inverse image of a compact
set is compact. In particular, X is proper over C if and only if X h is compact.
One can also compare the cohomology of coherent sheaves on X and X h.
There is a continuous map cp: X h --+ X of the underlying topological spaces,
which sends Xh bijectively onto the set of closed points of X, but of course
the topology is different. There is also a natural map of the structure
sheaves cp- 1mx--+ mxh' which makes cp into a morphism of locally ringed
spaces. It follows from our definitions that for any coherent sheaf :F of
mx-modules, :Fh ~ cp* :F. From this one can show easily that there are
natural maps of cohomology groups
a;: Hi(X,:F)--+ Hi(Xh,:Fh)
for each i. Here we always take cohomology in the sense of derived functors
(III, §2), but one can show that on the analytic space X h• this coincides with

439
Appendix B Transcendental Methods

the other cohomology theories in use in the literature-see the historical

note at the end of(III, §2).

2 Comparison of the Algebraic and

Analytic Categories

To stimulate our thinking about the comparison between schemes of finite

type over C and their associated complex analytic spaces, let us consider
five questions which arise naturally from contemplating the functor h.

Ql. Given a complex analytic space .r, does there exist a scheme X such
that xh ~ .I?

Q2. If X and X' are two schemes such that Xh ~ X~, then is X ~ X'?

Q3. Given a scheme X, and a coherent analytic sheaf 1Y on Xh, does

there exist a coherent sheaf ff on X such that ff h ~ 1J?

Q4. Given a scheme X, and two coherent sheaves Iff and ff on X such
that lffh ~ ffh on Xh, then is Iff ~ ff?

Q5. Given a scheme X, and a coherent sheaf ff, are the maps rx; on
cohomology isomorphisms?

As one might expect, when phrased in this generality, the answer to all
five questions is NO. It is fairly easy to give counterexamples to Q1, Q3,
and Q5-see (Ex. 6.1), (Ex. 6.3), (Ex. 6.4). Q2 and Q4 are more difficult,
so we mention the following example.

Example 2.0.1 (Serre). Let C be an elliptic curve, let X be the unique non-
trivial ruled surface over C with invariant e = 0 (V, 2.15), and let C 0 be
the section with C6 = 0 (V, 2.8.1). Let U = X - C 0 . On the other hand,
let U' = (A 1 - {0}) x (A 1 - {0}). Then one can show that Uh ~ U~, but
U ;t_ U', because U is not affine. In particular, Uh is Stein although U is not
affine. Furthermore, one can show that Pic U ~ Pic C, whereas Pic U h ~
Z. In particular, there are nonisomorphic invertible sheaves fi' and !£'' on
U such that fi'h ~ !£'~. For details see Hartshorne [5, p. 232].

In contrast, if one restricts one's attention to projective schemes, then

the answer to all five questions is YES. These results were proved by Serre
in his beautiful paper GAGA (Serre [4]). The main theorem is this.

Theorem 2.1 (Serre). Let X be a projective scheme over C. Then the functor
h induces an equivalence of categories from the category of coherent
sheaves on X to the category of coherent analytic sheaves on Xh. Further-

440
 3 When is a Compact Complex Manifold Algebraic?

more, for every coherent sheaf :F on X, the natural maps

rx;:H;(X,.'F)-> H;(Xh,:Fh)
are isomorphisms, for all i.

This answers questions Q3, Q4, and Q5. The proof requires knowing
the analytic cohomology groups H;(P~,(D(q)) for all i,n,q, which can be
computed using Cartan's Theorems A and B. The answer is the same as
in the algebraic case (III, 5.1). Then the result follows from the standard
technique of embedding X in P", and resolving :F by sheaves of the
form L@(q;), as in (III, §5). This theorem has also been generalized by
Grothendieck [SGA 1, XII] to the case when X is proper over C.
As a corollary, Serre obtains a new proof of a theorem of Chow [1 J:

Theorem 2.2 (Chow). If X is a compact analytic subspace of the complex

manifold P(:, then there is a sub scheme X c;; P" with X h = X.

This answers Q1 in the projective c-ase. We leave Q2 as an exercise

(Ex. 6.6).

3 When is a Compact Complex

Manifold Algebraic?

If X is a compact complex manifold, then one can show that a scheme X

such that X h ~ X, if it exists, is unique. So if such an X exists, we will simply
say X is algebraic. Let us consider the modified form of question 1:

Ql'. Can one give reasonable necessary and sufficient conditions for a
compact complex manifold X to be algebraic?

The first result in this direction is

Theorem 3.1 (Riemann). Every compact complex manifold of dimension 1

(i.e., a compact Riemann surface) is projective algebraic.

This is a deep result. To understand why, recall that the notion of com-
plex manifold is very local. It is defined by glueing small discs, with holo-
morphic transition functions. We make one global hypothesis, namely
that it is compact, and our conclusion is that it can be embedded globally
in some projective space. In particular, by considering a projection to P 1,
we see that it has nonconstant meromorphic functions, which is not at all
obvious a priori. One proves the theorem in two steps:
(a) One shows that X admits a global nonconstant meromorphic func-
tion. This requires some hard analysis. One proof, given by Weyl [1 ],
following Hilbert, uses Dirichlet's minimum principle to prove the existence

441
Appendix B Transcendental Methods

of harmonic functions, and hence of meromorphic functions. Another

proof, given by Gunning [1 ], uses distributions to first prove the finite-
dimensionality of cohomology of coherent analytic sheaves, and then de-
duce the existence of meromorphic functions.
(b) The second step is to take a nonconstant meromorphic function f
on X, and to regard it as giving a finite morphism of X to P 1 . Then one
shows that X is a nonsingular algebraic curve, hence projective.
Part (b), which is often called the "Riemann existence theorem," is ele-
mentary by comparison with part (a). It has been generalized to higher
dimensions by Grauert and Remmert [1]. Recently Grothendieck [SGA 1,
XII] has given an elegant proof of their generalization, using Hironaka's
resolution of singularities. The result is this:

Theorem 3.2 (Generalized Riemann Existence Theorem). Let X be a normal

scheme of finite type over C. Let X' be a normal complex analytic space,
together with a finite morphism f:X'--+ Xh. (We define a finite morphism
of analytic spaces to be a proper morphism with finite fibres.) Then there
is a unique normal scheme X' and a finite morphism g: X' --+ X such that
X~ ~ X' and gh = f.

One corollary of this theorem is that the algebraic fundamental group

of X, n~1 g(X), defined as the inverse limit of the Galois groups of finite etale
covers of X (IV, Ex. 4.8), is isomorphic to the completion n~0 P(Xhf of the
usual fundamental group of X h with respect to subgroups of finite index.
Indeed, if 'D is any finite unramified topological covering space of X h• then
'D has a natural structure of normal complex analytic space, so by the
theorem it is algebraic (and etale) over X.

In dimensions greater than 1, it is no longer true that every compact

complex manifold is algebraic. But we have the following result which
gives a necessary condition.

Proposition 3.3 (Siegel [ 1] ). Let X be a compact complex manifold of dimen-

sion n. Then the field K(X) of meromorphic functions on X has transcen-
dence degree ~ n over C, and (at least in the case tr.d. K(X) = n) it is a
finitely generated extension field of C.

If X is algebraic, say X ~ Xh, then one can show that K(X) ~ K(X),
the field of rational functions on X, so in this case we must have tr.d. K(X) =
n. Compact complex manifolds X with tr.d. K(X) = dim X were studied by
Moishezon [2], so we call them M oishezon manifolds.
In dimension n ?: 2, there are compact complex manifolds with no non-
constant meromorphic functions at all, so these cannot be algebraic. For
example, a complex torus Cn/A, where A ~ Z 2 n is a sufficiently general
lattice, for n ?- 2, will have this property. See for example Morrow and
Kodaira [1].

442
 3 When is a Compact Complex Manifold Algebraic?

Restricting our attention to Moishezon manifolds, we have the following

theorem in dimension 2.

Theorem 3.4 (Chow and Kodaira [1]). A compact complex manifold of di-
mension 2, with two algebraically independent meromorphic functions, is
projective algebraic.

In dimensions ~ 3, both Hironaka [2] and Moishezon [2] have given

examples of Moishezon manifolds which are not algebraic. They exist in
every birational equivalence class of algebraic varieties of dimension ~ 3
over C. However, Moishezon shows that any Moishezon manifold be-
comes projective algebraic after a finite number of monoidal transforma-
tions with nonsingular centers, so they are not too far from being algebraic.

Example 3.4.1 (Hironaka [2]). We describe two examples with a similar

construction. The first is a nonsingular complete algebraic three-fold over
C which is not projective. The second is a Moishezon manifold of dimen-
sion three which is not algebraic.
For the first example, let X be any nonsingular projective algebraic
three-fold. Take two nonsingular curves c,d s; X which meet transversally
at two points P,Q, and nowhere else (Fig. 24). On X - Q, first blow up the
curve c, then blow up the strict transform of the curve d. On X - P, first

Figure 24. A complete nonprojective variety.

443
Appendix B Transcendental Methods

blow up the curve d, then blow up the strict transform of the curve c. On
X - P - Q it doesn't matter in which order we blow up the curves c and
d, so we can glue our blown-up varieties along the inverse images of X -
P - Q. The result is a nonsingular complete algebraic variety X. We will
show that X is not projective. It follows, incidentally, that the birational
morphism f: X --+ X cannot be factored into any sequence of monoidal
transformations, because it is not a projective morphism.
To do this, we must examine what happens in a neighborhood of P
(Fig. 24). Let l be the inverse image in X of a general point of c. Let m be
the inverse image of a general point of d. Note that land mare projective
lines. Then the inverse image of P consists of two lines 10 and m 0 , and we
have algebraic equivalence of cycles l "' 10 + m0 and m "' m0 . Note the
asymmetry resulting from the order in which we blew up the two curves.
Now in the neighborhood of Q the opposite happens. So f- 1 (Q) is the
union of two lines I~ and m~, and we have algebraic equivalence l "' 10 and
m "' l~ + m~. Combining these equivalences, we find that /0 + m~ "' 0.
This would be impossible on a projective variety, because a curve has a
degree, which is a positive integer, and degrees are additive and are pre-
served bv algebraic equivalence. So X is not projective.

Example 3.4.2. For the second example, we start with any nonsingular pro-
jective algebraic threefold, as before. Let c be a curve in X which is non-
singular except for one double point P, having distinct tangent directions.
In a small analytic neighborhood of P, blow up one branch first, then the
other. Outside of that neighborhood, just blow up c. Then glue to obtain
the compact complex manifold X. Clearly X is Moishezon, because the
meromorphic functions on X are the same as those on X. We will show
that X is not an abstract algebraic variety.
Using the same notation as before (Fig. 25) we have homological equiva-
lences l "' 10 + m 0 , m "' m0 , and l "' m, because the two branches meet

Figure 25. A nonalgebraic Moishezon manifold.

444
 4 Kahler Manifolds

away from P. So we find that 10 ~ 0. But this is impossible if x is algebraic.

Indeed, let T be a point of 10 . Then T has an affine neighborhood U in x.
Let Y be an irreducible surface in U which passes through T but does not
contain 10 . Extend Y by closure to a surface Yin x. Now Y meets 10 in a
finite nonzero number of points, so the intersection number of Y with 10 is
defined and is =f. 0. But the intersection number is defined on homology
classes, so we cannot have 10 ~ 0. Hence x is not algebraic.

Remark 3.4.3. At this point we should also mention the algebraic spaces of
Artin [2] and Knutson [1]. Over any field k, they define an algebraic space
to be something which is locally a quotient of a scheme by an etale equiva-
lence relation. The category of algebraic spaces contains the category of
schemes. If X is an algebraic space of finite type over C, one can define its
associated complex analytic space Xh. Artin shows that the category of
smooth proper algebraic spaces over C is equivalent, via the functor h, to
the category of Moishezon manifolds. Thus every Moishezon manifold is
"algebraic" in the sense of algebraic spaces. In particular, Hironaka's
example (3.4.2) gives an example of an algebraic space over C which is not
a scheme.

4 Kahler Manifolds

The methods of differential geometry provide a powerful tool for the study
of compact complex manifolds, and hence of algebraic varieties over the
complex numbers. Notable among such applications of differential geome-
try are Hodge's [1] theory of harmonic integrals, and the resulting decom-
position of the complex cohomology into its (p,q)-components (see also
Wei! [5]); the vanishing theorems ofKodaira [1] and Nakano [1], recently
generalized by Grauert and Riemenschneider [ 1J; and the work of Griffiths
on the intermediate Jacobians and the period mapping. Here we will only
mention the definition of a Kahler manifold, and how that notion helps to
characterize algebraic complex manifolds.
Any complex manifold admits a Hermitian metric (in many ways). A
Hermitian metric is said to be Kahler if the associated differential 2-form of
type (1,1) is closed. A complex manifold with a Kahler metric is called a
Kahler manifold. One can show easily that complex projective space has a
natural Kahler metric on it, and hence that every projective algebraic
manifold is a Kahler manifold with the induced metric. A compact Kahler
manifold x is called a Hodge manifold if the cohomology class in H 2 (x,C)
of the 2-form mentioned above is in the image of the integral cohomology
H 2 (x,Z). Now a fundamental result is

Theorem 4.1 (Kodaira [2]). Every Hodge manifold is projective algebraic.

445
Appendix B Transcendental Methods

This can be thought of as a generalization of the theorem of Riemann

(3.1) quoted above, because every compact complex manifold of dimension
one is trivially seen to be a Hodge manifold. We also have the following

Theorem 4.2 (Moishezon [2]). Every Moishezon manifold which is Kahler is

projective algebraic.

Summing up, we have the following implications among properties of

compact complex manifolds, and there are examples to show that no further
implications are possible.

projective abstract
I Hodge I <=> algebraic
==>
algebraic

' A
I Kiihle' I
I
Moishezon

5 The Exponential Sequence

Let us give one simple example of the use of transcendental methods, by

looking at the exponential sequence. The exponential function f(x) = e2 "ix
gives an exact sequence of abelian groups
0 ---+ Z ---+ C ~ C* ---+ 0,
where C has its additive structure, and C* = C - {0} has its multiplicative
structure. If l: is any reduced complex analytic space, by considering holo-
morphic functions with values in the above sequence, we get an exact
sequence of sheaves

where Z is the constant sheaf, (!)I is the structure sheaf, and (!)~ is the sheaf
of invertible elements of(!) I under multiplication.
The cohomology sequence of this short exact sequence of sheaves is very
interesting. Let us apply it to X h' where X is a projective variety over C.
At the H 0 level, we recover the original exact sequence of groups 0 ---+ Z ---+
C ---+ C* ---+ 0, because the global holomorphic functions are constant. Then
starting with H\ we have an exact sequence
0---+ H 1 (X h,Z)---+ H 1(Xh,(!Jx_}---+ H 1(Xh,(!Jt)---+ H 2 (X h,Z)---+ H 2 (X h,(!Jxh)---+ .. ..
By Serre's theorem (2.1), we have H;(Xh,(!Jxh) ;::; H;(X,(!Jx). On the other
hand

446
 5 The Exponential Sequence

by (III, Ex. 4.5), which is valid for any ringed space. But Serre's theorem
(2.1) also gives an equivalence of categories of coherent sheaves, so m
particular, Pic Xh ~ Pic X. So we can rewrite our sequence as
0 ----+ H 1(Xh,Z) ----+ H 1(X,@x) ----+ Pic X ----+ H 2 (Xh,Z) ----+ H 2 (X,(()x) ----+ •••
The only nonalgebraic part is the integral cohomology of X h. Since any
algebraic variety is triangulable (see for example, Hironaka [5]), the coho-
mology groups Hi(Xh,Z) are finitely generated abelian groups. From this
sequence we can deduce some information about the Picard group of X.
First of all, one sees easily that algebraically equivalent Cartier divisors
give the same element in H 2 (Xh,Z). Therefore the Neron-Severi group of
X is a subgroup of H 2 (Xh,Z), and hence is finitely generated (V, Ex. 1.7).
On the other hand, the group Pico X of divisors algebraically equivalent to
zero modulo linear equivalence is isomorphic to H 1 (X,(()x)/H 1(Xh,Z). One
shows that this is a complex torus, and in fact it is an abelian variety, the
Picard variety of X.
If X is a nonsingular curve of genus g, we can see even more clearly what
is happening. In that case Xh is a compact Riemann surface of genus g.
As a topological space, it is a compact oriented real 2-manifold which is
homeomorphic to a sphere with g handles. So we have
H 0 (Xh,Z) = H 2 (Xh,Z) = Z, and
On the other hand, H 1 (X,(()x) ~ C 9 , so
Pica X ~ C9 jZ 29 •
This is the Jacobian variety of X (IV, §4), which is an abelian variety of
dimension g. Of course NS(X) ~ Z in this case, the isomorphism being
given by the degree function.

EXERCISES

6.1. Show that the unit disc inC is not isomorphic to Xh for any scheme X.

6.2. Let z1 ,z 2 , . . . be an infinite sequence of complex numbers with lz.l---> oo as n---> oo.
Let :1 <:; (I) c be the sheaf of ideals of holomorphic functions vanishing at all of the
z•. Show that there is no coherent algebraic sheaf of ideals§ <:; (l)x, where X =
A~ is the affine line, such that 3 = §has an ideal in (l)c· Show on the other hand
that there is a coherent sheaf$' on X such that $'h : : : : ~as coherent sheaves.

6.3. (Serre [12].) On C 2 - [0, 0}, we define an invertible analytic sheaf i.' as follows:
£ : : : : (I) when z =1 0; £ : : : : (I) when w =1 0, and when both z,w =1 0, the two copies
of (I) are glued by multiplication by e-t;zw in the local ring at the point (z,w).
Show that there is no invertible algebraic sheaf!£' on A 2 - {0,0} with !i'h : : : : £.

6.4. Show directly that if X is a scheme which is reduced and proper over C, then
H 0 (X,(I)x) : : : : H 0 (Xh,(l)xJ. Conversely, show that if X is not proper over C, then
there is a coherent sheaf$' on X with H 0 (X,$') =I H 0 (Xh,$'h).

447
Appendix B Transcendental Methods

6.5. If X,X' are nonsingular affine algebraic curves, with Xh ~ X~, show that X ~ X'.

6.6. Show that if X and Yare projective schemes over C, and f:Xh-> Y, is a morphism
of analytic spaces, then there exists a (unique) morphism f:X -> Y with fh = f.
[Hint: First reduce to the case Y = P". Then consider the invertible analytic
sheaf 5! = f*(l)(l) on Xh, use (2.1), and the techniques of(II, §7).]

448
APPENDIX C

The W eil Conjectures

In 1949, Andre Weil [ 4] stated his now famous conjectures concerning the
number of solutions of polynomial equations over finite fields. These con-
jectures suggested a deep connection between the arithmetic of algebraic
varieties defined over finite fields and the topology of algebraic varieties
defined over the complex numbers. Weil also pointed out that if one had a
suitable cohomology theory for abstract varieties, analogous to the ordinary
cohomology of varieties defined over C, then one could deduce his conjec-
tures from various standard properties of the cohomology theory. This
observation has been one of the principal motivations for the introduction
of various cohomology theories into abstract algebraic geometry. In 1963,
Grothendieck was able to show that his /-adic cohomology had sufficient
properties to imply part of the Wei! conjectures (the rationality of the zeta
function). Deligne's [3] proof in 1973 of the remainder of the Wei! conjectures
(specifically the analogue of the "Riemann hypothesis") may be regarded as
the culmination of the study of /-adic cohomology begun by Grothendieck,
M. Artin, and others in the Paris seminars [SGA 4], [SGA 5], and [SGA 7].

1 The Zeta Function and the Weil Conjectures

Let k = Fq be a finite field with q elements. Let X be a scheme of finite type

over k. For example, X could be the set of solutions in affine or projective
space over k of a finite number of polynomial equations with coefficients in
k. Let ]( be an algebraic closure of k, and let X = X x k k be the corre-
sponding scheme over k. For each integer r ~ 1, let N, be the number of
points of X which are rational over the field k, = Fqr of q' elements. In
other words, N, is the number of points of X whose coordinates lie in k,.

449
Appendix C The Wei! Conjectures

The numbers N 1 ,N 2 ,N 3 , . . . are clearly of great importance in studying

arithmetical properties of the scheme X. To study them, we form the zeta
function of X (following Weil), which is defined as

Z(t) = Z(X; t) = exp ( L00 t')

N,- .
r= 1 r
Note that by definition, it is a power series with rational coefficients:
Z(t) E Q[[t]J.
For example, let X = P 1 . Over any field, P 1 has one more point than
the number of elements of the field. Hence N, = q' + 1. Thus

Z(P\t) = exp ,~ 1 (q'

(
GO
+ t')
1)-;: .

It is easy to sum this series, and we find that

1 1
Z(P ,t) = (1 - t)(1 - qt)

In particular, it is a rational function oft.

Now we can state the Weil conjectures. Let X be a smooth projective

variety of dimension n defined over k = Fq. Let Z(t) be the zeta function
of X. Then

1.1. Rationality. Z(t) is a rational function oft, i.e., a quotient of poly-

nomials with rational coefficients.

1.2. Functional equation. Let E be the self-intersection number of the

diagonal L1 of X x X (which is also the top Chern class of the tangent
bundle of X (App. A, Ex. 6.6) ). Then Z(t) satisfies a functional equation,
namely

1.3. Analogue of the Riemann hypothesis. It is possible to write

where P 0 (t) = 1 - t; P 2 n(t) = 1 - qnt; and for each 1 ~ i ~ 2n - 1, P;(t)

is a polynomial with integer coefficients, which can be written

where the r:xii are algebraic integers with lrxiil = qi12 . (Note that these con-
ditions uniquely determine the polynomials P;(t), if they exist.)

450
 2 History of Work on the Wei! Conjectures

1.4. Betti numbers. Assuming (1.3), we can define the ith Betti number
B; = B;(X) to be the degree of the polynomial P;(t). Then we have E =
L( -1)iB;. Furthermore, suppose that X is obtained from a variety Y
defined over an algebraic number ring R, by reduction modulo a prime ideal
p of R. Then B;(X) is equal to the ith Betti number of the topological space
Y, = (Y x R C)h (App. B), i.e., B;(X) is the rank of the ordinary cohomology
group Hi(Y,,Z).

Let us verify the conjectures for the case X = P 1 . We have already seen
that Z(t) is rational. The invariant E ofP 1 is 2, and one verifies immediately
the functional equation which says in this case

z Gt) = qt 2 Z(t).

The analogue of the Riemann hypothesis is immediate, with P 1 (t) = 1.

Hence B 0 = B 2 = 1 and B 1 = 0. These are indeed the usual Betti numbers
ofP~, which is a sphere, and finally we have E = 2:( -l)iB;.

2 History of Work on the Weil Conjectures

Weil was led to his conjectures by consideration of the zeta functions of

some special varieties. See his article Weil [ 4] for number-theoretic back-
ground, and calculations for the "Fermat hypersurfaces" l:a;x7 = 0. One
of Weil's major pieces of work was the proof that his conjectures hold for
curves. This is done in his book Weil [2]. The rationality and the func-
tional equation follow from the Riemann-Roch theorem on the curve. The
analogue of the Riemann hypothesis is deeper (V, Ex. 1.10). He deduces it
from an inequality of Castelnuovo and Severi about correspondences on a
curve (V, Ex. 1.9). This proof was later simplified by Mattuck and Tate [1 J
and Grothendieck [2]. Weil [3] also gave another proof using the /-adic
representation of Frobenius on abelian varieties, which inspired the later
cohomological approaches. Recently a completely independent elementary
proof of the Riemann hypothesis for curves has been discovered by Stepanov,
Schmidt and Bombieri (see Bombieri [1 ]).
For higher-dimensional varieties, the rationality of the zeta function and
the functional equation were first proved by Dwork [1 ], using methods of
p-adic analysis. See also Serre [8] for an account of this proof.
Most other work on the Weil conjectures has centered around the search
for a good cohomology theory for varieties defined over fields of charac-
teristic p, which would give the "right" Betti numbers as defined in (1.4)
above. Furthermore, the cohomology theory should have its coefficients in
a field of characteristic zero, so that one can count the fixed points of a
morphism as a sum of traces on cohomology groups, a la Lefschetz.

451
Appendix C The Wei! Conjectures

The first cohomology introduced into abstract algebraic geometry was

that of Serre [3] using coherent sheaves (Ch. III). Although it could not
satisfy the present need, because of its coefficients being in the field over
which the variety is defined, it served as a basis for the development of later
cohomology theories. Serre [ 6] proposed a cohomology with coefficients
in the Witt vectors, but was unable to prove much about it. Grothendieck,
inspired by some of Serre's ideas, saw that one could obtain a good theory
by considering the variety together with all its unramified covers. This
was the beginning of his theory of etale topology, developed jointly with
M. Artin, which he used to define the 1-adic cohomology, and thus to obtain
another proof of the rationality and functional equation of the zeta function.
See Grothendieck [ 4] for a brief announcement; Artin [1] and Grothendieck
[SGA 4] for the foundations of etale cohomology; Grothendieck [ 6] for the
proof of rationality of the zeta function, modulo general facts about 1-adic
cohomology which are supposed to appear in [SGA 5] (as yet unpublished).
Lubkin [1] more or less independently developed a p-adic cohomology
theory which led also to a proof of rationality and the functional equation,
for varieties which could be lifted to characteristic zero. The crystalline
cohomology of Grothendieck [8] and Berthelot [1] gives another similar
cohomological interpretation of the Weil conjectures.

The analogue of the Riemann hypothesis has proved more difficult to

handle. Lang and Weil [ 1] established an inequality for n-dimensional
varieties, which is equivalent to the analogue of the Riemann hypothesis if
n = 1, but falls far short of it if n :,;;, 2. Serre [9] established another analogue
of the Riemann hypothesis for the eigenvalues of certain operators on the
cohomology of a Kahler manifold, using the powerful results of Hodge
theory. This suggests that one should try to establish in abstract algebraic
geometry some results known for varieties over C via Hodge theory, in
particular the "strong Lefschetz theorem" and the "generalized Hodge index
theorem." Grothendieck [9] optimistically calls these the "standard con-
jectures," and notes that they immediately imply the analogue of the
Riemann hypothesis. See also Kleiman [2] for a more detailed account of
these conjectures and their interrelations.
Until Deligne's proof [3] of the general analogue of the Riemann hy-
pothesis, only a few special cases were known: curves (above), rational
threefolds by Manin [2]-see also Demazure [1 ], K3 surfaces by Deligne
[2], and certain complete intersections by Deligne [5].
In addition to the references given above, I would like to mention Serre's
survey article [10] and Tate's companion article [1] suggesting further (as
yet untouched) conjectures about cycles on varieties over fields of charac-
teristic p. Also, for the number-theorists, Deligne's article [1] which shows
that the analogue of the Riemann hypothesis implies the conjecture of
Ramanujan about the r-function.

452
 3 The /-adic Cohomology

3 The 1-adic Cohomology

In this and the following section we will describe the cohomological inter-
pretation of the Weil conjectures in terms of the 1-adic cohomology of
Grothendieck. Similar results would hold in any cohomology theory with
similar forinal properties. See Kleiman [2] for an axiomatic treatment of
a "Weil cohomology theory."

-
Let X be a scheme of finite type over an algebraically closed field k of
characteristic p :;:,: 0. Let l be a prime number l -=F p. Let Z 1 = lim Z/l'Z
be the ring of 1-adic integers, and Q 1 its quotient field. We consider the
etale topology of X (see Artin [1] or [SGA 4]), and then using etale coho-
mology, we define the 1-adic cohomology of X by
H;(X,Q 1) = (fu!! mt (X,Z/l'Z)) ®z, Qz.
We will not go into a detailed explanation of this definition here (see [SGA
41]). Rather, we will content ourselves with listing some of the main
properties of the 1-adic cohomology.

3.1. The groups H;(X,Q 1) are vector spaces over Q1• They are zero
except in the range 0 ~ i ~ 2n, where n = dim X. They are known to be
finite-dimensional if X is proper over k. (They are expected to be finite-
dimensional in general, but there is no proof yet because of the problem of
resolution of singularities in characteristic p > 0.)
3.2. H;(X,Q 1) is a contravariant functor in X.
3.3. There is a cup-product structure
H;(X,Q 1) x Hi(X,Q 1) ~ Hi+ i(X,Q 1)
defined for all i,j.
3.4. Poincare duality. If X is smooth and proper over k, of dimension n,
then H 2 "(X,Q 1) is !-dimensional, and the cup-product pairing
Hi(X,Qz) X H2n-i(X,Qz) ~ H2"(X,Qz)
is a perfect pairing for each i, 0 ~ i ~ 2n.
3.5. Lefschetz fixed-point formula. Let X be smooth and proper over k.
Let f; X ~ X be a morphism with isolated fixed points, and for each fixed
point x EX, assume that the action of 1 - df on Q} is injective. This last
condition says that the fixed point has "multiplicity 1." Let L(f,X) be the
number of fixed points off Then
L(f,X) = 2) -1); Tr(f*; H;(X,Q1))
where f* is the induced map on the cohomology of X.

453
Appendix C The Wei! Conjectures

3.6. Iff: X -.. Y is a smooth proper morphism, with Y connected, then

dim Hi( X Y,Q 1} is constant for y E Y. In particular, dim Hi(X,Q 1) is constant
under base field extension.
3.7. Comparison theorem. If X is smooth and proper over C, then
Hi(X,Q 1) ®Q, C ~ Hi(Xh,C)
where X his the associated complex manifold in its classical topology (App. B).
3.8. Cohomology class of a cycle. If X is smooth and proper over k, and
if Z is a subvariety of codimension q, then there is associated to Z a coho-
mology class tJ(Z) E H 2 q(X,Q 1). This map extends by linearity to cycles.
Rationally equivalent cycles have the same cohomology class. Intersection
of cycles becomes cup-product of cohomology classes. In other words, fJ
is a homomorphism from the Chow ring A(X) to the cohomology ring
H*(X,Q 1}. Finally, it is non-trivial: if P EX is a closed point, then rJ(P) E
H 2 n(X,Q 1) is nonzero.

This list of properties has no pretensions to completeness. In particular,

we have not mentioned sheaves of twisted coefficients, higher direct images,
Leray spectral sequence, and so forth. For further properties, as well as
for the proofs of the properties given above, we refer to [SGA 4] for the
corresponding statement with torsion coefficients, and to [SGA 5] for the
passage to the limit of Z 1 or Q1 coefficients.

4 Cohomological Interpretation of the Weil Conjectures

Using the 1-adic cohomology described above, we can give a cohomological

interpretation of the Weil conjectures. The main idea, which goes back to
Weil, is very simple. Let X be a projective variety defined over the finite
field k = Fq, and let X = X x k k be the corresponding variety over the
algebraic closure k of k. We define the Frobenius morphism f:X -..X by
sending the point P with coordinates (a;), ai E k, to the point f(P) with
coordinates (a'!). This is the k-linear Frobenius morphism, where Xq is
identified with X (IV, 2.4.1). Since X is defined by equations with coeffi-
cients in k, f(P) is also a point of X. Furthermore, P is a fixed point of
f if and only if its coordinates lie in k. More generally, P is a fixed point of
the iterate f' if and only if it has coordinates in the field k, = F qr· Thus
in the notation of §1, we have
N, = # {fixed points off'} = L(f',X).
If X is smooth, we can calculate this number by the Lefschetz fixed-point
formula (3.5). We find
2n
N, = L (-1)i Tr(f'*; Hi(X,Q 1) ).
i=O

454
 4 Co homological Interpretation of the Wei! Conjectures

Substituting in the definition of the zeta function, we have

Z(X,t) =I\2n [ exp ( oo 1 Tr(f'*; H;(X,QJ)-;:t')](-1)'

,~

To simplify this expression, we need an elementary lemma.

Lemma 4.1. Let cp be an endomorphism of a finite-dimensional vector space

V over a field K. Then we have an identity of formal power series in t,
with coefficients in K,

exp ( L
00

r= 1
t')
Tr(cp'; V)- = det(1 - cpt; V)- 1 .
r

PROOF. If dim V = 1, then cp is multiplication by a scalar A E K, and it says

exp ( .LA'-t')
00

r= 1r
= -- 1 .
1 - At
This is an elementary calculation, which we already did in computing the
zeta function of P 1 . For the general case, we use induction on dim V.
Furthermore, we may clearly assume that K is algebraically closed. Hence
cp has an eigenvector, so we have an invariant subspace V' s:::: V. We use
the exact sequence
0 ~ V' ~ V ~ V jV' ~ 0
and the fact that both sides of the above equation are multiplicative for
short exact sequences of vector spaces. By induction, this gives the result.

Using the lemma, we immediately obtain the following result.

Theorem 4.2. Let X be projective and smooth over k = F q• of dimension n.

Then
P1(t) · · · P2n-1(t)
Z(X,t) = ,
P 0 (t) · · · Pzn(t)
where
P;(t) = det(1 - f*t; H;(X,Q 1))
and f* is the map on cohomology induced by the Frobenius morphism f:
x~x.

This theorem shows immediately that Z(t) is a quotient of polynomials

with Q1 coefficients. One can show by an elementary argument on power
series (Bourbaki [2, Ch. IV §5, Ex. 3, p. 66]) that Q [[ t ]] n Q 1( t) = Q( t). Since
we know that Z(t) is a power series with rational coefficients, we deduce that
Z(t) is a rational function, which proves (1.1). Notice, however, that we do
not know yet whether the P;(t) have rational coefficients, and we do not know
whether they are the polynomials referred to in (1.3) above.

455
Appendix C The Wei! Conjectures

We can extract a bit more information from this theorem. Since f* acts
on H 0 (X,Q 1) as the identity, P 0 (t) = 1 - t. Furthermore, we can determine
P zn(t). The Frobenius morphism is a finite morphism of degree qn. Hence
it must act as multiplication by qn on a generator of H 2 n(X,Q 1). So P 2 it) =
1 - qnt. If we provisionally define the ith Betti number B; as dim H;(X,Q 1),
then B; = degree P;(t), and one can show easily that the invariant E of X is
given by

So we call E the "topological Euler-Poincare characteristic" of X. We do not

yet know that this definition of the Betti numbers agrees with the one in (1.4)
above. However, once we do know this, the statement (1.4) will follow from
the general properties (3.6) and (3. 7) of the l-adic cohomology.

Next, we will show that the functional equation follows from Poincare
duality. Again we need a lemma from linear algebra.

Lemma 4.3. Let V x W ~ K be a perfect pairing of vector spaces V,W of

dimension r over K. Let A E K, and let qJ: V ~ V and t/J: W ~ W be
endomorphisms such that
(((Jv,t/Jw) = A(v,w)
for all v E V, wE W. Then
( -1)'A't' ( ((J )
det(1 - t/Jt,W) = det(((J; V) det 1 - At; V
and
X
det(t/J; W) = (det(((J; V)

Theorem 4.4. With the hypotheses of (4.2), the zeta function Z(X,t) satisfies
the functional equation (1.2).
PROOF. One applies the lemma (whose proof is elementary) to the pairings
H;(X,Q 1) x H 2n-i(X,Q1) ~ H 2n(X,Q 1) given us by Poincare duality (3.4).
Using the fact that f* is compatible with cup-product, and that it acts by
multiplication by qn on H 2n(X,Q 1), we get an expression for Pzn-i in terms of
P;, namely
qnB;tB; ( 1\
Pzn-Jt) = (-1)B' det(f*;Hi)P; qntj"
Furthermore, we have
nBi
d et(f *·' H2n-i) -_ -de-t-=(f-*-;
q
-H--,---;)"

Substituting these in the formula of(4.2) and using E = 2_) -l);B;, we obtain
the functional equation.

456
 4 Cohomological Interpretation of the Wei! Conjectures

So we see that the conjectures (1.1), (1.2), and (1.4) follow from the formal
properties of 1-adic cohomology once we have interpreted the zeta function
as in (4.2). The analogue of the Riemann hypothesis is much deeper.

Theorem 4.5 (Deligne [3]). With the hypotheses of(4.2), the polynomials P;(t)
have integer coefficients, independent of l, and they can be written
P;(t) = 0(1 - rx;/)
where the r:xii are algebraic integers with lrxiil = qi1 2 •
This result completes the solution of the Weil conjectures. Note that it
implies that the polynomials P;(t) of (4.2) are the same as those of (1.3), and
hence the two definitions of the Betti numbers agree.
We cannot describe the proof of Deligne's theorem here, except to say
that it relies on the deeper properties of 1-adic cohomology developed in
[SGA 4], [SGA 5] and [SGA 7]. In particular it makes use of Lefschetz's
technique of fibering a variety by a "Lefschetz pencil," and studying the
monodromy action on the cohomology near a singular fibre.

EXERCISES

5.1. Let X be a disjoint union oflocally closed subschemes X;. Then show that
Z(X,t) = fJZ(X;,t).
5.2. Let X = P~, where k = Fq, and show from the definition of the zeta function
that
1
Z(P",t) = .
(1 - t)(l - qt) 0 0 0
(1 - q"t)
Verify the Wei! conjectures for P".

5.3. Let X be a scheme of finite type over Fq, and let A 1 be the affine line. Show that
Z(X x A 1 ,t) = Z(X,qt).
5.4. The Riemann zeta function is defined as
1
((s) = f1 1 -p s'

for s E C, the product being taken over all prime integers p. If we regard this
function as being associated with the scheme Spec Z, it is natural to define, for
any scheme X of finite type over Spec Z,
(x(s) = [1{1 - N(x)-s)- 1
where the product is taken over all closed points x EX, and N(x) denotes the
number of elements in the residue field k(x). Show that if X is of finite type over
Fq, then this function is connected to Z(X,t) by the formula
(x(s) = Z(X,q-s).
[Hint: Take dlog of both sides, replace q-s by t, and compare.]

457
Appendix C The Wei! Conjectures

5.5. Let X be a curve of genus g over k. Assuming the statements (1.1) to (1.4) of the
Wei! conjectures, show that N 1 ,N 2 , •.. ,N9 determine N, for all r ~ 1.
5.6. Use (IV, Ex. 4.16) to prove the Wei! conjectures for elliptic curves. First note
that for any r,
N, = q' - (f' + ]') + 1,

where f = F'. Then calculate Z(t) formally and conclude that

(1 - ft)(1 - ]t)
Z(t) = (1 - t)(1 - qt)
and hence
1 -at+ qt 2
Z(0 = ,
(1 - t)(1 - qt)

where f + J = ax. This proves rationality immediately. Verify the functional

equation. Finally, if we write
1 - at + qt 2 = (1 - IJ(t)(1 - {Jt),

show that lal ~ 2.jq if and only if 11)(1 = lfJI = Jq. Thus the analogue of the
Riemann hypothesis is just (IV, Ex. 4.16d).
5.7. Use (V, Ex. 1.10) to prove the analogue of the Riemann hypothesis (1.3) for any
curve C of genus g defined over Fq. Write N, = 1 - a, + q'. Then according to
(V, Ex. 1.10),
la,l ~ 2gJ(l.
On the other hand, by (4.2) the zeta function of C can be written

Z(t) = p 1 (t)
(1 - t)(l - qt)
where
2g
P l(tl = IT (1 - IJ(;t)
i= 1

is a polynomial of degree 2g = dim H 1 (C,Q 1).

(a) Using the definition of the zeta function and taking logs, show that
2g

a,= L (I)(;)'
i= 1
for each r.
(b) Next show that
Ia, I ~ 2g J(l for all r
[Hint: One direction is easy. For the other, use the power series expansion

for suitable t E C.]

(c) Finally, use the functional equation (4.4) to show that II)(;! ~ Jq for all i implies
that IIJ(d = Jq for all i.

458
Bibliography

Altman, A. and Kleiman, S.

1. Introduction to Grothendieck Duality Theory, Lecture Notes m Math. 146
Springer-Verlag, Heidelberg (1970), 185 pp.
Artin, M.
1. Grothendieck topologies, Harvard Math. Dept. Lecture Notes (1962).
2. The implicit function theorem in algebraic geometry, in Algebraic Geometry,
Bombay 1968, Oxford Univ. Press, Oxford (1969), 13-34.
3. Some numerical criteria for contractibility of curves on algebraic surfaces, Amer.
J. Math. 84 (1962) 485-496.
4. Algebraization of formal moduli II: Existence of modifications, Annals of Math.
91 (1970) 88-135.
Atiyah, M. F.
1. Vector bundles over an elliptic curve, Proc. Lond. Math. Soc. (3) VII 27 (1957),
414-452.
Atiyah, M. F. and Macdonald, I. G.
1. Introduction to Commutative Algebra. Addison-Wesley, Reading, Mass. (1969),
ix + 128 pp.
Baily, W. L., Jr.
1. On the automorphism group of a generic curve of genus >2, J. Math. Kyoto
Univ. I (1961/2) 101-108; correction p. 325.
Biinicii, C. and Stiina~ilii, 0.
1. Algebraic Method.~ in the Global Theory of Complex Spaces, John Wiley, New
York (1976) 296 pp.
Berthelot, P.
1. Cohomologie Crista/line des Schemas de Caracteristique p > 0, Lecture Notes in
Math. 407, Springer-Verlag, Heidelberg (1974), 604 pp.
Birch, B. J. and Kuyk, W., ed.
1. Modular Functions of One Variable, IV (Antwerp), Lecture Notes in Math. 476,
Springer-Verlag, Heidelberg (1975).

459
Bibliography

Bombieri, E.
1. Counting points on curves over finite fields (d'apn!s S. A. Stepanov), Seminaire
Bourbaki 430 (1972/73).
Bombieri, E. and Husemoller, D.
1. Classification and em beddings of surfaces, in Algebraic Geometry, Arcata I974,
Amer. Math. Soc. Proc. Symp. Pure Math. 29 (1975), 329-420.
Borel, A. and Serre, J.-P.
1. Le theon!me de Riemann-Roch, Bull. Soc. Math. de France 86 (1958), 97-136.
Borelli, M.
1. Divisorial varieties, Pacific J. Math. I3 (1963), 375-388.
Bourbaki, N.
1. Algebre Commutative, Elements de Math. 27, 28, 30, 3I Hermann, Paris (1961-
1965).
2. Algebre, Elements de Math. 4, 6, 7, II, I4, 23, 24, Hermann, Paris (1947-59).
Brieskom, E.
1. Beispiele zur Differentialtopologie von Singularitiiten, Invent. Math. 2 (1966)
1-14.
2. Rationale Singularitiiten Komplexer Fliichen, Invent. Math. 4 (1968) 336-358.
Burnside, W.
1. Theory of Groups of Finite Order, Cambridge Univ. Press, Cambridge (1911);
reprinted by Dover, New York.
Cartan, H. and Chevalley, C.
1. Geometrie Algebrique, Seminaire Cartan-Chevalley, Secretariat Math., Paris
(1955/56).
Cartan, H. and Eilenberg, S.
1. Homological Algebra, Princeton Univ. Press, Princeton (1956), xv + 390 pp.
Cassels, J. W. S.
1. Diophantine equations with special reference to elliptic curves, J. Land. Math.
Soc. 4I (1966), 193-291. Corr. 42 (1967) 183.
Cassels, J. W. S. and Frohlich, A., ed.
1. Algebraic Number Theory, Thompson Book Co, Washington D.C. (1967).
Castelnuovo, G.
1. Sui multipli di una serie lineare di gruppi di punti appartenente ad una curva
algebrica, Rend. Circ. Mat. Palermo 7 (1893). Also in Memorie Scelte, pp.
95-113.
Chevalley, C.
1. Intersections of algebraic and algebroid varieties, Trans. Amer. Math. Soc. 57
(1945), 1-85.
2. Anneaux de Chow et Applications, Seminaire Chevalley, Secretariat Math., Paris
(1958).
Chow, W. L.
1. On compact complex analytic varieties, Amer. J. of Math. 7I (1949), 893-914;
errata 72, p. 624.
2. On Picard varieties, Amer. J. Math. 74 (1952), 895-909.
3. The Jacobian variety of an algebraic curve, Amer. J. of Math. 76 (1954), 453-476.
Chow, W. L. and Kodaira, K.
1. On analytic surfaces with two independent meromorphic functions, Proc. Nat.
Acad. Sci. USA 38 (1952), 319-325.

460
 Bibliography

Clemens, C. H. and Griffiths, P. A.

I. The intermediate Jacobian of the cubic threefold, Annals of Math. 95 (1972),
281-356.
Deligne, P.
I. Formes modulaires et representations /-adiques, Seminaire Bourbaki 355,
Lecture Notes in Math. 179, Springer-Verlag, Heidelberg (1971), 139-172.
2. La conjecture de Wei! pour les surfaces K3, Invent. Math. 15 (1972), 206-226.
3. La conjecture de Wei!, I, Pub!. Math. IHES 43 (1974), 273-307.
4. Theon!mes de Lefschetz et criteres de degenerescence de suites spectrales, Pub!.
Math. IHES 35 (1968) 107-126.
5. Les intersections completes de niveau de Hodge un, Invent. Math. 15 (1972)
237-250.
Deligne, P. and Mumford, D.
I. The irreducibility of the space of curves of given genus, Pub!. Math. IHES 36
(1969), 75-110.
Demazure, M.
I. Motifs des varietes algebriques, Seminaire Bourbaki 365, Lecture Notes in Math.
180, Springer-Verlag, Heidelberg (1971), 19-38.
Deuring, M.
I. Die Typen der Multiplikatorenringe elliptischer Funktionenkorper, Abh. Math.
Sem. Univ. Hamburg 14 (1941), 197-272.
2. Die Klassenkorper der Komplexen Multiplikation, Enz. der Math. Wiss., 2nd ed.,
12 , Heft 10, II, §23 (1958) 60 pp.
Dieudonne, J.
I. Cours de Geometrie Algebrique, I. Aper9u Historique sur le Developpement de Ia
Geometrie Algebrique, Presses Univ. France, Collection Sup. (1974), 234 pp.
Dwork, B.
I. On the rationality of the zeta function of an algebraic variety, Amer. J. Math. 82
(1960), 631-648.
Freyd, P.
I. Abelian Categories, an Introduction to the Theory of Functors, Harper & Row,
New York (1964), 164 pp.
Fulton, W.
I. Algebraic Curves, W. A. Benjamin, New York (1969), xii + 226 pp.
2. Riemann-Roch for singular varieties, in Algebraic Geometry, Arcata 1974, Amer.
Math. Soc. Proc. Symp. Pure Math 29 (1975), 449-457.
Godement, R.
I. Topologie Algebrique et Theorie des Faisceaux, Hermann, Paris (1958).
Grauert, H.
I. Ein Theorem der analytischen Garbentheorie und die Modulraume komplexer
Strukturen, Pub. Math. IHES 5 (1960), 233-292.
2. Uber Modifikationen und exzeptionelle analytische Mengen, Math. Ann. 146
(1962), 331-368.
Grauert, H. and Remmert, R.
I. Komplex Raume, Math. Ann. 136 (1958), 245-318.
Grauert, H. and Riemenschneider, 0.
I. Verschwindungssatze fur analytische Kohomologiegruppen auf komplexen
Raumen, Inv. Math. 11 (1970), 263-292.

461
Bibliography

Grothendieck, A.
I. Sur quelques points d'algebre homologique, Tohoku Math. J. 9 (1957), 119~221.
2. Sur une note de Mattuck-Tate, J. Reine u. Angew. Math. 200 (1958), 208~215.
3. La theorie des classes de Chern, Bull. Soc. Math. de France 86 (1958), 137 ~ 154.
4. The cohomology theory of abstract algebraic varieties, Proc. Int. Cong. Math.,
Edinburgh (1958), 103~ 118.
5. Fondements de !a Geometrie Algebrique, Seminaire Bourbaki 1957 ~62, Secretariat
Math., Paris (1962).
6. Formule de Lefschetz et rationalite des fonctions L, Seminaire Bourbaki 279
(1965).
7. Local Cohomology (notes by R. Hartshorne), Lecture Notes in Math. 41, Springer-
Verlag, Heidelberg (1967), 106 pp.
8. Crystals and the De Rham cohomology of schemes (notes by I. Coates and 0.
Jussila), in Dix Exposes sur !a Cohomologie des Schemas, North-Holland, Amster-
dam (1968), 306~358.
9. Standard conjectures on algebraic cycles, in Algebraic Geometry, Bombay 1968,
Oxford University Press, Oxford (1969), 193~199.
Grothendieck, A. and Dieudonne, J.
Elements de Geometrie Algebrique.
EGA I. Le langage des schemas, Pub!. Math. IHES 4 (1960).
EGA II. Etude globale elementaire de quelques classes de morphismes, Ibid. 8
(1961).
EGA III. Etude cohomologique des faisceaux coherents, Ibid. 11 (1961), and 17
(1963).
EGA IV. Etude locale des schemas et des morphismes de schemas, Ibid. 20 (1964),
24 (1965), 28 (1966), 32 (1967).
EGA I. Elements de Geometrie Algebrique, I, Grundlehren 166, Springer-Verlag,
Heidelberg (new ed., 1971), ix + 466 pp.
Grothendieck, A. et a!.
Seminaire de Geometrie Algebrique.
SGA I. Revetements etales et Groupe Fondemental, Lecture Notes in Math. 224,
Springer-Verlag, Heidelberg (1971).
SGA 2. Cohomologie Locale des Faisceaux Coherents et Theoremes de Lefschetz
Locaux et Globaux, North-Holland, Amsterdam (1968).
SGA 3. (with Demazure, M.) Schemas en Groupes I, II, Ill, Lecture Notes in Math.
151, 152, 153, Springer-Verlag, Heidelberg (1970).
SGA 4. (with Artin, M. and Verdier, J. L.) Theorie des Tapas et Cohomologie Etale
des Schemas, Lecture Notes in Math. 269, 270, 305, Springer-Verlag,
Heidelberg (1972~ 1973).
SGA 4!. (by Deligne, P., with Boutot, J. F., Illusie, L., and Verdier, J. L.) Coho-
mologie Etale, Lecture Notes in Math. 569, Springer-Verlag, Heidelberg
(1977).
SGA 5. Cohomologie l-adique et fonctions L (unpublished).
SGA 6. (with Berthelot, P. and Illusie, L.) Theorie des Intersections et Theoreme de
Riemann~Roch, Lecture Notes in Math. 225, Springer-Verlag, Heidelberg
(1971).
SGA 7. (with Raynaud, M. and Rim, D. S.) Groupes de Monodromie en Geometrie
Algebrique, Lecture Notes in Math. 288, Springer-Verlag, Heidelberg
(1972). Part II (by Deligne, P. and Katz, N.) 340 (1973).
Gunning, R. C.
I. Lectures on Riemann Surfaces, Princeton Math. Notes, Princeton U. Press,
Princeton (1966), 256 pp.

462
 Bibliography

Gunning, R. C. and Rossi, H.

1. Analytic Functions of Several Complex Variables, Prentice-Hall (1965), xii +
317 pp.
Halphen, G.
1. Memoire sur Ia classification des courbes gauches algebriques, J. Ec. Polyt. 52
(1882), 1-200.
Hartshorne, R.
1. Complete intersections and connectedness, Amer. J. of Math. 84 (1962), 497-508.
2. Residues and Duality, Lecture Notes in Math. 20, Springer-Verlag, Heidelberg
(1966).
3. Cohomological dimension of algebraic varieties, Annals of Math. 88 (1968),
403-450.
4. Curves with high self-intersection on algebraic surfaces, Pub!. Math. IHES 36
(1969), 111-125.
5. Ample Subvarieties of Algebraic Varieties, Lecture Notes in Math. 156, Springer-
Verlag, Heidelberg (1970), xiii + 256 pp.
6. Equivalence relations on algebraic cycles and subvarieties of small codimension,
in Algebraic Geometry, Arcata 1974, Amer. Math. Soc. Proc. Symp. Pure Math. 29
(1975), 129-164.
7. On the DeRham cohomology of algebraic varieties, Pub!. Math. IHES 45 (1976),
5-99.
Hartshorne, R., ed.
1. Algebraic Geometry, Arcata 1974, Amer. Math. Soc. Proc. Symp. Pure Math.
29 (1975).
Hilbert, D. and Cohn-Vossen, S.
1. Geometry and the Imagination, Chelsea Pub. Co., New York (1952), ix + 357 pp.
(translated from German Anschauliche Geometrie (1932) ).
Hilton, P. J. and Stammbach, U.
1. A Course in Homological Algebra, Graduate Texts in Mathematics 4, Springer-
Verlag, Heidelberg (1970), ix + 338 pp.
Hironaka, H.
1. A note on algebraic geometry over ground rings. The invariance of Hilbert
characteristic functions under the specialization process, Ill. J. Math. 2 (1958),
355-366.
2. On the theory of birational blowing-up, Thesis, Harvard (1960) (unpublished).
3. On resolution of singularities (characteristic zero), Pro c. Int. Cong. Math. (1962),
507-521.
4. Resolution of singularities of an algebraic variety over a field of characteristic
zero, Annals of Math. 79 (1964). I: 109-203; II: 205-326.
5. Triangulations of algebraic sets, in Algebraic Geometry, Arcata 1974, Amer.
Math. Soc. Proc. Symp. Pure Math. 29 (1975), 165-184.
Hironaka, H. and Matsumura, H.
1. Formal functions and formal embeddings, J. of Math. Soc. of Japan 20 (1968),
52-82.
Hirzebruch, F.
1. Topological Methods in Algebraic Geometry, Grundlehren 131, Springer-Verlag,
Heidelberg (3rd ed., 1966), ix + 232 pp.
Hirzebruch, F. and Mayer, K. H.
1. O(n)-Mannigfaltigkeiten, Exotische Sphiiren und Singularitiiten, Lecture Notes in
Math. 57, Springer-Verlag, Heidelberg (1968).

463
Bibliography

Hodge, W. V. D.
1. The Theory and Applications of Harmonic Integrals, Cambridge Univ. Press,
Cambridge (2nd ed., 1952), 282 pp.
Horrocks, G. and Mumford, D.
1. A rank 2 vector bundle on P 4 with 15,000 symmetries, Topology 12 (1973), 63-81.
Hurwitz, A.
1. Ober algebraische Gebilde mit eindeutigen Transformationen in sich, Math. Ann.
41 (1893), 403-442.
Hurwitz, A. and Courant, R.
1. Allgemeine Funktionentheorie und elliptische Funktionen; Geometrische Funktion-
entheorie, Grundlehren 3, Springer-Verlag, Heidelberg (1922), xi + 399 pp.
lgusa, J.-1.
1. Arithmetic genera of normal varieties in an algebraic family, Proc. Nat. A cad.
Sci. USA 41 (1955) 34-37.
2. Class number of a definite quaternion with prime discriminant, Pro c. Nat. A cad.
Sci. USA 44 (1958) 312-314.
Iskovskih, V. A. and Manin, Ju. I.
1. Three-dimensional quartics and counterexamples to the Liiroth problem, Math.
USSR-Sbornik 15 (1971) 141-166.
Jouanolou, J.P.
1. Riemann-Roch sans denominateurs, Invent. Math. 11 (1970), 15-26.
Kleiman, S. L.
1. Toward a numerical theory of ampleness, Annals of Math. 84 (1966), 293-344.
2. Algebraic cycles and the Wei! conjectures, in Dix Exposes sur Ia Cohomologie des
Schemas, North-Holland, Amsterdam (1968), 359-386.
3. The transversality of a general translate, Compos. Math. 28 (1974), 287-297.
Kleiman, S. L. and Laksov, D.
1. Another proof of the existence of special divisors, Acta Math. 132 (1974), 163-
176.
Klein, F.
1. Ueber die Transformationen siebenter Ordnung der elliptischen Funktionen,
Math. Ann. 14 (1879). Also in Klein, Ges. Math. Abh. 3 (1923), 90-136.
2. Lectures on the Icosahedron and the Solution of Equations of the Fifth Degree,
Kegan Paul, Trench, Triibner, London (1913). Dover reprint (1956).
Knutson, D.
1. Algebraic spaces, Lecture Notes in Math. 203, Springer-Verlag, Heidelberg
(1971) vi + 261 pp.
Kodaira, K.
1. On a differential-geometric method in the theory of analytic stacks, Proc. Nat.
Acad. Sci. USA 39 (1953), 1268-1273.
2. On Kahler varieties of restricted type. (An intrinsic characterization of algebraic
varieties.), Annals of Math. 60 (1954), 28-48.
Kodaira, K. and Spencer, D. C.
1. On arithmetic genera of algebraic varieties, Proc. Nat. Acad. Sci. USA 39 (1953),
641-649.
Kunz, E.
1. Holomorphe Differentialformen auf algebraischen Varietaten mit Singularitaten,
I, Manus. Math. 15 (1975) 91-108.

464
 Bibliography

Lang, S.
1. Abelian Varieties, Interscience Pub., New York (1959), xii + 256 pp.
2. Algebra, Addison-Wesley (1971), xvii + 526 pp.
Lang, S. and Neron, A.
1. Rational points of abelian varieties over function fields, Amer. J. Math. 81 (1959),
95~118.

Lang, S. and Trotter, H.

I. Frobenius Distributions in GL 2 -Extensions, Lecture Notes in Math. 504, Springer-
Verlag, Heidelberg (1976).
Lang, S. and Wei!, A.
1. Number of points of varieties in finite fields, Amer. J. Math. 76 (1954), 819~827.

Lascu, A. T., Mumford, D., and Scott, D. B.

1. The self-intersection formula and the "formule-clef," Math. Proc. Camb. Phil.
Soc. 78 (1975), 117~123.
Lichtenbaum, S. and Schlessinger, M.
1. The cotangent complex of a morphism, Trans. Amer. Math. Soc. 128 (1967),
41~70.

Lipman, J.
I. Introduction to resolution of singularities, in Algebraic Geometry, Arcata 1974,
Amer. Math. Soc. Proc. Symp. Pure Math. 29 (1975), 187~230.
2. Rational singularities with applications to algebraic surfaces and unique fac-
torization, Pub!. Math. IHES 36 (1969) 195~279
3. Unique factorization in complete local rings, in Algebraic Geometry, Arcata 1974,
Amer. Math. Soc. Proc. Symp. Pure Math. 29 (1975) 531~546.
Lubkin, S.
I. Ap-adic proof ofWeil's conjectures, Annals of Math. 87 (1968), 105~ 194, and 87
(1968), 195~255.
Macbeath, A. M.
1. On a theorem of Hurwitz, Proc. Glasgow Math. Assoc. 5 (1961) 90~96.

Manin, Yu. I.
1. Lectures on the K-functor in Algebraic Geometry, Russian Mathematical Surveys
24 (5) (1969), 1~89.
2. Correspondences, motifs, and monoidal transformations, Math USSR-Sbornik
6 (1968) 439-470.
3. Cubic forms: Algebra, Geometry, Arithmetic, North-Holland, Amsterdam (1974),
vii + 292 pp.
Maruyama, M.
1. On Classification of Ruled Surfaces, Kyoto Univ., Lectures in Math. 3, Kino-
kuniya, Tokyo (1970).
Matsumura, H.
1. Geometric structure of the cohomology rings in abstract algebraic geometry,
Mem. Col!. Sci. Univ. Kyoto (A) 32 (1959), 33~84.
2. Commutative Algebra, W. A. Benjamin Co., New York (1970), xii + 262 pp.
Mattuck, A. and Tate, J.
1. On the inequality of Castelnuovo-Severi, Abh. Math. Sem. Univ. Hamburg 22
(1958), 295-299.
Moishezon, B. G.
I. A criterion for projectivity of complete algebraic abstract varieties, Amer. Math.
Soc. Translations 63 (1967), 1~50.

465
Bibliography

2. On n-dimensional compact varieties with n algebraically independent meromor-

phic functions, Amer. Math. Soc. Translations 63 (1967), 51-177.
Morrow, J. and Kodaira, K.
I. Complex Manifolds, Holt, Rinehart & Winston, New York (1971), vii + 192 pp.
Mumford, D.
I. Geometric Invariant Theory, Ergebnisse, Springer-Verlag, Heidelberg (1965),
vi+ 146 pp.
2. Lectures on Curves on an Algebraic Surface, Annals of Math. Studies 59, Princeton
U. Press, Princeton (1966).
3. Pathologies, III, Amer. J. Math. 89 (1967), 94-104.
4. Varieties defined by quadratic equations (with an Appendix by G. Kempf), in
Questions on Algebraic Varieties, Centro Internationale Matematica Estivo,
Cremonese, Rome (1970), 29-100.
5. Abelian Varieties, Oxford Univ. Press, Oxford (1970), ix + 242 pp.
6. The topology of normal singularities of an algebraic surface and a criterion for
simplicity, Pub!. Math. IHES 9 (1961) 5-22.
Nagata, M.
1. On the embedding problem of abstract varieties in projective varieties, Mem. Colt.
Sci. Kyoto (A) 30 (1956), 71-82.
2. A general theory of algebraic geometry over Dedekind domains, Amer. J. ofMath.
78 (1956), 78-116.
3. On the imbeddings of abstract surfaces in projective varieties, Mem. Colt. Sci.
Kyoto (A) 30 (1957), 231-235.
4. Existence theorems for non-projective complete algebraic varieties, Ill. J. Math. 2
(1958), 490-498.
5. On rational surfaces I, II, Mem. Colt. Sci. Kyoto (A) 32 (1960), 351-370, and 33
(1960), 271-293.
6. Imbedding of an abstract variety in a complete variety, J. Math. Kyoto Univ. 2
(1962), 1-10.
7. Local Rings, Interscience Tracts in Pure & Applied Math. 13, J. Wiley, New
York (1962).
8. On self-intersection number of a section on a ruled surface, Nagoya Math. J. 37
(1970), 191-196.
Nakai, Y.
1. A criterion of an ample sheaf on a projective scheme, Amer. J. Math. 85 (1963),
14-26.
2. Some fundamental lemmas on projective schemes, Trans. Amer. Math. Soc. 109
(1963), 296-302.
Nakano, S.
1. On complex analytic vector bundles, J. Math. Soc. Japan, 7 (1955) 1-12.
Narasimhan, M. S. and Seshadri, C. S.
I. Stable and unitary vector bundles on a compact Riemann surface, Annals ofMath.
82 (1965), 540-567.

Noether, M.
1. Zur Grundlegung der Theorie der Algebraischen Raumcurven, Verlag der Konigli-
chen Akademie der Wissenschaften, Berlin (1883).
Olson, L.
1. An elementary proof that elliptic curves are abelian varieties, Ens. Math. 19
(1973), 173-181.

466
 Bibliography

Ramanujam, C. P.
1. Remarks on the Kodaira vanishing theorem, J. Indian Math. Soc. (N.S.) 36
(1972), 41-51.
Ramis, J.P. and Ruget, G.
1. Complexes dualisants et theon':me de dualite en geometrie analytique complexe,
Pub. Math. IHES 38 (1970), 77-91.
Ramis, J. P., Ruget, G., and Verdier, J. L.
l. Dualite relative en geometrie analytique complexe, Invent. Math. 13 (1971),
261-283.
Roberts, J.
l. Chow's moving lemma, in Algebraic Geometry, Oslo 1970 (F. Oort, ed.), Wolters-
Noordhoff (1972), 89-96.
Raquette, P.
1. Abschiitzung der Automorphismenanzahl von Funktionenkorpern bei Prim-
zahlcharakteristik, Math. Zeit. 117 (1970) 157-163.
Rotman, J. J.
l. Notes on Homological Algebra, Van Nostrand Reinhold Math. Studies 26, New
York (1970).
Saint-Donat, B.
l. On Petri's analysis of the linear system of quadrics through a canonical curve,
Math. Ann. 206 (1973), 157-175.
Sally, J.
l. Regular overrings of regular local rings, Trans. Amer. Math. Soc. 171 (1972)
291-300.
Samuel, P.
l. Methodes d'Algebre Abstraite en Geomhrie Algebrique, Ergebnisse 4, Springer-
Verlag, Heidelberg (1955).
2. Lectures on old and new results on algebraic curves (notes by S. Anantharaman),
Tata Inst. Fund. Res. (1966), 127 pp.
3. Anneaux Fact oriels (redaction de A. Micali), Soc. Mat. de Silo Paulo (1963) 97 pp.
Schlessinger, M.
l. Functors of Artin rings, Trans. Amer. Math. Soc. 130 (1968), 208-222.
Serre, J.-P.
l. Cohomologie et geometrie algebrique, Proc. ICM (1954), vol. III, 515-520.
2. Un theon!me de dualite, Comm. Math. Helv. 29 (1955), 9-26.
3. Faisceaux algebriques coherents, Ann. of Math. 61 (1955), 197-278.
4. Geometrie algebrique et geometrie analytique, Ann. Inst. Fourier 6 (1956), 1-42.
5. Sur Ia cohomologie des varietes algebriques, J. de Maths. Pures et Appl. 36
(1957), 1-16.
6. Sur Ia topologie des varietes algebriques en caracteristique p, Symposium Int.
de Topologia Algebraica, Mexico (1958), 24-53.
7. Groupes Algebriques et Corps de Classes, Hermann, Paris (1959).
8. Rationalite des fonctions (des varietes algebriques (d'apn!s B. Dwork) Seminaire
Bourbaki 198 (1960).
9. Analogues kahleriens de certaines conjectures de Wei!, Annals of Math. 71
(1960), 392-394.
10. Zeta and L functions, in Arithmetical Algebraic Geometry (Schilling, ed.),
Harper & Row, New York (1965), 82-92.
11. Algebre Locale-Multiplicites (redige par P. Gabriel), Lectures Notes in Math.
11, Springer-Verlag, Heidelberg (1965).

467
Bibliography

12. Prolongement de faisceaux analytiques coherents, Ann. Inst. Fourier 16 (1966),

363-374.
13. Critere de rationalite pour les surfaces algebriques (d'apres K. Kodaira),
Seminaire Bourbaki 146 (1957).
14. A Course in Arithmetic, Graduate Texts in Math. 7, Springer-Verlag, Heidelberg
(1973) 115 pp.
Severi, F.
I. Intorno ai punti doppi impropri di una superficie generale dello spazio a quattro
dimensioni, e a suoi punti tripli apparenti, Rend. Circ. Matern. Palermo 15 (1901),
33-51.
2. Vorlesungen iiber Algebraische Geometrie (trans!. by E. Loffler), Johnson Pub.
(rpt., 1968; I st. ed., Leipzig 1921).
3. Uber die Grundlagen der algebraischen Geometrie, Hamb. Abh. 9 (1933) 335-364.
Shafarevich, I. R.
I. Algebraic surfaces, Proc. Steklov Inst. Math. 75 (1965) (trans. by A.M.S. 1967).
2. Basic Algebraic Geometry, Grundlehren 213, Springer-Verlag, Heidelberg (1974),
XV+ 439 pp.

Shannon, D. L.
I. Monoidal transforms of regular local rings, Amer. J. Math. 95 (1973) 294-320.
Shioda, T.
I. An example of unirational surfaces in characteristic p, Math. Ann. 211 (1974)
233-236.
Siegel, C. L.
J. Meromorphe Funktionen auf kompakten analytischen Mannigfaltigkeiten,
Nach. Akad. Wiss. Gottingen (1955), 71-77.
Singh, B.
1. On the group of automorphisms of a function field of genus at least two, J. Pure
Appl. Math. 4 (1974) 205-229.
Spanier, E. H.
I. Algebraic Topology, McGraw-Hill, New York (1966).
Stichtenoth, H.
1. Uber die Automorphismengruppe eines algebraischen Funktionenkorpers von
Primzahlcharakteristik, Archiv der Math. 24 (1973) 527-544.
Suominen, K.
I. Duality for coherent sheaves on analytic manifolds, Ann. Acad. Sci. Fenn (A)
424 (1968), 1-19.
Tate, J. T.
I. Algebraic cycles and poles of the zeta function, in Arithmetical Algebraic
Geometry (Schilling, ed.), Harper & Row, New York (1965), 93-110.
2. Residues of differentials on curves, Ann. Sci. de l'E.N.S. (4) 1 (1968), 149-159.
3. The arithmetic of elliptic curves, Inv. Math. 23 (1974), 179-206.
Tjurin, A. N.
1. On the classification of two-dimensional fibre bundles over an algebraic curve
of arbitrary genus (in Russian) Izv. Akad. Nauk SSSR Ser. Mat. 28 (1964) 21-52;
MR 29 (1965) # 4762.
2. Classification of vector bundles over an algebraic curve of arbitrary genus,
Amer. Math. Soc. Translations 63 (1967) 245-279.
Verdier, J.-L.
1. Base change for twisted inverse image of coherent sheaves, in Algebraic Geometry,
Bombay 1968, Oxford Univ. Press, Oxford (1969), 393-408.

468
 Bibliography

Vitushkin, A. G.
I. On polynomial transformation of C", in Manifolds, Tokyo 1973, Tokyo Univ.
Press, Tokyo (1975), 415-417.
Walker, R. J.
I. Algebraic Curves, Princeton Univ., Princeton (1950), Dover reprint (1962).
van der Waerden, B. L.
I. Modern Algebra, Frederick Ungar Pub. Co, New York: I (1953), xii + 264 pp.;
II (1950), ix + 222 pp.
Weil,A.
I. Foundations of Algebraic Geometry, Amer. Math. Soc., Colloquium Pub!. 29
(1946) (revised and enlarged edition 1962), xx + 363 pp.
2. Surles Courbes Algebriques et les Varietes qui s'en Deduisent, Hermann, Paris
(1948). This volume and the next have been republished in one volume, Courbes
Algebriques et Varietes Abeliennes, Hermann, Paris (1971), 249 pp.
3. Varietes Abeliennes et Courbes Algebriques, Hermann, Paris (1948).
4. Number of solutions of equations over finite fields, Bull. Amer. Math. Soc. 55
(1949), 497-508.
5. Varietes Kiihleriennes, Hermann, Paris (1958), 175 pp.
6. On the projective embedding of abelian varieties, in Algebraic Geometry and
Topology (in honor of S. Lefschetz), Princeton Univ., Princeton (1957) 177-181.
Wells, R. 0., Jr.
1. Differential Analysis on Complex Manifolds, Prentice-Hall (1973), x + 252 pp.
Weyl, H.
I. Die Idee der Riemannschen Fliiche, Teubner (3rd ed., 1955), vii + 162 pp. (1st
ed., 1913).
Zariski, 0.
I. The concept of a simple point on an abstract algebraic variety, Trans. Amer.
Math. Soc. 62 (1947), 1-52.
2. A simple analytical proof of a fundamental property of birational transforma-
tions, Proc. Nat. Acad. Sci. USA 35 (1949), 62-66.
3. Theory and Applications of Holomorphic Functions on Algebraic Varieties over
Arbitrary Ground Fields, Memoirs of Amer. Math. Soc. New York (1951).
4. Complete linear systems on normal varieties and a generalization of a lemma
of Enriques-Severi, Ann. of Math. 55 (1952), 552-592.
5. Introduction to the Problem of Minimal Models in the Theory of Algebraic
Surfaces, Pub. Math. Soc. of Japan 4 (1958), vii + 89 pp.
6. The problem of minimal models in the theory of algebraic surfaces, Amer. J.
Math. 80 (1958), 146-184.
7. The theorem of Riemann-Roch for high multiples of an effective divisor on an
algebraic surface, Ann. Math. 76 (1962), 560-615.
8. Collected papers. Vol. I. Foundations of Algebraic Geometry and Resolution
of Singularities, ed. H. Hironaka and D. Mumford, M.I.T. Press, Cambridge
(1972), xxi + 543 pp. Vol. II, Holomorphic Functions and Linear Systems, ed.
M. Artin and D. Mumford, M.I.T. Press (1973), xxiii + 615 pp.
9. On Castelnuovo's criterion of rationality Pa = P 2 = 0 of an algebraic surface,
Ill. J. Math. 2 (1958) 303-315.
10. Algebraic Surfaces, 2nd suppl. ed., Ergebnisse 61, Springer-Verlag, Heidelberg
(1971).
Zariski, 0. and Samuel, P.
I. Commutative Algebra (Vol. I, II), Van Nostrand, Princeton (1958, 1960).

469
Results from Algebra

Chapter I

(1.3A) Hilbert's Nullstellensatz

(1.8A) Krull dimension of integral domains
(l.llA) Krull's Hauptidealsatz
(1.12A) Unique factorization domains
(3.9A) Finiteness of integral closure
(4.6A) Theorem of the primitive element
(4. 7A) Separating transcendence base
(4.8A) Separably generated field extension
(5.2A) dim m/m 2 ;;::: dim A for a local ring
(5.4A) Completion of local rings
(5.5A) Cohen structure theorem
(5. 7A) Elimination theory
(6.1A) Valuation rings
(6.2A) Discrete valuation rings
(6.3A) Extension of Dedekind domains

Chapter II

(4.11A) Integral closure is intersection of valuation rings

(6.3A) A = nAp taken over ht p = 1
(6.11.1A) Regular local~ UFD
(8.1A) Kahler differentials
(8.2A) Base change
(8.3A) First exact sequence

470
 Results from Algebra

(8.4A) Second exact sequence

(8.6A) Case of a field extension
(8.14A) Localization of regular local rings
(8.21A) Cohen-Macaulay rings
(8.22A) Normal=- R 1 + S2
(8.25A) Existence of a field of representatives
(9.3A) J-adic completion

Chapter III

(l.lA) Derived functors

(1.2A) Acyclic resolutions
(1.3A) Universal 6-functors
(2.1A) Existence of enough injective modules
(3.1A) Krull's theorem
(6.10A) Projective dimension and Ext
(6.11A) Regular local= finite projective dimension
(6.12A) pd + depth = dim
(7.10A) Regular sequences and Koszul complex
(9.1A) Flat modules
(10.3A) Local criterion of flatness

471
Glossary of Notations

s- 1A localization by a multiplicative system, xvi

Av localization by a prime ideal, xvi
A! localization by an element, xvi
k a field, I
A'k affine n-space over k,
k[xb···•xnJ polynomial ring, 2
Z(f) zero set, 2
l(Y) ideal of a set of points, 3
Yo radical of an ideal, 3
A(Y) affine coordinate ring, 4
R real numbers, 4
dim X dimension of a topological space, 5
pk projective n-space, 8
S(Y) homogeneous coordinate ring, lO
tl(Y) ring of regular functions on a variety, 16
tfp,y local ring of a point, 16
K(Y) function field, 16
s<vJ degree zero localization of a graded ring, 18
s«OJ) ditto, 18
s(f) ditto, 18
Ga additive group, 23
Gm multiplicative group, 23
SingY set of singular points, 33
A completion of a local ring, 33
CK abstract nonsingular curve, 42
Aut group of automorphisms, 46
PGL(l) group of fractional linear transformations, 46
AnnM annihilator of a module, 50
!Lv(M) multiplicity, 51
i(Y,H;Z) intersection multiplicity, 53

472
 Glossary of Notations

RegY set of nonsingular points, 54

Q the rational numbers, 58
Spec A spectrum of a ring, 59
:.top( X) category of open sets of X, 61
~b category of abelian groups, 61
f( U,Y) sections of a sheaf, 61
ker kernel, 63
coker cokernel, 63
im image, 63
j*Y direct image sheaf, 65
~b(X) category of sheaves of abelian groups on X, 65
YEB;§ direct sum of sheaves, 66
limY; direct limit of sheaves, 66
--+
limY; inverse limit of sheaves, 67
spe(Y) espace etale of a presheaf, 67
Supp s support of a section of a sheaf, 67
Supp Y support of a sheaf, 67
:YI'om(Y, .~) sheaf of local morphisms, 67
i!(Y) extension of a sheaf by zero, 68
f z(X,Y) sections with support in Z, 68
:YI"~ (Y) subsheaf with support in Z, 68
tJx sheaf of regular functions on a variety, 68
-'r sheaf of ideals of a subvariety, 69
SpecA spectrum of a ring, 70
tJ sheaf of rings, 70
V(a) closed subset of an ideal, 70, 76
D(j). open subset of Spec A, 70
(X, tJx) scheme, 74
sp(X) space of X, 74
A} affine line (as a scheme), 74
s+ ideal of positive elements, 76
ProjS Proj of a graded ring, 76
D+(f) open subset of Proj, 76
P1 projective n-space over a ring, 77
®c~(S) category of schemes over S, 78
~ar(k) category of varieties over k, 78
t( V) scheme associated to a variety, 78
V'x )red associated reduced sheaf of rings, 79
A red reduced ring, 79
X red reduced scheme, 79
tJX local ring of a point on a scheme, 80
mx maximal ideal of local ring at x, 80
k(x) residue field at x, 80
Tx Zariski tangent space at x, 80
c complex numbers, 80
FP finite field of p elements, 80
X! open set defined by j, 81
nil A nilradical of a ring, 82
Yn nth infinitesimal neighborhood, 85

473
Glossary of Notations

dim X dimension of a scheme, 86

codim(Z,X) codimension of a subscheme, 86
XXsY fibred product of schemes, 87
XXY product of schemes, 87
K(X) function field of an integral scheme, 91
ks separable closure of a field, 93
kp perfect closure of a field, 93
xi-x2 specialization, 93
.:l diagonal morphism, 96
rt graph morphism, 106
Homt'x(Y, ;:1) group of morphisms of sheaves of t'x-modules, 109
d'Y'om t1x (Y, ;:§' ) sheaf Hom, 109
Y}i!)tlx~ tensor product, I 09
f*'!J inverse image sheaf, 110
M sheaf associated to an A-module, 110
fy ideal sheaf of a closed subscheme, 115
M sheaf associated to a graded S-module, 116
M(V) degree zero localization, 116
M(f) degree zero localization, 117
tlx (I) twisting sheaf, 117
Y(n) twisted sheaf, 117
r.(Y) graded module associated to a sheaf, 118
XI open set defined by a section of an invertible sheaf, 118
iJ dual of a locally free sheaf, 123
Annm annihilator of an element of a module, 124
f 0 (M) submodule with supports in a, 124
s<d> the graded ring EBn;;.oSnd, 126
T(M) tensor algebra of M, 127
S(M) symmetric algebra of M, 127
1\(M) exterior algebra of M, 127
Speed spectrum of a sheaf of algebras, 128
V(6') vector bundle associated to a locally free sheaf, 128
Div(X) group of divisors, 130
Vy valuation of a prime divisor, 130
(f) divisor of a rational function, 131
K* multiplicative group of a field, 131
D-D' linear equivalence of divisors, 131
CIX divisor class group, 131
degD degree of a divisor, 132
[K(X):K(Y)] degree of a finite field extension, 137
f* inverse image for divisors, 137
Cia X divisor class group of degree 0, 139
,'%* sheaf of invertible elements, 141
CaCIX Cartier divisor class group, 142
Pic X Picard group of X, 143
./(D) sheaf associated to a Cartier divisor, 144
K(X) Grothendieck group, 148
y(Y) image of a sheaf in the Grothendieck group, 148
PGL(n,k) projective general linear group, 151
GL(n,k) general linear group, 151

474
 Glossary of Notations

Proj 160
P(6') projective space bundle, 160
rlf· t!x inverse image ideal sheaf, 163
QB/A module of relative differential forms, 172
d the derivation B~QB/A• 172
tr. d. transcendence degree, 174
dimx dimension of a K-vector space, 174
QX/Y sheaf of relative differentials, 175
Yx tangent sheaf, 180
wx canonical sheaf, 180
Pg geometric genus, 181
fjf2 conormal sheaf, 182
AY;x normal sheaf, 182
gr1 A associated graded ring, 185
pn nth plurigenus, 190
hq,O Hodge number, 190
Qi/k sheaf of regular q-forms, 190
'771 fundamental group, 190
(ML) Mittag-Leffler condition, 191
A completion of a ring with respect to an ideal, 193
x formal completion of a scheme, 194
:T completion of a coherent sheaf, 194
(I, t'l) formal scheme, 194
M6 sheaf on an affine formal scheme associated to a module M, 195
Ob~ set of objects of a category, 202
~b category of abelian groups, 202
~b(X) category of sheaves of abelian groups on X, 202
!mob( X) category of sheaves of modules on a ringed space, 202
Gco(X) category of quasi-coherent sheaves, 202
(Iol)(X) category of coherent sheaves, 202
R;F right derived functor, 204
H;(X,Y) cohomology group, 207
fy(X,Y) sections with support in Y, 212
H~(X, ·) cohomology with support in Y, 212
C(U,Y) Cech complex, 218
IfP(U,Y) Cech cohomology group, 219
6'"(U,Y) sheafified Cech complex, 220
Exti(Y, ·) Ext group, 233
flxt; (Y, ·) Ext sheaf, 233
hd(Y) homological dimension, 238
K 1(X) Grothendieck group, 238
w~ dualizing sheaf, 241
q irregularity of a surface, 247
resp residue map, 247
dlog logarithmic derivative, 250
Rj* higher direct image functor, 250
T 1(A) functor of Lichtenbaum and Schlessinger 267
T;(M) functor associated to a projective morphism, 282
r; localization of T;, 287

475
Glossary of Notations

l(D) dimension of H 0 (X,_I'(D)), 295

K canonical divisor on a curve, 295
X reg set of regular points of a curve, 298
8p measure of a curve singularity, 298
ep ramification index, 299
dtjdu quotient of differentials, 300
R ramification divisor, 301
xp modified scheme in characteristic p, 302
Sec X secant variety, 310
Tan X tangent variety, 310
j the )-invariant of an elliptic curve, 317
Aut X group of automorphisms of an elliptic curve, 318
~n symmetric group on n letters, 318
w a cube root of 1, 320
Aut(X,P0 ) group of automorphisms leaving P0 fixed, 320
nx multiplication by n, 322
End(X,P0 ) ring of endomorphisms, 323
Pic (X / T)
0
relative Picard group, 323
e identity section of a group scheme, 324
p inverse morphism of a group scheme, 324
p. multiplication of a group scheme, 324
p(z) Weierstrass $)-function, 327
J(r) ]-invariant of a lattice, 328
Z[i] ring of Gaussian integers, 331
'IT•( X) fundamental group, 338
Zt /-adic integers, 338
gd a linear system of dimension rand degree d, 341
Wlg set of all curves of genus g, 345
C.D intersection number, 357
degc degree of an invertible sheaf on a curve, 358
(C.D)p intersection multiplicity at P, 360
D2 self-intersection number, 360
K2 self-intersection of the canonical divisor, 361
l(D) dimension of H 0(X,.:I'(D)), 362
s(D) superabundance, 362
c2 second Chern class, 363
D=oE numerical equivalence of divisors, 364
Picn X divisor classes numerically equivalent to 0, 364
NumX group of divisors modulo numerical equivalence, 364
c(D) cohomology class of a divisor, 367
'IT:x-c a ruled surface, 369
f a fibre of a ruled surface, 369
e invariant of a ruled surface, 372
e divisor of 1\ 21], 373
'IT:i-x a monoidal transformation, 386
E the exceptional curve, 386
JJ.p( C) multiplicity of a curve C at a point P, 388
r•<n reduced inverse image divisor, 391
grm(A) associated graded ring of a local ring, 394

476
 Glossary of Notations

p.(A) multiplicity of a local ring, 394

'1T:x~p2 projection of a cubic surface, 401
E~>···,£6 the exceptional curves, 40 I
eh···,e6 their linear equivalence classes, 401
I the class of a line, 40 I
~ a Weyl group, 405
An a Weyl group, 408
T(Z) total transform of Z by T, 410
elmpX elementary transformation of a ruled surface, 416
tc(X) Kodaira dimension, 421
p2 second plurigenus, 422
J.(Y) direct image cycle, 425
A'(X) cycles modulo rational equivalence, 426
A(X) Chow ring, 426
Y.Z intersection cycle class, 426
i(Y,Z; Wj) local intersection multiplicity, 427
C;(6') ith Chern class, 429
c(6') total Chern class, 429
c,( 6') Chern polynomial, 429
ch(6') exponential Chern character, 431
td(6') Todd class, 432
17( Y) cohomology class of Y, 435
Y -homO homological equivalence, 435
!' inverse image in K(X), 435
!! direct image in K(X), 436
xh associated complex analytic space, 439
N, number of points of X rational over F q'• 449
Z(t) the zeta function, 450
E self-intersection of the diagonal, 450
B; ith Betti number, 451
z, /-adic integers, 453
Q, /-adic numbers, 453
H;(X,Q 1) /-adic cohomology, 453
L(f,X) number of fixed points of a morphism, 453
ns) Riemann zeta function, 457

477
Index

Abelian category, 202-206 Algebraic equivalence of divisors, 140, 367,

with enough injectives, 204, 217 369
Abelian variety, 105, 140,422, 437,447,451. Algebraic family. See Family
See also Jacobian variety Algebraic integer, 330-332, 450, 457
Abhyankar, Shreeram, 391 Algebraic set, 2, 4, 5, 9, 47
Abstract nonsingular curve, 42-46, 108, 136 Algebraic space, 445
is quasi-projective, 45 Algebraizable formal scheme, 195
Abstract variety, 58, 105 Algebraizable sheaf, 281
Acyclic resolution, 205 Altman, Allen, 268
Additive functor, 203 Ample divisor. See also Ample invertible
-Adic completion, 193, 254 sheaf
-Adic topology, 33, 213, 279 Nakai-Moishezon criterion, 356, 365, 382,
Adjoint functors, 68, 110, 124 405, 434
Adjunction formula, 361, 368, 374, 382, 387, on a curve, 156, 307, 308, 372
390 on a surface, 365, 368, 405
Affine coordinate ring, 4, 8, 17, 20, 23 Ample invertible sheaf, 150, 153-156, 161,
Affine curve, 4, 7, 8, 47 248
any noncomplete curve is, 297, 298 cohomological criterion of, 229
locally free sheaves on, 385 existence of ~separated, 169
Affine formal scheme, 195 not very ample, 156
global section functor is exact, 198, 199 on a curve, 307, 308, 365
Affine line, A1, 74 on a quadric surface, 156
is not proper, 100 on P", 155
with a point doubled, 76, 96, 169 properties of, 169, 232
Affine morphism, 128, 222, 252 scheme without an, 169, 171
Affine n-space, An, I Analytically isomorphic singularities, 34,
automorphisms of, 23 38, 298, 393-395
Affine plane, A2 , 75 (Fig) Annihilator, 50, 124
Affine scheme, 59, 74, 124 Arithmetic genus,p0 , 54, 57, 201, 230, 232,
cohomology of, 114, 213-216 246, 268
criterion for, 81, 215, 216 constant in a family, 263, 265, 289
global section functor is exact, 113 invariant by monoidal transformation,
is quasj_-compact, 80 388, 394
sheaf M on, 110-113 is a birational invariant, 230, 409, 413
Affine variety, 1-8, 20, 21, 25 of a complete intersection, 54, 231
Algebraic equivalence of cycles, 444 of a curve, 54, 181, 246, 294, 298

478
 Index

of a curve on a surface, 366, 389, 401 Birch, B. J., 317, 334

of a P(lf), 253 Bitangent, 305
of a product, 54 Blowing down lines, 398, 416 (Fig)
of a surface, 188,247, 362, 371,409,421 Blowing up, 28-31, 163-171, 356. See also
Artin, Michael, 417, 445, 449, 452, 453 Monoidal transformation
Artin ring, 266 a curve on a surface, 388-394
Assigned base point of a linear system, 395, a nonsingular subvariety, 186, 188, 394,
399,400 443
Associated point (of a scheme), 257 canonical sheaf of, 188
Atiyah, Michael F., 378, 383 curve singularities, 29 (Fig), 37, 390
Automorphisms is birational, 29, 166
of a curve of genus > 2, 305, 348, 349, is intrinsic, 166
368 is not flat, 258
of a curve of genus 3, 348 Picard group of, 188
of A", 23 strict transform under, See Strict trans-
of an elliptic curve, 318, 321, 336 form
of k(x), 46 surfaces, 56, 395
of P 1, 46 to construct valuation rings, 108, 420
of P", 151, 158, 347 universal property of, 164, 412
of the configuration of 27 lines, 405, 408 vertex of cone, 37, 374, 381
Bombieri, Enrico, 356, 421, 451
Baily, W. L., Jr., 349 Borel, Armand, 149, 239, 432, 435
Biinicii, Constantin, 439 Borelli, Mario, 238
Base extension, 89, 254, 265 Branch divisor, 306
behavior of cohomology, 282, 290, 326, Branched covering, 293
369 Branch of a curve, 30
behavior of differentials, 175 Branch point, 299, 317
flat, 255, 287 Brieskom, Egbert, 420, 421
stable under, 90 Burnside, W., 349
Base for topology, 25, 71
Base-point free, 158, 307, 318 Canonical curve, 341,346,348,353,385
Base points Canonical divisor
of a linear system, 158, 307, 395 on a curve, 293, 295, 299, 373
scheme of, 169 on a hyperelliptic curve, 343
Base scheme, 89 on a singular curve, 298
Baum, Paul, 436 on a surface, 361, 373, 387, 421, 424
Berthelot, Pierre, 452 Canonical embedding, 293, 340-349
Bertini, E., 316 Canonical linear system, 340, 341
Bertini's theorem, 179 (Fig), 183, 187, 188, Canonical morphism, 341, 422
245, 250, 281, 425 Canonical sheaf, wx, 180, 239, 246
gives nonsingular curves, 183, 231, 314, of a blowing-up, 188
342, 358, 381, 406 of a complete intersection, 188, 315
in characteristic 0, 268, 274 of a curve, 294, 295
with singularities, 180 of a hypersurface, 183, 184
Betti numbers, 451, 456 of a nonsingular subvariety, 182
Bezout's theorem, 47, 54, 146, 361, 400, 417 of a product, 187
Bilinear form, 364 of P", 182
Binomial coefficient, 49, 52 Cardinality of a variety, 31
Birational equivalence, 24,26-31,45, 55, Cartan, Henri, 172
181, 314, 370, 418 Cartan's theorem B, 211, 441
Birational invariant, 56, 181, 190, 387,409, Cartesian product (of graded rings), 125
421 Cartier divisor, 140-146, 231. See also
Birationally ruled surface, 370, 383, 419. See Divisor
also Ruled surface algebraic family of, 261
Birational morphism, 56, 166,280, 310, 313 associated Wei! divisor of, 141
Birational transformation, 409-420 effective, 145, 427
defined at a point, 410 linear equivalence of, 141
factorization of, 386, 409, 411-413, 416 on a singular curve, 142, 148, 298
fundamental point of, 410 principal, 141
of a ruled surface, 416 (Fig) sheaf ~(D) associated to, 144

479
Index

Cartier, Pierre, 172 Class of a curve, 305

Cassels, J. W. S., 331, 335 Clemens, C. Herbert, 184, 304
Castelnuovo, G., 351, 368, 422 Clifford's theorem, 341, 343, 350, 423
Castelnuovo's criterion for contracting a Closed immersion, 85, 92. See also Closed
curve, 356, 410, 414-417 subscheme
Category criterion for, 151, 152, 158, 307
abelian, 202-206 is finite, 124
inverse limit in, 192 is of finite type, 93
of abelian groups, mo, 61 is proper, 102
of open sets, ~op(X), 61 is separated, 99
of schemes overS, ®c!)(S), 78 Closed morphism, 91, 100
of sheaves on X, mo(X), 65 Closed points are dense, 93
of varieties over k, ~ar(k), 78 Closed subscheme. See also Closed immer-
• product in, 22 sion
<;ech cocyle, 232, 240 associated Cartier divisor, 145, 149
Cech cohomology, 201, 211, 218-225, 255, associated cycle, 425
262 completion along, 86, 190, 194, 279
computation of, 219, 220, 225-227 criterion to be nonsingular, 178
computes R1•. 252, 282 ideal sheaf of, 115
limit over coverings, 223 image of a morphism, 92
of a plane curve, 224 locally principal, 145
Center of a valuation, 106, 108, 137 of an affine scheme, 85, 92, Ill, 116
Center of blowing up, 163 of a Proj, 92, 119, 125
Characteristic p, 21, 31, 33, 80, 89, 276, 293, of codimension one, 145
312, 316, 317, 332-335, 339, 385, 391, reduced induced structure, 86, 92
422. See also Frobenius morphism with nilpotents. See Nilpotent element
funny curve in, 305, 385 Coarse moduli variety, 347
reduction mod p, 89, 334, 340, 451 Codimension, 86, 87
search for good cohomology, 451 Coeffaceable functor, 206, 238, 240, 243
Characteristic 2, 312, 317 Cohen, I. S., 34, 187
Characteristic zero, 232, 268, 271-275, 304, Cohen-Macaulay ring, 184
305, 348, 382, 391. See also Complex Cohen-Macaulay scheme, 185, 239,243,
numbers 276
Chern class, 363, 425,429-431, 433,435, Coherence of direct image sheaf, 115
437, 450 for a finite morphism, 124
Chern polynomial, 429, 435 for a projective morphism, 123, 152, 252,
Chern, S. S., 423 280
Chevalley, Claude, 48, 94, 95, 172, 222,427, Coherent analytic sheaf, 439, 440, 447
428 Coherent sheaf, 111-115. See also Quasi-
Chinese remainder theorem, 138 coherent sheaf
Chow ring, 426, 428, 429, 433, 437, 454 completion along a closed subscheme,
Chow's Lemma, 107 194
Chow's moving lemma, 427, 434 extending from an open set, 126
Chow variety, 349 Grothendieck group of. See Grothendieck
Chow, Wei-Liang, 105, 324, 441, 443 group
Circle, 213, 220 old definition, 124
Class field theory, 331 on a formal scheme, 194-200
Classical projective geometry, 403, 407 on Proj S, 116-123, 125
Classification on Spec A, 110-116
of curves, 56, 341, 345-347 quotient of a locally free sheaf, 121, 238
of curve singularities, 38, 393-395 Cohn-Vossen, S., 403
of curves in P 3 , 349-355, 354 (Fig), 409 Cohomological dimension, 224
of curves of genus 2, 304 Cohomology. See also Cech cohomology
of elliptic curves, 317, 345 and base change, 282, 290, 326, 369
of elliptic ruled surfaces, 377 as a derived functor, 207, 211, 439
of locally free sheaves of rank 2 on an Cech process, 218-225
elliptic curve, 378 characterizes ample sheaf, 154, 229
of rational ruled surfaces, 376, 419 class of a cycle, 435, 454
of surfaces, 56, 421-423 class of a divisor, 367, 418
problem, 39, 55-57, 181, 293, 345 class of a subvariety, 249

480
 Index

commutes with flat base extension, 255, Complex multiplication, 330-332, 334,
287 337-339
commutes with lim, 209 Complex numbers, C, 106, 317, 326-332,
crystalline, 452 ~ 367,391,414,417,420,422,432.See
etale, 307, 452, 453 also Transcendental methods
functor, 207 Condition (t), 160
gives numerical invariants, 57, 230, 246, 247 Condition (ML) of Mittag-Leffler, 191, 192,
group, 207, 424 200, 278, 290
1-adic, 435, 449, 452-457 Condition (*), 130, 426
of a circle, 213, 220 Condition S 2 of Serre, 185
of a complete intersection, 231 Conductor, 331, 340
of a complex, 203 Cone, 12 (Fig), 13, 38, 49, 266
of an affine scheme, 114, 213-216 blowing up vertex of, 37, 374, 381
of fibres, 250, 255, 281, 290 divisor class group of, 146
of P 1, 219 quadric. See Quadric cone
of projective space, 225-230 ruling on, 134 (Fig)
of sheaves, 206-211 Configuration of 27 lines, 404
p-adic, 452 Conic, 7, 20, 30, 183, 312
Picard group as, 143, 224, 367, 446 as a real form of P 1, 107
theories, 211, 449 determined by 5 points, 397
with supports, 212, 217 linear system of, 170, 396-398
Cokemel, 63, 65, 109 Pascal's theorem, 407 (Fig)
Complete intersection, 14, 188, 452 Conical double point, 36 (Fig)
a divisor on a quadric hypersurface is, Connected. See also Zariski's Main Theorem
147 ample divisor is, 244
arithmetic genus of, 54, 231 complete intersection is, 188, 231
canonical sheaf of, 188, 315 fibre is, 279, 280
cohomology of, 231 total transform of a point is, 410
curve, 38, 342, 346, 352, 355 when Spec A is, 82
elliptic quartic curve is, 38 Connectedness principle, 190, 281
geometric genus of, 188 Conormal sheaf, .fj.P, 182, 245, 246. See
hyperelliptic curve is not, 348 also Normal sheaf
hyperplane section of, 267 of a nonsingular subvariety, 178
is connected, 188, 231 of a twisted cubic curve, 385
is every curve?, 14 relation to differentials, 175, 176
local, 184-186, 245, 428 Constant sheaf, 62, 65
normal ~ projectively normal, 188 is flasque, 67
not a local, 8 of function field, 69, 112, 145
set-theoretic, 14, 224 Constructible set, 94, 266
strict, 14, 188 Continuous family. See Family
surface, 409, 423, 437 Contractible curve, 417, 419, 420
Complete linear system, 157, 159, 170, 294. Contraction
See also Linear system of exceptional curves, 410, 414-416
dimension of. See Riemann-Roch prob- problem, 417, 419
lem Coordinates
Complete variety, 105, 106, 136 affine, I
any variety is contained in a, 168 homogeneous, 9
nonprojective, 171, 443 (Fig) Correspondence on a curve, 451
Completion Cotangent complex, 185
along a subscheme, 86, 190, 194, 279 Counting constants, 401
ofalocalring,33-35, 187,275,278,420 Courant, R., 326
of a ring with respect to an ideal, 193 Cousin problem, 69
Complex analytic space, 96, 252, 291, 417 Cremona transformation, 30. See also
associated to a scheme over C, 439, 440 Quadratic transformation
cohomology of, 439 Crystalline cohomology, 452
definition, 438 Cubic curve. See also Cuspidal cubic curve;
Complex cohomology, H; (·,C), 435 Elliptic curve; Nodal cubic curve;
Complex in an abelian category, 203, 282 Nonsingular cubic curve; Twisted cubic
Complex manifold, 249, 289, 435, 438, 454 curve
when is it algebraic?, 441-445 linear system of, 399

481
Index

Cubic curve (cont.) of genus 2, 298, 304, 309, 315, 341, 347,
through 8 points determines a 9th, 400 355. See also Hyperelliptic curve
Cubic surface in pJ, 356, 395-409 of genus 3, 342, 346-349
ample divisors on, 405 of genus 4, 342, 346, 348
as P 2 with 6 points blown up, 400 of genus 5, 346, 348, 353, 406
canonical sheaf of, 184, 401 of genus 6, 348, 409
curves on, 401, 406-409 of genus 10, 354, 409
Picard group of, 136, 401 of genus II, 355, 409
27 lines on, 402-406 on a cubic surface, 401, 406-409
Cubic threefold is not rational, 184 on a quadric surface. See Quadric surface,
Cup-product, 453, 454, 456 curves on
Curve, 105, 136. See also Abstract nonsingu- over a finite field, 339, 368, 458. See also
lar curve; Nonsingular curve; Plane Riemann hypothesis
curve over R, 4
affine, 4, 7, 8, 47, 297, 298, 385 product of, 44, 338, 367, 368
ample divisor on, 156, 307, 308, 372 rational. See Rational curve
any two homeomorphic, 31 ruled surface over. See Ruled surface
behavior under monoidal transformation, singularities, 35, 36 (Fig), 38, 386, 393.
388-394 See also Cusp; Node; Tacnode
birational to a plane curve with nodes, singular, Picard group of, 148
314 strange, 311, 316
can be embedded in p3, 310 trigonal, 345
classification of, 56, 341, 345-347 twisted cubic. See Twisted cubic curve
complete intersection, 38, 342, 346, 352, twisted quartic. See Rational quartic curve
355 zeta function of, 458
complete~ projective, 44, 136, 232, 294 Cusp, 36 (Fig), 37, 39, 298, 305, 392 (Fig),
cubic. See Cubic curve 394. See also Cuspidal cubic curve
definition for Ch. II, 105 higher order, 37
definition for Ch. IV, 294 Cuspidal cubic curve, 21, 171, 276
definition for Ch. V, 357 as a projection, 22, 266
divisors on, 129, 136-140, 294 blown up, 31, 392 (Fig)
dual, 54, 304 divisor class group of, 142, 148
elliptic. See Elliptic curve is not normal, 23
elliptic quartic, 38, 315, 353 is rational, 30
equivalence of singularities, 393 Cycle, 425
exceptional. See Exceptional curve associated to a closed subscheme, 425
existence for all g ;;. 0, 294, 385, 394 cohomology class of, 435, 454
genus bounded by degree, 315,351,407, homological equivalence of, 435, 444
408 of dimension zero, 437
genus of. See Genus rational equivalence of, 425, 426, 436, 454
genus of normalization, 393
hyperelliptic. See Hyperelliptic curve Decomposable locally free sheaf, 376, 378,
in P 3, classification of, 349-355, 354 (Fig), 383, 384
409 Dedekind domain, 40, 41, 58, 132
invariant ~P of a singularity, 298, 393-395 Deformation, 89, 90 (Fig), 188, 267. See also
locally free sheaf on, 369, 370, 372, 376, Family
378, 379, 384, 385 Deformation theory, 265
of degree 2, 315, 353. See also Conic Degree
of degree 3, 159, 315, 353. See also Cubic of a coherent sheaf on a curve, 149, 372
curve of a divisor on a curve, 137, 142, 294
of degree 4, 159, 309, 315, 342, 353, 355, of a divisor on a projective variety, 132,
407. See also Quartic curve 146
of degree 5, 348, 353, 355 of a finite morphism of curves, 137, 298
of degree 6, 342, 350, 353 of a hypersurface, 52
of degree 7, 353, 406 of a linear system on a curve, 159
of degree 8, 346, 409 of an intersection, 53
of degree 9, 354, 355, 409 of a plane curve, 4, 54
of genus 0, 297, 345. See also Rational of a projective variety, 47, 52, 57, 250, 309,
curve 366
of genus I. See Elliptic curve of a zero-cycle, 426, 428

482
 Index

Deligne, Pierre, 217,249, 289, 317, 347, 449, Diophantine equation, 335, 340
452 Direct image
Del Pezzo surface, 401, 408 cycle, 425
6-functor, 205, 219, 234, 238, 240, 243, 282 divisor, 306, 436
Demazure, M., 452 sheaf, f.:F, 65, 109, 115, 123, 124, 250.
Density of a set of primes, 334, 339 See also Higher direct image sheaf
Depth, 184, 237, 243, 264 Direct limit, lim, 66, 72, 109, 208, 209
-->
cohomological interpretation of, 217 Direct product, 82, 109. See also Product
Derivation, 172, 189 Direct sum, $, 66, 109
Derivative, 31, 300 Dirichlet's minimum principle, 441
Derived functor, 201-206 Dirichlet's theorem, 335, 339
cohomology, 207, 211, 439. See also Discrete valuation ring, 40, 42, 45, 107, 108,
Cohomology 258, 325. See also Valuation; Valuation
Ext, 233 ring
higher direct images, R'i., 250. See also center of, on a curve, 137
Higher direct image sheaf of a prime divisor, 130
local cohomology modules, H~(M), 217 set of. See Abstract nonsingular curve
Determinant of a coherent sheaf, 149, 306 spectrum of, 74, 93, 95, 124
Deuring, M., 331, 334 Disjoint union, 80
Diagonal, 24, 48 Divisor, 57, 129-149. See also Cartier
closed ~ scheme separated, 96 divisor; Divisor class group; Invertible
homomorphism, 96, 173 sheaf; Wei) divisor
morphism, 96, 99, 107, 175, 427 algebraic equivalence of, 140, 367, 369
reduction to the, 427, 428 associated to an invertible sheaf, 144, 145,
self-intersection of, 367, 368, 437, 450 157, 294, 425
Diagonalized bilinear form, 364 cohomology class of, 367, 418
Diagram-chasing, 203 degree of, 132, 137, 142, 146, 294
Difference polynomial, 49 effective, 130
Differentiable structures on a sphere, 421 group of all, Div(X), 130, 357
Differential form. See Differentials inverse image of, j*, 135, 137, 299
Differential geometry, 311, 438, 445 linear equivalence of, 57, 131, 141, 294,
Differentials, 57, 172-190. See also Canoni- 367, 425, 426
cal sheaf locally principal, 142
Kahler, 172-175 numerical equivalence of, 364, 367, 369
module free ~ regular local ring, 174 of an elliptic function, 327
of a polynomial ring, 173 of a rational function, 130, 131, 294
on An, 176 on a curve, 129, 136-140, 294
on a product, 187 on a surface, 135, 357
on P", 176 prime, 130
residues of, 247 principal, 131, 132, 138, 141
sheaf locally free~ nonsingular variety, special, 296
177, 178, 276 very ample. See Very ample divisor
sheaf of, 175-177, 219, 247, 268, 295, 300 with normal crossings, 391
sheaf of q-forms, 01/k• 190, 247, 249 Divisor class group, CI(X), 131, 145. See
Dilatation, 386 also Picard group
Dimension exact sequence of an open subset, 133
equal to transcendence degree, 6 is zero~ UFO, 131
of a linear system, 157, 295, 357, 424. See of a cone, 146
also Riemann-Roch problem of a cubic surface, 136
of An is n, 6 of a curve, 139, 140, 142, 148
of a projective variety, 10, 57 of a Dedekind domain, 132
of a ring, 6, 86 of a product, 134, 146
of a scheme, 86, 87, 94 of a quadric hypersurface, 147
of a special linear system, 341. See also of a quadric surface, 133, 135
Clifford's theorem of a variety in P", 146
of a topological space, 5, 8, 208 of P", 132
of fibres of a morphism, 95, 256, 257, 269 Dominant morphism, 23, 81, 91, 137
of intersections, 48 Dominant rational map, 24, 26
of P" is n, 12 Domination (of local rings), 40, 98
relative, 95 Double line, 36 (Fig), 90 (Fig)

483
Index

Double point, 36 (Fig), 37, 38, 393. See also Elliptic scroll, 385
Cusp; Tacnode Elliptic surface, 422
ordinary. See Node Embedded point, 85, 259
Doubly periodic function, 327 Embedded resolution of singularities, 390,
Dual curve, 54, 304 391, 392 (Fig), 419
Duality, 239-249. See also Serre duality Embedding
for a finite flat morphism, 239, 306 a curve in projective space, 307-316
Dualizing sheaf, 239, 241, 242, 246, 249, 298 a variety in a complete variety, 168
Dual locally free sheaf, if, 123, 143, 235, Enough injectives, 204, 217
430 Enough locally frees, 238, 239
Dual numbers, ring of, 80, 265, 267, 324 Enough projectives, 235
Dual projective space, (P")•, 54, 55, 130, Enriques, Federigo, 348
304, 316 Enriques-Severi-Zariski, lemma of, 244
d-Uple embedding. See -Uple embedding Enriques surface, 422
Dwork, Bernard M., 451 Equidimensional, 243
Dynkin diagram, 420 J?:space etale of a presheaf, 67
Etale
Effaceable functor, 206 cohomology, 307, 452, 453
Effective divisor, 130, 145, 157, 294, 363 covering, 303, 306, 338, 340, 442
Elements de Geometrie Algebrique (EGA), equivalence relation, 445
89, 100, 462 morphism, 268, 275, 276 (Fig), 299
Elimination theory, 35, 39 neighborhood, 275
Elliptic curve, 46, 56, 293, 316-340. See also topology, 452, 453
Cubic curve Euler characteristic, 230, 295, 360, 362, 366,
as a plane cubic curve, 309, 319 424
automorphisms of, 318, 321, 336 Euler-Poincare characteristic, topological,
canonical divisor, 297 456
classified by }-invariant, 317, 345 Euler's lemma, 37
complex multiplication on, 330-332, 334, Exact in the middle, 204, 282
337-339 Exact sequence of sheaves, 64, 66, 68, 109
defined over Q, 335 Exceptional curve, 29, 31, 108, 386, 392, 395,
dual of a morphism, 337 408, 437
group structure, 297, 316, 321-323. See contraction of, 410, 414-416
also Group, law on cubic curve infinitely many, 409, 418
group structure over C, 329 of the first kind, 410, 414, 418
Hasse invariant of, 317, 322, 332-335, 339, self-intersection of, 386
340 Excision, 212
in characteristic p, 317, 332-335 Exotic sphere, 421
isogeny of, 338 Exponential Chern character, 431, 435
Jacobian variety of, 316, 323-326, 338 Exponential sequence, 446, 447
}-invariant, 316-321, 331, 336, 345, 347 Extending
locally free sheaves on, 378 a function to a normal point, 23, 217
over C, 326-332 a morphism, 43, 44, 97, 370
Picard group of, 297, 323 a section of a sheaf, 67, 112, 118
points of order n, 322, 323, 329, 337, 340 a sheaf by zero, 68, Ill, 149
points of order p, 339 coherent sheaves, 126
points with integer coordinates, 340 Extension
quartic, 38, 315, 353 of invertible sheaves, 372, 375, 376, 383,
rational points over Fq• 339 430
rational points over Q, 317,335,336 (Fig) of £T!x-modules, 237
ring of endomorphisms, 323, 329, 330, of quasi-coherent sheaves, 114
338, 340 Exterior algebra, 1\M, 127
supersingular, 332 Exterior power, NM, 127, 149, 181, 430
withj=O, 320, 321, 331, 334 Ext group, 233-240,375,376
with}= 1728, 320, 321, 331, 334 Ext sheaf, 233-239, 241
zeta function of, 458
Elliptic function, 316, 326-332, 338 Faithful functor, 290
Elliptic ruled surface, 369, 375, 384, 385, Family
440 flat, 253, 256-266, 260 (Fig), 289, 315
classification of, 377 of curves of genus g, 347

484
 Index

of divisors, 261, 367, 384 Fixed point of a morphism, 451,453,454

of elliptic curves, 340, 347 Flasque resolution, 201, 208, 212, 248
of hypersurfaces, 291 Flasque sheaf, 67, 207
of invertible sheaves, 323 cohomology vanishes, 208, 221, 251
of locally free sheaves, 379 direct limit of, 209
of plane curves, 39 injective sheaf is, 207
of plane curves with nodes, 314 Flat
of schemes, 89, 90 (Fig), 202, 250, 253 base extension, 255, 287 ·
of varieties, 56, 263 family, 253,256-266, 260 (Fig), 289, 315.
Fermat curve, 320, 335, 339 See also Family
Fermat hypersurface, 451 module, 253
Fermat's problem, 58, 335 morphism, 239, 253-267, 269, 299, 340,
Fermat surface, 409 436
Fibre morphism is open, 266
cohomology of, 250, 255, 281, 290 sheaf, 254, 282
dimension of, 95, 256, 257, 269 Flatness
is connected, 279, 280 is an open condition, 266
of a morphism, 89, 92 local criterion of, 270
with nilpotent elements, 259, 277, 315 Formal completion. See Completion
Fibred product, 87, 100. See also Product Formal functions, 276-281, 290, 387, 415
Field Formal neighborhood, 190
algebraically closed, I, 4, 22, 152 Formal power series, 35
of characteristic p. See Characteristic p Formal-regular functions, 199,279. See also
of complex numbers. See Complex Holomorphic functions
numbers, C Formal scheme, 190-200, 279
of elliptic functions, 327 Picard group of, 281
of meromorphic functions, 442 Fractional linear transformation, 46, 328
of quadratic numbers, 330-332, 334, Free module, 174
340 Freshman calculus, 408
of rational numbers. See Rational num- Freyd, Peter, 203
bers, Q Frobenius morphism, 21,272, 301,332, 340.
of real numbers, R, 4, 8, 80, 106 See also Characteristic p
of representatives, 187, 275 fixed points of, 454
perfect, 27, 93, 187 k-linear, 302, 339, 368, 385
separable closure of, 93 1-adic representation of, 451
spectrum of, 74 trace on cohomology, 455
transcendence degree of, 6, 27 Frohlich, A., 331
uncountable, 409, 417 Fulton, William, 297, 436
Field extension Functional equation, 450, 458
abelian, 332 Function field, 16
purely inseparable, 302, 305, 385 determines birational equivalence class,
pure transcendental, 303 26
separable, 27, 300, 422 of an integral scheme, 91
separably generated, 27, 174, 187, 271 of a projective variety, 18, 69
Final object, 79 of dimension I, 39, 44
Fine moduli variety, 347 transcendence degree of, 17
Fine resolution, 201 valuation rings in, 106, 108
Finite field, 80, 339. See also Characteristic p Functor
number of solutions of polynomial equations additive, 203
over, 449 adjoint, 68, 110, 124
Finite morphism, 84, 91, 124, 280, 456 coeffaceable, 206, 238, 240, 243
a projective, quasi-finite morphism is, 280, derived. See Derived functor
366 effaceable, 206
is affine, 128 exact in the middle, 204, 282
is closed, 91 faithful, 290
is proper, 105 left exact, 113, 203, 284, 286
is quasi-finite, 91 of global sections. See Global sections
of curves, 137, 148, 298-307, 299 (Fig) representable, 241, 324
Finite type, morphism locally of, 84, 90 right exact, 204, 286
Finite type, morphism of, 84, 91, 93, 94 satellite, 206

485
Index

Fundamental group, '17 1, 190, 338, 420, 442 of a morphism, 368, 426
Fundamental point, 410 Grauert, Hans, 249, 252, 288, 291, 369, 417,
Funny curve in characteristic p, 305, 385 438, 442, 445
Griffiths, Phillip A., 184, 304, 423, 445
GAGA, 330, 440 Grothendieck, A., 57, 59, 60, 87, 89, 100,
Galois extension, 318 115, 120, 172, 190, 192, 201, 208, 217,
Galois group, 147, 320, 338, 442 249,252,259,279,281,282,291,324,
Gaussian integers, 331, 335 338,363,366,368,429,432,435,436,
General position, 409, 418 441, 442, 449, 451-453
Generically finite morphism, 91, 436 Grothendieck group, K(X), 148, 149,230,
Generic point, 74, 75 (Fig), 80, 294 238, 385, 435
in a Zariski space, 93 Group
local ring of, 91, 425 additive, G0 , 23, 142, 148, 171
Generic smoothness, 272 fundamental, '17 1, 190, 338, 420, 442
Generization, 94 Galois, 147, 320, 338, 442
Genus general linear, GL, 151
arithmetic. See Arithmetic genus Grothendieck. See Grothendieck group
bounded by degree, 315, 351,407,408 law on cubic curve, 139 (Fig), 142, 147,
geometric. See Geometric genus 148, 297, 299, 407, 417
of a curve, 54, 56, 140, 183, 188, 294, 345, multiplicative, Gm, 23, 148, 149
421 Neron-Severi, See Neron-Severi group
of a curve on a surface, 361, 362, 393, 401, of automorphisms. See Automorphisms
407, 408 of cycles modulo rational equivalence.
Geometrically integral scheme, 93 See Chow ring
Geometrically irreducible scheme, 93 of divisor classes, See Divisor class group
Geometrically reduced scheme, 93 of divisors, Div(X), 130, 357
Geometrically regular, 270 of divisors modulo algebraic equivalence,
Geometrically ruled surface. See Ruled See Neron-Severi group
surface of divisors modulo numerical equivalence,
Geometric genus,p8 , 181, 190, 246,247, 294, Num X, 364, 367-369
421 of invertible sheaves. See Picard group
is a birational invariant, 181 of order 6, 318, 321
of a complete intersection, 188 of order 12, 321
Geometry on a surface, 357-368 of order 60, 420
Germ, 62, 438 of order 168, 349
Global deformation, 265, 267. See also of order 51840, 405
Deformation; Family projective general linear, PGL, 46, 151,
Global sections. See also Section 273, 347 .
finitely generated, 122, 156, 228 scheme, 324
functor of, 66, 69, 113 symmetric, 304, 318, 408
restricted to open set D(f), 112 variety, 23, 139 (Fig), 142, 147, 148, 272,
sheaf generated by, 121, 150-156, 307, 321, 323, 324
358, 365 Weyl, 405, 408
Glueing Gunning, Robert C., 69, 201, 249, 438,442
analytic spaces, 439
morphisms, 88, 150 Halphen, G., 349
schemes, 75, 80, 91, 171, 439, 444 Harmonic function, 442
sheaves, 69, 175 Harmonic integrals, 435, 445
Godement, Roger, 61, 172, 201 Hartshorne, Robin, 14, 105, 140, 144, 190,
Graded module, 50 193, 195, 199,224, 249, 281, 366, 367,
associated to a sheaf, r .(JF), 118 383, 419, 428, 440
quasi-finitely generated, 125 Hasse, H., 339
sheaf M associated to, 116 Hasse invariant, 317, 322, 332-335, 339, 340
Graded ring, 9, 394, 426 Hausdorff toplogy, 2, 8, 95, 439
associated to an elliptic curve, 336 Height of a prime ideal, 6
graded homomorphism of, 80, 92 Hermitian metric, 445
Proj of, 76 Higher direct image sheaf, Ri.(JF), 250, 276,
Graph 282, 290, 371, 387, 436
morphism, 106, 107 locally free, 288, 291
of a birational transformation, 410 Hilbert, David, 51, 403, 441

486
 Index

Hilbert function, 51 Ideal of a set of points, 3, 10

Hilbert polynomial, 48, 49, 52, 57, 170, 230, Ideal of definition, 196
231, 294, 296, 366 Ideal sheaf, .fy, 109
constant in a family, 256, 261, 263 blowing up, 163, 171
Hilbert-Samuel polynomial, 394 of a closed subscheme, I 15, I 16, 120, 145
Hilbert scheme, 258, 349 of a subvariety, 69
Hilbert's Nullstellensatz, 4, II of denominators, 167
Hironaka, Heisuke, 105, 168, 195, 264, 391, Idempotent, 82
413, 417, 442, 443, 445, 447 Igusa, Jun-ichi, 265, 334
Hinebruch, Friedrich,57,363,421,431,432 Image
Hodge index theorem, 356, 364, 366, 435, direct See Direct image
452 inverse. See Inverse image
Hodge manifold, 445 of a morphism of sheaves, 63, 64, 66
Hodge numbers, hP,q, 190, 247 of a proper scheme is proper, I 06
Hodge spectral sequence, 289 scheme-theoretic, 92
Hodge theory, 414, 435, 445, 452 Immersion, 120
Holomorphic functions, 60, 190, 279, 330, closed, See Closed immersion
438, 447. See also Formal-regular open, 85
functions Indecomposable locally free sheaf, 376, 384
Homeomorphism, 21, 31 Index of a bilinear form, 364
Homogeneous coordinate ring, 10, II, 18, Index of speciality, 296
23, 49, 128, 132 Induced structure
criterion to be UFD, 147 of scheme, 79, 86, 92
depends on embedding, 2 I of variety, 21
Proj of, 81 Inequality of Castelnuovo and Severi, 368,
Homogeneous coordinates, 9 451
Homogeneous element, 9 Infinitely near point, 392, 395
Homogeneous ideal, 9, 10, 92, 125 Infinitesimal deformation. See Deformation
Homogeneous space, 273 Infinitesimal extension, 189, 225, 232, 265
Homological dimension, 238 Infinitesimal lifting property, 188
Homological equivalence of cycles, 435, 444 Infinitesimal neighborhood, 86, 190, 276,
Homotopy of maps of complexes, 203 393
Homotopy operator, 203, 221 Inflection point, 139, 148, 304, 305, 335, 337
Hopf map, 386 cubic curve has 9, 305, 322
Horrocks, G., 437 Initial object, 80
Hurwitz, Adolf, 301, 305, 326 Injective module, 206, 207, 213, 214, 217
Hurwitz's theorem, 293, 299-307, 31 I, 313, Injective object (of a category), 204, 217, 233
317, 337, 382 Injective resolution, 204, 242
Husemoller, Dale, 356, 421 Injective sheaf, 207, 213, 217
Hyperelliptic curve, 298, 306, 341, 345, 384. Inseparable morphism, 276, 311, 312
See also Nonhyperelliptic curve Integers, Z, 79, 340
canonical divisor of, 342, 343 Integral closure, 40, 91, 105
existence of, 298, 394 finiteness of, 20, 38, 123
moduli of, 304, 347 Integrally closed domain, 38, 40, 126, 132,
not a complete intersection, 315, 348 147, See also Normal
Hyperelliptic surface, 422 Integral scheme, 82, 91
Hyperosculating hyperplane, 337 Intersection divisor, 135, 146, 261
Hyperosculation point, 337, 348 Intersection multiplicity, 36, 47, 53, 233, 304,
Hyperplane, 10, 429 357, 360, 394, 427
corresponds to sheaf Ill (I), I 45 Intersection number, 357-362, 366, 394, 445
Hyperplane section, 147, 179 (Fig) Intersection of affines is affine, I06
Hypersurface, 4, 7, 8, 12 Intersection of varieties, 14, 21, 47-55
any variety is birational to, 27 Intersection, proper, 427
arithmetic genus of, 54 Intersection, scheme-theoretic, 171, 358
canonical sheaf of, I 83, 184 Intersection theory, 47-55, 58, 424-437
complement of is affine, 21, 25 axioms, 426
existence of nonsingular, 183 on a cubic surface, 401
on a monoidal transform, 387
Icosahedron, 420 on a nonsingular quasi-projective variety,
Ideal class group, 132 427

487
Index

Intersection theory (cont.) Knutson, Donald, 445

on a quadric surface, 361, 364 Kodaira dimension, 421, 422
on a ruled surface, 370 Kodaira, Kunihiko, 249, 266,409,414,422,
on a singular variety, 428 442,443
on a surface, 356-362, 366, 425 Kodaira vanishing theorem, 248, 249, 408,
on p2, 361. See also Bezout's theorem 423, 424, 445
Invariant theory, 420 Koszul complex, 245, 389
Inverse image Krull-Akizuki, theorem of, 108
ideal sheaf,j- 1..1·1Vx, 163, 186 Krull dimension, 6, 86. See also Dimension
of cycles, 426 Krull's Hauptidealsatz, 7, 48
of divisors,j*, 135, 137, 299 Krull, Wolfgang, 213, 279
sheaf, f- 1$"", 65 K3 surface, 184, 422, 423, 437, 452
sheaf, j*$"", I 10, 115, 128, 299 Kunz, E., 249
Inverse limit, lim Kuyk, W., 317, 334
in a category':I92
of abelian groups, 190-192, 277 1-Adic cohomology, 435, 449, 452-457
of rings, 33 1-Adic integers, 338, 453
of sheaves, 67, 109, 192 Laksov, D., 345
Inverse system, 190 Lang, Serge, 140, 334, 364, 367, 452
Invertible sheaf, 109, 117, 118, 124, 143-146, Lascu, A.T., 431
169 Lattice, 326 (Fig)
ample. See Ample invertible sheaf Lefschetz fixed-point formula, 453, 454
associated to a divisor, 144, 145, 157, 294, Lefschetz pencil, 457
425 Lefschetz, Solomon, 451
determines a morphism toP", 150-153, Lefschetz theorem, 190, 281
158, 162, 307, 318, 340 strong, 452
extension of, 372, 375, 376, 383, 430 Left derived functor, 205
generated by global sections, 150-156, Left exact functor, 113, 203, 284, 286
307, 358, 365 Length of a module, 50, 290, 360, 394
group of. See Picard group Leray spectral sequence, 252
on a family, 291 Leray, theorem of, 211
very ample. See Very ample invertible Lichtenbaum, Stephen, 185, 267
sheaf Lie group, 328
Involution, 106, 306 Line, 22, 28, 129, 183. See also Linear
Irreducible variety; Secant line; Tangent line
closed subset, 78, 80 on a surface, 13, 136, 367, 402-406, 408
component, 5, 7, II, 47, 365 Linear equivalence, 57, 131, 141, 294, 425,
scheme, 82 426. See also Divisor
topological space, 3, 4, 8, II =>algebraic equivalence, 367
Irregularity, 247, 253, 422 Linear projection. See Projection
Irrelevant ideal, II Linear system, 130, 150, 156-160, 274
Iskovskih, V.A., 184, 304 complete, 157, 159, 170, 294
Isogeny of elliptic curves, 338 determines a morphism toP", 158, 307,318
Isolated singularity, 420 dimension of, 157, 295, 357, 424
Italians, 391 not composite with a pencil, 280
of conics, 170, 396-398
Jacobian matrix, 32 of plane cubic curves, 399, 400
Jacobian polynomial, 23 on a curve, 307
Jacobian variety, 105, 140, 316, 323-326, separates points, 158, 308, 380
338, 445, 447. See also Abelian variety separates tangent vectors, 158, 308, 380
j-Invariant of an elliptic curve, 316-321, very ample, 158, 307, 308, 396
331, 345, 347 with assigned base points, 395
Jouanolou, J. P., 436 without base points, 158, 307, 341
Linear variety, 13, 38, 48, 55, 169, 316. See
Kahler differentials. See Differentials also Line
Kahler manifold, 445, 446, 452 Lipman, Joseph, 249, 391, 420
Kernel, 63, 64, I 09 Local cohomology. See Cohomology, with
Kleiman, Steven L., 144, 238, 268, 273, 345, supports
434, 435, 452, 453 Local complete intersection, 8, 184-186, 245,
Klein, Felix, 147, 349, 420 428. See also Complete intersection

488
 Index

Local criterion of flatness, 270 free, 174

Local homomorphism, 72, 74, 153 graded. See Graded module
Locally closed subset, 21, 94 of finite length, 50, 290, 360, 394
Locally factorial scheme, 141, 145, 148, .238 Moduli, variety of, 56, 58, 266, 317, 346, 421
Locally free resolution, 149, 234, 239 Moishezon, Boris, 366, 434, 442, 443, 446
Locally free sheaf, 109, 124, 127, 178 Moishezon manifold, 442-446
as an extension of invertible sheaves, 372, nonalgebraic, 444 (Fig)
375, 376, 383, 430 Monodromy, 457
Chern classes of. See Chern class Monoidal transformation, 356, 386-395,
decomposable, 376, 378, 383, 384 409, 410, 414, 443. See also Blowing-up
dual of, 123, 143, 235, 430 ample divisor on, 394
indecomposable, 376, 384 behavior of a curve, 388-394
of rank 2, 356, 370, 376, 378, 437 behavior of arithmetic genus, 387-389
on a curve, 369, 370, 379, 384 behavior of cohomology groups, 387, 419
on an affine curve, 385 canonical divisor of, 387
projective space bundle of. See Projective Chow ring of, 437
space bundle intersection theory on, 387
resolution by, 149, 234, 239 local computation, 389
stable, 379, 384 Picard group of, 386
trivial subsheaf of, 187 Morden, L. J., 335
vector bundle V(lf) of, 128, 170 Morphism
zeros of a section, 157, 431 affine, 128, 222, 252
Locally noetherian scheme, 83 closed, 91, 100
Locally principal closed subscheme, 145 determined by an open set, 24, I 05
Locally principal Weil divisor, 142 dominant, 23, 81, 91
Locally quadratic transformation, 386 etale, 268, 275, 276 (Fig), 299
Locally ringed space, 72, 73, 169. See also finite. See Finite morphism
Ringed space flat. See Flat, morphism
Local parameter, 137, 258, 299 Frobenius. See Frobenius morphism
Local ring generically finite, 91, 436
complete, 33-35, 187, 275, 278, 420 glueing of, 88, 150
local homomorphism of, 72, 74 injective, of sheaves, 64
of a point, 16-18, 22, 31, 41, 62, 71, 80 inseparable, 276, 311, 312
of a subvariety, 22, 58 locally of finite type, 84, 90
regular. See Regular local ring of finite type, 84, 91, 93, 94
Local space, 213 of Spec K to X, 80
Logarithmic differential, dlog, 250, 367, 457 projective, 103, 107, 123, 149-172, 277,
Lubkin, Saul, 452 281,290
Liiroth's theorem, 303, 422 proper. See Proper morphism
quasi-compact, 91
Macbeath, A.M., 306 quasi-finite, 91, 280, 366
MacPherson, R.D., 436 quasi-projective, 103
Manifold, 31. See also Complex manifold ramified, 299, 312
Manin, Ju. 1., 149, 184, 304, 401, 405, 408, separated. See Separated morphism
436,452 smooth, 268-276, 303
Maruyama, Masaki, 383 smjective, of sheaves, 64, 66
Matsumura, Hideyuki, 123, 172, 184, 195, toP", determined by an invertible sheaf,
250, 268, 418 150-153, 158, 162, 307, 318, 340
Mattuck, Arthur, 451 universally closed, I 00 .
Maximal ideal, 4 unramified, 275, 299. See also Etale
Mayer, K.H., 421 morphism
Mayer-Vietoris sequence, 212 Morrow, James, 266, 442
Meromorphic function, 327, 442 Moving lemma, 425, 427, 434
Minimal model, 56, 410, 418, 419, 421 Moving singularities, 276
Minimal polynomial, 147 Multiple tangent, 305
Minimal prime ideal, 7, 50, 98 Multiplicity, 36, 51, 388, 393, 394. See also
Mittag-Leffler condition, (ML), 191, 192, Intersection multiplicity
200, 278, 290 Multisecant, 310, 355
Module Mumford, David, 140, 249, 291, 324, 336,
flat, 253 347, 356, 366, 379,417,419-421, 431,437

489
Index

Nagata, Masayoshi, 58, 87, 105, 108, 168, 268, 424. See also Regular scheme;
381, 383, 384, 394, 401, 409, 419 Smooth morphism
Nakai-Moishezon criterion, 356, 365, 382, hyperplane section of. See Bertini's
405, 434 theorem
Nakai, Yoshikazu, 144, 366, 434 infinitesimal lifting property, 188
Nakano, Shigeo, 249, 445 fl is locally free, 177, 178, 276
Nakayama's Lemma, 125, 153, 175, 178, 288 Nonspecial divisor, 296, 343, 349
Narasimhan, M.S., 379 Norm (of a field extension), 46
Natural isomorphism, 240 Normal. See also Projectively normal
Negative definite, 368, 419 bundle. See Normal, sheaf
Neron, Andre, 367 crossings, divisor with, 391
Neron-Severi group, 140, 367, 418 cuspidal cubic curve is not, 23
is finitely generated, 447 point, 23, 38
Neron-Severi theorem, 364, 367, 368 quadric surface is, 23, 147
Nilpotent element. See also Dual numbers, ring, 185, 264
ring of ~ R 1 +S2, 185
in a ring, 79-81 scheme, 91, 126, 130, 244, 280
in a scheme, 79, 85, 190, 259, 277, 315 sheaf, X y /X' 182, 361, 386, 431, 433, 436.
Nilradical of a ring, 82 See also Conormal sheaf
Nodal cubic curve, 259, 263, 276 variety, 23, 263, 410
blown up, 29 (Fig) Normalization, 23, 91, 426
divisor class group of, 148 of a curve, 148, 232, 258, 298, 343, 365,
is rational, 30 382
Node, 36 (Fig), 37, 258, 293, 298, 392, 394. Number theory, 58, 316, 451, 452
See also Nodal cubic curve Numerical equivalence, 364, 367, 369, 435
analytic isomorphism of, 34, 38 Numerical invariant, 56, 256, 361, 372, 379,
plane curve with, 310-316 425, 433
Noetherian formal scheme, 194 Numerical polynomial, 49
Noetherian hypotheses, 100, 194, 201,
213-215, 218 Olson, L., 140
Noetherian induction, 93, 94, 214 Open affine subset, 25, 106
Noetherian ring, 80 Open immersion, 85
Noetherian scheme, 83 Open set X1, 81, 118, 151
Noetherian topological space, 5, 8, II, 80, Open subscheme, 79, 85
83. See also Zariski space Ordinary double point. See Node
Noether, Max, 349 Ordinary inflection point, 305. See also
Nonalgebraic complex manifold, 444 (Fig) Inflection point
Nonhyperelliptic curve, 340. See also Ordinary r-fold point, 38, 305
H yperelliptic curve Osculating hyperplane, 337
existence of, 342, 345, 385
Nonprojective scheme, 232 p-Adic analysis, 451
Nonprojective variety, 171, 443 (Fig) p-Adic cohomology, 452
Nonsingular cubic curve Parameter space, 56. See also Variety, of
ample sheaves on, 156 moduli
canonical sheaf of, 183 Parametric representation, 7, 8, 22, 260
divisor class group of, 139 Paris seminar, 436, 449
group law on, 139 (Fig), 147, 297, 417 Pascal's theorem, 407 (Fig)
has 9 inflection points, 305, 322 Pencil, 280, 422
is not rational, 46, 139, 183, 230 Perfect field, 27, 93, 187
Nonsingular curve, 39-47, 136 Period mapping, 445
abstract. See Abstract nonsingular curve Period parallelogram, 326 (Fig)
divisors on, 129 Petri, K., 348
existence of, 37, 231, 352, 406 Picard group, Pic X, 57, 143, 232, 250, 357,
Grothendieck group of, 149 428. See also Divisor class group
morphism of, 137 as H 1(X, £!! 1), 143, 224, 367, 446
projective ~ complete, 136 Lefschetz theorem, 190
Nonsingular in codimension one, 130 of a blowing-up, 188
Nonsingular points, 32, 37 of a cubic surface in P 3, 401
form an open subset, 33, 178, 187 of a family, 323
Nonsingular variety, 31-39, 130, 177-180, of a formal scheme, 200, 281

490
 Index

of a line with point doubled, 169 from a point, 22, 151, 259, 309-316
of a monoidal transformation, 386 morphism, 87
of an elliptic curve, 297, 323 of a twisted cubic curve, 22
of a nonprojective variety, 171 Projective closure, 12
of a P(<n, 170, 291 Projective cone. See Cone
of a product, 292, 338, 367 Projective dimension, 237
of a projective variety over C, 447 Projective general linear group, PGL, 46,
of a ruled surface, 370 151, 273, 347
of a singular curve, 148 Projectively normal, 23, 126, 147, 159, 188,267
torsion elements of, 306 canonical curve is, 348
Picard variety, 140, 447 curve, 23, 231, 315, 336, 354, 355
Pinch point, 36 (Fig) d-uple embedding is, 126
Plane curve, 7, 35-39, 304, 305, 319,407 Projective module, 238, 284
birational to a curve with ordinary Projective morphism, 103, 107, 123, 149-172,
singularities, 407 277, 281, 290
with nodes, 310-316, 337, 348 Projective n-space, pn, 8, 10, 77, 103, 120,
p-Linear map, 332 151, 155, 225-230
Plurigenus, Pn, 190, 422 Projective object (in a category), 205, 237
Poincare duality, 453, 456 Projective resolution, 205, 236
Point Projective scheme, 103, 120, 121, 232
closed, 74, 75 (Fig), 81, 93 Projective space bundle, 170, 171, 186
embedded, 85, 259 associated to a locally free sheaf, P( lf ),
generic. See Generic point 162-169
open, 74 canonical sheaf of, 253
rational over k. See Rational points Chow ring of, 429
Pole of a rational function, 130, 131, 297 cohomology of, 253
Polynomial equations, 55 Picard group of, 170, 291
over finite fields, 449 ruled surface as, 370
Presheaf, 61, 62, 109, 193 Projective variety, 8-14
Prime divisor, 130 Proper intersection, 427
Prime ideal, 4, II, 22, 70, 76, 132 Proper morphism, 95-108, 161, 252, 279
Primes in an arithmetic progression, 335 is closed, I00, 152
Primitive element, theorem of, 27 Purely inseparable, 302, 305, 385
Principal divisor, 131, 141, 367 Pure transcendental field extension, 303
has degree zero, 132, 138
Principal ideal domain, 254 Quadratic number field, 330-332, 334, 340
Product. See also Direct product Quadratic transformation, 30, 31, 397, 398
arithmetic genus of, 54 (Fig), 403, 406, 408, 409. See also
canonical sheaf of, 187 Birational transformation
Cartesian, of graded rings, 125 Quadric cone, 23, 133, 134 (Fig), 142, 346,
differentials on, 187 352, 428
fibred, 87, 100 curves on, 384
in a category, 22, 66, 87 Quadric hypersurface, 38, 55
morphism, 99 divisor class group of, 147
of curves, 44, 338, 367 Quadric surface, 13, 14 (Fig), 23, 30
of schemes, 87 ample sheaves on, 156, 366
of varieties, 22 as a ruled surface, 369, 381
Picard group of, 292, 338, 367 canonical sheaf of, 183
Segre embedding of. See Segre embedding cohomology of, 231
topological space of, 91 curves on, 231,294,346,351,352,354,
Zariski topology on, 7 362, 381
Proj, 76, 77 divisor class group of, 133, 135
closed subschemes of, 92, 119, 125 intersection theory on, 361, 364
of a graded homomorphism, 80, 92 2-uple embedding of, 401
sheaves of modules on, 116-123, 125 Quartic curve. See also Curve, of degree 4
Proj, 160-169 elliptic, 38, 315, 353
Projection has 28 bitangents, 305
birational, 30, 31, 310 rational. See Rational quartic curve
formula, 124, 253, 380, 419, 426, 431 Quartic threefold is not rational, 184
from a linear space, 169 Quasi-affine variety, 3, 21, 223

491
Index

Quasi-coherent sheaf, 111-115, 126. See also Refinement of a covering, 223

Coherent sheaf Regular function, 15, 21, 23, 30, 42, 62, 70
Quasi-compact morphism, 91 Regular in codimension one, 130, 185
Quasi-compact scheme, 80, 119 Regular local ring, 32, 38, 40, 130, 136, 237,
Quasi-compact topological space, 8, 80 238, 243, 415
Quasi-finite morphism, 91, 280, 366 complete, 34
Quasi-projective morphism, 103 is Cohen-Macaulay, 184
Quasi-projective variety, I 0 is UFD, 142
Quotient sheaf, 65, 109, 141 localization of is regular, 177
Quotient topology, 75 ~ 0 is free, 174, 187
Regular q-forms, sheaf of, Oi;k• 190, 247,
Radical ideal, 4 249
Radical of an ideal, 3 Regular scheme, 142, 170, 179,238, 268, 275.
Ramanujam, C. P., 249 See also Nonsingular variety; Smooth
Ramanujan T-function, 452 morphism
Ramification divisor, 301, 382 Regular sequence, 184, 217, 245, 261
Ramification index, 299 Relative cohomology, 250
Ramification point, 299 Relative differentials. See Differentials
Ramified morphism, 299, 312 Relative dimension, 268
Ramis, J. P., 249 Relatively minimal model. See Minimal
Rank, 109, 148 model
Rational curve, 46, 55, 138, 143, 297, 315, Relative projective space, 162
343, 345. See also Rational quartic Relative tangent sheaf, 436
curve; Twisted cubic curve Remmert, Reinbold, 442
elliptic curve is not, 46, 139, 183 Repartitions, 248
Rational equivalence of cycles, 425, 426, Representable functor, 241, 324
436, 454 Residue, 239, 247, 248
Rational function, 16, 30, 46, 297, 419 Residue field, 22, 80
divisor of, 130, 131, 294 Resolution
Rationality of plane involutions, 422 acyclic, 205
Rationality of the zeta function, 449, 450, by Cech c0mplex, 220
452 fine, 201
Rational map, 24-31, 168 flasque, 201, 208, 212, 248
Rational normal curve, 315, 343 injective, 204, 242
Rational numbers, Q, 58, 148, 335 locally free, 149, 234, 239
Rational points, 80 projective, 205, 236
on an elliptic curve, 317, 335, 336 (fig), Resolution of singularities, 28, 55, 168, 386,
339 390, 391, 413, 420, 442, 453
over a finite field, 368, 449 embedded, 390, 391, 392 (Fig), 419
Rational quartic curve, 23, 159, 231,291, Restriction
315, 353, 386 of a sheaf to a subset, §'lz, 65, 112
Rational ruled surface, 369, 379-381. See of sections of a sheaf, 61, 118
also Ruled surface Riemann, Bernhard, 446
blow-up of P 2, 375, 381, 397 Riemann existence theorem, 442
classification of, 376, 419 Riemann hypothesis, 339, 368, 449-452, 457,
embedded in pn, 381, 385, 437 458
of degree 3 in P 4 , 170,348,381, 397,437 Riemann-Roch problem, 170, 296, 424
Rational surface, 56, 170, 422. See also Riemann-Roch theorem
Rational ruled surface; Veronese generalized, 57, 149, 170, 363, 424,
surface 431-436
blow-up of P 2 , 375, 381, 395, 397,400, on a curve, 156, 249, 293-299, 317, 325,
406, 408 341, 343, 349, 362, 376, 424, 432
Rational variety, 30, 183, 184 on a singular curve, 298
Raynaud, Michel, 249 on a surface, 356, 357, 362-364, 366, 423,
Real numbers, R, 4, 8, 80, 106 424,433
Reduced induced structure, 86, 92 on a 3-fold, 437
Reduced inverse image divisor, 391 Riemann surface, 338, 441, 447
Reduced scheme, 79, 82, 365 Riemann zeta function, 457
Reduction mod p, 89, 334, 340, 451 Riemenschneider, Oswald, 249, 445
Reduction to the diagonal, 427, 428 Right derived functor, 204, 207, 233, 250

492
 Index

Right exact functor, 204, 286 geometrically irreducible, 93

Rigid algebra, 267 geometrically reduced, 93
Rigid scheme, 267 glueing of, 75, 80, 91, 171, 439, 444
Ring integral, 82, 91
graded. See Graded ring irreducible, 82
local. See Local ring locally factorial, 141, 145, 148, 238
of dual numbers, 80, 265, 267, 324 locally noetherian, 83
of endomorphisrns of an elliptic curve, noetherian, 83
323, 329, 330, 338, 340 nonprojective, 171, 232, 443 (Fig)
of integers in a number field, 330, 331 nonseparated, 76, 96
of regular functions, 16-18, 21 nonsingular in codimension one, 130
spectrum of, 70-75, 110-113 normal. See Normal
Ringed space, 72, 143. See also Locally of finite type over a field, 93
ringed space over another scheme, 78
sheaf of modules on, 109, 110, 123, 124, over R, 106
127, 233 reduced, 79, 82, 365
Roberts, Joel, 427 regular. See Regular scheme
Root systems, 408 regular in codimension one, 130, 185
Roquette, P., 306 separated. See Separated scheme
Rossi, Hugo, 69, 201, 438 Scheme-theoretic closure, 258
Ruget, G., 249 Scheme-theoretic image, 92
Ruled surface, 56, 170, 253, 356, 369-385, Scheme-theoretic intersection, 171, 358
373 (Fig), 422. See also Rational ruled Schlafli's double-six, 403
surface Schlessinger, Michael, 185, 266, 267
ample divisors on, 380, 382, 383 Schwarz's inequality, 402
arithmetic genus of, 371 Scott, D. B., 431
birationally, 370, 419 Scroll, 381, 385
blown-up cone, 374, 381 Secant line, 309, 316
canonical divisor of, 373 not a multisecant, 314, 316, 351
elementary transformation of, 416 (Fig), with coplanar tangents, 310
419 Secant variety, 310, 316
elliptic, 369, 375, 377, 384, 385, 440 Section. See also Global sections
embedded in P", 381, 385 discontinuous, sheaf of, 67
geometric genus of, 371 of a morphism, 129, 170, 369
intersection theory on, 370 of a presheaf, 61
invariant e, 372, 374-377, 384 Segre embedding, 13, 22, 54, 107, 125, 126,
invariant K 2, 374 156
nonsingular curves on, 380 Self-intersection, 360, 386, 40 I, 437, 450
normalized, 373 formula, 431, 433
Picard group of, 370 of the canonical divisor, K 2 , 361, 367, 374,
sections of, 371, 373 (Fig), 379, 383-385 387, 421
stable, 379 Semicontinuity theorem, 281-292, 354, 369
very ample divisors on, 379, 380, 383, 385 Semi-linear automorphism, 106
zero-cycles on, 437 Semistable, 384
Ruling on a quadric cone, 133, 134 (Fig), Separable field extension, 27, 300, 422
428 Separable morphism, 300
Separably generated field extension, 27, 174,
Saint-Donat, Bernard, 344, 348 187,271
Sally, Judith, 412 Separated morphism, 95-108
Samuel, Pierre, 312, 394, 420, 428 Separated scheme, 76, 96, 106, 119, 130, 169
Satellite functor, 206 Separate points, 152, 158, 308, 380
Saturation of a homogeneous ideal, 125 Separate tangent vectors, 152, 158, 308, 380
Scheme, 59, 69-82 Separating transcendence base, 27
affine. See Affine scheme Serre duality, 181, 239-249, 293-295, 362,
associated complex analytic space. See 363, 367, 419, 424
Complex analytic space Serre, Jean-Pierre, 51, 57, 121, 149, 154, 185,
associated to a variety, 78, 84, 104, 136 201,211,215,228,239,247,249,297,
connected, 82. See also Connected 307,330,331,335,366,422,427,428,
formal. See Formal scheme 432, 435, 440, 446, 447, 451, 452
geometrically integral, 93 Seshadri, C. S., 379

493
Index

Set-theoretic complete intersection, 14, 224 Spectrum of a ring, Spec A, 70-75, 82

Severi, Francesco,47,244, 314,316,368 sheaf M on, 110-113
SGA, 436, 449, 462 Spectrum of a sheaf of algebras, Spec, 128,
Shafarevich, I. R., 356, 369, 383,410,421 280, 306
Shannon, David L., 412 Spencer, D. C., 414
Sheaf, 60-69 Splitting principle, 430
associated to a presheaf, 64, 250 Square-free polynomial, 147
coherent. See Coherent sheaf Stable image, 191
constant. See Constant sheaf Stable locally free sheaf, 379, 384
flasque. See Flasque sheaf Stable ruled surface, 379
free, 109 Stable under base extension, 90, 93, 99, 102,
generated by global sections, 121, 107
150-156, 307, 358, 365 Stable under generization, 94, 266
glueing, 69, 175 Stable under specialization, 94, 98
Horn, 67, 109 Stalk, 62, 71, 76, 124
inverse image of, 65, 110, 115, 128, 299 Stanasila, Octavian, 439
invertible. See Invertible sheaf Standard conjectures of Grothendieck, 435
locally free. See Locally free sheaf Steiner prize, 349
M associated to a module, 110-1 13, 116 Stein factorization, 276, 280, 366
of differentials. See Differentials, sheaf Stein space, 216, 440
of Stepanov, S. A., 451
of discontinuous sections, 67 Stereographic projection, 420
of graded algebras, 160 Stichtenoth, H., 306
of ideals. See Ideal sheaf Strange curve, 311, 316
of modules, 108-129, 233 Strict transform, 30, 31, 165, 171, 381, 388,
of (!) y -algebras, 128 390, 406. See also Blowing up
of rings, 62, 70 Strong Lefschetz theorem, 452
of total quotient rings, 141 Structure sheaf, 74, Ill
quasi-coherent. See Quasi-coherent sheaf Subrnodule with supports f.,(M), 124,213
Sh-ioda, T., 422 Subscherne
Siegel, Carl Ludwig, 442 closed. See Closed subscherne
o- Process, 3 86 open, 79, 85
Signature, 364 Subsheaf, 64
Simple group with supports£'~ (Y), 68
of order 60, 420 Subvariety, 21, 425
of order 168, 349 cohomology class of, 249
of order 25920, 405 local ring of, 22, 58
Simply connected scheme, 303, 304, 340 Suorninen, Kalevi, 249
Singh, Balwant, 306 Superabundance, 362
Singular locus, 390 Supersingular elliptic curve, 332
Singular point, 32, 33, 35, 36 (Fig) Support, 67, 124, 212, 215, 217
analytically isomorphic, 34, 38, 298, 393 Surface, 4, 105
moving in a linear system, 276 algebraic equivalence of divisors, 367, 369
multiplicity of. See Multiplicity complete intersection, 409, 423, 437
of a cone, blowing up, 37, 374, 381 cubic. See Cubic surface in P3
of a curve on a surface, 386 definition for Ch. II, 105
of a surface, 36 (Fig), 420 definition for Ch. V, 357
resolution of. See Resolution of singulari- Del Pezzo, 401, 408
ties divisor on, 135, 357
Skew commutative graded algebra, 127 elliptic, 422
Skyscraper sheaf, 68, 296 elliptic ruled. See Elliptic ruled surface
Slope, 30 Enriques, 422
Smooth morphism, 268-276, 303 factorization of birational transformations,
Space curves, 349 386, 409, 411-413, 416
Special divisor, 296 geometry on, 357-368
existence of, 345 group Nurn X, 364, 367, 368
Specialization, 93, 97, 354 in P 4, 170, 348, 381, 397,433, 434,437
Special linear system invariant K 2 , 361, 367, 374, 387, 421
dimension of. See Clifford's theorem K3, 184, 422, 423, 437, 452
Spectral sequence, 236, 252, 289 lines on, 13, 136, 367, 402-406, 408

494
 Index

minimal model of. See Minimal model Tor group, 427, 428
of degree din pd, 401 Torsion sheaf, 149, 372
of degree d in pd+ 1, 381 Torus, 328, 442, 447
of degree 3 in P\ 170,348, 381, 397,437 Total Chern class, 429
of general type, 184, 422, 423 Total quotient ring, 141
product of two curves, 338, 367, 368 Total transform, 410
P2 with I point blown up, 375, 381, 397 Trace map, 241, 247, 249, 307, 453, 455
P2 with 2 points blown up, 406 Trace of a linear system, 158
P2 with 3 points blown up, 397 Transcendence base, 27
P2 with 5 points blown up, 408 Transcendence degree, 6, 27, 421
P2 with 6 points blown up, 395, 400 Transcendental methods, 289, 326, 367,
quadric. See Quadric surface 438-448
rational. See Rational ruled surface; Transversal intersection, 357-360, 392, 425,
Rational surface 427
ruled. See Ruled surface Triangulable, 447
valuation rings of, 108, 420 Trigonal curve, 345
Veronese, 13, 170, 316, 381 Triple point, 36 (Fig)
with infinitely many exceptional curves, Trisecant, 346, 348
409, 418 Trotter, H., 334
Sylvester, J. J., 364 Tsen's theorem, 369
Symmetric algebra of a module, 127, 162, 374 Twenty-seven lines, 402-406
Symmetric function, 431 Twisted cubic curve, 7, 12-14, 159, 315, 353
Symmetric group, 304, 318, 408 conormal sheaf of, 385
Symmetric product of a module, 127, 185 not a complete intersection, 14, 136
projection of, 22, 259
Tacnode, 36 (Fig), 37, 39, 305, 348 Twisted module, M(n), 50
Tame ramification, 299 Twisted quartic curve. See Rational quartic
Tangent bundle, 57, 450. See also Tangent curve
sheaf Twisted sheaf, $'(n), 117
Tangent direction, 36 cohomology vanishes, 228
Tangent line, 54, 139, 148, 304, 309, 337 generated by global sections, 121
Tangent sheaf, Yx, 180, 182, 225, 265, 363, Twisting sheaf, IV(l), 117, 120, 225
425, 431, 437 generated by global sections, 150
Tangent variety, 310 generates Pic P", 145
Tate, John T., 247, 317, 335, 451, 452 not very ample, 171
Tensor algebra, 127 on P(<f), 162
Tensor operations, 127 on Proj .9', 160
Tensor product, ®, 22, 87, 109, 127, 153, 430 Type (a, b) of a divisor on a quadric surface,
Thickened fibre, 277 135
Three-fold, 105, 184, 437
Tjurin, Andrei, 383 Unassigned base point of a linear system, 395
Todd class, 432 Uncountable field, 409, 417
Topological covering space, 442 Underlying topological space, 74
Topological space Uniformizing parameter, 258. See also Local
axiom T0 , 93 parameter
dimension of, 5. See also Dimension Unique factorization domain, UFD, 7, 46,
disconnected, 82 131, 141, 147, 420
generic point of, 74, 75 (Fig), 80, 93 Union of two planes, 224, 266, 267, 428
irreducible, 3 Universal 11-functor, 205, 238, 243
noetherian. See Noetherian topological Universally closed morphism, 100
space Universal parameter space, 323, 347. See
quasi-compact, 8, 80 also Variety, of moduli
set of irreducible closed subsets of, 78 Unmixedness theorem, 188
underlying a scheme, 74 Unraplified morphism, 275, 299. See also
Topology, Etale morphism
etale, 452, 453 -Uple embedding, 13, 21, 54, 155, 156, 159,
Hausdorff, 2, 8, 95, 439 167, 183, 291, 343, 385
of algebraic varieties over C, 421 is projectively normal, 126, 315
quotient, 75 Upper semi-continuous function, 125, 287,
Zariski, 2, 5, 7, 10, 14, 70 291

495
Index

Valuation, 39, 43, 299 Very flat family, 266

center of, 106 Virtual arithmetic genus, 367
of a prime divisor, 130, 135 Vitushkin, A. G., 23
Valuation ring, 40, 97, 101, 105. See also
Discrete valuation ring Walker, R. J., 391, 393
examples of, 108, 420 Weierstrass ~-function, 327
nondiscrete, 108 Wei!, Andre, 48, 58, 105, 248, 368, 428, 435,
Valuative criterion of properness, 10 I, 107, 445, 449-452, 454
181, 259, 325, 410 Weil cohomology, 453
Valuative criterion of separatedness, 97, 107 Wei! conjectures, 449-458
Vanishing theorem cohomological interpretation of, 454-457
of Grothendieck, 208 Wei! divisor, 130-136, 294,426,428,436.
of Kodaira, 248, 249, 408, 423, 424, 445 See also Divisor
of Nakano, 445 Weyl group, 405, 408
of Serre, 228, 424 Weyl, Hermann, 441
Variety Wild ramification, 299
abelian. See Abelian variety Witt vectors, 452
abstract, 58, 105
affine. See Affine variety
algebraic family of. See Family Zariski, Oscar, 32, 57, 58, 60, 105, 170, 190,
complete. See Complete variety 244,276,356,391,393,394,410,419,
isomorphisms of, 16, 18 422
morphisms of, 14-23 Zariski's Main Theorem, 57, 276, 280, 410,
nonprojective, 171, 443 (Fig) 412
normal. See Normal Zariski space, 93, 94, 213
of moduli, 56, 58, 266, 317, 346, 421 Zariski tangent space, 37, 80, !58, 265, 270,
over k, 15, 78 324
projective. See Projective variety Zariski topology, 2, 5, 7, 10, 14, 70
projectively normal. See Projectively base of open affine subsets, 25
normal is not Hausdorff, 2, 95
quasi-affine, 3, 21, 223 weaker than usual topology, 439
rational. See Rational variety Zeros
scheme associated to, 78, 84, 104 common, 35
Vector bundle, 128, 170. See also Locally of a polynomial, 2, 9
free sheaf of a rational function, 130, 131
sheaf of sections of, .9"(X j Y), 129 of a section of a locally free sheaf, 157,
Verdier, Jean-Louis, 249 431
Veronese surface, 13, 170, 316, 381 Zeta function, 449-452, 455-458
Vertex of a cone, 37, 394. See also Cone functional equation of, 450
Very ample divisors, 307, 308, 358 of a curve, 458
form a cone, 359 of an elliptic curve, 458
Very ample invertible sheaf, 120, 126, of P 1, 450, 451
153-156, 228, 236 of P", 457
on a curve, 307 of Riemann, 457
on Proj [/, 161 rationality of, 449, 450, 452

496