Commutative algebra

Andrea Ferretti

July 24, 2020

AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 CONTENTS

introduction vii
1 basics 1
1.1 Rings and ideals 1
1.2 Quotients 11
1.3 Modules 15
1.4 More constructions with modules 24
1.5 Euclidean rings 30
1.6 Localization 35
1.7 Graded rings and modules 43
1.8 Exercises 47
2 finiteness conditions 53
2.1 Principal ideal domains 53
2.2 Artinian and Noetherian modules 56
2.3 Noetherian rings 61
2.4 Artinian rings 65
2.5 Length 69
2.6 Exercises 73
3 factorization 79
3.1 Unique factorization domains 79
3.2 Primary decomposition 87
3.3 Primary decomposition for modules 99
3.4 Factorization in Dedekind rings 105
3.5 Modules over Dedekind rings 110
3.6 Exercises 115
4 computational methods 119
4.1 The resultant 120
4.2 Discriminants 126
4.3 Gröbner bases 131
4.4 More algorithmic operations 140
4.5 Exercises 144
5 integral dependence 149
5.1 Integral extensions 149
5.2 Going up and down 157
5.3 Noether normalization 162

iii
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 Contents

5.4 Integral extensions of Dedekind rings 164

5.5 Exercises 168
6 lattice methods 173
6.1 Additive structure of number rings 174
6.2 Prime extensions in number rings 179
6.3 Prime extensions in Dedekind rings 184
6.4 Galois extensions of Dedekind rings 186
6.5 Discriminant and ramification 191
6.6 Computing prime factorizations 193
6.7 Geometry of ideal lattices 196
6.8 Cyclotomic rings 203
6.9 Exercises 209
7 metric and topological methods 215
7.1 Absolute values 216
7.2 Valuations and valuation rings 227
7.3 Discrete valuation rings 232
7.4 Direct and inverse limits 234
7.5 Completion of rings and modules 239
7.6 Hensel’s lemma 247
7.7 Witt vectors 250
7.8 Exercises 263
8 geometric dictionary 269
8.1 Affine varieties 270
8.2 The Nullstellensatz 272
8.3 The Ax-Grothendieck theorem 276
8.4 Morphisms 277
8.5 Local rings and completions revisited 279
8.6 Graded rings and projective varieties 280
8.7 A new idea: the dimension 284
8.8 The Zariski tangent space 288
8.9 Curves and Dedekind rings 293
8.10 Exercises 295
9 dimension theory 303
9.1 Dimension of rings and modules 304
9.2 Hilbert functions 306
9.3 The main theorem on dimension 310
9.4 Height 315
9.5 Properties of dimension 317
9.6 Dimension of graded rings 321

iv
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 Contents

9.7 Exercises 324

10 local structure 329
10.1 Regular rings 330
10.2 Multiplicity and degree 336
10.3 Formulas for multiplicity 343
10.4 Multiplicity and valuations 351
10.5 Superficial elements 357
10.6 Cohen’s structure theorem 363
10.7 Exercises 371
a fields 375
a.1 Algebraic elements 376
a.2 Finite fields 381
a.3 Separability 383
a.4 Normal extensions 394
a.5 The Galois correspondence 397
a.6 Some computations 403
a.7 The trace and norm 405
a.8 Abelian extensions 408
a.9 Exercises 418
Bibliography 425
Index of Notation 435
Index 439

v
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 INTRODUCTION

I have learned the basics of commutative algebra from the famous

book by Atiyah and MacDonald [AM69]. The present text is born
from the intention to expand the material therein, and give an
alternative organization, although at this point it has grown into
its own thing.
In writing this book, I have tried to follow a few principles.
First, I have tried not to skim on the basics too rapidly. While it
is true that some topics – such as Euclidean domains or unique
factorization – are often encountered in a first algebra course, I
feel that they belong in a basic text on commutative rings.
Second, and more important, I have tried to present the con-
nections with the most important applications of commutative
algebra, such as number theory, algebraic geometry and compu-
tational algebra. This approach makes for a less terse style, but I
hope that this is repaid by a greater sense of unity.
Commutative algebra is at the crossroad between many fer-
tile areas of mathematics, and I hope that this book conveys the
various points of view appropriately. The geometric interpreta-
tion often clarifies the sometimes obscure meaning of a result in
commutative algebra. In the present book I have not relied on
previous knowledge on the topic, but I have opted for a chapter
that translates the algebra in geometric terms. Number theory
is the other root of commutative algebra, and a constant source
of inspiration. Many important topics, such as completion, can
be better appreciated by first learning a special case of arithmetic
relevance – in this case, the construction of the field of p-adic
numbers.
Computational algebra has clearly seen an explosion since the
advent of computers. Chapter 4 is dedicated to it, but other com-
putational topics are scattered in the text, such as the algorithm
to compute Smith’s normal form in Chapter 1 or the LLL algo-
rithm in the exercises for Chapter 6. Many other topics could be
mentioned, but I could only make some small connections with

vii
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 introduction

model theory and invariant theory in the exercises. For the for-
mer, see [Sch99], while a good introduction to invariant theory is
[Dol03].
Unfortunately, many topics are not covered here, in particular
those that depend on homological algebra techniques. Among
them, we can mention flatness, spectral sequences, the study
of the Koszul complex and regular sequences, Cohen-Macaulay
rings, Gorenstein rings and duality theory. I plan to introduce
these topics in a sequel, tentatively called (without much fantasy)
“Homological methods in commutative algebra”.
I should remark that this is an introductory text, so I didn’t
even try to cover the material of more advanced books, such as
Mastumura’s books [Mat70] and [Mat86], or Eisenbud [Eis95]. To
give an idea, the latter was created as a reference for the famous
algebraic geometry book [Har77], and ended up being far thicker
than it. I advise the reader to consult [Eis95] for further reading,
and remark that it is actually quite enjoyable despite its appear-
ance; my scope here is far more limited than Eisenbud’s.

Here is a brief description of the contents of the book; see the

beginning of the various chapters for more details on the topics
covered therein.
In the first chapter, we introduce the basics notions of commu-
tative algebra, like rings, ideals and modules. Moreover, we treat
the two basic constructions that we will use throughout the book,
that is, quotients and localization. A section about Euclidean
rings is also present, to show an example of rings where the the-
ory is particularly simple. These topics will probably be familiar
from a first algebra course, and the knowledgeable reader can
just have a quick glance, as we do nothing fancy in this chapter.
The core of the book starts with Chapter 2, where we introduce
some particular finiteness conditions on our rings and modules.
These are strong enough to produce a lot of result, but no so
stringent, so that most of the objects that we will encounter will
satisfy these hypothesis. In particular most of the subsequent
results in the book will work for the class of Noetherian rings.
Chapter 3 is about factorization. We start from the simple case
of rings that admit a unique factorization for elements. This is
a rather special class of rings, hence we then present the theory
of primary decomposition of Lasker, which is a generalization

viii
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 introduction

that works over arbitrary Noetherian rings and modules. We

then specialize this case to study the important case of Dedekind
domains, which have a theory of prime factorization for ideals.
In Chapter 4, we focus on the case of polynomial rings, where
some explicit computational techniques are available. In partic-
ular, we introduce the resultant and use it to study the problem
of elimination, which is about solving polynomial systems in an
inductive way. We also use resultants to introduce discriminants
and show their basic properties. Later in the chapter, we intro-
duce Gröbner bases, which are special sets of generators of an
ideal in a polynomial ring. Again, after having proved their ba-
sic properties, we use them to study the problem of elimination
from a different angle.
Chapter 5 introduces the concept of integral elements. These
are the analogue of algebraic elements from field theory in the
setting of rings. In fact, much of the theory just mimics what one
does for fields. The parallel of an algebraic extension of fields is
an integral extension of rings. For such an extension A ⊂ B, we
present the Cohen-Seidenberg theory that relates prime ideals
in A and B in a precise way. We also use these results to give
another, more traditional, characterization of Dedekind rings.
Dedekind rings are studied in much more detail in Chapter 6.
In particular, we study the properties of factorization of ideals
and what happens in integral extensions of Dedekind rings. An
important special case here is the class of number rings, which
are obtained by taking elements integral over Z in a finite exten-
sion of Q. Such rings have a natural embedding in Rn that rep-
resents them as lattices. One can then use geometric techniques
to derive important bounds on the index of ideals. In particular,
we use these techniques to prove two important finiteness results
(finiteness of the class group of ideals and finite generation of the
unit group).
Chapter 7 introduces topological methods. In particular, we de-
fine absolute values over fields, which one can use to define the
completion of rings. This construction is analogue to the Cauchy
completion in real analysis. In particular, this allows us to define
p-adic numbers. As a special case, nonarchimedean absolute val-
ues can be obtained from valuations, which we study next. In
the last part of the Chapter, we study the problem of completion
from a more algebraic point of view. To do this, we introduce

ix
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 introduction

the machinery of direct and inverse limits. The algebraic point

of view on completion allows us to generalize the notion and
extend it to modules, as well as to prove the fundamental Artin-
Rees lemma, which relates the topology of a module and that of
its submodules.
In Chapter 8, while proving few new results, we give the basic
definitions about algebraic geometry (affine and projective vari-
eties, Zariski topology and so on) and show how most of the ma-
terial covered so far can be used to quickly obtain information
about these new objects. We have decided to put this chapter al-
most at the end of the book, so that most of the material can be
read independently of it. Still one can start reading this chapter
even at the beginning, and follow the dictionary while learning
new algebraic concepts. The main result from commutative alge-
bra that we introduce in this chapter is Hilbert’s Nullstellensatz,
which gives a deep link between points of an algebraic variety
and maximal ideals of its coordinate ring.
A note for the savvy reader: our treatment of algebraic geom-
etry is as elementary as possible; in particular we do not even
mention the machinery of schemes. We just want to show the
following principle: one can apply results from commutative al-
gebra to some simple rings (usually finitely generated reduced
k-algebras) to obtain results which are geometric in nature, and
conversely the geometry can suggest that a result may be true for
some special rings, and often the algebraic result can be proved
for a much more general class of rings.
The next chapter is about dimension theory. The concept of
dimension is introduced in Chapter 8 from a geometric point of
view, but Chapter 9 gives a wide algebraic generalization. We
show that a Noetherian local ring has a definite notion of dimen-
sion, the main result being that all reasonable definitions sug-
gested by the geometric intuition are actually the same. In most
of the chapter we work with local or graded rings, and the paral-
lelism here is very strict. The main technical tool for this theory
is the Hilbert polynomial, which estimates the order of growth
of the size of homogeneous components in a graded module, so
we begin the chapter by proving its existence and studying its
properties.
In Chapter 10 we define regular local rings, which correspond
to nonsingular points on a variety. We also study the nonregu-

x
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 introduction

lar case by introducing the concept of multiplicity of a local ring,

which is a simple measure of singularity. As it turns out, this is
related to the concept of degree of a graded ring. In the geomet-
ric case, where a graded ring corresponds to a projective variety,
the degree expresses the number of points of intersection with a
general linear space of complementary dimension. This chapter
develops the theory of multiplicity, and in doing so describes in
some detail the structure of local rings, culminating in the cele-
brated theorem of Cohen.
The Appendix consists of an exposition of the classical Galois
and Kummer theory of field extensions. It aims to give back-
ground for the field theoretic results used in the rest of the book
(in particular, Galois theory of finite extensions and the notion
of separability) but covers more ground than it is strictly needed.
In fact, it can be read independently of the rest of the book as
a short introduction to field theory. While fields are in many re-
spects simpler than general rings, they also present many new
phenomena, and a familiarity with fields is certainly part of the
study of commutative algebra in a broad sense.

The prerequisites for reading the book are not many. We as-
sume that the reader is more or less familiar with algebraic ob-
jects, and some acquaintance with linear algebra is assumed – for
instance, it is useful if the reader has some familiarity with the
tensor product construction in the context of vector spaces. Fi-
nally, from Chapter 7 we make some use of elementary topology
and some notions about metric spaces.
The book is suitable for a semester on algebra at the introduc-
tory graduate level. It could also be used to support a shorter
course on algebraic number theory, introducing global and local
fields, Dedekind rings with their factorization theory and com-
pletions.
To help the reader orient themselves, we suggest some possible
paths through the book, other than reading it cover to cover.
A standard introduction to commutative algebra, along the
lines of [AM69], would start from the basics in Chapters 1 to
3, then go through integral extensions in Chapter 5, topological
methods in Chapter 7 (going quickly over Sections 7.1 and 7.2),
prove the Nullstellensatz in Sections 8.1 and 8.2, then introduce

xi
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 introduction

dimension theory in Chapter 9 and at least the basics on regular

rings in Section 10.1.
An introduction to algebraic number theory could cover factor-
ization in Chapter 3, the basics of field theory from Appendix A,
the first half of Chapter 4 to introduce discriminants, and parts
of Chapter 5 to characterize Dedekind rings; then Chapter 6 cov-
ers the global theory and Chapter 7 the local theory, especially
Sections 7.1, 7.2 and 7.3.
The reader that wants a quick introduction to the methods of
computational algebra can just follow Chapter 2, Section 3.1 and
Chapter 4, although we advise to complement this with other
texts on the matter, such as [KR00]. In a similar way, Appendix A
could be used, together with other material, in a minicourse on
Galois theory.
Finally, a geometrically minded reader could learn the basics
in Chapters 1 and 2, then go through Chapter 8 and learn the nec-
essary commutative algebra along the way. Section 8.5 requires
learning about completions in Sections 7.4 and 7.5, while the de-
composition in irreducible components of Section 8.7 makes use
of the primary decomposition of Section 3.2. Finally, Section 8.9
makes use of the theory of Dedekind rings developed in Chap-
ters 5 and 6. The ideas about dimension are then expanded in
Chapter 9, while the notion of regularity comes again in Section
10.1. This way of reading the book will require tracking back the
prerequisites for some results, but has the advantage of giving
geometrical reasons to introduce algebraic constructs.

Examples in the text usually require only trivial verifications.

They are part of the core of the text and should not be skipped;
some definitions are actually given inside the examples (for in-
stance, the basic operations on ideals). When an example re-
quires more work, it should be considered as an exercise.
On the other hand, exercises vary from simple to hard, and
there is (intentionally) no indication to distiguish the level. So
give a try at as many exercise as you can, and don’t feel frustrated
if some of them look too hard. Maybe you can come back later,
when you are familiar with more techniques.
In general, I have tried to avoid depending on exercises for the
main body of the text. The cases where I have done so should

xii
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 introduction

be easy verifications. On the other hand, many important and

subtle counterexamples are presented as series of exercises.
No contribution in this book is original, except of course the
usual amount of errors, that should be attributed only to the au-
thor. If you spot some of them, you can send an email to ferret-
[email protected].
I hope that you will enjoy reading this book as much as I en-
joyed writing it!

If I was able to write this book, it is because Massimo Gobbino

and Paolo Tilli, when I was young and did not know better, be-
lieved in me and persuaded me to undertake the study of mathe-
matics. This turned out ot be one of the best choices I made, and
I have to really thank them for this. Thanks to Roberto Dvorni-
cich, who instilled in me a lasting love for algebra. Most of all,
I want to thank my wife Sbambi, who with her love shows me
everyday what is really important, and with her patience and un-
derstanding has given me the time and peace to do mathematics
and finish this book. ^ ¨

xiii
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 1
BASICS

In this chapter we introduce (or review) the basic notions of com-

mutative algebra, like rings, ideals and modules. The first three
sections cover these basics, then we treat the case of rings ad-
mitting a sort of Euclid’s algorithm. These are called Euclidean
rings and have a very simple structure: in particular we classify
(finitely generated) modules over these rings, generalizing the
classification of finitely generated abelian groups.
In the last two sections, we introduce the somewhat parallel
notions of local and graded rings, even though the parallelism
will emerge later, as the theory is developed. The notion of local-
ization of a ring is complementary to that of a quotient, and it is
a common technique in commutative algebra. In short the idea is
to enlarge a ring formally introducing multiplicative inverses of
its elements, in the very same way that one passes from Z to Q.

1.1 rings and ideals

Not surprisingly we start by defining rings, the objects that will
be central in our study. The idea of a ring is that of a set where
one is allowed to perform all usual operations, except for division.
The notion was born to identify the common features shared by
the set Z of integer numbers and things like polynomial rings
k[x1 , . . . , xn ], or a bit more exotic examples like Z[i], the set of
complex numbers that have integral real and imaginary part.

Definition 1.1.1. A ring A is a set endowed with two (binary) op-

erations, usually denoted + and · and called addition and multipli-

1
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 basics

cation, which are required to satisfy the following axioms. First,

the operation + should make A into an abelian group, that is
1. addition is associative, namely for every a, b, c ∈ A we have
(a + b) + c = a + (b + c);
2. there exists an element 0 ∈ A such that 0 + a = a + 0 = a
for every a ∈ A;
3. for every a ∈ A there exists an element −a ∈ A such that
a + (−a) = 0;
4. addition is commutative, that is for every a, b ∈ A we have
a + b = b + a.
Second, we ask for some properties of the multiplication, explic-
itly
5. the operation · is associative too, so for every a, b, c ∈ A we
have (a · b) · c = a · (b · c);
6. multiplication is distributive over addition, that is for every
a, b, c ∈ A we have a · (b + c) = a · b + a · c.
Remark 1.1.2. We have slightly abused notation giving a precise
name 0 to the neutral element in (2), since a priori it could be the
case that other such elements exist. So we check its uniqueness:
if another neutral element 0 0 exists, we must have

0 = 0 + 00 = 00

by two applications of (2).

In a similar way we can easily check that for a given a there
exist at most one element b such that a + b = 0, so we can safely
call it the inverse and denote it by −a.
Another standard identity can be derived from the axioms,
namely for every a ∈ A we have 0 · a = a · 0 = 0. Indeed from
0 = 0 + 0 and the distributivity axiom (6) it follows that

0·a+0·a = 0·a

and the desired identity follows by cancellation of 0 · a on both

sides, which is allowed since A is a group with respect to addi-
tion.

2
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 1.1 rings and ideals

Before giving the obvious examples, together with some slightly

less obvious ones, we give some definition, to restrict the atten-
tion to the rings we are actually interested in.
Definition 1.1.3. We say that a ring A is commutative if its multi-
plication is; explicitly
7. for every a, b ∈ A we have a · b = b · a.
We say that A is a ring with unit if there is a unit for multiplication,
namely
8. there exists an element 1 ∈ A such that 1 · a = a · 1 = a for
every a ∈ A.
Remark 1.1.4. All the rings that we shall consider in this book are
assumed to be commutative with unit. Of course noncommutative
rings are also of interest, but the theory is actually different in
many respects. So, starting from the next section, ring will be
shorthand for commutative ring with unit. Notice that some au-
thors include the existence of 1 into the definition of a ring, and
use the name rng for a ring without unit.
Example 1.1.5. (a) The set Z of integer numbers is a ring with
the usual operations. So is the set

Z[i] = {a + ib | a, b ∈ Z}

of Gaussian integers. Both these examples are commutative

rings with unit.
(b) Every field k is in particular a commutative ring with unit.
(c) If A is a commutative ring, then one can form the ring A[x]
of polynomials with coefficients in A. Clearly A[x] is again
commutative, and has a unit if and only if A has. In particular
if k is a field, then k[x1 , . . . , xn ] is a ring.
(d) Let E be an abelian group, and let

End(E) = {f : E → E | f is a group homomorphism}

denote the set of endomorphisms of E. Then End(E) inherits

an abelian group structure from E (in order to add up two

3
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 basics

functions you just sum their values); moreover one can de-
fine multiplication to be composition of endomorphisms. In
this way End(E) becomes a ring with unit, usually noncom-
mutative.

(e) In a similar fashion if V is a vector space, then End(V) is a

noncommutative ring with unit (here we are only considering
linear endomorphisms of V).

(f) If X is a topological space, the set C(X) of continuous func-

tions from X to R becomes a ring under point-wise addition
and multiplication.
Similarly if U is a smooth variety (if you don’t know what
these are, just let U be an open set in Rn ) one gets a ring
C∞ (X) of infinitely differentiable functions on U. Both rings
are commutative and have the constant function 1 as a unit.

(g) Let n ∈ Z. Then the set nZ of multiples of n is a subring of

Z. It is commutative, but it has no unit.

(h) If A1 , . . . , An are rings, then one can form in the obvious way
the direct product

A = A1 × · · · × An ,

with component-wise sum and multiplication. A is clearly a

ring; it is commutative if and only if all the Ai are, and it has
a unit if and only if all the Ai are rings with unit.

(i) For a fancier example, consider the set A of all functions

f : Z>1 → C,

that is, sequences of complex numbers. Addition in A is just

the usual addition, but for multiplication we take the Dirichlet
convolution defined by
X n
f ∗ g(n) = f(d)g .
d
d|n

4
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 1.1 rings and ideals

With these operations, A becomes a commutative ring with

unit. Everything is clear, except perhaps for the associativity
of multiplication, which follows from the fact that
X n X d X
f g h(e) = f(a)g(b)h(c).
d e
d|n e|d abc=n

The unit of this ring is the function


1 if n = 1
f(n) =
0 if n 6= 1.

Definition 1.1.6. Let A be a ring. We let A[x] be the ring of

polynomials in the indeterminate x, having coefficients in A, with
the usual operations.
We also define the ring A[[x]] of formal power series with coeffi-
cients in A. Elements of A[[x]] are formal linear combinations

a(x) = a0 + a1 x + a2 x2 + · · ·

with all ai ∈ A. Note that we allow an infinite number of ai

to be nonzero. There is no request of convergence on the series
(actually this does not even make sense unless A = R or C); still
we will usually write a(0) for a0 .
Operations are defined as follows: given

a(x) = a0 + a1 x + a2 x2 + · · · ,
b(x) = b0 + b1 x + b2 x2 + · · ·

we let

(a + b)(x) = (a0 + b0 ) + (a1 + b1 )x + (a2 + b2 )x2 + · · ·

while c(x) = a(x)b(x) is defined by the Cauchy product which

should be familiar from calculus

c0 = a0 b0
c1 = a1 b0 + a0 b1
c2 = a2 b0 + a1 b1 + a0 b2 .
..
.

5
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 basics

This is just the operation that comes from distributivity and the
request that xm · xn = xm+n .
By construction A[x] is a subring of A[[x]]. When it makes sense
we will also use the notation A{x} for the ring of convergent power
series.

Let us agree again that from now on all rings will be commutative
with unit. If B is a ring and {Ai } is a collection of subrings of B,
the intersection A = ∩i Ai is again a ring. In particular given a
set E ⊂ B there is a smallest ring containing E.

Definition 1.1.7. The smallest ring containing the set E is said to

be the ring generated by E. If A is a subring of B and b ∈ B, the
symbol A[b] denotes the ring generated by A and b; this is easily
seen the set of polynomial expressions in b with coefficients in
A.

We remark that the notation introduced above is consistent

with the notation A[x] for the polynomial ring with coefficients
in A in the indeterminate x. Indeed the latter can be seen as the
smallest ring containing A and the indeterminate x.
Inside the class of rings we identify a smaller subclass:

Definition 1.1.8. Let A be a ring. We say that A is an integral

domain if ab = 0 implies that a or b is 0. Unlike the identity
a · 0 = 0, this property does not follow from the ring axioms.

Example 1.1.9. (a) The rings Z, Z[i] are integral domains.

(b) Every field is an integral domain.

(c) Every subring of an integral domain is integral; in particular

every subring of a field is an integral domain.

(d) The ring C(X) of continuous functions on a topological space

X is usually not an integral domain. Similarly for the ring
C∞ (U), where U is a manifold (or an open set in Rn ).

(e) If A is an integral domain, A[x] is again an integral domain:

this is easily seen considering the monomials of highest (or
lowest) degree in a product.

6
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 1.1 rings and ideals

Definition 1.1.10. An element a ∈ A is called invertible if there

exists b ∈ A such that ab = 1. The element b is then uniquely
determined and denoted by a−1 . The set of invertible elements
of A is denoted by A∗ . Sometimes an invertible element is also
called a unit.

Example 1.1.11. An element a(x) ∈ A[[x]] is invertible if and only

if a(0) is invertible in A. Indeed the equations

1 = a0 b0
0 = a1 b0 + a0 b1
0 = a2 b0 + a1 b1 + a0 b2
..
.

can be solved for the bi inductively, provided a0 ∈ A∗ .

In particular if A = k is a field, a(x) is invertible if and only if
a(0) 6= 0; it follows that every a(x) ∈ k[[x]] can be written

a(x) = xr b(x)

for some b(x) ∈ k[[x]] invertible; this makes the algebra of power
series much easier than that of polynomials.

Definition 1.1.12. An element a ∈ A is called a zero divisor if

there exists b 6= 0 such that ab = 0. 0 itself is considered to be a
zero divisor. Of course nontrivial zero divisors are present only
if A is not an integral domain. If an = 0 for some n, a is called
nilpotent; of course this is stronger than being a zero divisor.

We next introduce the other players: the ideals.

Definition 1.1.13. Let I ⊂ A be an additive subgroup. We say that

I is an ideal if for every i ∈ I and a ∈ A we have ai ∈ I. Any ideal
different from A itself is called proper.
The proper ideal I is called prime if ab ∈ I implies that either
a ∈ I or b ∈ I. It is called maximal if it is not properly contained
in any other (proper) ideal.

Remark 1.1.14. For any subset E ⊂ A there is a smallest ideal

containing E, namely the intersection of all ideals containing E.
This is denoted by (E). When E = {a1 , . . . , an } we simply denote

7
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 basics

it by (a1 , . . . , an ). Explicitly, the ideal (a1 , . . . , an ) is the set of

elements a ∈ A which can be written in the form

a = x1 a1 + · · · + xn an .

To see this, one only has to check that the set thus defined is an
ideal; of course it is the smallest ideal containing a1 , . . . , an .

Definition 1.1.15. If I = (a) is generated by a single element, we

say that I is principal. When (a) is a prime ideal we say that a is
a prime element.

Remark 1.1.16. Any proper ideal is contained in a maximal one.

This follows directly from Zorn’s lemma, together with the fact
that the union of an ascending chain of proper ideals is an ideal
(it is not the whole ring because it does not contain 1).

Proposition 1.1.17. Any maximal ideal M is prime.

/ M and ab ∈ M; as M is maximal (M, a) must be

Proof. Let a ∈
the whole ring, so we can write

1 = m + xa

for some m ∈ M and x ∈ A. Multiplying by b we get

b = bm + xab ∈ M.

Example 1.1.18. (a) Z has the principal ideals (n); these are the
only ideals because they are the only additive subgroups.
This will be generalized in Proposition 1.5.5. The ideal (n)
is prime if and only if n = 0 or n is a prime of Z.

(b) The ideal (5) is not prime in Z[i]; indeed 5 = (2 + i)(2 − i).
So a prime element of a ring need not be prime in a bigger
ring.

(c) The ideal (x, y) of k[x, y] is not principal (why?).

(d) An ideal I is the whole ring if and only if it contains 1, if and

only if it contains any invertible element. In particular the
only ideals of a field k are 0 and k itself. Conversely, any ring
having just the trivial ideals is a field.

8
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 1.1 rings and ideals

(e) The ideal (0) is prime if and only if A is an integral domain.

(f) For any ring A the set of nilpotent elements is an ideal N(A),
called the nilradical of A. Recall that an element a ∈ A is
nilpotent if an = 0 for some n. We shall prove in Proposition
1.6.10 that it is the intersection of all prime ideals of A.
(g) We define the Jacobson radical J(A) as the intersection of all
maximal ideals of A. We then have N(A) ⊂ J(A).
(h) If I, J are ideals of A, then we can define the ideals
I + J = {i + j | i ∈ I, j ∈ J} = (I, J)
I · J = {i · j | i ∈ I, j ∈ J} .

Note that we always have I · J ⊂ I ∩ J.

(i) If I, J are ideals of A, then we can define the ideal
(I : J) = {a ∈ A | aJ ⊂ I}.
When J = (x) is principal, we will simply write (I : x) to mean
(I : J), and similarly when I is principal. This operation may
behave slightly differently from what the notation suggests.
For instance in Z we have
(8 : 4) = (2) but (8 : 5) = (8).
We easily see that I ⊂ (I : J) in any case.
(j) Let I ⊂ A be any ideal. Then the set
√
I = {a ∈ A | an ∈ I for some n}

√ the radical of I.√ The only nontrivial check

is an ideal, called
is that if a, b ∈ I, then a + b ∈ I. Let n be big enough, so
that an , bn ∈ I. Then
X
2n  
2n i 2n−i
(a + b)2n = a b ∈I
i
i=0

because each addend is in I. Note that by definition we have

√
0 = N(A).
√
The ideal I is called radical when I = I.

9
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 basics

The Jacobson radical admits the following characterization.

Proposition 1.1.19. Let A be a ring. Then a ∈ J(A) if and only if
1 − ab is invertible for all b ∈ A.
Proof. Let a ∈ J(A); then 1 − ab is not contained in any maximal
ideal, so (1 − ab) = A and finally 1 − ab is a unit.
Vice versa, let M be a maximal ideal. If a ∈
/ M, then (a, M) = A,
so we can write
1 = ab + m,
for some b ∈ A and m ∈ M; but then 1 − ab ∈ M is not invertible.

The defining property of prime ideals with respect to elements

extends in some way to ideals.
Proposition 1.1.20.
i) Let P1 , . . . , Pn ⊂ A be prime ideals and I ⊂ A be any ideal. If

I ⊂ P1 ∪ · · · ∪ Pn ,

then I ⊂ Pk for some k.

ii) Let I1 , . . . , In ⊂ A be ideals and P ⊂ A be a prime ideal. If

P ⊃ I1 ∩ · · · ∩ In ,

then P ⊃ Ik for some k.

Proof. i) By induction on n we can assume that I is not con-
tained in any union of n − 1 of these primes, so we can take
[
ai ∈ I \ Pj .
j6=i

Then we must have ai ∈ Pi . Let

X
n
a= a1 · · · · · a
ci · · · · · an ,
i=1

where aci means that ai is omitted. Then all addends but

one lie in Pi , so a ∈
/ Pi , for every i. This is a contradiction
because a ∈ I.

10
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 1.2 quotients

ii) Assume that each Ik 6⊂ P. Take ak ∈ Ik \ P; then

a = a1 · a2 · · · · · an ∈
/ P,

which contradicts the fact that a ∈ Ik for each k.

1.2 quotients
After one introduces rings, the next natural steps is to define the
class of maps between them.
Definition 1.2.1. Let A, B be rings. A ring homomorphism, or sim-
ply homomorphism, between them is a map

f: A → B

such that for every a, b ∈ A we have

f(a + b) = f(a) + f(b) and f(ab) = f(a)f(b).

We will always require that our homomorphisms are unital, that

is f(1) = 1.
Definition 1.2.2. Given a homomorphism f : A → B we define its
kernel
ker f := f−1 (0) = {a ∈ A | f(a) = 0}.
It is a straightforward check that ker f is an ideal of A.
Remark 1.2.3. More generally for any ideal I of B, f−1 (I) is an
ideal of A; the check is straightforward. If I is prime, then f−1 (I)
is prime too.
Remark 1.2.4. If f : A → B is a homomorphism, then f(A) is only
a subring of B.
As special cases we have
Definition 1.2.5. A homomorphism f : A → B is called an iso-
morphism when it admits an inverse g : B → A. In this case g is
necessarily a homomorphism, since f is.
When A = B we will say that f is an automorphism of A.

11
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 basics

Example 1.2.6. (a) Multiplication by n is not a ring homomor-

phism Z → Z. Indeed

nh · nk 6= n · hk

as soon as n 6= 1.

(b) If a ∈ A satisfies a2 = a (in which case a is called idempotent),

multiplication by a is a ring homomorphism A → A, but it is
not unital.

(c) For any ring A, there is a unique homomorphism φ : Z →

A. Its kernel ker φ is generated by a nonnegative number n,
possibly 0 when φ is injective. We call n the characteristic of
A, denoted char(A).

(d) Let a = (a1 , . . . , an ) ∈ kn , and consider the evaluation map

eva : k[x1 , . . . , xn ] → k

defined by eva (f) = f(a1 , . . . , an ). This is a homomorphism

of rings; indeed polynomial rings are constructed exactly in
such a way that this holds.
The same example works for any ring A in place of k.

(e) In the same way for every topological space X and for every
x ∈ X we have the valuation homomorphism

evx : C(X) → R

defined by evx (f) = f(x).

(f) The conjugation map

c : Z[i] → Z[i]

is a ring automorphism.
Similarly, let ω ∈ C be a primitive third root of unity, and
take A = Z[ω]. There is a unique automorphism A → A
exchanging the third roots of unity ω and ω2 .

12
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 1.2 quotients

(g) Let f : kn → kn be a polynomial function. Then composition

with f yields a homomorphism
cf : k[x1 , . . . , xn ] → k[x1 , . . . , xn ]
given by cf (g) = g ◦ f. This is an automorphism when f
admits a polynomial inverse, for instance if f is an invertible
linear function.
An example of such a function outside the realm of linear
maps is f(x, y) = (x + y2 , y).
There is an automatic way to produce homomorphism, actu-
ally one that will produce a homomorphism with an assigned
kernel.
Definition 1.2.7. Given a ring A and an ideal I we define a ring
A/I as follows. As a set A/I is just A modulo the equivalence
relation a ∼ b if there is some i ∈ I such that a = b + i. The
equivalence class of a in A/I is denoted a, or also aI when we
want to emphasize the dependence on I.
The operation are defined on representatives:
a + b := a + b and a · b := ab.
These are well-defined: for instance if a = a 0 + i we have
ab = (a 0 + i)b = a 0 b + ib = a 0 b
since ib ∈ I.
The set A/I, endowed with the above ring structure is called
the quotient of A by I.
As a notation we sometimes write
a≡b (mod I)
and say that a and b are congruent modulo I whenever aI = bI ,
that is, a − b ∈ I.
By construction we have a surjective homomorphism, which
we usually call projection,
πI : A → A/I
sending a to a. We see at once that ker πI = I. In a precise sense
these cover all the examples of homomorphisms.

13
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 basics

Proposition 1.2.8. Let f : A → B be a homomorphism, and let I =

ker f. Then we have an injective homomorphism fb: A/I → B such that
f(a)
b = f(a). Moreover fb is surjective (hence an isomorphism) if and
only if f is surjective.

Proof. One checks that fb given as above is well-defined. Then

ker fb = I/I = 0, so fb is injective. By definition, it has the same
image of f.

Proposition 1.2.9. The preimage map J → π−1 I (J) gives a bijective

correspondence between ideals of A/I and ideals of A containing I. This
correspondence preserves prime and maximal ideals.
Proof. For every ideal J of A/I, π−1
I (J) is an ideal of A. The in-
verse correspondence sends the ideal K ⊃ I to K/I = π(K). This is
an additive subgroup of A/I (this is true for any homomorphism
of groups). To check that it is an ideal we take a ∈ A/I and
k ∈ K/I and note that

a · k = ak ∈ K/I.

If J is prime then π−1 (J) is prime; this is true for any homo-
morphism. Vice versa, let P ⊃ I be a prime of A and assume
that
a · b = ab ∈ P/I.
Then ab ∈ P + I = P, so either a ∈ P or b ∈ P. It follows that P/I
is prime.
Finally, the correspondence preserves inclusions, hence it pre-
serve maximality of ideals.
Since 0 is prime in A if and only if A is an integral domain,
and is maximal if and only if A is a field, we deduce the
Corollary 1.2.10. The ideal I ⊂ A is prime if and only if A/I is an
integral domain. It is maximal if and only if A/I is a field.
From the corollary we can deduce again that maximal ideals
are prime.
Example 1.2.11. Consider the homomorphism va of Example
1.2.6. It is a surjective homomorphism whose kernel is

Ia := ker va = {f ∈ k[x1 , . . . , xn ] | f(a1 , . . . , an ) = 0}.

14
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 1.3 modules

From the isomoprhism

∼k
k[x1 , . . . , xn ]/Ia =

we deduce that Ia is a maximal ideal.

We end the section with the classical

Theorem 1.2.12 (Chinese remainder theorem). Let A be a ring,

I, J ⊂ A two ideals which are coprime, in the sense that I + J = A.
Then we have a canonical isomorphism

f : A/(I ∩ J) → A/I × A/J

such that f(aI∩J ) = (aI , aJ ).

Proof. Define a homomorphism

g : A → A/I × A/J

by g(a) = (aI , aJ ). If we prove that g is surjective, the thesis

follows from Proposition 1.2.8, since ker g = I ∩ J.
Then for any b ∈ A/I and c ∈ A/J we want to find x ∈ A such
that

b = x+i
c = x+j

for some i ∈ I, j ∈ J. This can be solved for x provided

b − c = i − j.

The existence of suitable elements i, j then follows from the hy-

pothesis I + J = A.

1.3 modules
Formally modules over a ring are defined exactly like vector
spaces over a field.

15
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 basics

Definition 1.3.1. Let A be a ring, M a set endowed with a (binary)

operation +, called addition, and a map

A×M →M
(a, m) → a · m,

called (scalar) multiplication. We will call M an A-module, or

simply a module, if the operations satisfy the following axioms.
First, the operation + should make M into an abelian group, that
is
1. addition is associative, namely for every m, n, p ∈ M we
have (m + n) + p = m + (n + p);
2. there exists an element 0 ∈ M such that 0 + m = m + 0 = m
for every m ∈ M;
3. for every m ∈ M there exists an element −m ∈ M such that
m + (−m) = 0;
4. addition is commutative, that is for every m, n ∈ M we have
m + n = n + m.
Second we ask for some properties of the multiplication, explic-
itly
5. the operation · is associative, in the sense that for every
a, b ∈ A and m ∈ M we have

(a · b) · m = a · (b · m)

(note that the two mutiplications involved are actually dif-

ferent operations);
6. multiplication is distributive over addition, that is for every
a ∈ A and m, n ∈ M we have a · (m + n) = a · m + a · n.
Definition 1.3.2. A submodule of M is just a subset N ⊂ M which
is closed under the operations, so that it inherits the structure of
an A-module itself.
Remark 1.3.3. Uniqueness of the neutral element and of the ad-
ditive inverse are proved exaectly in the same way as for rings,
so there is no ambiguity in using the symbols 0 and −m.

16
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 1.3 modules

Remark 1.3.4. Let M be an abelian group. Giving M the structure

of an A-module is the same as giving a homomorphism of rings
A → End(M).

Example 1.3.5. (a) When A = k is a field an A-module is just

a vector space over k. In general A-modules have a much
subtler structure, as the next examples show.

(b) An is a module over A with component-wise multiplication.

(c) The submodules of A are just its ideals.

(d) Let I ⊂ A be an ideal. Then A/I is an A-module.

(e) Every abelian group G has a unique structure of Z-module.

This is because there is exactly one homomorphism Z →
End(G) sending 1 to the identity. Explicitly we have

n·g = g+···+g.
| {z }
n times

(f) Let X be a topological space, E → M a vector bundle. Let

C(X, E) be the set of continous sections of E; then C(X, E) is a
module over C(X).
A similar example can be obtained when X = M is a man-
ifold, E is a differentiable vector bundle and only C∞ func-
tions and sections are considered.

(g) Let M be a manifold (if you prefer just take M an open set in
Rn ). Then C(M) is a module over C∞ (M).

(h) Let V be a k-vector space and choose any linear endomor-

phism L ∈ End(V). Then V becomes a module over k[x];
multiplication by x is defined by

x · v = L(v).

(i) Let M be an A-module, E ⊂ M. There is a smallest submod-

ule of M containing E; this is called the submodule generated
by E, and denoted hEiA .

17
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 basics

(j) Let M be an A-module, and denote M[x] the set of all formal
finite linear combinations

mn xn + · · · + m1 x + m0

with mn , . . . , m0 ∈ M; then M[x] is an A[x]-module in the

obvious way.

(k) Let M be an A-module, I ⊂ A an ideal. We define I · M as

the submodule of M generated by all products i · m for i ∈ I
and m ∈ M, in symbols

I · M = hi · m | i ∈ I, m ∈ MiA .

This agrees with our previous definition of product when M

is an ideal of A.

(l) If M, N ⊂ R are submodules, the sum

M + N := {m + n | m ∈ M, n ∈ N}

is a submodule of R; it coincides with hM, NiA .

(m) If M ⊂ N is a submodule, the set

(M : N) := {a ∈ A | aN ⊂ M}

is an ideal of A.

Remark 1.3.6. Let M be an A-module, I ⊂ A an ideal. Then M

has an induced structure of A/I-module if and only if I · M = 0.
In particular if I · M = 0, the A-submodules and A/I submod-
ules of M are the same. When M = A/I we conclude that ideals
of A/I are the same of A-submodules of A/I.
This fact will be used in many places, often implicitly.

Next we introduce some notions analogue to those already

seen for rings.

Definition 1.3.7. Let M, N be A-modules. A homomorphism be-

tween M and N is a map

f: M → N

18
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 1.3 modules

such that

f(m + n) = f(m) + f(n) and f(am) = af(m)

for all m, n ∈ M and a ∈ A. We will also say that f is a A-linear

map.
When M = N we say that f is an endomorphism of M. When f is
both injective and surjective, so that there is an inverse A-linear
map, we say that f is an isomorphism.
Definition 1.3.8. Given a homomorphism f : M → N we define
its kernel
ker f := f−1 (0) = {m ∈ M | f(m) = 0}
and its image im f := f(M) ⊂ N. Both ker f and im f are submod-
ules, of M and N respectively.
Not surprisingly every submodule of M arises as the kernel of
suitable homomorphism, which is the projection to the quotient.
Definition 1.3.9. Let N ⊂ M be A-modules. We define the quo-
tient A-module M/N as follows. As a set M/N is formed by
equivalence classes of elements of M modulo the equivalence re-
lation l ∼ m if there is some n ∈ N such that l = m + n. The
equivalence class of m is denoted m or sometimes mN .
Operations on M/N are defined on representatives:

m + n := m + n and a · m := am.

The verification that these are well-defined is identical to the one

we made for rings.
As a notation we sometimes write

m ≡ n (mod N)

and say that m and n are congruent modulo N whenever mN =

nN , that is, m − n ∈ N.
We have a surjective homomorphism, which we usually call
projection,
πN : M → M/N
sending m to m. By construction ker πN = N. As in the case of
rings we have (with the same proof)

19
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 basics

Proposition 1.3.10. Let f : M → R be a homomorphism, and let N =

ker f. Then we have an injective homomorphism fb: M/N → R such
that f(m)
b = f(m). Moreover fb is surjective (hence an isomorphism) if
and only if f is surjective.

Proposition 1.3.11. The preimage map R → π−1 N (R) gives a bijec-

tive correspondence between submodules of M/N and submodules of
M containing N.

Definition 1.3.12. Given a homomorphism f : M → N we define

the cokernel of f as
coker f := N/ im f.
In a sense which will be clearer later on, this is the specular no-
tion of the kernel.

Usually we use diagrams to keep track of kernels and coker-

nels.

Definition 1.3.13. Consider a sequence (finite or infinite) of A-

modules with maps between them:

αn−1 αn
M• : · · · → Mn−1 → Mn → Mn+1 → · · · .

If αn ◦ αn−1 = 0 for every n we say that M• is a complex of

A-modules. This means that

im fn−1 ⊂ ker fn .

If we have
im fn−1 = ker fn
we say that M• is an exact sequence. A three-term exact sequence
like
0 →M →N →P →0
is called a short exact sequence. In this case the first map is injective
and the last one is surjective.

Remark 1.3.14. If

α β
0 →M →N →P →0

20
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 1.3 modules

is a short exact sequence, then M is isomorphic to ker β and P to

coker α.
Vice versa, any quotient M/N fits into an exact sequence
0 →N →M → M/N → 0.
Remark 1.3.15. Any long exact sequence can be split into short
exact sequences like
0 → ker fn → Mn → ker fn+1 →0
0 → ker fn+1 → Mn+1 → ker fn+2 → 0;
..
.
this allows us to reduce many statements about arbitrary exact
sequences to the case of short ones.
Example 1.3.16. (a) For any ideal I of A we have the short exact
sequence of A-modules
0 →I →A → A/I → 0;
from the point of view of rings the modules I and A/I have
quite a different role (in particular A/I is a ring, while I is
not).
(b) Multiplication by n defines a homomorphism of Z-modules
Z → Z, which gives rise to the exact sequence
0 →Z →Z → Z/nZ → 0;
this is not a homomorphism of rings since we require the
latter to be unital.
(c) A homomorphism from Am to An is defined by a m × n
matrix with coefficients in A, as in the field case.
(d) The set of homomorphism from M to N is in the obvious way
an A-module, denoted Hom(M, N), or else HomA (M, N) if
the ring is not clear from the context. We also use the notation
End(M) := Hom(M, M).
(e) Let U ⊂ C be an open set and regard M = C∞ (U, C) as
a module over A = O(U). Then the derivative d/dz is an
endomorphism of M, while d/dz is not A-linear.

21
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 basics

After all these preliminary definitions, we actually begin to

prove something about modules.

Definition 1.3.17. An A-module M is said to be finitely generated

if there is a finite number of elements m1 , . . . , mr ∈ M such that

M = hm1 , . . . , mr iA .

Equivalently, every element of M is a linear combination of the

elements m1 , . . . , mr with coefficients in A.

Remark 1.3.18. Every finitely generated module is a quotient of

Ar for some r, and vice versa.

The following may be the most widely used result in commu-

tative algebra.

Theorem 1.3.19 (Nakayama’s lemma). Let A be a ring, J = J(A)

its Jacobson radical. Let M be a finitely generated module such that
J · M = M; then M = 0.

Proof. Assuming M 6= 0, let m1 , . . . , mr be a minimal set of gener-

ators of M. Since mr ∈ J · M we can write

mr = a1 m1 + · · · + ar mr ,

with a1 , . . . , ar ∈ J. Rewrite this as

(1 − ar )mr = a1 m1 + · · · + ar−1 mr−1 .

According to Proposition 1.1.19 we can invert 1 − ar to write

mr as a linear combination of the other mi , contradicting the
minimality.

The above result and its corollaries are often used when A is a
local ring, that is A has only one maximal ideal M. In this case of
course we just have J = M.
Applying Nakayama’s lemma to the quotient M/N we get the
seemingly stronger form

Corollary 1.3.20. Let A be a ring, J = J(A) its Jacobson radical. Let

M be a finitely generated module, N ⊂ M a submodule, such that
M = N + J · M; then M = N.

22
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 1.3 modules

There is a notion of finite generation for a different kind of

objects, which we now introduce.
Definition 1.3.21. Let A ⊂ B be rings. Then on B we have both
the structure of ring and that of A-module. We say that B is
a A-algebra. More generally we speak of a A-algebra when we
have a (not necessarily injective) homomorphism f : A → B. A
homomorphism between two A-algebras is a homomorphism of
rings which is also A-linear.
Remark 1.3.22. When A = k is a field, every homomorphism
k → B is injective, so the distiction above does not apply.
If B is an A-algebra and I ⊂ A an ideal we will write I · B to
denote the ideal generated by I in B. This is sometimes called the
extension of I in B.
Definition 1.3.23. Let B be an A-algebra, E ⊂ B a subset. We say
that E generates B if the smallest sub-A-algebra of B containing
E is B itself. In this case, if E is finite, we say that B is finitely
generated.
Remark 1.3.24. Of course, if B is finitely generated as an A-
module, it is also finitely generated as an A-algebra, but the con-
verse is in general false.
Remark 1.3.25. If B is a finitely generated A-algebra, there is a
surjective homomorphism
A[x1 , . . . , xn ] → B,
and conversely.
Example 1.3.26. (a) Just as any abelian group is a Z-module, ev-
ery ring B is a Z-algebra, via the only homomorphism Z → B
which sends 1 to 1.
(b) For any ring A we can consider the finitely generated A-
algebra A[x1 , . . . , xn ]; as A-module this is not finitely gen-
erated
(c) Q is not finitely generated as Z-algebra. If it was, only a finite
number of prime factors could appear in the denominators of
rational numbers.
(d) Every quotient A/I is a A-algebra via the canonical projection,
and is of course finitely generated.

23
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 basics

1.4 more constructions with modules

Next we describe more operations that we can perform on A-
modules.
Definition 1.4.1. Let {Mi }i∈I be a (not necessarily finite) collec-
tion of A-modules. We define the direct product of the Mi to be
Y 
Mi := (mi )i∈I | mi ∈ Mi

and their direct sum as the submodule


Mi := (mi )i∈I | mi ∈ Mi , mi 6= 0 for finitely many i .
L
Q L
Operations on Mi and Mi are defined component-wise.
We say that M is free if it is isomorphic to a direct sum of copies
of A.
Remark 1.4.2. Given a collection of homomorphisms
fi : N → Mi
there is a unique homomorphism
Y
f: N → Mi

such that fi = πi ◦ f, where πi is the projection on the factor Mi .

Symmetrically given homomorphisms
gi : Mi → N
there is a unique homomorphism
L
g: Mi → N
which agrees with gi on each summand.
Of course, when the index set is finite direct product and direct
sum agree, and enjoy both properties.
Remark 1.4.3. Let M be any A-module, E = {ei } a set of gener-
ators. Take any set S = {si } in bijective correspondence with E,
and consider the direct sum of copies of A, one for each si . This
is a free module with basis {si }, call it F. We have a surjective
homomorphism F → M sending si to ei . So every A-module is a
quotient of a free one.

24
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 1.4 more constructions with modules

Example 1.4.4. (a) Every vector space over k has a basis, hence
it is free.
(b) Regard A[x] as an A-module; then it is the direct sum of a
denumerable quantity of copies of A. Instead A[[x]] is their
direct product.
(c) If M = N ⊕ R we have an exact sequence

0 →N →M →R → 0.

Conversely, given an exact sequence

i p
0 →N →M →R → 0,

M is the direct sum of N and R (in such a way that i and p

are the natural inclusion and projection) if and only if there
is an A-linear map
s: R → M
such that p ◦ s = idR . In this case we say that the exact se-
quence splits.
(d) The condition above is not always verified; for instance we
have an exact sequence of Z-modules
α
0 → Z/2Z → Z/4Z → Z/2Z →0

where α(n) = 2n, which does not split. Indeed Z/4Z has an
element of order 4, while Z/2Z ⊕ Z/2Z does not.
(e) For another example of an exact sequence which is not split
take A = k[x]/(x2 ). Then we have the exact sequence
α
0 →k →A →k → 0,

where α(t) = tx. Can you see why it does not split?
We go on with some definitions.
Definition 1.4.5. Let M be an A-module, m ∈ M. The annihilator
of m is the ideal

Ann(m) = {a ∈ A | a · m = 0}.

25
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 basics

When Ann(m) 6= 0 we say that m is a torsion element. More

generally, if N ⊂ M, we denote

Ann(N) = {a ∈ A | a · n = 0 for all n ∈ N}.

Now assume that A is an integral domain. The torsion submod-

ule T(M) of M is the set of torsion elements. To check that this is
a submodule, let m, n be torsion, so that am = bn = 0 for some
nonzero a, b ∈ A. Then

ab(m + n) = 0 and a · cm = 0

for all c ∈ A, so both m + n and cm are torsion (here we are using

that ab 6= 0).
We say that M is torsion-free when T(M) = 0; equivalently mul-
tiplication by any nonzero a ∈ A defines an injective homomor-
phism M → M.

Remark 1.4.6. A free module is always torsion-free; the converse

holds only for some particular classes of rings.

Proposition 1.4.7. Let A be an integral domain, M an A-module.

Then M/ T(M) is torsion-free.

Proof. Assume am = 0 in M/ T(M); then am is torsion, so m

is.

The reader may already know the next construction in the con-
text of vector spaces.

Definition 1.4.8. Let M, N be A-modules. We define their tensor

product M ⊗ N as follows. First we consider the free module with
basis a symbol m } n for every pair m ∈ M, n ∈ N, call it M } N.
The tensor product M ⊗ N is the quotient of M } N by the
submodule generated by the relations

(m + m 0 ) } n − m } n − m 0 } n
(am) } n − a · (m } n)
m } (n + n 0 ) − m } n − m } n 0
m } (an) − a · (m } n)

for every choice of m, m 0 ∈ M, n, n 0 ∈ N and a ∈ A.

26
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 1.4 more constructions with modules

The equivalence class of m } n is denoted m ⊗ n, so elements

of M ⊗ N are formal linear combinations
X
mi ⊗ ni .

In particular if the elements {mi } generate M and the {ni } gener-

ate N, the products {mi ⊗ nj } generate M ⊗ N.
Remark 1.4.9. By construction we have a map
M×N → M⊗N
which is A-linear in each variable; we shall say that such a map
is A-bilinear. Namely it is the map sending (m, n) to m ⊗ n.
Moreover given any A-module R with a bilinear map
M×N → R
there is a unique A-linear map M ⊗ N → R making the diagram
M×N → M⊗N

→ ↓
R
commute. This is all immediate from the definition.
This property is the raison d’être of the tensor product. It allows
us to reduce the study of bilinear maps to that A-linear ones, at
the cost of changing the target. By iterating this construction, one
can also use tensor products to investigate multilinear maps, that
is, maps
M1 × · · · × Mn → R
which are A-linear on each factor. These are in bijective corre-
spondence with A-linear maps
M1 ⊗ · · · ⊗ Mn → R.
Remark 1.4.10. From another point of view, to specify a bilin-
ear map M × N → R is the same as to give a linear map M →
Hom(N, R). This observation yields a bijective correspondence
Hom(M ⊗ N, R) ↔ Hom(M, Hom(N, R)),
which is easily seen to be an isomorphism of A-modules. For
those who know some categorical nonsense, this means that · ⊗ N
and Hom(N, ·) are adjoint functors.

27
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 basics

The above remark can be used to prove the following

Proposition 1.4.11. Let
α β
0 → M1 → M2 → M3 →0
be a short exact sequence of A-modules and N another A-module. Then
the sequences
α∗ β∗
0 → Hom(N, M1 ) → Hom(N, M2 ) → Hom(N, M3 ),
β∗ α∗
0 → Hom(M3 , N) → Hom(M2 , N) → Hom(M1 , N)
and
α ⊗ idN β ⊗ idN
M1 ⊗ N → M2 ⊗ N → M3 ⊗ N →0
are exact.
We should say where the maps in the various exact sequences
come from. Given a homomorphism
α : M1 → M2
we obtain a corresponding homomorphism
α∗ : Hom(N, M1 ) → Hom(N, M2 )
defined by α∗ (f) = α ◦ f for f : N → M1 . The homomorphism α∗
is defined by composition in the other direction.
We also have a homomorphism
α ⊗ idN : M1 ⊗ N → M2 ⊗ N
sending m ⊗ n to α(m) ⊗ n. More generally given another homo-
morphism
β : N1 → N 2
there is an induced homomorphism
α ⊗ β : M1 ⊗ N1 → M2 ⊗ N2
defined by
α ⊗ β(m ⊗ n) = α(m) ⊗ β(n);
indeed the map M1 × N1 → M2 ⊗ N2 which sends (m, n) to
α(m) ⊗ β(n) is bilinear.

28
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 1.4 more constructions with modules

Proof. Let f ∈ ker α∗ , so f : N → M1 and α ◦ f = 0. Since α is

injective we have f = 0, showing that α∗ is injective.
Clearly
β∗ ◦ α∗ = (β ◦ α)∗ = 0,
so we need only to prove that ker β∗ ⊂ im α∗ . Let f ∈ kerβ∗ , so
f : N → M2 and β ◦ f = 0. Then f takes values in ker β = im α, so
it can be regarded as a map N → M1 . This shows that f ∈ im α∗ ,
proving that the Hom(N, ·) sequence is exact.
The proof of exactness of the Hom(·, N) sequence is similar.
By the above remark, for every A-module R we have
∼ Hom(N, Hom(Mi , R)).
Hom(Mi ⊗ N, R) =
By what we have already proved we obtain the exact sequence
0 → Hom(L3 , R) → Hom(L2 , R) → Hom(L1 , R),
where Li := Mi ⊗ N. So we end the proof if we show that this
implies the exactness of
L1 → L2 → L3 → 0.
This is really more complicated to write up than to prove by
yourself, so we leave it as an exercise.
Remark 1.4.12. As easy as it may seem, the above proposition is
not completely trivial. Try proving exactness at M2 ⊗ N directly!
Remark 1.4.13. The extent to which the above exact sequences
fail to be exact is the subject of homological algebra. We will not
treat this topic here, but see for instance [Rot79] or [Wei95] if you
are interested.
We can use the tensor product to prove that the rank of a free
A-module is well-defined.
∼ An . Then n = m.
Proposition 1.4.14. Assume that Am =
Proof. Let M be any maximal ideal of A and consider the field
k = A/M. Then
∼ km ,
Am ⊗ A/M = Am ⊗ k =
and the same for An . So we get km = ∼ kn as A-modules, and a
fortiori as k-vector spaces, and we obtain the conclusion by the
usual linear algebra.

29
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 basics

Remark 1.4.15. If A is an integral domain we can argue in the

same way, using the field of fractions of A (see Section 1.6) in-
stead.

Remark 1.4.16. The proposition works equally for the case of

infinite rank.

Definition 1.4.17. Let M be a finitely generated free A-module,

so M =∼ An for some n. We call n the rank of M. It is well defined
by the above proposition.

1.5 euclidean rings

We now introduce the simplest class of rings, those which ad-
mit an analogue of the Euclidean algorithm. This section is just
an example of application of the preceding concepts in a simple
situation.

Definition 1.5.1. Let A be an integral domain. We say that A is

Euclidean if it admits a function

N : A \ {0} → N,

called a norm, such that:

1. for all nonzero a, b ∈ A, N(a) 6 N(ab);

2. for each a, b ∈ A, b 6= 0, there exists q, r ∈ A such that

a = qb + r

and either r = 0 or N(r) < N(b).

In short Euclidean rings are those where we can perform the

division with remainder, in such a way that the remainder is
“smaller” than the divisor.

Remark 1.5.2. We do not ask that the decomposition a = qb + r

is unique, even up to invertible elements.

Remark 1.5.3. The invertible elements of A are exactly those with

minimal norm. Indeed the first condition tells us that N(1) 6

30
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 1.5 euclidean rings

N(b) for every nonzero b ∈ A. So k = N(1) is the minimal

possible norm. If a ∈ A is a unit, then N(a) 6 N(aa−1 ) = k, so
N(a) = k by minimality.
Vice versa, assume that N(a) = k. Then the division property
tells us that there are q, r ∈ A with

1 = qa + r;

since we cannot have N(r) < N(a) we must have r = 0, that is a

is invertible.

Example 1.5.4. (a) The ring Z is Euclidean; the norm is just the
usual absolute value: N(n) = |n|.

(b) For every field k, the ring k[x] is Euclidean. For the norm we
can take N(f) = deg(f); the division is just the usual division
between polynomials.

(c) Now a more subtle example: the ring Z[i] is Euclidean; here
the norm is the squared modulus for complex numbers, that
is
N(a + ib) = a2 + b2 .
Since multiplicativity is clear, let us verify that we can per-
form the division. Let z, w ∈ Z[i] and consider the usual
quotient u = z/w ∈ C. If u ∈ Z[i] we are done; otherwise u
lies in the interior of some 1 × 1 square with integer coordi-
nates. Let q be the nearest vertex of this square, and define r
by
r = z − qw.
Then N(r) = N((u − q)w) is less than N(w) since the modu-
lus of u − q is less than 1.
Note that in this example there may be other vertices that
work, so we really don’t have any kind of uniqueness.

(d) The ring k[[x]] is Euclidean too. For a power series a(x) we
let N(a) be the degree of the first nonzero monomial; that is,
if
a(x) = xk a1 (x)
with a1 (0) 6= 0, we let N(a) = k.

31
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 basics

The first condition is immediate. For the existence of the

division take any two power series a(x), b(x) with b 6= 0. If
N(a) > N(b) write

a(x) = xN(a) a1 (x)

b(x) = xN(b) b1 (x);

then we can divide exactly a by b taking

q(x) = xN(a)−N(b) a1 (x)b1 (x)−1 , r(x) = 0.

If N(a) < N(b) just take q = 0 and r = a.

If A is Euclidean we can perform the Euclidean algorithm. That
is, given a, b ∈ A we perform the repeated divisions

a = q1 b + r1
b = q2 r1 + r2
r1 = q3 r2 + r3
..
.

until some rn = 0. This must happen in a finite number of steps

because
N(b) > N(r1 ) > N(r2 ) > · · ·
is a decreasing sequence of natural numbers. So we have a last
step
rn−2 = qn rn−1 .
By induction we see that rn−1 divides both a and b. Moreover,
inverting the steps one at a time we find

rn−1 = rn−3 − qn−1 rn2 =

= rn−3 − qn−1 (rn−4 − qn−2 rn3 ) = · · · = xa + yb

for some x, y ∈ A. So if c divides both a and b, it must divide

rn−1 . In this case we say that rn−1 is a maximum common divisor
between a and b. The relation rn−1 = xa + yb is called Bezout’s
identity.
Proposition 1.5.5. Let A be a Euclidean ring. Then every ideal of A
is principal.

32
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 1.5 euclidean rings

Proof. Let I ⊂ A be an ideal, and take any nonzero a ∈ I with

minimal norm, say k = N(a). If b ∈ I, write

b = qa + r;

since we cannot have N(r) < k we must have r = 0, that is b ∈

(a).
We see from the preceding proposition that ideals of an Eu-
clidean ring have a very simple structure. This is true also for
more general modules.
Theorem 1.5.6. Let A be Euclidean and let F be a finitely generated
free module over A. If N ⊂ F is a submodule, then N is free. More
precisely, there exists a basis {f1 , . . . , fn } of F and elements a1 , . . . , ak
of A, k 6 n, such that {a1 f1 , . . . , ak fk } is a basis of N.
Proof. We give an algorithmic proof. Starting from any basis of
F=∼ An , we can write an element f ∈ F as a column vector with
entries in A. Let n1 , . . . , ns be generators for N, and write the
respective column vectors as a s × n matrix. We will allow our-
selves to perform some elementary operation on the matrix.
First, we can shuffle rows (this amounts to a permutation of the
vectors of the basis of F) or columns (a permutation of the gener-
ators of N). We can also take two rows r1 and r2 and substitute
r1 with
r10 = r1 + ar2
for any a ∈ A: this is an invertible change of basis for F. Similarly,
we can perform the corresponding column operation.
We claim that a repeated application of the elementary opera-
tions takes the matrix in the form
 
a1 0 ··· 0
 0 a2 ··· 0
 .. .. 
 
 . . ..
 .
 0 ··· ak · · · 0 
 , (1.5.1)
0
 · · · 0 

 . .. 
 .. .
0 ... 0

where a1 |a2 | · · · |ak . From this, the thesis follows.

33
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 basics

Let x1,1 and x2,1 be the first two elements of the first column.
Using row operations and the Euclidean algorithm we can sub-
stitute x1,1 with the maximum common divisor of the two. We
repeat the process until we have one element on the first column
that divides all the others. We put it in first position, and subtract
a suitable multiple of the first row from the others, so that all the
elements in the first column below the first are 0. So we get
 
x1 ∗ ··· ∗
0 ∗ ··· ∗
..  .
 
 ..
 . .
0 ∗ ··· ∗

Now we do the same for the first row. This may ruin our first
column, anyway we obtain
 
x2 0 ··· 0
∗ ∗ · · · ∗
..  .
 
 ..
 . .
∗ ∗ ··· ∗

We turn back to the first column and so on. Since x1 is multiple

of x2 , which is multiple of x3 and so on, the process will stop.
When it does we are left with
 
a1 0 · · · 0
 0 ∗ · · · ∗
..  .
 
 ..
 . .
0 ∗ ··· ∗

At this point we can conclude by induction on the number of

columns.

A matrix in the form of (1.5.1) (with a1 |a2 | · · · |ak ) is said to

be in Smith normal form. What we have essential proved is that a
matrix with entries in an Euclidean ring can be brought in Smith
normal form with a sequence of elementary row and column
operations.

34
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 1.6 localization

Corollary 1.5.7 (Structure of modules over Euclidean rings). Let

M be a finitely generated module over the Euclidean ring A. Then there
exist a1 , . . . , ak ∈ A such that
∼ Ar × A/(a1 ) × · · · × A/(ar ).
M=
Proof. Since M is finitely generated we have a surjective homo-
morphism φ : F → M, where F is finitely generated free. Apply
the proposition with N = ker φ.
Example 1.5.8. Since a Z-module is just an abelian group, we
recover the usual structure theorem for finitely generated abelian
groups.
Corollary 1.5.9. Let G be a finitely generated free abelian group of rank
n, H < G a subgroup. Then H is finitely generated of rank m 6 n.
Moreover, if m = n, the quotient is finite with |det M| elements, where
M is any matrix expressing the generators of H in coordinates with
respect to generators of G, that is,
X
n
hi = Mi,j gj ,
j=1

for generators g1 , . . . , gn of G and h1 , . . . , hn of H.

Proof. A special case of the above when A = Z. Fix generators
g1 , . . . , gn of G and identify G with Zn . A finite set of elements
in H can be represented with a matrix. By taking this matrix
in Smith normal form, we don’t change the generated subgroup,
which shows that H has at most rank m. If moreover m = n, the
moves that bring the matrix in Smith normal form preserve the
determinant. When H is generated by the columns of a matrix in
Smith normal form, the thesis is clear.
These results will be generalized for principal ideal domains
and for Dedekind rings in the following sections.

1.6 localization
In this section we introduce the process of enlarging a ring A by
the use of “fractions” with entries in A. Thing would be sim-
pler if A was an integral domain, but for our purpose we need

35
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 basics

the more general case. For reasons which will become apparent
when translated in geometric context, this process is called local-
ization.
Definition 1.6.1. Let A be a ring, S ⊂ A a subset. We say that S
is a multiplicative set if 1 ∈ S and for every s, t ∈ S we have st ∈ S.
Example 1.6.2. (a) For every nonzero a ∈ A the set

S = {1, a, a2 , . . . }

is a multiplicative set.
(b) If P ⊂ A is a prime ideal, then S = A \ P is a multiplicative
set. More generally if {Pi } is a family
S of prime ideals of A, we
have the multiplicative set S = A \ i Pi .
(c) The set of nonzero divisors of A is a multiplicative set. So is
the set of invertible elements.
(d) If I ⊂ A is an ideal, S = 1 + I is a multiplicative set.
(e) Take A = k[x1 , . . . xn ] and let V ⊂ kn be any subset. Then

S = {f ∈ k[x1 , . . . , xn ] | f(x) 6= 0 for all x ∈ V}

is a multiplicative set.
(f) Let A = k[x]; then

S = {f ∈ k[x] | f does not have any root in k}

is a multiplicative set.
Definition 1.6.3. Given a ring A with a multiplicative set S we
define the localization of A at S as the ring S−1 A obtained as
follows.
Elements of S−1 A are couples (a, s) ∈ A × S modulo the fol-
lowing equivalence relation. Two couples (a, s) and (b, t) are
equivalent if there exists u ∈ S such that

atu = bsu.

Apart from the presence of u this is the usual cross relation for
fractions. We cannot avoid u unless A is an integral domain.

36
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 1.6 localization

Not surprisingly we denote the class of equivalence of (a, s) by

a/s.
We define addition and multiplication in S−1 A by the familiar
rules

a/s + b/t = (at + bs)/(st), a/s · b/t = (ab)/(st).

It is a simple exercise to check that the operations are well-defined,

so S−1 A becomes a commutative ring, with unit 1/1.

By construction, we have a natural homomorphism

ι : A → S−1 A

sending a to a/1. In general this is not injective: indeed

ker ι = {a ∈ A | there exists u ∈ S such that au = 0}.

So ι is injective precisely when S does not contain any zero di-

visor; in this case we shall allow ourselves to denote ι(a) = a/1
simply by a. When we want to recall S in the notation we shall
write ι = ιS , or even ι = ιP if S = A \ P. For more about the
kernel of ιP , see 7.5.26.
Corresponding to some cases in Example 1.6.2, we have a spe-
cial notation for S−1 A. When S = {an , n > 0} we denote S−1 A =
Aa .
When S = A \ P, P a prime a ideal, we denote S−1 A = AP ; this
is the most important case of localization.
Finally when S is the set of nonzero divisors, we denote S−1 A =
F(A), and call it the total ring of fractions of A. If A is an integral
domain then S = A \ {0}, and F(A) is a field, which we shall call
the field of fractions of A.

Remark 1.6.4. Any localization of an integral domain A is con-

tained in the field F(A), hence it is again an integral domain.

The following universal property comes for free from the defi-
nition:

Proposition 1.6.5. Let f : A → B be a homomorphism such that f(S) ⊂

B∗ . Then there exists a unique homomorphism fb: S−1 A → B such that
f = fb◦ ι.

37
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 basics

So S−1 A together with the map ι is universal for such homo-

morphisms.
Ideals behave well under localization; in particular we have the
following proposition, which should be compared to Proposition
1.2.9.
Proposition 1.6.6.
i) For any ideal I ⊂ S−1 A, I is generated by ι(ι−1 (I)); in particular
every ideal of S−1 A is the extension of an ideal of A.
ii) For any ideal I ⊂ A we have
[
ι−1 (I · S−1 A) = (I : s).
s∈S

iii) The ideal I · S−1 A is the whole S−1 A if and only if I ∩ S 6= ∅.

For instance, let P ⊂ A be a prime ideal and take S = A \ P.
According to the proposition, S−1 P is the only maximal ideal of
AP , which is then a local ring:
Definition 1.6.7. A ring A is called local if it has only one maximal
ideal. It is called semilocal if it has finitely many maximal ideals.
A local ring A with maximal ideal M will sometimes be denoted
simply by the pair (A, M).
Unfortunately we don’t have a bijective correspondence be-
tween ideals of S−1 A and ideals of A not meeting S.
Example 1.6.8. In the ring Z(2) , the ideals (2) and (6) of Z extend
to the same ideal, since 3 becomes invertible after localization.
The situation is better for prime ideals:
Corollary 1.6.9. There is a bijective correspondence between prime ide-
als of S−1 A and prime ideals of A which don’t meet S.
Proof of Proposition 1.6.6.
i) Let
J := ι−1 (I) · S−1 A.
The inclusion J ⊂ I is obvious. For the converse, let a/s ∈ I;
then a ∈ ι−1 (I), so a/s ∈ J.

38
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 1.6 localization

ii) Let
J := ι−1 (I · S−1 A)
We first prove that (I : s) ⊂ J for any s ∈ S. Indeed let
a ∈ (I : s), so as ∈ I; then
 as 
a ∈ ι−1 ⊂ J.
s

For the other inclusion let a ∈ J; this means that a/1 = b/s
for some b ∈ I, s ∈ S. So there is u ∈ S such that

asu = bu ∈ I.

This shows that a ∈ (I : su).

iii) The ideal I · S−1 A is trivial if an only if it contains 1. This is

equivalent to 1 = i/s for some i ∈ I and s ∈ S; finally this
means i = su ∈ S.

Proof of Corollary 1.6.9. It remains to show that if P ⊂ A is prime,

then P = ι−1 (P · S−1 A). According to Proposition 1.6.6 we have
[
ι−1 (P · S−1 A) = (P : s)
s∈S

and (P : s) = P for all s ∈ S, since P is prime and s ∈

/ P.

We can use our new tool to prove

Proposition 1.6.10. The nilradical N(A) of A is the intersection of all

prime ideals of A.

Recall from Example 1.1.18 that N(A) is the ideal formed by

nilpotent elements of A.

Proof. Let P be a prime ideal and let a ∈ N(A), then some power
an = 0 ∈ P, so a ∈ P. This proves one inclusion.
For the other, let a be contained in every prime ideal. Then Aa
is ring without prime ideals, that is, Aa must be the 0 ring. Then
ι(1) = 0 means that 1 · an = 0 for some n, so a is nilpotent.

By considering the quotient A/I we obtain the

39
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 basics
√
Corollary 1.6.11. Let I be an ideal of A. Then I is the intersection
of all prime ideals containing I.

It is useful to know that localization and quotients commute,

in the following sense.

Proposition 1.6.12. Let S be a multiplicative set, I ⊂ A an ideal not

meeting S. Then there is a natural isomorphism between S−1 A/S−1 I
and T −1 (A/I), where T is the image of S inside A/I.

Proof. It is clear that T is a multiplicative set. The composition of

the natural maps

A → S−1 A → S−1 A/S−1 I

has kernel ι−1 (S−1 I) = I by Proposition 1.6.6. So we have an

injective homomorphism

A/I → S−1 A/S−1 I.

Since this sends T into invertible elements, the universal property

of localization yields a homomorphism

φ : T −1 (A/I) → S−1 A/S−1 I.

Vice versa, consider the composition

A → A/I → T −1 (A/I).

This sends S into invertible elements, so we get a homomorphism

S−1 A → T −1 (A/I),

whose kernel contains S−1 I. Hence we have a map in the other

direction
ψ : S−1 A/S−1 I → T −1 (A/I).
It is immediate to check that φ and ψ are mutual inverses.

Corollary 1.6.13. Let P ⊂ A be a prime. Then the fields of fractions

of the integral domain A/P is canonically isomorphic to AP /PAP , the
quotient of the local ring AP by its maximal ideal.

40
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 1.6 localization

Definition 1.6.14. The field

k(P) = F(A/P) = AP /PAP

is called the residue field of A at P.

Modules can be localized in the same way as rings.
Definition 1.6.15. Let A be a ring, S a multiplicative set and M
an A-module. Then we define the S−1 (A)-module S−1 M as fol-
lows. Elements of S−1 M are couples (m, s) ∈ M × S, modulo an
equivalence relation. The couples (m, s) and (n, t) are equivalent
when there exists u ∈ S such that

mtu = nsu.

The equivalence class of (m, s) is denoted by m/s.

As in the ring case, we have a homomorphism of A-modules

ιM : M → S−1 M

given by ιM (m) = m/1.

The same proof of Proposition 1.6.6 yields:
Proposition 1.6.16. Let M be an A-module.
i) For any submodule N ⊂ S−1 M, N is generated by ι(ι−1 (N)); in
particular every submodule of S−1 M is the extension of a submod-
ule of M.
ii) For any submodule N ⊂ M we have
[
ι−1 (S−1 N) = (N : s).
s∈S

Here S−1 N denotes the extension of N in S−1 M, that is, the

S−1 A submodule generated by N. The submodule (N : s) is
defined by
(N : s) := {m ∈ M | ms ∈ N}.
Finally we have the following useful results.
Proposition 1.6.17. Localization of modules preserve exact sequences.

41
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 basics

Proof. Let
n αn−1 α
· · · → Mn−1 −−−→ Mn −−→ Mn+1 → · · ·

be an exact sequence of A- modules; then we have an induced

sequence of S−1 A-modules
αn−1 α
· · · → S−1 Mn−1 −−−→ S−1 Mn −−→
n
S−1 Mn+1 → · · · .

Let us check exactness at Mn .

First for any m/s ∈ Mn−1 we have

αn (αn−1 (m/s)) = αn (αn−1 (m))/s = 0.

Vice versa, let m/s ∈ Mn and assume that αn (m/s) = αn (m)/s =

0. Then there exists u ∈ S such that

αn (um) = uαn (m) = 0.

So we have um = αn−1 (m 0 ) for some m 0 ∈ Mn−1 and finally

m/s = αn−1 (m 0 /(su)).
Proposition 1.6.18. Let M be an A-module, m ∈ M. Then m = 0
if and only if ιP (m) = 0 ∈ MP for every maximal (or prime) ideal
P ⊂ A.
Proof. One implication is obvious. For the other assume that
ιP (m) = 0 for all maximal ideals P ⊂ A. This means that for
every such P we find some s ∈ / P such that sm = 0. In other
words the ideal Ann(m) is not contained in any maximal ideal,
so it must be the whole A.
This has some useful corollaries
Corollary 1.6.19. Let M be an A-module; then M = 0 if and only if
MP = 0 for every maximal (or prime) ideal P ⊂ A.
Corollary 1.6.20. Let f : M → N be an A-module homomorphism.
Then f is injective (resp. surjective) if and only if the induced homomor-
phism fP : MP → NP is injective (resp. surjective) for every maximal
(or prime) ideal P ⊂ A.
Proof. Use the fact that localization preserves exact sequences
and apply the above corollary to ker(f) (resp. coker(f)).

42
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 1.7 graded rings and modules

Properties like the above, which hold true for a ring (or a mod-
ule, or a homomorphism. . . ) if and only if they hold for the
localization at every prime ideal are called local properties. We
will see more examples in the next chapters.

1.7 graded rings and modules

We begin with some routine definitions.
Definition 1.7.1. An abelian monoid is a set G endowed with a
binary operation +, called addition, such that
1. addition is associative, namely for every g, h, l ∈ G we have
(g + h) + l = g + (h + l);
2. there exists an element 0 ∈ G such that 0 + g = g + 0 = g
for every g ∈ G;
3. addition is commutative, that is for every g, h ∈ G we have
g + h = h + g.
In down to earth terms, an abelian monoid is just an abelian
group without the requirement of the existence of inverse ele-
ments.
Remark 1.7.2. Since inverses don’t exist in G, one does not have
a cancellation law!
Example 1.7.3. (a) Any abelian group is an abelian monoid.
(b) The set N of natural numbers is a monoid which is not a
group; it will be our main example. More generally Nk is a
monoid.
(c) We can add to N a symbol ∞ and give N ∪ {∞} the structure
of an abelian monoid by declaring that
n+∞ = ∞+∞ = ∞
for every n ∈ N.
(d) For an example quite different from the monoids we will con-
sider, A∗ is an abelian monoid under multiplication for every
ring A.

43
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 basics

The reason why we have introduced monoids is the following

Definition 1.7.4. Let A be a ring. We say that A is graded over the

monoid G if we have a decomposition
L
A = g∈G Ag

as abelian groups such that

ab ∈ Ag+h for a ∈ Ag , b ∈ Ah .

Elements of some Ag will be said homogeneous. Every element

can be written uniquely as a finite sum of nonzero homogeneous
elements, which are called it homogeneous components.
If we speak of a graded ring without further specification we
always mean graded over N.

Remark 1.7.5. If A is a graded ring, A0 is a subring and A has

the structure of A0 -algebra.
In the case where A is graded over N, the subset
L
A+ := n>1 An

is an ideal of A, which – for geometric reasons which we shall see

later (Exercise 2 in Chapter 8) – is sometimes called the irrelevant
ideal.

Example 1.7.6. (a) The most fundamental example of a graded

ring is k[x1 , . . . , xn ]. Usually it is graded over N by con-
sidering the total degree, but it has also the structure of a
Nn -graded ring by considering a separate degree for each
variable.

(b) For any topological space X the cohomology ring H∗ (X) is

graded over N. The complex K-theory ring K∗ (X) is graded
over Z/2Z by Bott periodicity.

(c) Let A be any ring, I ⊂ A an ideal. The graded ring associated

to I is

GrI (A) := ∞ k k+1 = A/I ⊕ I/I2 ⊕ · · · .

L
k=0 I /I

Check that the operations are well-defined!

44
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 1.7 graded rings and modules

In the case where (A, M) is a local ring, we simply write

Gr(A) for GrM (A). This association is at the source of a deep
connection between the structure of graded and local rings,
which will emerge when we study the concept of dimension
in Chapter 9.
In the context of graded rings there are some ideals and mod-
ules which are particularly well-behaved.
Definition 1.7.7. Let A be a G-graded ring. The ideal I ⊂ A is
called homogeneous if L
I = g∈G Ig ,
where Ig := I ∩ Ag . This amounts to saying that if a ∈ I, the
homogeneous components of a are still in I.
Note that if I and J are homogeneous, so is I ∩ J.
Proposition 1.7.8. The ideal I is homogeneous if and only if it gener-
ated by homogeneous elements.
Proof. Assume I is homogeneous. Take any set of generators of I;
then the set of their homogeneous components still generates I.
Vice versa assume I = (E), where E is composed of homoge-
neous elements. Let a ∈ I and write

a = e1 f1 + · · · + ek fk

for some ei ∈ E and fi ∈ A. Taking the homogeneous component

of both members and taking in account that the ei are homoge-
neous shows that each homogeneous component of a is again a
combination of the ei , so it lies in I.
Corollary 1.7.9. If I and J are homogeneous ideals, so are I + J and
I · J.
Definition 1.7.10. Let A be a ring graded over G, M an A-module.
We say that M is graded if we have a decomposition
L
M = g∈G Mg

as abelian groups such that

am ∈ Mg+h for a ∈ Ag , m ∈ Mh .

45
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 basics

As for rings we have the definition of homogeneous elements,

homogeneous components and so on. A graded submodule N ⊂ M
is a submodule such that
L
N = g∈G Ng ,
where Ng := N ∩ Mg .
As before, if A is any ring, I ⊂ A an ideal and M an A-module
we can define the associated graded module
GrI (M) := ∞ k k+1 M = M/IM ⊕ IM/I2 M ⊕ · · · ;
L
k=0 I M/I

this is a module over GrI (A). When (A, M) is local we will write
Gr(M) for GrM (M).
Definition 1.7.11. More generally, we call a nested sequence of
submodules of M ⊃ M1 ⊃ · · · ⊃ Mn ⊃ · · · a filtration and we de-
fine the associated graded module
Gr(M) := ∞
L
k=0 Mn /Mn+1 .

Definition 1.7.12. Let f : M → N be a homomorphism of G-

graded A-modules. We say that f is graded of degree h if f(Mg ) ⊂
Ng+h .
In the most common case h = 0, so we simply ask that f(Mg ) ⊂
Ng ; that is, f preserves the degree of homogeneous elements.
In this case we will simply speak of a graded homomorphism,
without mentioning the degree.
Example 1.7.13. (a) Write
Hom(M, N)g := {f : M → N | f is homogeneous of degree g}.
Then we have the submodule
L
Hom(M, N)gr := g∈G Hom(M, N)g
of Hom(M, N), which is a graded module.
(b) If M and N are graded over N, the tensor product is also
graded. Indeed we define
L
(M ⊗ N)n := p+q=n Mp ⊗ Nq .
Then one can check for exercise that
L
M ⊗ N = n∈N (M ⊗ N)n .

46
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 1.8 exercises

When f : M → N is a graded homomorphism, the image f(N) is

a graded submodule of M. If N ⊂ M is a graded submodule, the
quotient M/N has a natural structure of graded module, since
L
M/N = g∈G Mg /Ng .

In particular the quotient of a graded ring for a homogeneous

ideal is again a graded ring.

Remark 1.7.14. This does not mean that m ∈ (M/N)g if and only
m ∈ Mg ! A homogeneous class can have various inhomogeneous
representatives, and even homogeneous representatives of differ-
ent degrees. The only case where the preceding assertion is true
is when N = N0 .

The following result is often useful.

Proposition 1.7.15. Let A be a graded ring (over N). If the ideal A+

is finitely generated, A is finitely generated as A0 -algebra.

Proof. Let A+ = (a1 , . . . , an ); since A+ is homogeneous we can

take the ai to be homogeneous. We shall show that

A = A0 [a1 , . . . , an ].

Let B := A0 [a1 , . . . , an ]; we show that Ak ⊂ Bk by induction, the

case k = 0 being trivial.
If k > 0, Ak ⊂ A+ , so given a ∈ Ak we can write

a = b1 a1 + · · · + bn an

for some b1 , . . . , bn ∈ A. We can choose the bi homogeneous,

since the components in degree different from k will cancel out;
moreover the bi will have degree less than k. By induction
b1 , . . . , bn ∈ B, so we conclude that a ∈ B.

1.8 exercises
1. Any (not necessarily commutative) ring is a subring of a ring
with unit. (Give the abelian group Z ⊕ A a suitable ring struc-
ture.)

47
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 basics

2. Let A be a (not necessarily commutative) ring. Prove that A

is a subring of End(E) for some abelian group E. (Consider the
action of A on itself by multiplication.)

3. Let V be a vector space of infinite countable dimension, A =

End(V) the (noncommutative) ring of linear endomorphisms of
V. Show that for any m, n > 1, An and Am are isomorphic as
modules over A. Hence Proposition 1.4.14 fails in the noncom-
mutative case.

4. Give an example of a module – which is not finitely generated

– that does not satisfy the conclusion of Nakayama’s lemma.

5. Let A be a ring such that for all a ∈ A there exists n (depend-

ing on a) for which an = a. Prove that every prime ideal of A is
maximal.

6. Show that a finite integral domain is a field.

7. Let A be the ring of Example 1.1.5 i. Prove that an element

f ∈ A is invertible if and only if f(1) 6= 0.
Let µ be the inverse of the constant function 1 (µ is usually
called the Möbius function). Let r(n) be the number of primes
dividing n. Prove that

(−1)r(n) if n is squarefree
µ(n) =
0 if n is not squarefree.

Prove the Möbius inversion formula: if

X
g(n) = f(d),
d|n

then X n
f(n) = g(d)µ .
d
d|n

8. The only maximal ideals of A = C([0, 1]) are those of the form

Ix = {f ∈ A | f(x) = 0}

for some x ∈ [0, 1].

48
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 1.8 exercises

9. Formulate and prove the Chinese remainder theorem for more

than two ideals.

10. Show that if the ideals I and J are coprime then I ∩ J = I · J.

11. Let I ⊂ A = k[x1 , . . . , xn ] be an ideal generated by monomials;

we shall say that I is a monomial ideal. Let f ∈ I; prove that any
monomial appearing in f is in I. Assume

I = (xαi ),

where each αi is a multiindex. A monomial xβ is in I if and only

if there is some αi such that xαi divides xβ .

12. Let
0 → M1 → M2 → M3 →0

f1 f2 f3
↓ ↓ ↓
0 → N1 → N2 → N3 →0
be a map of exact sequences of A-modules, in the sense that all
displayed maps are A-linear, the rows are exact sequences and
the diagram commutes. Show that we have a long exact sequence

0 → ker f1 → ker f2 → ker f3 →

→ coker f1 → coker f2 → coker f3 → 0.

This result is sometimes called the snake lemma.

13. Let P be a finitely generated A-module. The following condi-

tions are equivalent:

i) there exists an A-module N such that

∼ An
P⊕N =

for some n;

ii) for all morphisms of A-modules

f: S → T, g: P → T

with f surjective there exists h : P → S such that g = f ◦ h.

49
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 basics

In this case M is called projective. In categorical style, the second

condition means that one can always complete the dashed arrow
in the diagram

h
g
↓
f
←
S →T → 0.
14. Show that a finitely generated torsion-free module over A =
k[x] is free.
15. For a converse, show that the ideal (x, y) is a torsion-free
module over A = k[x, y], but it is not free.
16. This exercise uses some topology. Let X be a topological
space, E a vector bundle on X. The continuous sections of E
are a module over the ring A = C(X).
Now take X = S2 , the 2-sphere, and let E = T S2 be its tangent
bundle. Prove that the associated module M is not free, but M ⊕
A is free.
17. We work out an algebraic version of the above exercise.Let

A = R[x1 , x2 , x3 ]/(x21 + x22 + x23 − 1).

Inside A3 consider the submodules

M = {(f1 , f2 , f3 ) | x1 f1 + x2 f2 + x3 f3 = 0}
N = {(x1 f, x2 f, x3 f) | f ∈ A}.
∼ A and that
Prove (possibly algebraically) that N =

∼ A3 ;
M⊕N =

still M is not free.

Let M ⊂ A be the ideal generated by x1 − 1, x2 , x3 . Prove that
M is maximal, in particular prime, and that MM is free over AM .
18. This exercise uses a little field theory. Let k 0 /k be a finite
separable extension of fields (this means that for all α ∈ k 0 the

50
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 1.8 exercises

minimum polynomial of α over k has distinct roots). Prove that

k 0 = k[α] for some α ∈ k 0 .
(Reduce to the case k 0 = k(α, β), and prove that k 0 = k(γ)
where γ = α + tβ for a suitable t ∈ k. For this let f and g be the
minimum polynomial of α and β respectively. What is the maxi-
mum common divisor of f(γ − tx) and g(x)? Use the Euclidean
algorithm to prove that this maximum common divisor is still in
k(γ)[x].)
h √ i
19. Prove that the ring A = Z 1+ 2−19 is not Euclidean. We
shall see later (Exercise 5 of Chapter 6) that it has nevertheless
the property that all its ideals are principal. (First describe A∗ .
Assuming that A has a norm function, let a ∈ A \ A∗ with min-
imum norm; prove that A/(a) is formed by the classes of 0 and
the invertible elements. How big can A/(a) be? Derive a contra-
diction.)
20. Let k be a field of characteristic 0, and let A = k[x, y]. Find
generators for (x2 , y) ∩ (x, y2 ).
21. Let A be a ring, S a multiplicative set, and consider the family
F = {I ⊂ A ideal | I ∩ S = ∅}.
Let I ∈ F be a maximal element; prove that I is prime.
22. Let A be an integral domain, so that all localizations AP at
prime ideals can be seen as subgrings of k = F(A). Show that
\ \
A= AM = AP .
M maximal P prime

23. [vDdB] Let k be a field, and K := k(x1 , x2 , . . . ) the field of

rational functions in infinitely many indeterminates. Consider
the rings A := K[y1 , y2 , . . . ] and B := Ay1 = A[y−1
1 ]. Show that
A is a localization of B, and conversely, but A is not isomorphic
to B. (Notice that A∗ ∪ {0} is closed under addition.)
24. Show that the ring C{x1 , . . . , xn } of power series in n variables
which have a positive radius of convergence is local.
25. Let A be a ring graded over Z, and assume that A does not
have any nonzero prime homogeneous ideals. Then A0 is a field
and either A = A0 or A =∼ A0 [x, x−1 ].

51
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 basics

26. Consider the ring A = k[x, x−1 ], where deg xn = n. Any

graded module over A is free. (Imitate the corresponding proof
for vector spaces.)

27. Let A = m∈N Am beLa ring graded over N with A0 a field,

L
and take a truncation B = m∈N Amk . Assume that A is finitely
generated as an A0 -algebra and it is an integral domain. Then B
is generated in degree 1 for k large enough.
28. Let P ⊂ A be a homogeneous ideal. Then P is prime if and
only if for every homogeneous a, b ∈ A such that ab ∈ P we have
a ∈ P or b ∈ P.
29. The radical of a homogeneous ideal is again homogeneous.
Q∞ ∼
30. Let A = ∞ n=0 Z and A = n=0 Z. Prove that Hom(A, Z) =
L
∼
B and Hom(B, Z) = A. (The first one is easy. For the second
one, show that the map Hom(B, Z) → B sending f to (f(ei )) is
injective with image A. For this, it will be useful to notice that
every integer x can be written in the form x = an 2n + bn 3n , for
any n ∈ N.)

52
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 2
FINITENESS CONDITIONS

In this chapter we introduce the most common finiteness condi-

tions one can put on rings or modules, the so called chain condi-
tions. Basically, we ask that our modules don’t have infinite (on
the right or on the left) chains of submodules. In the Section 2.2,
we study this conditions on modules; here the direction is not
relevant as the situation is symmetric.
For rings this symmetry breaks up. First we study the class of
Noetherian rings, that is rings that don’t have infinite ascending
chains of ideals. This is by far the most common class of rings
one encounters in commutative algebra. The condition turns out
to be equivalent to the request that ideals are finitely generated;
indeed, to give some motivation, we start with the case where all
ideals are actually principal.
Then we study Artinian rings, that is rings that don’t have
infinite descending chains of ideals. This condition turns out to
be much more restrictive, and indeed we show that an Artinian
ring is Noetherian as well. Moreover we can give a quite precise
structure theorem for this class of rings.
Finally, we turn to modules which are both Artinian and Noe-
therian: for these modules we have a notion of length, which
generalizes the concept of dimension for vector spaces. We also
prove the classical Jordan-Hölder theorem, which compares dif-
ferent series of composition for the same module.

2.1 principal ideal domains

Much of the complication in studying rings more general than
say Z or k[x] comes from the fact that ideals can have a nontrivial

53
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 finiteness conditions

structure. From this point of view, the most fortunate case is that
of rings whose ideals are all generated by a single element.
Definition 2.1.1. Let A be an integral domain. We say that A is a
principal ideal domain if all the ideals of A are principal.
Remark 2.1.2. If I ⊂ A is any ideal of a principal ideal domain,
∼ A as A-modules. Indeed let I = (a); then the map
then I =
A →I
x → ax
is both injective and surjective.
Example 2.1.3. We shall not give many examples. We have seen
that Euclidean rings are principal ideal domains, in particular Z,
Z[i], k[x] and k[[x]].
On the other hand, it is not easy to distinguish the class of prin-
cipal ideal domains from that of Euclidean rings. The simplest
example
h √ of ia principal ideal domain which is not Euclidean is
1+ −19
Z 2 , see Exercise 19 of Chapter 1 and 5 of Chapter 3.

In this section we shall not say much about principal ideal do-
mains, the main result being Theorem 3.1.6, which states that in
a principal ideal domain we have unique factorization. We start
with a simple but important fact.
Proposition 2.1.4. Let A be a principal ideal domain. Then every
nonzero prime ideal of A is maximal.
Proof. Let P = (p) be prime, and take a ∈ / P; we only need to
show that (a, p) = A.
Let (a, p) = (b), so p = bh for some h ∈ A. Note that b ∈
/ P,
since a ∈ (b), so we have h ∈ P. This means that

p = bh = bkp

for some k ∈ A, from which we deduce bk = 1, since p 6= 0.

Hence b is invertible and (a, p) = A.
We can give a generalization of Theorem 1.5.6, characterizing
modules over a principal ideal domain. Since the proof is not
too much different, we leave this as Exercise 2. Here we give

54
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 2.1 principal ideal domains

a different argument, which yields a weaker result, but works

equally well for the infinite case. The reader who is not familiar
with ordinals and transfinite induction may try to translate the
following in a proof by ordinary induction (in the finite case).
The idea is to proceed by induction, observing that the defini-
tion of a principal ideal domain is just the thesis of the theorem
for F = A. This proof can be found in [Rot79].

Theorem 2.1.5. Let A be a principal ideal domain and let F be a free

module over A. If N ⊂ F is a submodule, then N is free.

Proof. Let
L
F= s∈S As ,

where S is any set and As is a copy of A; call fs a generator of

As . Choose a well-ordering 6 of S, and put
L
F6s := Ft
Lt6s
F<s := t<s Ft

We also let N6s := N ∩ F6s and similarly for N<s .

Then the quotient N6s /N<s is a submodule of As , hence it is
either 0 or it is generated by a single element as fs . In the latter
case, choose a vector ns ∈ N6s having such a component in As .
Call S 0 the set for which the quotient is not 0. We want to show
that {ns }s∈S 0 is a basis of N.
First, assume there is a linear combination

b1 ns1 + · · · + bk nsk = 0

with bj not all 0. We can assume k is minimal, and we can ar-

range the order so that sk is the biggest index. Then projecting
into Ask gives
bk ask fsk = 0,
thereby showing that bk = 0, contradiction. Hence the vectors
are linearly independent.
Assume by contradiction that these vectors do not generate.
Every n ∈ N lies in N6s for some s. Since S is well-ordered we
can choose a minimal s such that there exists n ∈ N6s which is
not in the span of the {ni }.

55
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 finiteness conditions

Let afs be the component of n in As ; then as |a, say a = bas ,

so
n − bns ∈ N<s .
By minimality of s, n − bns is in the span of the {nj }, and so n
is.

We cannot hope to get the stronger result that there exists a

basis {fi } of F and elements {ai } of A, such that {ai fi } is a basis
of N. Since any A-module is the quotient of a free module, this
would imply that every A-module would be the direct sum of
addends, each one isomorphic to A/(a) for some a ∈ A. This is
false, as shown by the following

Example 2.1.6. We take the ring A = Z, so A-modules are just

abelian groups. The additive group Q is divisible, that is, for
every x ∈ Q and m ∈ Z we can find y ∈ Q such that my = x. So
Q cannot be isomorphic to the direct sum of cyclic groups.

Still the conclusion is true for finitely generated free modules,

as you are asked to prove in Exercise 2, so we obtain:

Theorem 2.1.7. Every finitely generated module over a principal ideal

domain A is isomorphic to

Ar ⊕ A/(a1 ) ⊕ · · · ⊕ A/(ak )

for a suitable choice of r ∈ N and aj ∈ A.

2.2 artinian and noetherian modules

One can try to generalize the notion of a principal ideal domain
considering rings whose ideals are finitely generated. In even
greater generality one could consider modules having the prop-
erty that all their submodules are finitely generated. We start
with the following characterization.

Proposition 2.2.1. Let M be an A-module. Then the following proper-

ties are equivalent:

i) every submodule of M is finitely generated;

56
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 2.2 artinian and noetherian modules

ii) every ascending chain of submodules

N1 ⊂ N2 ⊂ N3 ⊂ · · ·

is eventually stationary.
S
Proof. Assume the first property, and let N := Ni . Since N
is generated by a finite number of elements, these all appear in
some Nk ; after k the inclusions become equalities.
Conversely, assume the second property, and let N ⊂ M be a
submodule. Take n1 ∈ N; if N = (n1 ) we are done. Otherwise
take n2 ∈ N \ (n1 ), and so on. We cannot have an infinite chain

(n1 ) ( (n1 , n2 ) ( (n1 , n2 , n3 ) ( · · · ,

hence for some k we have N = (n1 , . . . , nk ).

Definition 2.2.2. Condition ii) of Proposition 2.2.1 is called the

ascending chain condition. We can also formulate a descending chain
condition:
every descending chain of submodules

N1 ⊃ N2 ⊃ N3 ⊃ · · ·

is eventually stationary.

Definition 2.2.3. An A-module M is called Noetherian (respec-

tively Artinian) if it satisfies the ascending (respectively descend-
ing) chain condition.
The ring A is said to be Noetherian or Artinian if it is as a mod-
ule over itself, that is, it satisfies the appropriate chain condition
for ideals.

Remark 2.2.4. The characterization by finite generation has no

analogue for Artinian modules.
Although the definitions look rather similar, we will see that
being Artinian is a very special condition, at least for rings, while
the Noetherian hypothesis is at the basis of most interesting re-
sults in commutative algebra.

Example 2.2.5. (a) Every finite A-module is trivially both Artin-

ian and Noetherian.

57
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 finiteness conditions

(b) Every principal ideal domain is a Noetherian ring, in par-

ticular k[x]. Note that any ring of the form A[x] cannot be
Artinian, because of the descending sequence of ideals

(x) ⊃ (x2 ) ⊃ (x3 ) ⊃ · · · .

(c) The ring k[x1 , x2 , x3 , . . . ] is neither Artinian nor Noetherian.

(d) Consider the abelian group µn formed by all n-th roots of 1,
say inside C∗ . We let
∞
[
µp∞ := µpn .
n=1

Then µp∞ is an Artinian Z-module, but it is not Noetherian.

The fact that it is not Noetherian is easily seen by the ascend-
ing chain of subgroups
µp ⊂ µp2 ⊂ µp3 ⊂ · · · .

To prove that it is Artinian, note that any subgroup of µp∞

which contains roots of 1 of arbitrarily high order must be the
whole µp∞ . So any proper subgroup contains roots of unity
of bounded order, hence it is finite (actually it must be one of
the µpn ).
(e) The ring A = C([0, 1]) is not Noetherian. Indeed let
Ik := {f ∈ A | f(x) = 0 on [0, 1/k]};
this is a strictly ascending sequence of ideals.
In fact the Noetherian hypothesis is rather strong: we will see
that it holds often for rings coming from algebraic objects,
like polynomial rings, but it is seldom (if ever) verified for
rings of functions on topological objects. In a vague sense
it is the condition that distinguishes the good properties of
polynomial-like rings from the intractable zoo of arbitrary
rings.
There is a useful reformulation of the chain conditions. To
avoid duplication, we formulate the result for a general partially
ordered set; we will then apply it to the set of submodules of a
given module M, ordered by ascending or descending inclusion.

58
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 2.2 artinian and noetherian modules

Proposition 2.2.6. Let (S, 6) be a partially ordered set. The following

conditions are equivalent:
i) every ascending chain

s1 6 s2 6 s3 6 · · ·

is eventually stationary;
ii) every subset of S has a maximal element.
Proof. Assume ii) and take an ascending chain

s1 6 s2 6 s3 6 · · · ;

if sk is maximal in the set {si }, the sequence becomes stationary

after k.
Conversely, assume i), and let T ⊂ S be any subset. Choose
s1 ∈ T ; if s1 is maximal, we are done. If not there is some s2 > s1
in T . If this is not maximal, proceed to find s3 > s2 and so on.
Since we cannot have any infinite strictly ascending sequence, we
must eventually find a maximal element of T .
So we see that every family of submodules of a Noetherian (re-
spectively Artinian) module has a maximal (respective minimal)
element; this is an equivalent characterization.
We finish this section by studying the behaviour of Noetherian
and Artinian modules under quotients and localization. We start
with a simple
Remark 2.2.7. If M is Noetherian (respectively Artinian) and
N ⊂ M, both N and M/N are Noetherian (respectively Artin-
ian). This is just because submodules of N are also submodules
of M, and submodules of M/N correspond to submodules of M
by Proposition 1.3.11.
A useful converse is the following
Proposition 2.2.8. Let N ⊂ M. If both N and M/N are Noetherian
(respectively Artinian) then M is.
Proof. Let {Mi } be any ascending (respectively descending) se-
quence of submodules. Then both Mi ∩ N and πN (Mi ) must
eventually be constant.

59
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 finiteness conditions

The thesis follows from the simple remark that for any sub-
modules M 0 , M 00 ⊂ M the equalities

M 0 ∩ N = M 00 ∩ N and πN (M 0 ) = πN (M 00 )

imply that M 0 = M 00 , provided M 0 ⊂ M 00 .

The proof is straightforward: if m 00 ∈ M 00 , find some m 0 ∈ M 0
such that m 0 ≡ m 00 modulo N. Then m 00 − m 0 ∈ N ∩ M 00 ⊂ M 0 ,
so m 00 ∈ M 0 .
Remark 2.2.9. The request that M 0 ⊂ M 00 in the proof is nec-
essary, otherwise the conclusion of the remark is false, even for
vector spaces!!
Corollary 2.2.10. If M1 , . . . , Mk are Noetherian (respectively Artin-
ian) then M1 ⊕ · · · ⊕ Mk is.
Corollary 2.2.11. If A is Noetherian (respectively Artinian) and M
is a finitely generated A-module, then M is Noetherian (respectively
Artinian).
Proof. Indeed M is the quotient of some Ak , which is Noetherian
(respectively Artinian) by Corollary 2.2.10.
We then treat the case of localization. Let M be an A-module,
S ⊂ A a multiplicative set. Then S−1 M can be regarded both
as A-module and S−1 A-module. Clearly every S−1 A-submodule
is a fortiori an A-submodule. It follows that if M is Noetherian
(resplectively Artinian) as A-module, it is as S−1 A-module. How-
ever the converse is false, so it is important to distinguish what
is the base ring when we speak about S−1 M.
Proposition 2.2.12. Let M be a Noetherian (respectively Artinian)
A-module, S ⊂ A a multiplicative set. Then S−1 M is Noetherian
(respectively Artinian) as S−1 A-module.
Proof. By Proposition 1.6.16 every submodule of S−1 M is the ex-
tension of a submodule of M. Hence an infinite ascending (or
descending) chain of submodules of S−1 M induces an infinite
chain of submodules of M.
In particular the proposition says that every localization of
a Noetherian (respectively Artinian) ring is Noetherian (respec-
tively Artinian).

60
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 2.3 noetherian rings

2.3 noetherian rings

In this section, we study the Noetherian condition for rings. As
it turns out, the apparent symmetry that exists between Artin-
ian and Noetherian modules breaks up for rings. The condition
that a ring is Noetherian is pretty general – in fact it is a stan-
dard finiteness condition that will underlie many of our results.
On the opposite side, Artinian rings are pretty special: we will
eventually characterize them as Noetherian rings of dimension 0.
The crucial result that distinguishes the two is Hilbert’s basis
theorem, which states that if A is a Noetherian ring, the ring A[x]
is Noetherian as well. This is today a pretty routine result (we
will see many other properties that can be carried from A to the
polynomial ring A[x]), but when it was first devised it caused a
diatribe about effective and abstract methods in algebra.
Around the beginning of the 20th century, many algebraist
were concerned with rings of invariants, which are obtained by
taking an action of a finite or linear group G on k[x1 , . . . , xn ]
and considering the ring of elements that are invariant under
G. Typical examples are the determinant or the Pfaffian, seen as
polynomials over the entries of a matrix. Many results of that
period concerned finding explicit finite set generators for such
rings, that were seen as a sort of universal invariants.
By using Hilbert’s basis theorem, one can prove that many
rings constructed in this fashion are Noetherian (see for instance
Exercise 7 of Chapter 5), and in particular their ideals are finitely
generated. This nonconstructive approach is in sharp constrast
with the explicit computations in vogue at the time, and was re-
garded as highly controversial. It was the start of a revolution in
algebra that eventually lead to abstract algebra as we know today.
The key condition was then abstracted by Noether, and today we
call this property by her name. In particular, she proved that
the Noetherian property is enough to carry over the theory of
primary decomposition, that was originally developed by Lasker
for polynomial rings and their quotients. This is the approach
that we will follow in Section 3.2.
After this historical parenthesis, we begin this section by prov-
ing a partial converse of Proposition 2.2.12.

61
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 finiteness conditions

Proposition 2.3.1. Let A be a ring and take elements a1 , . . . , an ∈ A

such that (a1 , . . . , an ) = A. If every localization Aaj is Noetherian, A
is itself Noetherian.

Proof. Let Aj = Aaj , and let

ιj : A → A j

be the localization map. We claim that for any ideal I ⊂ A we

have
n
\
I= ι−1
j (I · Aj ).
j=1

One inclusion is obvious.

For the other, take and element b in the right-hand side inter-
section. Then for each j we can write
n
ιj (b) = bj /aj j ,

so in A we have
m n
aj j (aj j b − bj ) = 0.
By increasing the nj and the mj and changing the bj accordingly,
it is not restrictive to assume that all nj are equal to a fixed n and
all mj to a fixed m.
So ajm+n b ∈ I for all j. The powers am+n j generate the whole
ring, so we can write

X
n
1= cj am+n
j ;
j=1

multiplying both sides by b we deduce that b ∈ I.

Now take any ascending chain {Ik } of ideals of A. Every chain
{Ik · Aj }k∈N becomes stationary; since there are finitely many lo-
calizations Aj , the chain {Ik } becomes stationary by the claim.

Remark 2.3.2. The hypothesis of the proposition easily implies

that every localization AP at a prime ideal is Noetherian. So the
result can be seen as something just a bit weaker than saying that
being Noetherian is a local property (which is in fact false, see
Exercise 16).

62
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 2.3 noetherian rings

Both the rings Aa and AP have a geometric meaning, and the

relation between the usual notion of local property and the vari-
ant used in the proposition is made clear in the theory of schemes,
see [Har77, Prop. II.3.2].
The main result about Noetherian rings is Hilbert’s basis the-
orem. This is a cornerstone of commutative algebra, so we will
give two proofs.
Theorem 2.3.3 (Hilbert’s basis theorem). Let A be a Noetherian ring.
Then A[x] is Noetherian.
Remark 2.3.4. The analog for Artinian rings is definitely false!
Before giving the proofs, it will be convenient to establish some
notation.
Definition 2.3.5. Let A be a ring, f ∈ A[x] a polynomial. The
leading coefficient of f is the coefficient of the highest order term
in f; it is denoted by LC(f). If I ⊂ A[x] is any ideal, we let

LC(I) := {LC(f) | f ∈ I};

it is an ideal of A.
First Proof. Let I ⊂ A[x] be any ideal. Since LC(I) is finitely gen-
erated we can choose f1 , . . . , fk ∈ I such that

LC(I) = (LC(f1 ), . . . , LC(fk )).

Let n be the maximum degree of some fi , and take any f ∈ I.

If deg f > n, we can write

f = g1 f1 + · · · + gk fk + r,

where r has degree smaller than f. Just choose the gi so to cancel

the leading term of f.
Repeating this step, we see that I is generated by f1 , . . . , fn
and by the intersection I ∩ A[x]<n , where A[x]<n denotes the
polynomials of degree less than n.
The A-module A[x]<n is finitely generated, hence Noether-
ian by Corollary 2.2.11, so I ∩ A[x]<n is finitely generated as A-
module, and a fortiori as A[x]-module. We conclude that I is
finitely generated as A[x]-module, which is the thesis.

63
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 finiteness conditions

Second Proof. This second proof does not use the notion of mod-
ule. Let I ⊂ A[x] be an ideal, and choose f1 ∈ I to be an element
of I of least degree. If I = (f1 ) we are done, otherwise choose f2
of minimal degree in I \ (f1 ). If I 6= (f1 , f2 ), repeat this process to
choose f3 and so on.
We only need to show that this process eventually stops. Let
k be the first integer such that LC(f1 ), . . . , LC(fk ) generate LC(I).
We claim that I = (f1 , . . . , fk ). Indeed take any f ∈ I; if deg f <
deg fk we must have f ∈ (f1 , . . . , fk−1 ) by the choice of fk .
Otherwise deg f > deg fk . As in the first proof we can write
f = g1 f1 + · · · + gk fk + r,
where r has degree smaller than f; moreover r is in (f1 , . . . , fk ) if
and only if f is. By repeating this step enough times we find that
f ∈ (f1 , . . . , fk ).
The Hilbert’s basis theorem has some corollaries, which can be
seen as generalizations of the theorem itself. First, if A is Noether-
ian, then A[x1 , . . . , xn ] is again Noetherian. More generally every
finitely generated A-algebra is a quotient of some A[x1 , . . . , xn ],
so we get
Corollary 2.3.6. Let A be a Noetherian ring, B a finitely generated
A-algebra. Then B is Noetherian.
We conclude this section with some facts on ideals of Noether-
ian rings. These results will be generalized by the theory of pri-
mary decomposition in Section 3.2, but we will need them in next
section to compare the Artinian and Noetherian property.
Proposition 2.3.7. Any ring A has minimal prime ideals. More pre-
cisely any prime P ⊂ A contains a minimal prime ideal.
Proof. The intersection of a descending chain of prime ideals is a
prime ideal; now apply Zorn’s lemma.
By considering the quotient A/I we see that indeed there are
minimal primes in the family of primes containing a fixed ideal
I. These will be simply called the minimal primes of I. For Noe-
therian ring we also have:
Proposition 2.3.8. Let A be a Noetherian ring, I ⊂ A an ideal. Then
there are finitely many minimal primes of I.

64
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 2.4 artinian rings

Proof. Assume that this is not true, and let F be the family of
ideals which don’t satisfy the thesis. Let I ∈ F be maximal; by
taking the quotient A/I we can assume that I = 0.
In particular 0 is not prime, so there are nonzero a, b ∈ A
such that ab = 0. By maximality, (a) and (b) have finitely many
minimal primes. Any minimal prime of (0) either contains a or b,
hence it is a minimal prime of (a) or (b). So 0 has finitely many
minimal primes, contradiction.

Corollary 2.3.9. In a Noetherian ring, the nilradical is a finite inter-

section of prime ideals.

Proposition 2.3.10. Let A be a Noetherian ring, I ⊂ A an ideal, J =

√
I. Then there exists m such that Jm ⊂ I.

Proof. Let
J = (a1 , . . . , an )
and take k big enough so that ak i ∈ I for every i. Let m = kn;
then every monomial of degree m in a1 , . . . , an contains at least
one ai at the power k, hence is in I.

In particular we note that the nilradical of a Noetherian ring is

nilpotent, that is, N(A)m = 0 for some m.

2.4 artinian rings

We now turn to the study of Artinian rings. Unlike the Noether-
ian case, we will be able to obtain a very precise picture. We start
with the following easy fact.

Proposition 2.4.1. Let A be an integral domain. If A is Artinian, then

it is a field.

Proof. Let a ∈ A be nonzero and considering the descending se-

quence
(a) ⊃ (a2 ) ⊃ (a3 ) ⊃ · · · .
The sequence stabilizes, hence (am+1 ) = (am ) for m big enough.
In particular am = bam+1 for some b ∈ A; cancelling am we get
ab = 1, so a is invertible.

65
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 finiteness conditions

Corollary 2.4.2. Let A be an Artinian ring. Every prime ideal of A is

maximal.

This is already a sign that the structure of Artinian rings is

rather simple. We then investigate maximal ideals.

Proposition 2.4.3. Let A be an Artinian ring. Then A has only finitely

many maximal ideals.

Proof. Assume we have a sequence of distinct maximal ideals Mk ,

and consider the descending chain

M1 ⊃ M1 ∩ M2 ⊃ M1 ∩ M2 ∩ M3 ⊃ · · · .

The chain stabilizes, hence for some n

n
\
Mn+1 ⊃ Mi .
i=1

By Proposition 1.1.20 ii) we get that Mn+1 contains some Mk ,

hence they coincide, contradiction.

So we see that in an Artinian ring A we have J(A) = N(A), and

moreover the nilradical is a finite intersection of maximal ideals.

Lemma 2.4.4. Let A be an Artinian ring. Then N(A)k = 0 for some

k.

Proof. Consider the descending chain {N(A)k }; by the Artinian

property we have N(A)k+1 = N(A)k for k big enough. We claim
that then N(A)k = 0.
If this is not the case, consider the set if ideals I such that
I · N(A)k 6= 0; we can choose I minimal in this set. In particular
I = (a) for some a ∈ I such that a · N(A)k 6= 0. We have

a · N(A)k+1 = a · N(A)k 6= 0,

hence a · N(A) is one of the ideals considered above. By minimal-

ity we have (a) = a · N(A), so a = a · n for some n ∈ N(A).
But then
a = an = an2 = · · · = 0,
since n is nilpotent, contradiction.

66
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 2.4 artinian rings

The above lemma allows us to reduce the theory of arbitrary

Artinian rings to the local case.
Theorem 2.4.5. Let A be an Artinian ring, M1 , . . . , Mk its maximal
ideals. Then we have
∼ A/Mm1 × · · · × A/Mmk
A= 1 1
for suitable integers m1 , . . . , mk . Moreover each ring A/Mm i
i is local.
Proof. By Lemma 2.4.4 we see that
Mm mk
1 · · · M1
1
=0
for some integers m1 , . . . , mk .
The only maximal ideal which contains a power of Mi is Mi
mj
itself, by Proposition 1.1.20 ii). So for i 6= j the ideal Mm
i + Mj
i

m
is not contained in any maximal ideal, hence Mm i + Mj
i j
= A.
We can then apply the Chinese remainder theorem 1.2.12, or
better the refined version given in Exercises 9 and 10 of Chapter
1, to obtain the first part of the thesis.
The ideals of A/Mm i
i
correspond to ideals of A which contain
mi
Mi , hence the only maximal ideal of the quotient is the image
of Mi .
Corollary 2.4.6. With the notations of the previous theorem, we have
i ∼
A/Mm
i = A Mi .
Proof. Let π : A → A/Mmi
i
and ι : A → AMi be the natural maps.
mi
Since A/Mi is local with maximal ideal π(Mi ), the projection
of any element outside Mi is invertible in A/Mm i
i . This yields a
map
AMi → A/Mm i
i .
To give the inverse map we only need to check that ι(a) = 0
for a ∈ Mm i
i . From the previous proof we have
Mm mk
1 · · · M1
1
= 0.
Note that
I = Mm mi mk
1 · · · Mi · · · M1
1 [ 6⊂ Mi ,
mi
where M
[
i means that this factor has been omitted. This follows
from Proposition 1.1.20 ii). Take any i ∈ I \ Mi , and let a ∈ Mm i
i .
Then ai = 0, whence ι(a) = 0.

67
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 finiteness conditions

Remark 2.4.7. In the last corollary, we found a localization map

which coincides with a quotient map. In particular, we see that a
localization map can look drastically different from the inclusion
in a bigger ring. We are now rewarded for having developed
localization for rings more general than integral domains.
Corollary 2.4.8. Every Artinian ring is the direct product of its local-
ization at maximal ideals.
We finish this section with the connection between Artinian
and Noetherian rings. We start with a simple
Remark 2.4.9. A vector space V over k is Artinian (as k-module)
if and only if it is Noetherian. Indeed both conditions are equiv-
alent to having finite dimension over k.
A simple induction then gives the
Lemma 2.4.10. Let A be a ring, and assume that there are maximal
ideals M1 , . . . , Mr , not necessarily distinct, such that M1 · · · Mr = 0.
Then A is Noetherian if and only if it is Artinian.
Proof. Consider the A-modules

Ni = (M1 · · · Mi−1 )/(M1 · · · Mi ).

Then A is Noetherian (respectively Artinian) if and only if all Ni

are. But each Ni is annihilated by Mi , so can be regarded as a
module over ki = A/Mi .
Moreover each Ni is Noetherian (respectively Artinian) as A-
module if and only if it is as ki -module. So we conclude by the
previous remark.
We finally get our comparison theorem.
Theorem 2.4.11. Let A be a ring. Then A is Artinian if and only if it
is Noetherian and every prime of A is maximal.
Proof. If A is Artinian, then every prime of A is maximal by Corol-
lary 2.4.2. Moreover the nilradical is the intersection of finitely
many maximal ideals, and it is nilpotent by Lemma 2.4.4. So
Lemma 2.4.10 implies that A is Noetherian.
Vice versa, assume that A is Noetherian; then N(A) is nilpotent
by Proposition 2.3.10, and it is the intersection of finitely many

68
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 2.5 length

primes by Corollary 2.3.9. By our assumption these primes are all

maximal, so the hypothesis of Lemma 2.4.10 are again satisfied
and A is Artinian.

Example 2.4.12. (a) The typical picture of a local Artinian ring is

k[x]/(xn ).

(b) For more examples, one can take any Noetherian ring A and
localize it at a prime P ⊂ A. The ring AP is local with max-
imal ideal M = P · AP . Then the ring AP /Mn is local and
Artinian.
Indeed it is Noetherian, and its prime ideals correspond to
primes Q ⊂ A such that Pn ⊂ Q ⊂ P. The only such prime
ideal is P itself, so every prime of AP /Mn is maximal.

(c) Let A be an infinite direct sum of fields. Then every prime of

A is maximal, but A is not Artinian (or Noetherian).

2.5 length
We begin with a

Definition 2.5.1. Let M be an A-module. A chain for M is a

sequence
0 = M0 ( M1 ( · · · ( Mn = M
of nested submodules. The integer n is called the length of the
chain. If one cannot insert another submodule to increase the
length, the chain is called a series of composition for M.

Example 2.5.2. It is not the case that a module always has a series
of composition. The simplest example is the case where A is a
field and M is a vector space of infinite dimension. For another
example Z as a module over itself will do, since any nonzero
ideal (n) of Z properly contains another ideal, for instance (2n).

Remark 2.5.3. A chain (Mi ) is a series of composition if and only

if every quotient Mi+1 /Mi is simple, that is, it does not have any
nontrivial submodule.

69
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 finiteness conditions

Remark 2.5.4. If M is both Artinian and Noetherian, a series of

composition can be obtained as follows. Let M0 = 0 and M1 be
a minimal nonzero submodule; this exists because M is Artinian.
Then let M2 be minimal among those containing M1 and so on.
The chain (Mi ) will stabilize because M is Noetherian, so we
have the desired series of composition. We shall see soon that
this condition is also necessary.

Definition 2.5.5. If M admits a series of composition, we say that

M has finite length, and we define the length of M, denoted `(M),
to be the minimum length of a series of composition for M. After
the following result, it turns out that every series of composition
is long `(M).

Theorem 2.5.6. All series of compositions for M (if any) have the same
length. If `(M) is finite, any chain be enlarged to a series of composition.

Proof. Let (Mi ) be a series of composition for M having length

`(M). We first prove that a submodule N ( M has stricly smaller
length. Indeed let
Ni = Mi ∩ N
and consider the chain (Ni ). Since Mi+1 /Mi is simple, we either
have Ni+1 /Ni = Mi+1 /Mi or 0. So, if we omit from (Ni ) the
repeated terms, we have a series of composition for N, long at
most `(M).
Actually the length must be smaller, otherwise by induction
Ni = Mi for each i, and so N = M.
Now let (Mi ) be any chain for M, of length k. Then

`(M) > `(Mk−1 ) > . . . `(M1 ) > `(M0 ) = 0,

so `(M) > k. In particular if this is a series of composition we

must have `(M) = k by minimality, so all series of composition
have the same length.
If (Mi ) is any chain, one can enlarge it only a finite number of
times, because the length is bounded by `(M). So after a finite
number of steps we reach a series of composition.

Corollary 2.5.7. M has finite length if and only if it is both Artinian

and Noetherian.

70
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 2.5 length

Proof. We have already noted one implication in Remark 2.5.4.

For the other, assume that M has finite length. Then every chain,
either ascending or descending, is long at most `(M), hence it
stabilizes.
Remark 2.5.8. A ring has finite length over itself if and only if it
is Artinian, by Theorem 2.4.11.
We can be more precise than that.
Theorem 2.5.9 (Jordan-Hölder). Let M be an A-module of finite
length and (Mi ), (Mi0 ) two series of composition. There is a bijec-
tive correspondence between the sets of subquotients {Mi+1 /Mi } and
{Mj+1
0 /Mj0 } such that corresponding couples are isomorphic.

Proof. Let k be the minimal index such that M1 ⊂ Mk0 . Then

0
M1 ∩ Mk−1 = 0 because M1 is simple, hence we have an injective
homomorphism
M1 → Mk0 /Mk−10
.
This must be an isomorphism because Mk0 /Mk−1 0 is simple.
Let N := M/M1 , π : M → N the projection. Then N has the
filtration {Mi /M1 }i>2 and the filtration {π(Mj0 )}j6=k−1 , so we can
conclude by induction on `(M).
The length of modules behaves additively on exact sequences:
Proposition 2.5.10. Let

0 → M1 → M2 → M3 →0

be a short exact sequence of A-modules. Then

`(M2 ) = `(M1 ) + `(M3 )

(including the statement that the left hand side is finite if and only if
the right hand side is).
Proof. Assume that M1 and M3 have finite length, and take two
series of composition for them. Using the bijective correspon-
dence between submodules of M3 and submodules of M2 con-
taining the image of M1 , the two fit together to give a chain for
M2 . The associated subquotients are simple, so we get a series of
composition for M2 , hence the thesis.

71
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 finiteness conditions

In the case where A is Noetherian and M is finitely generated

we have a clear description of when the length of M is finite and
the shape of the factors that can appear. First we need a
Definition 2.5.11. Let M be an A-module. The set of associated
primes Ass(M) is defined to be
Ass(M) = {P ⊂ A prime | P = Ann(m) for some m ∈ M}.
Remark 2.5.12. The simple fact that if N ⊂ M then Ass(N) ⊂
Ass(M) will be used often in the sequel.
Proposition 2.5.13. Let P be a maximal element in the set
{Ann(m) | m ∈ M}.
Then P is a prime ideal.
Proof. Let P = Ann(m) and take a, b ∈ A such that ab ∈ P. Then
abm = 0, hence a ∈ Ann(bm). Of course P ⊂ Ann(bm), so
by maximality we must have a ∈ P or Ann(bm) = A. In the
first case we are done; in the second we have 1 ∈ Ann(bm), so
b ∈ Ann(m) = P.
Corollary 2.5.14. If A is Noetherian and M 6= 0, then Ass(M) 6= ∅.
Lemma 2.5.15. A prime P belongs to Ass(M) if and only if M has a
submodule N isomorphic to A/P.
Proof. Let P = Ann(m) be a prime, and consider the homomor-
phism
φ: A → M
given by φ(a) = am. Then the image of φ is Am, and the kernel
is P, so
A/P =∼ Am ⊂ M.
For the reverse implication we can assume that M = N = A/P;
then P = Ann(1).
We can now state the result for Noetherian rings.
Theorem 2.5.16. Let A be Noetherian and let M be a finitely generated
A-module. Then there exists a chain (Mi ) for M such that
∼ A/Pi
Mi+1 /Mi =
for some primes Pi ⊂ A.

72
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 2.6 exercises

Proof. Since M is Noetherian we can find a maximal submodule

N which satisfies the conclusion of the Theorem. Let us prove
that N = M. If N ( M, then by Corollary 2.5.14, M/N has some
associated prime P. By Lemma 2.5.15 we find some N 0 ⊂ M
containing N such that
∼ A/P.
N 0 /N =

But then also N 0 satisfies the conclusion of the Theorem, contra-

diction.
Corollary 2.5.17. Under the hypothesis of the Theorem, M has finite
length if and only if each Pi is maximal. In this case (Mi ) is a series of
composition.
Proof. By Proposition 2.5.10 it is enough to prove that `(A/P) is
finite if and only if P is maximal. Each A-submodule of A/P is
also an A/P-submodule; hence A/P has finite length if and only
if it is Artinian as a ring. By Proposition 2.4.1 an Artinian domain
is a field, so we have the first conclusion.
If all Pi are maximal, the factors Mi+1 /Mi are fields, so they
are simple; it follows that (Mi ) is a series of composition.
Corollary 2.5.18 (Generic freeness). Let A be a Noetherian integral
domain, M a finitely generated A-module. Then there exists a nonzero
f ∈ A such that Mf is free over Af .
Proof. Let (Mi ) be a series of composition as in Theorem 2.5.16,
so that Mi+1 /Mi = ∼ A/Pi . For each prime Pi 6= 0, choose a
Q
nonzero element fi ∈ Pi . Then f := fi 6= 0 because A is an
integral domain.
If Pi 6= 0, we have f ∈ Pi . This implies (A/Pi )f = 0, so the
locallization of the factor Mi+1 /Mi at f becomes trivial. This
means that Mf has a series of composition where each successive
quotient is isomorphic to Af . It is easy to check that a short
exact sequence where the last element is free splits, hence Mf is
isomorphic to a direct sum of some copies of Af .

2.6 exercises
1. Assume that A[x] is a principal ideal domain. Then A is a field.

73
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 finiteness conditions

2. Prove the analogue of Theorem 1.5.6 for principal ideal do-

mains.
3. Give examples to show that the conclusion of Proposition
2.2.12 cannot be strengthened to say that S−1 M is Noetherian
(or Artinian) as an A-module.
4. If A is Noetherian, then A[[x]] is.
5. Let M be a Noetherian A-module, f : M → M a surjective
homomorphism. Then f is an isomorphism.
6. Let M be an Artinian A-module, f : M → M an injective homo-
morphism. Then f is an isomorphism.
7. Let M be an A-module of finite length, f : M → M a homo-
morphism. Show that for m big enough there is a decomposition
M = ker fm ⊕ im fm .
8. Let A be a ring and assume that every prime ideal of A is
finitely generated. Then A is Noetherian. This result is known
as Cohen’s theorem. (Prove that there are maximal ideals among
those which are not finitely generated.)
9. Let A be a domain and assume that every prime ideal of A is
principal. Then A is a PID.
10. [Pol] A subring of a Noetherian ring is not necessarily Noe-
therian. Namely, let A ⊂ k[x, y] be the ring of those polynomials
that do not involve monomials of the form xk for some k > 0.
Show that A is a ring and exhibit an ideal that is not finitely
generated.
11. Let B = A[b1 , . . . , bn ] be a finitely generated A-algebra,
p and
M a finitely generated B-module. If b1 , . . . , bn ∈ AnnB (M),
show that M is finitely generated as A-module.
12. Let A ⊂ Q[x] be the ring of polynomials f such that f(x) ∈ Z
for all x ∈ Z. Show that Z[x] ( A ( Q[x]. Show that A is not
Noetherian – namely, the ideal

I := {f ∈ A | f(0) ∈ 2Z}

is not finitely generated.

74
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 2.6 exercises

For the next exercises, we need a little terminology. A Boolean

ring A is a ring such that a2 = a for all a ∈ A.

13. [Cla15, Prop 9.4] Let A be a Boolean integral domain. Show

∼ Z/2Z. Deduce that a prime
that A is a field, and in fact A =
ideal in a Boolean ring is maximal.

14. [Cla15, Prop 9.5] Let A be a local Boolean ring. Show that
A= ∼ Z/2Z.

15. [Cla15, Prop. 9.6] Let A be a Noetherian boolean ring. Show

that A is finite.

16. Using the previous exercises, find an example of a non-Noe-

therian ring such that all its localizations are Noetherian. Hence,
being Noetherian is not a local property.

17. Let A be a graded ring, M a graded A-module (over Z).

Prove that every prime P ∈ Ass(M) is graded, and moreover
P = Ann(m) for some homogeneous element m ∈ M.

18. Let k be a field. The ring A = k[x1 , x2 , . . .] with the relations

x21 = 0 and x2k+1 = xk for k > 0 on has a single prime (hence
maximal) ideal. Still it is not Noetherian, hence it is not Artinian.
Exhibit an explicit infinite descending chain of ideals. Show that
its analogue with a finite number of variables is Artinian.

19. Let k be a field and A a local Artinian k-algebra, with maxi-

mal ideal MA and residue field kA = A/MA . Let d be the degree
of the field extension kA /k, and let M be any finitely generated
A-module. Then
dimk M = d`A (M).

20 (By Eric Wofsey on Math Stack Exchange, see [Wof]). Let A be

a Noetherian ring. Prove that a localization S−1 A is a quotient
of A if, and only if, there is a decomposition as a direct product
A = ∼ B × C such that the B-coordinate of every element of S is
a unit in B and there is some element of S whose C-coordinate
is 0. In this case the map A → S−1 A can be identified with the
projection to B.

21 ([Jen80]). This exercise requires some model theory, see for

instance [CK90]. Let U be any ultrafilter on N that contains all

75
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 finiteness conditions

cofinite sets – in particular U is not principal. The ultraproduct

A := ZN /U satisfies all first-order propositions as Z (say over
the language {0, 1, +, ×}) by Loś’s Theorem – in particular it is a
ring and integral domain. Prove that A is not Noetherian, hence
being Noetherian is not a property expressible in first order logic.
(The element (1, 2, 4, 8, · · · ) is divisible by all powers of 2.)

22 ([Leq85]). Let A be a ring, M ⊂ A a maximal ideal, and assume

that every ideal I ⊂ M is finitely generated. Then A is Noetherian.
(Show that for an ideal I 6⊂ M we have I ⊕ M = ∼ A ⊕ (I ∩ M).)

23. Prove the following lemma of Artin and Tate ([AT51]). Let
A ⊂ B ⊂ C be rings, where A is Noetherian. Assume that C is
finitely generated as an A-algebra and as a B-module. Prove that
B is also finitely generated as an A-algebra.

The next exercises, up to Exercise 28, follow [Cla15] in laying

out a beautiful thread of related results about conditions which
guarantee Noetherianity.

24. Let M be an A-module, N a submodule of M maximal with

respect to the condition of not being finitely generated. Let P =
Ann(M/N). Prove that P is prime. (Let a, b ∈ A such that ab ∈ P.
If a, b ∈
/ P, then N + aM is finitely generated, say by k elements
{ni + ami }. Define L := {m ∈ M | am ∈ N}) and prove that
N = (n1 , . . . , nk ) + aL. Derive a contradiction.)

25. Prove Jothilingam’s theorem ([Jot00], [Nag05]): let A be a ring,

M a finitely generated module such that P · M is finitely gen-
erated for all prime ideals P of A. Then M is Noetherian. This
generalizes Cohen’s theorem from Exercise 8. (Assuming it is not,
find a submodule N maximal with respect to the condition of not
being finitely generated, and let P be the prime ideal of the pre-
vious exercise. Prove that P = Ann(m) for some m ∈ M. Show
that N + Am is finitely generated, say by k elements {ni + ai m},
and prove that N = (n1 , . . . , nk ) + PM. Deduce that P · M is not
finitely generated.)

26. Let M be an A-module, EM the family of submodules of the

form I · M for ideals I ⊂ A. Prove that submodules in EM are
finitely generated if and only if EM satisfies the ascending chain
condition.

76
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 2.6 exercises

27. Prove Formanek’s theorem ([For73]): let A be a ring, M a finitely

generated, faithful A-module. Let EM be the family of submod-
ules of the form I · M for ideals I ⊂ A. If EM satisfies the ascend-
ing chain condition, then M is Noetherian, and A is Noetherian
as well. (See the previous two exercises.)
28. Prove the Eakin-Nagata theorem ([Eak68], [Nag68]): let A ⊂ B
be rings, with B Noetherian, and assume that B is finitely gener-
ated as A-module. Then A is Noetherian as well.

77
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 3
FA C TO R I Z AT I O N

In this chapter we study how the analogue of the unique factor-

ization theorem for the ring Z can be generalized to different
rings. In the first section we study the rings that satisfy the most
obvious generalization of the unique factorization theorem. In
particular we show that principal ideal domains belong to this
class.
The next step would be to study the class of Dedekind do-
mains: those for which unique factorization holds for ideals. It
turns out that this is in many respect a more interesting class of
rings than unique factorization domains, and their introduction
is really a cornerstone of algebraic number theory. Before doing
so, however, we study the theory of primary decomposition. This
gives a factorization theorem for ideals much less precise that the
one achieved in Dedekind domains, but has the advantage that it
holds with the only assumption that the ring is Noetherian.
The section about Dedekind rings benefits from the primary
decomposition. Using this theory, we can derive the factoriza-
tion in a quick and efficient way. After having introduced Dede-
kind rings, we prove a structure theorem for modules over them,
generalizing our previous result for modules over Euclidean do-
mains.

3.1 unique factorization domains

One of the basic tools in elementary number theory is the unique
factorization theorem, which shows that prime numbers are really
the building bricks of all integers. Let us study the rings that
satisfy simplest generalization of this result.

79
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 factorization

Definition 3.1.1. Let A be a ring. We say that an element x ∈ A

is irreducible if whenever x = ab for some a, b ∈ A, at least one
among a and b is invertible.
We say that x is prime if the ideal (x) is.
Remark 3.1.2. We note that x is prime explicitly means that if
x divides a product ab, then it must divide at least one of the
factors.
Remark 3.1.3. The other implication is true for any integral do-
main; namely if x is prime it is always irreducible. Indeed assume
that x = ab; in particular x divides ab, so it must divide one of
the factors, say a. So a = ξx, and x = ξbx. Factoring x out we
see that ξb = 1, so b is invertible.
Definition 3.1.4. Let A be an integral domain. We say that A is a
unique factorization domain if
1. every irreducible element of A is prime and
2. every infinite sequence of principal ideals

(x1 ) ⊂ (x2 ) ⊂ · · · ⊂ (xn ) ⊂ · · ·

is stationary. This means that eventually we have xi+1 =

ξi xi for some invertible element ξi .
We will refer to such a sequence as a divisibility chain, so we
can rephrase 2 saying that every infinite divisibility chain stabi-
lizes. Note that this is just a weaker form of the ascending chain
condition; in particular it is satisfied by Noetherian rings.
We can now show that unique factorization domains satisfy a
unique factorization theorem. This is not surprising (and justi-
fies the name), since our requests for a ring A to be a unique
factorization domain are precisely the hypothesis used to prove
the classical theorem.
Theorem 3.1.5 (unique factorization). Let A be a unique factoriza-
tion domain, and let x ∈ A. Then we have a factorization

x = p1 · · · pn

where p1 , . . . , pn ∈ A are prime. Moreover the factors pi are unique

up to the order and up to multiplication by a unit of A.

80
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 3.1 unique factorization domains

Proof. Let us first show the existence of a factorization. If x is

prime we have nothing to do. Otherwise it must be reducible, so
x = x1 y1 where both x1 and y1 are not units. If both x1 and
y1 are prime, then we are done, otherwise we repeat the process.
This eventually must stop, otherwise we would have an infinite
divisibility chain in A.
As to the uniqueness, assume that x has two decompositions

x = p1 · · · pn = q1 · · · q m .

Then p1 divides x: since it is prime it must divide one of the

factors qi , and up to reordering we assume that p1 divides q1 .
So q1 = ξp1 , and since q1 is irreducible, ξ must be a unit.
We can fit it into q2 , so it is safe to assume that p1 = q1 . Since
A is an integral domain, we can factor them out and obtain that

p2 · · · pn = q 2 · · · qm .

Now we can conclude by induction on n.

At this point it is natural to investigate which rings are unique

factorization domains. Our first result says that principal ideal
domains are, so we have the unique factorization for a familiar
class of rings.

Theorem 3.1.6. Let A be a principal ideal domain. Then A is a unique

factorization domain.

Proof. We begin by showing that every irreducible element is

prime, so let x ∈ A be irreducible. Assume that x divides a
product ab but x does not divide a. The ideal (a, x) is principal
so (a, x) = (c) for some c ∈ A. This means in particular that
x = cd for some d ∈ A.
Since x is irreducible and d is not a unit (otherwise x would
divide a), c must be invertible. In other words (a, x) is the whole
ring, so we find m, n ∈ A such that

ma + nx = 1. (3.1.1)

Multiplying (3.1.1) by b we get

mab + nxb = b;

81
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 factorization

since x divides ab, we see that it must divide b.

For the other part, assume that

x1 , x2 , . . . , xn , . . .

is an infinite divisibility chain, and let I = (x1 , x2 , . . . ). Since A is

a principal ideal domain, we must have I = (a) for some a ∈ A.
So a admits a finite writing as

a = x1 y1 + · · · xn yn .

By the divisibility condition, every summand is a multiple of xn ,

so a is. Since vice versa a divides xn we see that I = (xn ).
But then if m > n both xm divides xn and conversely; this
means that the chain stabilizes.

By now we have a reasonable amount of examples of unique

factorization domains, like Z, Z[i] or k[x] for every field k. But
these are all principal ideal domains and we would like to extend
the unique factorization to more general rings. In order to do
this we need to generalize the classical Gauss’ lemma, which is
of interest itself. First let us give some definitions.

Definition 3.1.7. Let A be a unique factorization domain, and

take x, y ∈ A. The greatest common divisor of x and y, denoted by
GCD(x, y) is the product of all common irreducible factors of x
and y. As usual if a prime appears both in x and y, its multiplicity
within GCD(x, y) is the minimum of the multiplicities that it has
in x and y.

Of course the greatest common divisor is determined up to a

unit.

Remark 3.1.8. It is easy to see that GCD(x, y) is determined by

the following properties:

1. GCD(x, y) divides both x and y;

2. if a divides both x and y, then it divides GCD(x, y).

The generalization to the greatest common divisor of three or

more elements is straightforward.

82
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 3.1 unique factorization domains

Definition 3.1.9. Let A be a unique factorization domain and take

f(x) ∈ A[x]. The content of f, denoted by cont(f), is the greatest
common divisor of its coefficients. We say that f is primitive if
cont(f) = 1.
In order to prove Gauss’ lemma, we start with some lemmas
about the behaviour of the content in products.
Lemma 3.1.10. Let f, g ∈ A[x] be primitive polynomials. Then the
product fg is primitive too.
Proof. Assume that a prime p of a divides all the coefficients of
fg; then fg = 0 in A/(p)[x]. Since p is prime, A/(p) is an integral
domain, so A/(p)[x] is an integral domain too. This means that
either f = 0 or g = 0, so f and g cannot be both primitive.
Corollary 3.1.11. Let f, g ∈ A[x]. Then

cont(fg) = cont(f) cont(g).

Proof. Write f = cont(f)f1 and g = cont(g)g1 , where f1 and g1

are primitive.
We are now ready to prove Gauss’ lemma.
Theorem 3.1.12 (Gauss’ lemma). Let A be a unique factorization
domain with field of fractions k. Take f, g ∈ A[x] with f primitive, and
assume that f divides g in k[x]. Then f divides g in A[x].
Proof. Assuming the hypothesis we have

g(x) = f(x)h(x)

where h has coefficients in k. Clearing out denominators we find

mg(x) = f(x) · (mh(x))

for some m ∈ A, where now mh(x) ∈ A[x].

By the corollary we find

m cont(g) = cont(f) cont(mh) = cont(mh),

since f is primitive. In particular m divides cont(mh), so that

h(x) actually has coefficients in A.

83
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 factorization

Corollary 3.1.13. Let A be a unique factorization domain with field of

fractions k. Let f ∈ A[x] be a primitive polynomial. If f is irreducible
over A, it is irreducible also over k.

The above corollary is most effective when coupled with the

following result.

Proposition 3.1.14 (Eisenstein’s criterion, [Eis50]). Let A be a ring,

P ⊂ A a prime ideal. Take a polynomial f(x) ∈ A[x] such that

f(x) = an xn + · · · + a1 x + a0 ,

where an ∈ / P2 . Then f is irreducible

/ P, all other ai ∈ P and a0 ∈
over A.

Proof. Assume that f(x) = g(x)h(x) and reduce this modulo P.

The reduction of f(x) is just an xn , so

an xn = g(x) · h(x).

Since A/P is an integral domain we see that g(x) = bxk for some
k < n, so
g(x) = bk xk + · · · + b1 x + b0 ,
with bi ∈ P for i < k, in particular b0 ∈ P. A similar remark
holds for h, so we see that a0 ∈ P2 .

More generally reduction modulo P, with an ad hoc argument,

may prove irreducibility of a given polynomial even if Eisen-
stein’s criterion doesn’t apply.
Our next theorem, together with Theorem 3.1.6 is the main tool
to find examples of unique factorization domains. Before proving
it we need a lemma.

Lemma 3.1.15. Let A be a unique factorization domain with field of

fractions k, and take f(x) ∈ A[x]. Then f is irreducible in A[x] if and
only if it is primitive and irreducible in k[x].

Proof. Assume that f is irreducible in A[x]. Then of course it must

be primitive, otherwise it has a constant factor. To show that it is
irreducible in k[x] assume that

g(x) = f(x)h(x)

84
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 3.1 unique factorization domains

for some g, h ∈ k[x] of positive degree. Multiplying g by a con-

stant and dividing h by the same constant, we can arrange that g
is in A[x] and primitive. By Gauss’ lemma we find that h ∈ A[x],
so f is reducible in A[x] too. The converse implication is obvi-
ous.
Theorem 3.1.16. Assume that A is a unique factorization domain.
Then A[x] is.
Proof. Let f(x) ∈ A[x] be an irreducible polynomial. To show
that it is prime assume that f(x) divides a product g(x)h(x), with
g, h ∈ A[x]. Let us call k the field of fractions of A. By Lemma
3.1.15 f is irreducible in k[x], so it is prime since we already know
that k[x] is a unique factorization domain.
It follows that f divides one of the factors in k[x], say g. Ap-
plying Lemma 3.1.15 again we see that f is primitive, so Gauss’
lemma tells us that f divides g in A[x].
Next we consider an infinite divisibility chain

f1 , f2 , . . . , fn , · · · ∈ A[x]. (3.1.2)

Let ci = cont(fi ); then we easily see that c1 , c2 , . . . is an infinite

divisibility chain in A. Hence it must stabilize, so up to trimming
a finite number of terms in the sequence (3.1.2) we can assume
that the fi have all the same content.
On the other hand we know that the chain in (3.1.2) will stabi-
lize in k[x], so there exist constants ξi ∈ k such that eventually
fi+1 = ξi fi . Since the fi are in A[x] and all have the same content
we easily see that the constants ξi ∈ A∗ , so the chain stabilizes in
A[x].
Corollary 3.1.17. If A is a unique factorization domain, then also
A[x1 , . . . , xn ] is.
Example 3.1.18. With this result at hand we see that rings like
k[x, y] or Z[x] are unique factorization domains, while they are
not principal ideal domains.
Another nontrivial example is A = k[x1 , x2 , . . . , xn , . . . ], that is,
a polynomial ring in an infinite number of indeterminates. To see
that A is a unique factorization domain we just have to remark
that both the conditions defining unique factorization domains
will involve only a finite number of variables.

85
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 factorization

Explicitly, take f ∈ A irreducible, and assume that f divides

a product gh. Then the polynomials f, g and h will live inside
a smaller ring k[x1 , . . . , xn ], and f will remain irreducible there,
hence prime. It follows that f divides one among g and h in the
smaller ring, and so a fortiori in A.
The argument for the divisibility chain is similar. Given a chain

f1 , f2 , . . . , fn , · · · ∈ A

we see that, since all fi divide f1 , they all live inside a ring
k[x1 , . . . , xn ], so we are done like before.

We end this section with other characterizations of unique fac-

torization domains. They are all consequences of the simple ob-
servation that an irreducible element of a ring that admits a factor-
ization into primes is itself prime.
The following proposition justifies the name of unique factor-
ization domains.

Proposition 3.1.19. An integral domain A is a unique factorization

domain if and only if every element a ∈ A factorizes uniquely as a
product of primes, up to the order and units.

Proof. One implication is Theorem 3.1.5. For the other, note that
any irreducible element is prime, by the above observation. More-
over, the unique factorization into primes easily implies that a
divisibility chain stabilizes. Indeed in any divisibility chain the
number of prime factors decreases.

The following result gives a much more useful criterion to es-

tablish that a ring is a UFD.

Theorem 3.1.20 (Kaplansky). Let A be an integral domain. Then A

is a unique factorization domain if and only if every nonzero prime ideal
contains a nonzero prime element.

Proof. Assume A is a unique factorization domain and take a

prime P ⊂ A. For any a ∈ P, a 6= 0, factorize a as

a = p1 · · · pn ,

with the pi primes. Then one of the pi belongs to P.

86
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 3.2 primary decomposition

Conversely, let S be the multiplicative set consisting of all ele-

ments which factorize as a product of nonzero primes. We want
to show that S = A \ {0}.
Assume we can find a nonzero element x ∈ / S. Then first note
that S ∩ (x) = ∅. Indeed assume

xy = p1 · · · pn ;

each prime divides either x or y, and assuming y is not invertible,

in which case x ∈ S, at least one prime divides y, say pn . Then
we can divide both sides by pn and proceed by induction. After
a finite number of steps we find x ∈ S, which we have excluded.
So the ideal (x) remains nontrivial in S−1 A, in particular it is
contained in a maximal ideal. The inverse image of this maximal
ideal is a maximal ideal M of A such that M ∩ S = ∅. This is a
contradiction, since by hypothesis M contains a prime element.

As a consequence of Kaplansky theorem, we recover again the

result that every principal ideal domain is a unique factorization
domain.

3.2 primary decomposition

In this section we will see a far reaching generalization of unique
factorization, which is valid for arbitrary Noetherian rings. Our
treatment follows [AM69], which is slightly more general than
other sources, proving the uniqueness theorems for arbitrary rings.

Example 3.2.1.√The simplest ring in which unique factorization

fails is A = Z[ −5]. Indeed in this ring we have
√ √
6 = 2 · 3 = (1 + −5)(1 − −5).

It is not difficult to check by hand that all elements involved in

these two factorization are irreducible, and that the two
√ factoriza-
tions are essentially distinct, meaning that 2 and 1 ± −5 do not
differ by an invertible element.

87
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 factorization

A clever way to check this without computations is the follow-

√
ing. The ring√A has an automorphism · sending u = a + b −5
to u := a − b −5. Therefore the map

N: A →Z
u → uu
is multiplicative. Assume that for instance 2 factors in a nontriv-
ial way in A; since N(2) = 4 we see that a factor α must have
N(α) = 2, which is easily seen to be impossible in A. Similarly
all other factors are irreducible.
√
To check that 2 and 1 ± −5 do not differ by an invertible
element observe first that if α is invertible, it must
√ satisfy N(α) =
±1. But then the fact that N(2) = 4 and N(1 ± −5) = 6 shows
that they are indeed distinct irreducible factors.

By the above example, we see that unique factorization can fail

even in nice rings. Note that when we speak about unique fac-
torization we always consider equivalent elements which differ
by an invertible; in other words generators of the same principal
ideal. This might lead to the idea that the right generalization
of unique factorization is unique factorization for ideals. That is,
we might hope that under mild hypothesis every ideal could be
written uniquely as the product of prime ideals.1
This is actually true in a class of rings called Dedekind rings, but
for our purposes they are still too special.

Example 3.2.2. (a) Consider the ring A = k[x, y], and the ideals
P = (x, y) and I = (x, y2 ). Then P is prime and P2 ⊂ I ⊂ P so
we cannot hope to factorize I as the product of prime ideals.

(b) In the ring A = k[x, y, z] consider the prime ideals P = (x, y)

and Q = (y, z), and let I = P ∩ Q. Then y ∈ I, but y ∈ / P · Q.
Again, since
P · Q ⊂ I ⊂ P, Q
there is no way to factorize I as a product.
1 By the way, this is the historical origin of the concept
√ of an ideal. Dedekind
proved that such a unique factorization holds in Z[ −5], and so thought that
unique factorization could be restored provided one allowed the use of “ideal
elements”.

88
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 3.2 primary decomposition

The examples above show three phenomena. First, even if in a

unique factorization domain, unique factorization for ideals can
fail. Second, one of the obstacles in the factorization is the exis-
tence of ideals lying between two consecutive powers of a prime.
Third, even for radical ideals the fact that the intersection and
product differ can be an obstacle.
The solution is the following: we enlarge the class of “building
blocks” from prime powers to something better behaved, and we
take intersection instead of product as the main operation.

Definition 3.2.3. Let Q ⊂ A be an ideal. We say that Q is primary

if xy ∈ Q implies that either x ∈ Q or some power yn ∈ Q.
√
Remark 3.2.4. If Q is primary its radical P = Q is a prime. In
this case we say that Q is P-primary.

Remark 3.2.5. As for prime and maximal ideals, there is charac-

terization in terms of the quotient ring. Namely a proper ideal
Q ( A is primary if and only if every zero divisor in A/Q is
nilpotent.

Although this concept is our replacement for the prime powers,

it is not true that a prime power is necessarily primary. Still
√
Proposition 3.2.6. Let I ⊂ A an ideal and assume that M = I is
maximal. Then I is primary.

Proof. Every prime ideal containing I must contain its radical, so

the only prime containing I is M. In particular A/I is local with
maximal ideal m = M/I. So every non-invertible element of A/I
lies in m, hence it is nilpotent.

In particular this applies to powers of a maximal ideal.

Example 3.2.7. (a) The example above of I = (x, y2 ) inside A =

k[x, y] shows that a primary ideal is not necessarily a prime
power. Indeed I is primary since A/I =∼ k[y]/(y2 ).

(b) The example of a prime power which is not primary is more

subtle. First we have to take a prime ideal which is not maxi-
mal. Take A = k[x, y]/(xy) and the prime P = (x). Then P2 is
not primary. Indeed xy = 0 ∈ P2 and x ∈ / P2 , so there should
2
be a power of y in P , which is clearly not the case.

89
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 factorization

(c) In a principal ideal domain the primary ideals are exactly the
prime powers, which is easily checked using factorization of
elements.

Definition 3.2.8. We say that I admits a primary decomposition if

there exist primary ideals Q1 , . . . , Qn such that
n
\
I= Qi .
i=1

We shall see that primary decomposition always exists in Noe-

therian rings, and that it satisfies a sort of uniqueness (this is
independent of the Noetherian hypothesis).
For Noetherian rings there is an easier concept to work with.

Definition 3.2.9. The ideal I ⊂ A is said irreducible if whenever

I = J ∩ K we must have I = J or I = K.

Proposition 3.2.10. Let A be a Noetherian ring, I an irreducible ideal.

Then I is primary.

Proof. Working in A/I we reduce to the case where I = 0. Let x be

a zero divisor, say xy = 0; we must show it is nilpotent. Consider
the chain
Ann(x) ⊂ Ann(x2 ) ⊂ · · · ;
this must stabilize, hence at some point Ann(xn ) = Ann(xn+1 ).
We claim that (xn ) ∩ (y) = 0. Indeed assume axn ∈ (y); then
a ∈ Ann(xn+1 ) = Ann(xn ). Since 0 is irreducible and y 6= 0 we
must have xn = 0.

Now we have the following easy result.

Proposition 3.2.11. Every ideal of a Noetherian ring is the intersection

of finitely many irreducible ideals.

Proof. Let X be the set of ideals which do not satisfy the conclu-
sion, and let I ∈ X be maximal. In particular I is not irreducible,
so we can write
I = J∩K
for bigger ideals J and K. But then J and K satisfy the thesis,
hence I does.

90
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 3.2 primary decomposition

Corollary 3.2.12. Every ideal of a Noetherian ring admits a primary

decomposition.
As in the case of the unique factorization we see that the ex-
istence part is easy, and it is a simple finiteness argument. The
uniqueness is more subtle, and at first it is not even clear what is
the right statement.
First, it is clear that we should group together primary ideals
with the same radical, as we do for prime powers is the case of
unique factorization domains. This is achieved by the following
Lemma 3.2.13. Let Q1 , . . . , Qn be P-primary ideals. Then the inter-
section
Q = Q1 ∩ · · · ∩ Qn
is P-primary.
Proof. First note that
p p p
Q = Q1 ∩ · · · ∩ Qn = P.

Now let xy ∈ Q √and assume

√ x∈
/ Q. Then x does not belong to
some Qi , so y ∈ Qi = Q.
Definition 3.2.14. Let
n
\
I= Qi .
i=1
be a primary decomposition. We call it minimal if
√
1. The radicals Pi = Qi are all distinct, and
2. no Qi includes an intersection Qi1 ∩ · · · Qik .
Clearly, starting from any primary decomposition we can get
a minimal one. Namely, we group together all primary factors
with the same radical, thanks to Lemma 3.2.13, and then simply
discard the primary factors which do not satisfy 2. So we treat
the uniqueness problem only for minimal decompositions.
Remark 3.2.15. With the usual factorization over Z the case 2
does not arise. But in more general rings it can happen that two
different prime ideals are included one in the other, so redundant
decompositions actually appear.

91
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 factorization

Recall that for a module M the set Ass(M) consists of those

prime ideals of the form Ann(m) for some m ∈ M.
Definition 3.2.16. Let I ⊂ A be an ideal. The associated primes of
I are the associated primes of the module A/I.
The definition is standard, and hopefully will not cause confu-
sion, even though I is of course a module itself.
Remark 3.2.17. For x ∈ A/I we have
Ann(x) = (I : x),
hence Ass(I) is equally described as the set of primes of the form
(I : x) for x ∈ A.
Theorem 3.2.18 (Uniqueness theorem, 1st form). Let
n
\
I= Qi .
i=1
√
be a minimal primary decomposition, Pi = Qi . Then the set {Pi } is
determined
p by I. More precisely the ideals Pi are exactly the primes of
the form (I : x) for x ∈ A.
Proof. For x ∈ A we have
p n p
\
(I : x) = (Qi : x).
i=1
p
Now p if x ∈ Qi we have (Qi : x) = A, while for x ∈ / Qi we
have (Qi : x) = Pi by definition of a primary ideal. Hence this
intersection can be rewritten as
p \
(I : x) = Pi . (3.2.1)
∈Qi
x/
p p
If moreover (I : x) is prime we must have (I : x) = Pi for
some i, by Proposition 1.1.20 ii).
In the other direction
T by minimality we can find and element
xi such that xi ∈ j6=i Qj but xi ∈ / Qi . Then
p
(I : xi ) = Pi
by (3.2.1).

92
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 3.2 primary decomposition

Corollary 3.2.19. Assume that A is Noetherian and let

n
\
I= Qi .
i=1
√
be a minimal primary decomposition, Pi = Qi . Then the set {Pi } =
Ass(I).
Proof. If P ∈ Ass(I)
p we have P = (I : x) for some x ∈ A, and P is
radical, so P = (I : x).
In the other direction take a prime Pi and write
\
R= Qi ,
j6=i

so that I = Qi ∩ R. Then
R R ∼ R + Qi ⊂ A .
= =
I Qi ∩ R Qi Qi
Note that A/Qi has the only associated prime Pi by the first part
of the proof. Since Ass(R/I) ⊂ Ass(A/Qi ) and is not empty, we
see that Pi is an associated prime of R/I, and a fortiori of A/I.
Note that the above results only give uniqueness of the radicals
of the factors of a primary decomposition. We shall say that
the primes Pi appearing in a minimal primary decomposition
of I belong to I. The factors themselves are not determined by I
only, as the following examples show, but there is a weak sort of
uniqueness, which we shall investigate.
Example 3.2.20. (a) It is very easy to find different minimal pri-
mary decompositions for ideals in a polynomial ring. A typ-
ical example is I = (x2 , xy) in the ring k[x, y], which can be
written
I = (x) ∩ (x2 , xy, y2 ) = (x) ∩ (x2 , y).
All ideals appearing in the decomposition are indeed pri-
mary: (x) is prime, while both (x2 , xy, y2 ) and (x2 , y) have
the radical (x, y), which is maximal.
(b) For another example in the same ring take

I = (y − x2 ) ∩ (x2 , xy, y2 ) = (y − x2 ) ∩ (x, y2 ).

93
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 factorization

(c) Consider2 the non-Noetherian ring A = k[x1 , x2 , . . . ], and let

I = (x21 , x22 , . . . ). Then, as
√
I = M = (x1 , x2 , . . . ),

which is maximal, I is M-primary. In particular I is certainly

decomposable, and M belongs to I. Still M ∈
/ Ass(I), showing
that the Noetherian hypothesis in Corollary 3.2.19 is neces-
sary.
To see this assume M = (I : f) for some f ∈ A, say f =
f(x1 , . . . , xn ). Clearly f ∈
/ I; but then

xn+1 f(x1 , . . . , xn ) ∈
/ I,

contradiction.

(d) Take the ring A = C(R) of continous functions on R. For a

closed interval D, say of length `, we shall see that the ideal

I= f∈A|f =0
D

does not admit a primary decomposition.

First note that I is radical, hence each primary component is
actually prime. Take a prime P ⊃ I. We can choose a function
f1 ∈ I which vanishes exactly on D and is positive outside,
say by taking it piecewise linear. Then write

f1 = f2 g2 ,

where f2 vanishes exactly on the left half of I and g2 vanishes

exactly on the left half of I. So either f2 or g2 belongs to P,
say f2 . We can repeat the process writing

f2 = f3 g3

and so on. The conclusion is that P contains functions which

vanish on arbitrarily small subintervals of D.
Now assume
I = P1 ∩ · · · Pn .
2 This example was suggested by Georges Elencwajg on MathOverflow, see [Ele].

94
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 3.2 primary decomposition

For each i we can choose a function hi ∈ Pi which vanishes

exactly on an interval of length less that `/n. But then the
product h1 · · · hn ∈ I and cannot be zero on the whole D,
contradiction.

Remark 3.2.21. Assume I admits a primary decomposition; then

any prime containing I contains a prime belonging to I. Indeed
assume
P ⊃ I = Q1 ∩ · · · ∩ Qn ;
then P contains some Qi by Proposition 1.1.20 ii), and taking
radicals we see that P ⊃ Pi .
In particular the minimal primes of I are the primes which
are minimal among those belonging to I. (Recall that a minimal
prime of I is a prime which is minimal among those containing
I.)

With this language we can prove the following result, which

looks somewhat unrelated to primary decomposition.

Corollary 3.2.22 (of Theorem 3.2.18). Let D be the set of zero divisors
of A. If 0 admits a primary decomposition, then D is the union of the
primes belonging to 0.

Proof. More or less by definition we have

[ p
D= (0 : x).
x6=0

Let
n
\
0= Qi
i=1
√
be a minimal decomposition of 0, and let pPi = Qi . Then by
(3.2.1) we see that each term of the union (0 : x) is contained in
some prime Pi .
pOn the other hand any prime belonging to 0 has the form
(0 : x), thereby proving the reverse inclusion.

Definition 3.2.23. Let P be a prime belonging to I. If P is not

minimal we say that P is an embedded prime of I.

95
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 factorization

It will turn out that only the primary factors corresponding to

minimal components are uniquely determined. To see this it is
convenient to understand how primary decomposition behaves
with respect to localization.
Let S be a multiplicative set. We know that there is a bijective
correspondence between prime ideals P such that P ∩ S = ∅ and
ideals of S−1 A, given by P → S−1 P (Corollary 1.6.9). We want to
extend this to primary ideals.
Proposition 3.2.24. There is a bijective correspondence between pri-
mary ideals Q ⊂ A such that Q ∩ S = ∅ and primary ideals of S−1 A.
Proof.√ First we remark that for any ideal I the conditions I ∩ S 6= ∅
and I ∩ S 6= ∅ are equivalent, since S is multiplicatively
√ closed.
So we need only consider primary ideals Q such that Q ∩ S = ∅.
The correspondence is the usual one. Denote by

ι : A → S−1 A
the canonical homomorphism. Then Q corresponds to the ex-
tended ideal S−1 Q and in the other direction Q 0 ⊂ S−1 A corre-
sponds to ι−1 (Q 0 ).
By Proposition 1.6.6 we know that for an ideal Q 0 ⊂ S−1 A we
have Q 0 = S−1 Q, where Q := ι−1 (Q 0 ). In the other direction
assume Q is primary. By the same Proposition
[
ι−1 (Q · S−1 A) = (Q : s).
s∈S
√
and if S ∩ Q 6= ∅ we have (Q : s) = Q by definition.
It remains to check that this correspondence sends primary ide-
als to primary ideals, which is easy and left to the reader.
Remark 3.2.25. With quotients the matters are even simpler. Since
the fact that Q is primary can be detected looking at the quotient
A/Q, the bijective correspondence between ideals containing I
and ideals of A/I restricts to a bijective correspondence between
primary ideals.
For the purpose of the rest of the section, denote bI := ι−1 (S−1 I),
where
ι : A → S−1 A
is the canonical homomorphism.

96
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 3.2 primary decomposition

Corollary 3.2.26. Let I an ideal admitting the minimal primary decom-

position
n
\
I= Qi
i=1
√
and let Pi = Qi . Then a minimal primary decomposition for bI is
\
bI = Qi .
Pi ∩S=∅

Definition 3.2.27. Let X be a subset of the primes belonging to I.

We say that X is isolated if whenever P1 ⊂ P2 are two prime ideals
belonging to I and P2 ∈ X, then also P1 ∈ X.
We can now prove
Theorem 3.2.28 (Uniqueness theorem, 2nd form). Let I an ideal
admitting the minimal primary decomposition
n
\
I= Qi
i=1
√
and let Pi = Qi . For any isolated set X of primes belonging to I the
intersection \
Qi
Pi ∈X

is independent of the decomposition. In particular if Pi is minimal, Qi

is determined by I.
Proof. Consider the multiplicative set
!
[
S := A \ P .
P∈X

The definition of isolated set is made so that Pi ∩ S = ∅ if and

only if Pi ∈ X. So by Corollary 3.2.26
\
Qi = bI
Pi ∈X

is determined by I.

97
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 factorization

Having established the two uniqueness theorems, we treat the

question non-uniqueness. For this purpose it is convenient to
introduce a tool of independent interest.
Definition 3.2.29. Let P be a prime ideal of A and let S := A \ P.
The ideal
P(n) := Pc
n = ι−1 (S−1 P n )

is called the n-th symbolic power of P.

These symbolic powers, among other things, give a measure of
how far is Pn from being P-primary.
Proposition 3.2.30. The symbolic power P(n) is P-primary. More
precisely if Pn is decomposable, P(n) is the P-primary component of
any primary decomposition of Pn .
Proof. Since S−1 P is maximal in AP , S−1 Pn is S−1 P-primary,
hence its contraction P(n) is P-primary by Proposition 3.2.24.
Now assume
\k
Pn = Qi ;
i=1
by Corollary 3.2.26 we know that
\
P(n) = Qi .
Pi ⊂P

But the assumptions Pi ⊂ P and Pn ⊂ Qi ⊂ Pi show, taking

radicals, that Pi = P. Hence this intersection only has one factor
Qi for Pi = P, and we are done.
The following result shows that non-uniqueness of embedded
primary components is not pathological, but is really what one
has to expect.
Theorem 3.2.31. Assume A is Noetherian and let I ⊂ A be an ideal
with a minimal decomposition
n
\
I= Qi ;
i=1
√
as usual let Pi = Qi . If Pi is an embedded prime, there are infinitely
many different choices for the Pi -primary component of I.

98
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 3.3 primary decomposition for modules

Proof. Passing to the quotient we can assume I = 0. We show

(k)
that Pi ⊂ Qi for k big enough, and that all symbolic powers
(k)
Pi are distinct. The two claims together imply the thesis.
The first claim follows immediately by Proposition 2.3.10 ap-
plied to the extension of Qi in the localization APi .
For the second one, observe that APi is not Artinian since it
contains prime ideals which are not maximal (the extension of
any prime strictly contained in Pi ). Let Mi := Pi APi . Since APi
is Noetherian, no power Mk i can be 0, otherwise we could apply
Lemma 2.4.10. By Nakayama’s lemma we deduce that

Mk+1 ( Mk .

So their contractions are also distinct by Proposition 3.2.24.

3.3 primary decomposition for modules

We now generalize the results of the previous section to the case
of submodules of a given module. Since the proofs are almost
identical, we shall be a little sketchy. This generalization is fruit-
ful: we have started studying the problem of factorization of ele-
ments and we end up with a sort of structure theorem for mod-
ules!
Our first remark is that the primes P containing an ideal I are
exactly those for which the extension I · AP is not trivial. In other
words the localization (A/I)P should not be zero. This condition
makes sense also for modules (we should remark that primary
decomposition for the ideal I will correspond to primary decom-
position for the module A/I). We isolate these primes with a
definition.

Definition 3.3.1. Let M be an A-module. The support of M is the

set
Supp(M) := {P ⊂ A prime | MP 6= 0}.

Remark 3.3.2. Assume that M is finitely generated; then the con-

dition P ∈ Supp(M) is equivalent to saying that Ann(M) ⊂ P. In
fact, P ∈ Supp(M) is equivalent to having some nonzero m ∈ MP .

99
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 factorization

This means that there is no s ∈ A \ P such that sm = 0, that is,

Ann(m) ⊂ P.
If M is finitely generated, say M = (m1 , . . . , mr ), then
r
\
Ann(M) = Ann(mi ),
i=1

so some Ann(mi ) ⊂ P if and only if Ann(M) ⊂ P by Proposition

1.1.20 ii.

The support of a module behaves nicely in exact sequences. In

fact, from Proposition 1.6.17 it follows immediately:

Proposition 3.3.3. Let

0 → M0 →M → M 00 →0

be an exact sequence. Then Supp(M) = Supp(M 0 ) ∪ Supp(M 00 ).

The concept of support is strictly related to that of associated

primes. Recall that in 2.5.11 we have defined associated primes
for a module M as

Ass(M) = {P ⊂ A prime | P = Ann(m) for some m ∈ M}.

Moreover, for an ideal I in a Noetherian ring the minimal primes

among those containing I are the minimal associated primes of
A/I. We are going to work out the corresponding facts for mod-
ules. To start, the following fact is reassuring.

Lemma 3.3.4. Any associated prime of M belongs to Supp(M).

Proof. Let P ∈ Ass(M), so we have an injective homomorphism

A/P ,→ M.

Then MP contains the residue field k(P) = F(A/P), so is not

zero.

Next we define primary modules. As we have remarked, the

property of Q being primary can be detected by the quotient
A/Q.

100
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 3.3 primary decomposition for modules

Definition 3.3.5. Let M be an A-module, a ∈ A. We say that a is

a zero divisor in M if the multiplication by a

a· : M →M
m → am
is not injective. We say that a is nilpotent in M if this map is
nilpotent.
We say that a submodule Q ⊂ M is primary if every zero divisor
in M/Q is nilpotent in M/Q.

Lemma 3.3.6. If Q is primary in M, then (Q : M) is a primary ideal.

Proof. Let xy ∈ (Q : M) with x ∈

/ (Q : M). Then multiplication by
y is a zero divisor on M/Q, so it is nilpotent. That is, a power of
y is in (Q : M).

Definition 3.3.7. For a submodule Q ⊂ M we define the radical

of Q in M as the ideal

(Q : M) = {a ∈ A | an M ⊂ Q for some n}.

p

p
If Q is primary in M and (Q : M) = P we say that Q is P-
primary. Throughout the rest of the section the ambient module
M is fixed.
A primary decomposition for N ⊂ M is a decomposition

N = Q1 ∩ · · · ∩ Qn ,

where the Qi are primary submodules of M.

Definition 3.3.8. Let N ⊂ M be a submodule. We say that N is

irreducible if whenever N = R ∩ S is written as the intersection of
two submodules, we must have N = R or N = S.

Proposition 3.3.9. Let M be a Noetherian module. Then every sub-

module of M is a finite intersection of irreducible submodules.

Proof. Noetherian induction, as in the proof of Proposition 3.2.11.

Proposition 3.3.10. Let M be a Noetherian module. Then an irre-

ducible submodule N ⊂ M is primary.

101
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 factorization

Proof. We can assume N = 0. Let a be a zero divisor on M, say

am = 0. Denote φx the multiplication by x endomorphism of M.
We look at the ascending chain

ker φa ⊂ ker φ2a ⊂ · · ·

This eventually stabilizes, so we must have ker φn n+1 .

a = ker φa
n n 0
Then a M ∩ Am = 0. Indeed assume a m ∈ Am; then

n 0 ∈ ker φn+1
a = ker φn
a,

so an m 0 = 0. Since 0 is irreducible and Am 6= 0 we must have

an M = 0, proving that φa is nilpotent.
Corollary 3.3.11. Let M be a Noetherian module. Then every submod-
ule of M admits a primary decomposition.
This settles the existence part. The uniqueness, again, is very
similar to the case of ideals. First, it is easy to see that the intersec-
tion of two P-primary submodules is again P-primary, so we can
group together all primary components with the same radical in
M. Then we remove components which do not cut out anything
and find a minimal primary decomposition.
Theorem 3.3.12 (Uniqueness theorem, 3rd form). Let
n
\
N= Qi .
i=1
p
be a minimal primary decomposition, Pi = (Qi : M). Then the set
{Pi } is determined p
by N. More precisely the ideals Pi are exactly the
primes of the form (N : m) for m ∈ M.
We omit the proof, which is identical to that of Theorem 3.3.12.
Corollary 3.3.13. Assume that A is Noetherian and M is finitely gen-
erated, and let
\n
N= Qi .
i=1
p
be a minimal primary decomposition, Pi = (Qi : M). Then the set
{Pi } = Ass(M/N).

102
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 3.3 primary decomposition for modules

Proof. If P ∈ Ass(M/N)p we have P = (N : m) for some m ∈ M,

and P is radical, so P = (N : m).
In the other direction, take a prime Pi and write
\
R= Qi ,
j6=i

so that N = Qi ∩ R. Then

R R ∼ R + Qi ⊂ M .
= =
N Qi ∩ R Qi Qi

Note that M/Qi has the only associated prime Pi . Since Ass(R/N)
is not empty and contained in Ass(M/Qi ), we see that Pi is an
associated prime of R/N, and a fortiori of M/N.

Corollary 3.3.14. If A is Noetherian and M is finitely generated, every

prime P ∈ Supp(M/N) contains a prime associated to M/N.

Proof. We can assume N = 0. Write

n
\
0= Qi .
i=1

Since MP 6= 0 there exists m ∈ M such that Ann(m) ⊂ P; in

particular
n
\
(Qi : m) ⊂ P.
i=1

But then P contains some (Qi : m) by Proposition 1.1.20 ii), and

a fortiori P ⊃ (Qi : M). Taking radicals we obtain P ⊃ Pi .

Definition 3.3.15. This ensures that the minimal primes in the

sets Supp(M/N) and Ass(M/N) are the same; we shall call them
the minimal primes of N in M.

We keep following closely the previous section.

Proposition 3.3.16. There is a bijective

p correspondence between pri-
mary submodules Q ⊂ M such that (Q : M) ∩ S = ∅ and primary
submodules of S−1 M.

103
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 factorization
p
Proof. First note that if (Q : M) ∩ S = ∅, then (Q : s) = Q for
any s ∈ S. Now apply Proposition 1.6.16 to obtain the bijective
correspondence. Again, one has to do the easy check that Q is
primary if and only if S−1 Q is.

Let
ι : M → S−1 M
be the canonical homomorphism and for N ⊂ M denote N
b :=
−1 −1
ι (S N).

Corollary 3.3.17. Let N ⊂ M be a submodule admitting the minimal

primary decomposition
\n
N= Qi
i=1
p
and let Pi = (Qi : M). Then a minimal primary decomposition for
b is
N \
N
b = Qi .
Pi ∩S=∅

We can now prove

Theorem 3.3.18 (Uniqueness theorem, 4th form). Let N ⊂ M be a

submodule admitting the minimal primary decomposition
n
\
N= Qi
i=1

(Qi : M). For any isolated subset X of {Pi } the intersec-

p
and let Pi =
tion \
Qi
Pi ∈X

is independent of the decomposition. In particular if Pi is minimal, Qi

is determined by N.

Proof. Consider the multiplicative set

!
[
S := A \ P .
P∈X

104
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 3.4 factorization in dedekind rings

By definition, Pi ∩ S = ∅ if and only if Pi ∈ X. So, by Corollary

3.3.17, \
Qi = N
b
Pi ∈X

is determined by N.

3.4 factorization in dedekind rings

We now look at the class of rings where the theory of primary
decomposition takes its simplest form.

Definition 3.4.1. Let A be a Noetherian integral domain. We

say that A is a Dedekind ring (or Dedekind domain) if for every
prime P ⊂ A the only ideals of the ring AP are the powers of the
maximal ideal.

Remark 3.4.2. The local rings having as only ideals the powers of
the maximal ideal are called discrete valuation rings (DVR). We
will study them in detail in Section 7.3.

Remark 3.4.3. Since localization commutes with quotients, it is

immediate to see that a nontrivial quotient of a Dedekind ring
that is integral is in fact a field.

The definition above is in some sense temporary. We will give

a more useful characterization of Dedekind rings in Theorem
5.1.19, after studying integral extensions. For now we are in-
terested in Dedekind rings because the theory of factorization of
ideals gets much simpler there.

Example 3.4.4. (a) Every principal ideal domain is a Dedekind

ring. Indeed Noetherianity is clear, and the other condition
follows from unique factorization.

(b) We are not able at the present point to give more examples,
but we shall see in√Chapter 5 that√many rings of interest in
arithmetics, like Z[ 19] or Z[(1 + 5)/2] are Dedekind rings.

We now apply the theory of primary decomposition to Dede-

kind rings. We start with two simple remarks.

105
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 factorization

Remark 3.4.5. Every nonzero prime ideal of a Dedekind ring is

maximal. Indeed if P ( Q is a proper inclusion of nonzero prime
ideals, both PAQ and QAQ are prime, hence the first cannot be
a power of the second.
Remark 3.4.6. Let I ⊂ A be a nonzero ideal of a Dedekind ring.
Since A is Noetherian, I admits a primary decomposition

I = Q1 ∩ · · · ∩ Qn ;

assume this is minimal. Since every nonzero prime of A is max-

imal, there cannot be embedded primes; so the primary compo-
nents Qi are uniquely determined by I.
This is good enough, but there is more.
Proposition 3.4.7. Let P be a nonzero prime of a Dedekind ring. The
P-primary ideals are exactly the powers of P.
Proof. Since P is maximal, Pn is primary by Proposition 3.2.6.
Vice versa, assume Q is P-primary. Let

ι : A → AP

be the localization morphism. By Proposition 3.2.24,

Q = ι−1 (QAP );

since QAP is a power of PAP , we are done.

Corollary 3.4.8 (Factorization of ideals). Let A be a Dedekind ring.
Any nonzero ideal I ∈ A can be uniquely written as

I = P1a1 · · · Pn
an
,

where the Pi are primes (up to reordering the factors).

Proof. By Proposition 3.4.7 and the previous remarks, it is clear
that I can be uniquely written as an intersection

I = P1a1 ∩ · · · ∩ Pn
an
.

Recall from the proof of the Chinese remainder theorem that

J∩K = J·K

106
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 3.4 factorization in dedekind rings

whenever J and K are coprime ideals, that is J + K = A. By

induction, we only need to check that this is satisfied for J =
ak+1
P1a1 · · · Pkak and K = Pk+1 . It is an easy check that two ideals are
coprime if and only if their radicals are. So we must prove that
J 0 := P1 · · · Pk and Pk+1 are coprime.
Since Pk+1 is maximal, this is true unless J 0 ⊂ Pk+1 . By Propo-
sition 1.1.20 ii) this implies that some Pi ⊂ Pk+1 , which is a
contradiction since the two are distinct and both maximal.

In the rest of this section, we explore the consequences of

unique factorization for ideals. In particular, we will obtain some
results which compare Dedekind rings with unique factorization
domains.
Let A be a Dedekind ring; if A is a principal ideal domain,
unique factorization for ideals just becomes unique factorization
for elements. This is unsurprising, since we know that every
principal ideal domain is a unique factorization domain. A little
less trivial is the converse.

Proposition 3.4.9. Let A be a Dedekind ring. Then A is a principal

ideal domain if and only if it is a unique factorization domain.

Proof. One direction is clear. So assume A is both a Dedekind

ring and a unique factorization domain. Let P be a nonzero prime
ideal of A; then P contains a prime element p by Theorem 3.1.20.
Since every prime of A is maximal, the inclusion (p) ⊂ P must be
an equality, showing that P is principal.
By unique factorization, every ideal of A is principal.

A way to restate this is that the intersection of the classes of

unique factorization domains and Dedekind rings consist exactly
of the principal ideal domains. We will give a slightly stronger
form in Corollary 5.1.20.
We now introduce a measure of how far a Dedekind ring is
from being a principal ideal domain.

Definition 3.4.10. Let A be an integral domain, k = F(A) its field

of fractions. Any A-submodule of k of the form I/a, for some
ideal I ⊂ A and a ∈ A, is called a fractional ideal. The fractional
ideal I/a is called principal if I is.

107
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 factorization

Remark 3.4.11. Assume M ⊂ k is a finitely generated A-submod-

ule. Then there is a common denominator for elements in M, so
for some a ∈ A we have aM ⊂ A. Then I := aM is an ideal of A,
so M = I/a is a fractional ideal.
The converse holds assuming A is Noetherian. So for Noether-
ian rings we can think of fractional ideals as finitely generated
submodules of F(A).
Fractional ideals of A can be multiplied by defining

(I/a) · (J/b) := IJ/ab.

With this operation, fractional ideals form a commutative monoid,

whose unit is the ring A itself. Principal ideals form a sub-
monoid.
In the case of Dedekind rings, things are better behaved.
Proposition 3.4.12. Let A be a Dedekind ring, I ⊂ J ideals. Then there
exists an ideal K such that I = JK.
Proof. Using factorization for I and J it is enough to prove the
following claim. Factorize I as

I = P1a1 · · · Pkak

and let P be a prime, say P = P1 (here we allow a1 to be 0). If

I ⊂ Pb , then b 6 a1 .
To see this, remark that every prime Pj with j 6= 1 is not con-
tained in P by maximality. So all such primes become trivial in
the localization AP , and

IAP = Pa1 AP .

Since this is contained in Pb AP , we have b 6 a1 . If this was not

the case, we would have

Pa1 +1 AP = Pa1 AP ,

which by Nakayama’s lemma gives Pa1 AP = 0; this is a contra-

diction since AP is an integral domain.
Corollary 3.4.13. Let A be a Dedekind ring. For every ideal I ⊂ A
there exists another ideal J such that IJ is principal.

108
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 3.4 factorization in dedekind rings

Proof. Take any nonzero a ∈ I. Then (a) ⊂ I and we can apply

the proposition.

The important consequence of this corollary is that if A ia Dede-

kind ring, the quotient of the monoid of fractional ideals by the
submonoid of principal ideals is actually a commutative group.
This group is an important invariant of A, and is called the ideal
class group of A.
We shall denote the ideal class group of A by G(A). If it is
finite, its order is denoted h(A).
By Proposition 3.4.9, G(A) is trivial if and only if A is a prin-
cipal ideal domain, or equivalently A is a unique factorization
domain.

Remark 3.4.14. As trivial as it can seem, the following observa-

tion is very important. If A is Dedekind ring, and I is an ideal
of order k in G(A), Ik is a principal ideal. In particular, if we
know that G(A) is finite, many questions about ideals in A can
be reduced to the case of principal ideals.

Another result which gives a hint how far Dedekind rings are
from principal ideal domains is the following.

Proposition 3.4.15. Let A be Dedekind ring, I ⊂ A a nonzero ideal.

Then A/I is a principal ideal domain.

Proof. Let J/I be an ideal of A/I; we must prove that it is principal.

Using unique factorization for ideals we can assume J is prime.
Since I ⊂ J, we know that J divides I by Proposition 3.4.12.
Factorize I as
I = P1a1 · · · Pkak ,
say J = P1 . By the Chinese remainder theorem we can find x ∈ A
such that

x≡0 (mod P1 )
x 6≡ 0 (mod P12 )
x≡1 (mod P2 )
..
.
x≡1 (mod Pk ).

109
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 factorization

We claim P1 = I + (x), from which the thesis follows. The

inclusion I + (x) ⊂ P1 is clear. On the other hand, by Proposition
3.4.12, I + (x) divides both I and (x); the factorization of I and the
choice of x show that the only common factor of these two ideals
is indeed P1 .

Corollary 3.4.16. Every ideal in a Dedekind ring is generated by at

most two elements.

Proof. Let J be a nonzero ideal and take any nonzero x ∈ J. Ap-

plying the proposition to (x) we see that the ideal J/(x) in A/(x)
is generated by a single element y, which we can lift to y ∈ J.
Then J = (x, y).

Remark 3.4.17. More precisely, we see from the proof that given
any nonzero x ∈ I, we can find y ∈ I such that I = (x, y).

3.5 the structure of modules over de-

dekind rings
We now see, following [Sch03], how to generalize the classifica-
tion of finitely generated modules over a principal ideal domain
to the case of Dedekind rings. We start by understanding what
ideals look like as modules.

Remark 3.5.1. Let I, J be two ideals of the Dedekind ring A. If

I = J in G(A), we have J = xI for some x ∈ F(A). Hence I and J
are isomorphic as A-modules.

Lemma 3.5.2. Let A be a Dedekind ring, I and J two nonzero ideals.

Then
I⊕J =∼ A ⊕ IJ

as A-modules.

Proof. First assume I + J = A. Then the natural map

t: I⊕J →A
(i, j) → i+j

110
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 3.5 modules over dedekind rings

∼ I ∩ J. But
is surjective. Moreover ker t =
∼ IJ
I∩J =

since I and J are coprime, hence we have the short exact sequence

0 → IJ → I⊕J →A → 0.

We only need to check that this exact sequence splits, that is, there
exist an A-linear map s : A → I ⊕ J such that t ◦ s = idA . But this
is clear since A is free as an A-module.
For the general case, using the above remark, it is enough to
show that given any two ideals I and J there exists an ideal J 0
with same class of J in G(A), and such that I and J 0 are coprime.
Start from any x ∈ J; by Proposition 3.4.12 there exists an ideal
K such that JK = (x). Moreover by Proposition 3.4.15 there exists
y ∈ A such that
IK + (y) = K.
Multiplying by J we get

IJK + (y)J = JK = (x). (3.5.1)

Again by Proposition 3.4.12 we have

(y)J = (x)J 0

for some ideal J 0 . This last relation tells us that J and J 0 are in the
same class in G(A).
Plugging this last relation into (3.5.1) yields

(x)I + (x)J 0 = (x),

whence I + J 0 = A.
Recall from Exercise 13 of Chapter 1 that an A-module is called
projective if it is a direct summand of a free module. As shown in
that exercise, projective modules are equivalently characterized
by a universal property. In particular this implies that if

0 →M →N →P →0

is a short exact sequence and P is projective, the sequence splits,

so N =∼ M ⊕ P.

111
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 factorization

Proposition 3.5.3. Let A be Dedekind ring. Every ideal of A is a

projective A-module.
Proof. We can assume I 6= 0. Take any x ∈ I; by Proposition 3.4.12
there exists an ideal J such that IJ = (x). Then by Lemma 3.5.2
we have
I⊕J =∼ A ⊕ (x) =∼ A ⊕ A.

Before going into the proof of the main result, we need to gen-
eralize our notion of rank to modules which are not free.
Definition 3.5.4. Let A be an integral domain, k := F(A), and let
M be an A-module. We define the rank of M as the dimension of
the k-vector space
M ⊗A k.
Remark 3.5.5. This notion of course agrees with our previous
definition of rank, but note that it is only defined for integral
domains.
Remark 3.5.6. If M is finitely generated we have
rk M = dimk M ⊗A k = dimk HomA (M, k),
since the latter vector space is dual of the former.
Lemma 3.5.7. Let A be a Dedekind ring, M a finitely generated torsion-
free A-module. Then M is isomorphic to nonzero ideal if, and only if,
rk M = 1.
Proof. Let I be a nonzero ideal; for x ∈ I we have
(x) ⊂ I ⊂ A.
∼ A as A-modules, we have
Since (x) =
1 = rk(x) 6 rk I 6 rk A = 1.
In the other direction assume rk M = 1. In particular there
exists a nontrivial homomorphism

φ : M → k.
Since M is finitely generated, the image of φ is a fractional ideal;
multiplying φ by a constant we can assume φ(M) ⊂ A is an ideal,
call it I.

112
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 3.5 modules over dedekind rings

The exact sequence

0 → ker(φ) →M →I →0

splits, since I is projective, so

rk ker(φ) + rk I = rk M;

by the first part of the proof we conclude that rk ker(φ) = 0. Since

ker(φ) is torsion-free, it must be zero, hence M = ∼ I.

Example 3.5.8. The lemma is false if we don’t assume that M is

finitely generated. For instance, Q has rank 1 as a Z-module, but
of course it is not isomorphic to any ideal. To compute the rank
of Q, just note that the map

m : Q ⊗Z Q → Q

given by m(a ⊗ b) = ab is an isomorphism.

We are now ready to generalize the classification of finitely

generated modules over principal ideal domains, from Exercise 2
of Chapter 2, and its consequence, Theorem 2.1.7.

Theorem 3.5.9. Let A be a Dedekind ring, M a finitely generated

module which is isomorphic to a direct sum of ideals. For every sub-
module N ⊂ M we can find m1 , . . . , mn ∈ M and ideals Jk ⊂ Ik for
k = 1, . . . , n such that
∼ I1 m1 ⊕ · · · ⊕ In mn
M=
∼ J1 m1 ⊕ · · · ⊕ Jn mn .
N=

Proof. As in the proof of Lemma 3.5.7, we can find a homomor-

phism
φ: M → A
with image an ideal I; let K := ker φ, J := φ(N).
We have a commutative diagram of exact sequences

0 → K∩N →N →J →0

↓ ↓ ↓
0 →K →M →I → 0.

113
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 factorization

Moreover, the rows are split exact sequences by Proposition 3.5.3.

First we argue that N is itself isomorphic to a direct sum of
ideals. This is done by induction on rk N, the induction basis
being trivial. Since the first row splits,

rk(K ∩ N) = rk N − 1

by Lemma 3.5.7. By induction, K ∩ N is isomorphic to a direct

sum of ideals, so N is.
We then prove the thesis by induction on rk M. As above we
have
rk K = rk M − 1,
and by the first part of the proof K is isomorphic to a direct sum
of ideals. By induction, K ∩ N ⊂ K satisfies the thesis. Since the
rows are split, the thesis for N follows.
Corollary 3.5.10. Let A be a Dedekind ring, M a finitely generated
A-module. Then M = ∼ T ⊕ P, where T is torsion and P is projective.
More precisely
T=∼ A/I1 ⊕ · · · ⊕ A/Ik
for some ideals Ij ⊂ A, and
∼ Ar ⊕ L
P=

for some ideal L and some r > 0.

Of course here L can be A itself.
Proof. By Theorem 3.5.9 we know that M is a direct sum of mod-
ules of the form I/J for some ideals J ⊂ I ⊂ A.
If J 6= 0 we have that
I = J + (x)
for some x ∈ A by Proposition 3.4.15. Hence
I ∼ J + (x) ∼ (x)
= = ;
J J (x) ∩ J
the latter is a quotient of (x), hence isomorphic to a quotient of
A. All this summands go into T .
The remaining summands are ideals of A, and we can group
them together by Lemma 3.5.2, to obtain a module of the form
Ar ⊕ L.

114
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 3.6 exercises

Corollary 3.5.11. Let A be a Dedekind ring, M a finitely generated

A-module. Then M is projective if, and only if, it is torsion-free.

3.6 exercises
1. Let k be a field of characteristic 0 and let A = k[x2 , x3 ]. Prove
that every element of A is a product of irreducible elements; still
A is not a unique factorization domain because there exists f ∈ A
irreducible but not prime.

2. Let A be a unique factorization domain, S ⊂ A a multiplicative

set. Prove that S−1 A is itself a unique factorization domain.

3. Find a subring of k[x1 , . . . , xn ] that is not a unique factoriza-

tion domain.

4. Prove Nagata’s lemma: let A be an integral domain such that

every element of A is a finite product of irreducible elements,
S ⊂ A a multiplicative set generated by prime elements. Then A
is a unique factorization domain if and only if S−1 A is a unique
factorization domain. Derive from this another proof of Theorem
3.1.16.

5. Let k be a field. The only homomorphisms

φ : k[x, y]/(y2 − x3 − x) → k[t]

are those given by φ(f) = f(α, β), where α, β ∈ k are such that

β2 = α3 + α.

6. Let A be a Noetherian ring and assume a ∈ A is not contained

in any minimal prime of A. If ab = 0, then b is nilpotent.

7. Let k be a field and r ∈ k(x) a rational function, say r(x) =

a(x)/b(x), where a and b are coprime polynomials. Show that
the element a(y) − rb(y) is irreducible over the ring A := k[r],
and that x is a root for it.

As a weaker condition than unique factorization, one can con-

sider rings such that every (nonzero, non-invertible) element can

115
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 factorization

be written (not uniquely) as a product of irreducible ones. A

ring that satisfies this condition is called an atomic ring. One may
expect that this condition is equivalent to the ascending chain
condition for principal ideals (ACCP), that is, condition 2 in Def-
inition 3.1.4. In fact, being atomic is a strictly weaker condition.
In the next exercises, up to Exercise 13, we will give a counterex-
ample from [Gra74]. For more on atomic rings, including other
examples, see also [Zak82].
8. Let A be a ring that satisfies ACCP. Show that A is atomic.
9. Let M be the additive monoid generated by all rationals of the
form 2k1p , where p0 = 3, p1 = 5, and so on, is an enumeration
k
of the odd primes. Show that each element m ∈ M can be written
uniquely as
a n ns
m = u + 0 +···+ s , (3.6.1)
2 3 2 ps
where a, u, ni ∈ N, ni < pi and either s = 0 or ns 6= 0. Deduce
that for each m ∈ M there is a maximum dyadic rational (that is,
a rational of the form a/2k ) ψ(m) such that ψ(m) 6 m.
10. Let k be a field of characteristic 0, and consider the set T of
elements having the form

X
n
f0 + fi xmi , (3.6.2)
i=1

where mi ∈ M and fi ∈ k. Show that T is a ring and the set

S ⊂ T of elements having f0 6= 0 is a multiplicative set.
11. Let T , S be as in the previous exercise and consider the ring
1
A = S−1 T . Show that each element of the form x 2k pk is an
irreducible element of A.
12. Let A be the ring of the previous exercise. Show that A does
not satisfy ACCP.
13. Let A be the ring of Exercise 11. Show that A is atomic. (Take
any element a/b ∈ A, with a in form (3.6.2) – we want to factorize
it as a product of irreducibles. By factoring out a suitable element,
show that it is not restrictive to assume that ψ(mi ) = 0 for some

116
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 3.6 exercises

mi , where ψ is the function of Exercise 9. Writing mi as in (3.6.1),

show that a factorization for a contains at most n0 + · · · + ns
nontrivial factors.)
14. Let A be a Dedekind ring. Prove that the principal frac-
tional ideals form a free abelian group. Deduce that Q∗ /±1 and
k(x)∗ /k∗ are free abelian groups.
15. Let A be a Dedekind ring, I and J two ideals. Prove that if
A/I is finite
#(A/I) = #(J/IJ).
16. Let I1 , . . . , In ⊂ A be ideals
T such thatQnIi + Ij = A for all
distinct pairs of i, j. Show that n k=1 Ik = k=1 Ik .
17. Let A be an integral domain. The following conditions are
equivalent:
(a) A is a Dedekind ring;
(b) every nonzero ideal of A factorizes uniquely as a product of
prime ideals;
(c) the monoid G(A) of fractional ideals modulo principal frac-
tional ideals of A is a group.
The two following exercises are taken from the beautiful pre-
sentation [GT02].
18. Assume you have a solution to the Fermat’s equation
an + bn = cn ,
where a, b, c ∈ k[x] for some field k of characteristic 0. Prove
that either all polynomials are multiple of each other, or else n 6
2. (Derive the equation and rearrange to find that an−1 divides
cb 0 − bc 0 , then make this symmetric.)
19. By the same strategy as the previos exercise, prove the follow-
ing theorem of Mason. Let k be a field of characteristic 0. For a
polynomial f ∈ k[t], let r(f) the number of distinct roots of f in
an algebraic closure of k.
Let a, b, c ∈ k[x] be polynomials without common roots such
that a + b = c; then max{deg a, deg b, deg c} 6 r(abc). The gener-
alization of this fact to integer numbers is the famous abc conjec-
ture – see [GT02].

117
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 factorization

20. Let A be a Noetherian ring, I ⊂ A an ideal. Prove that if

IAP = 0 for all primes P ∈ Ass(A), then I = 0.

21. Let A be a Noetherian ring, M a finitely generated A-module,

and S a multiplicative subset of A. Prove that there is a natural
bijection between AssS−1 A (S−1 M) and AssA (S−1 M), and that in
fact the latter is the set of primes in AssA (M) that do not meet S.
22. Prove some analogue of Proposition 3.3.3 for associated primes.
Namely, let

0 → M0 →M → M 00 →0

be an exact sequence of A-modules. Then Ass(M) ⊂ Ass(M 0 ) ∪

Ass(M 00 ). Find a counterexample to show that we do not always
have equality.

23. Let A be a Noetherian ring, I ⊂ A an ideal, not necessarily

prime. Define the n-th symbolic power of I as the intersection
\
I(n) := ι−1 n
P (I AP ),
P∈Ass(I)

where P runs through the associated primes of I, and ιP : A → AP

is the localization map. Prove that this definition agrees with our
previous Definition 3.2.29 when I is a prime ideal.

24. Let A be a Noetherian ring, I ⊂ A an ideal. Prove that for any

m > 1, Im ⊂ I(m) . Moreover, for m, n > 1, Im ⊂ I(n) if and only
if m > n ([BDRH+ 09, Lemma 8.1.4]).
25. Let M, N be two finitely generated A-modules. Prove that
Supp(M ⊗ N) = Supp(M) ∩ Supp(N).

118
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 4
C O M P U TAT I O N A L M E T H O D S

In this chapter we introduce some computational techniques that

allow us to work out explicitly some interesting examples. These
techniques are most well-suited for polynomial rings and their
quotients. This chapter is indeed composed of two mostly inde-
pendent parts.
The first one introduces the resultant. This is a tool that al-
lows one to find the common factors of two polynomials in A[x],
where A is a unique factorization domain. When we take A =
k[x1 , . . . , xn ] a polynomial ring itself, the resultant is helpful to
find the common zeroes of a set of polynomials in an inductive
way. The resultant is the core technique of classical elimination
theory, which is concerned with solving systems of polynomial
equations by eliminating one variable at a time. In Section 4.2,
we give a link between the resultant and some invariants of field
extensions.
In the second part we introduce the Gröbner bases. These are
special set of generators of an ideal I in a polynomial ring, which
make computable the operation of deciding wether a given el-
ement f ∈ I. Most of the foundations on Gröbner bases were
laid in Buchberger’s thesis [Buc65], and they have become the
most used tool in computational algebra since then. In Section
4.3 we give the definition and study the simplest properties; in
particular we solve the ideal membership problem. Then in Section
4.4 we show that Gröbner bases can be used to solve some other
common problems in commutative algebra, in particular getting
a different approach to elimination theory. Our presentation of
this part is inspired by [CLO96] and [KR00].
A word of caution about the implementation of the algorithms.
For brevity, we shall be a bit sloppy when we describe algorithms.

119
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 computational methods

In particular, we will only give a textual description, without a

pseudocode implementation. We leave to the reader the task of
formalizing and implementing the algorithms.
Moreover, we don’t strive for effectiveness. The algorithms
that we give are guaranteed to terminate, but we do not make
any study of the expected (or worst case) time of execution. We
do not even try to give the quickest and most efficient known
algorithms, and we shall be content of presenting the basics.

4.1 the resultant

Let A be a unique factorization domain. We know by Theo-
rem 3.1.16 that A[x] is again a unique factorization domain, so
it makes sense to ask, given f, g ∈ A[x], whether they have a
common divisor of positive degree.
This is essentially a problem in linear algebra. Let m := deg f,
n := deg g and write

f(x) = am xm + · · · + a1 x + a0
g(x) = bn xn + · · · + b1 x + b0 .

The whole idea of the resultant stems from the simple remark
that f and g have a common factor of positive degree if and only
if their minimum common multiple has smaller degree than their
product. Or, equivalently, if and only if they have any common
multiple of degree smaller then their product.
So we ask if we can find nonzero polynomials r, s ∈ A[x], with
deg r < n and deg s < m, such that

f(x)r(x) + g(x)s(x) = 0. (4.1.1)

Let k = F(A); then Equation (4.1.1) has solutions r, s ∈ A[x]

if and only if it has solutions r, s ∈ k[x], because we can always
clear denominators. But finding solutions in k[x] is a problem in
linear algebra.

120
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 4.1 the resultant

Namely, we can assign indeterminate coefficients to r and s

and observe that (4.1.1) becomes a linear system in these indeter-
minates. The matrix of this system is just
 
am · · · a1 a0 0 ··· 0
 0 am · · · a1 a0 0 ··· 0
..
 
 

 . 

 0 ··· 0 am · · · a1 a0 
Syl(f, g) := 
 ,
 bn · · · b1 b0 0 ··· 0 
 0
 bn · · · b1 b0 0 ··· 0  
 .. 
 . 
0 · · · 0 bn · · · b1 b0

where the ai appear in n lines and the bi in m lines.

Definition 4.1.1. The matrix Syl(f, g) is called the Sylvester matrix
of f and g. Its determinant is the resultant of f and g, and is
denoted R(f, g).
Since a homogeneous square linear system over a field has a
nontrivial solution if and only if the associated matrix has deter-
minant 0, we deduce from our earlier remarks:
Theorem 4.1.2. The polynomials f and g have a common factor in A[x]
of positive degree if and only if R(f, g) = 0.
The theorem has some interesting corollaries when one takes A
a polynomial ring itself. The first observation is that the resultant
of f and g is just a polynomial whose coefficients can be explicitly
written in terms of those of f and g. These expressions do not
depend on the ring where we are working, hence we have the
following specialization result.
Proposition 4.1.3. Let A be a unique factorization domain, take f, g ∈
A[x, y] and let R = R(f, g) ∈ A[x] denote the resultant with respect to
the y variable. Write

f(x, y) = am (x)ym + · · · + a1 (x)y + a0 (x)

g(x, y) = bn (x)yn + · · · + b1 (x)y + b0 (x).

Finally let α ∈ A. Assume am (α) 6= 0 and bn (α) 6= 0; then the

resultant of f(α, y) and g(α, y) is R(α).

121
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 computational methods

Remark 4.1.4. The request that am (α) 6= 0 and bn (α) 6= 0 is

necessary to be sure that the degrees of f(α, y) and g(α, y) do not
drop; if the degrees change, the expression for the coefficients of
the resultant also changes.

Corollary 4.1.5. With the same notation, f(α, y) and g(α, y) have a
common factor of positive degree in y if and only if R(f, g)(α) = 0.

By induction, this result can be generalized to specialization

of more than one variable. We leave this to the reader, and only
note the following particular case.

Corollary 4.1.6. Let k be an algebraically closed field, take f, g ∈

k[x1 , . . . , xs , y], and consider the resultant with respect to the y vari-
able R(f, g) ∈ k[x1 , . . . , xs ]. Let α1 , . . . , αs ∈ ks such that

R(f, g)(α1 , . . . , αs ) = 0.

Then either

i) the highest coefficient of f or g vanishes at α1 , . . . , αs , or

ii) f(α1 , . . . , αs , y) and g(α1 , . . . , αs , y) have a common root in k.

Proof. If the degree of f and g does not drop, the resultant of

f(α1 , . . . , αs , y) and g(α1 , . . . , αs , y) is R(f, g)(α1 , . . . , αs ). This is
0 if and only if f(α1 , . . . , αs , y) and g(α1 , . . . , αs , y) have a com-
mon factor of positive degree. Since k is algebraically closed, this
is equivalent to having a common root.

The above corollary can be used to solve the main problem of

elimination theory: given a system of polynomial equations


 f (x , . . . , xn ) = 0
 1 1
..
 .

f (x , . . . , x ) = 0
m 1 n

find the solutions in k, by eliminating one variable at a time. If we

let I = (f1 , . . . , fm ), then the problem reduces to finding genera-
tors for I ∩ k[x1 , . . . , xn−1 ]. If we are able to do this, we eventually
reduce to a problem in k[x1 ], solve for one variable, and finally
backtrack to find the general solution.

122
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 4.1 the resultant

We will only consider the case m = n = 2, and leave the gen-

eral case to Exercise 2. In this case, the system becomes just

f(x, y) = 0
g(x, y) = 0

and we can apply Corollary 4.1.6. If (x0 , y0 ) is a solution of the

system, then f(x0 , y) and g(x0 , y) have a common linear factor,
hence x0 is a root of R(f, g).
Example 4.1.7. Consider the system given by

f(x, y) = x2 y2 − 25x2 + 9 = 0
g(x, y) = 4x + y = 0.

Then R(f, g) = 16x4 − 25x2 + 9, which is biquadratic and has roots

x = ±3, ±4. Setting these values for x in g and solving for y, we
find the four solutions (1, −4), (−1, 4), ( 43 , −3) and (− 43 , 3).
The next results give some alternative expressions for the re-
sultant. They are all based on the slogan that to prove an identity
between polynomials, it is enough to prove it in a universal case.
Proposition 4.1.8. Assume

f(x) = a(x − α1 ) · · · (x − αm )
g(x) = b(x − β1 ) · · · (x − βn )

in A[x] (the roots can be repeated). Then

Y
R(f, g) = an bm · (αi − βj ).
i,j

Proof. It is enough to prove the identity in the ring

A 0 = A[x1 , . . . , xm , y1 , . . . , yn ],

in the case where αi = xi and βj = yj . Indeed once we know

this case, the identity remains true when we specialize xi to αi
and yj to βj , thanks to Proposition 4.1.3.
To prove this case, let k = F(A) and isolate one of the vari-
ables, say y1 . We regard the resultant R(f, g) as an element of

123
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 computational methods

k(x1 , . . . , xm , y2 , . . . , yn )[y1 ]. As a polynomial in y1 it becomes 0

for y1 = xi , hence it must be divisible by y1 − xi in that ring. By
Gauss’ lemma it remains divisible by y1 − xi in A 0 .
Since the same Q is true for y2 , . . . , yn we conclude that R(f, g)
is divisible by i,j (xi − yj ). Both this product and R(f, g) are
bihomogeneous of degree n in the xi and m in the yj , so
Y
R(f, g) = c (xi − yj )
i,j

for some c ∈ A. Finally comparing the coefficient of any given

monomial we find c = an bm .

Corollary 4.1.9. With notation as above we have

Y
m
R(f, g) = an g(αi ).
i=1

Corollary 4.1.10. For any f, g, h ∈ A[x] we have

R(f, gh) = R(f, g) · R(f, h).

Proof. It is enough to prove the identity in a bigger ring. So we

can substitute A by the algebraic closure of F(A). Hence we can
assume that f, g and h factor completely.
In this case the assertion follows by Proposition 4.1.8.

The above corollary is very useful, because computations with

determinants of high rank easily become very cumbersome.
Assuming that f is monic, we can give yet another character-
ization of the resultant. In this case the ring B = A[x]/(f) is a
free A-module with basis {1, x, . . . , xn−1 }, since one can use f to
express any monomial of higher degree in terms of this basis.
Any A-linear endomorphism of B is then represented by a ma-
trix, hence has a well defined determinant, which a posteriori is
independent of the choice of a basis. Let

φg : B → B

be the multiplication by g.

124
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 4.1 the resultant

Proposition 4.1.11. With the above notation we have

R(f, g) = det φg .

Proof. As in the previous proof, we can substitute A by the alge-

braic closure of F(A), hence we can assume that g factors com-
pletely. Note that both members of the identity are multiplicative
in g: the left hand side by Corollary 4.1.10, and the right hand
side thanks to the properties of the determinant and the fact that
φgh = φg ◦ φh .
So it is enough to prove the assertion when g has degree 1, say
g(x) = bx + c. As usual set

f(x) = xm + am−1 xm−1 + · · · + a1 x + a0 .

Since
φg (xi ) = bxi+1 + cxi
for i 6 m − 2 and

φg (xm−1 ) = bxm + cxm−1 =

= (c − bam−1 )xm−1 − bam−2 xm−2 − · · · − ba0 ,

a matrix for φg is the transpose of

 
c b 0 ··· 0
 0
 c b ··· 0 

 ..
M= . .

 
 0 ··· 0 c b 
−ba0 ··· −bam−2 c − bam−1

We have to prove that det M is the same as the determinant of

 
1 am−1 · · · a1 a0
b
 c 0 ··· 0
0
Syl(f, g) =  b c 0 ··· 0.
 .. 
. 
0 ··· 0 b c

Write the last row of M as

−b(a0 , . . . , am−1 ) + (0, . . . , 0, c).

125
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 computational methods

Using the multilinearity of the determinant we obtain

det M = cm − b det(N),

where  
c b 0 ··· 0

 0 c b ··· 0 

..
N= .
 
 . 
0 ··· 0 c b 
a0 ··· am−2 am−1
By Laplace expansion on the first column we get

det Syl(f, g) = cm − b det(N),

hence the thesis.

Remark 4.1.12. It is not a priori clear that the right hand side is
multiplicative in f; this would simplify the proof. Anyway this
follows as a byproduct.

4.2 discriminants
In this section we look at a particular case of resultant, called
the discriminant. Then we review some constructions from field
theory, and we see how the discriminant of a basis of a field
extension gives a generalization of this concept.
Recall that for a polynomial over any ring we can define the
formal derivative by requiring that
d k
x = kxk−1
dx
for monomials, and then extending by linearity.
Definition 4.2.1. Let A be an integral domain, f ∈ A[x], say

f(x) = an xn + an−1 xn−1 · · · + a0 ,

with an 6= 0. Then the discriminant of f is defined as

n 1
disc(f) := (−1)( 2 ) R(f, f 0 ).
an

126
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 4.2 discriminants

Remark 4.2.2. The change of sign and the fraction may appear
arbitrary, but this is the established definition.

As an example, let us compute the discriminant of a quadratic

polynomial f(x) = ax2 + bx + c over a field of characteristic 6= 2.
The derivative is f 0 (x) = 2ax + b, hence the Sylvester matrix is
 
a b c
Syl(f, g) := 2a b 0 ,
0 2a b

so
R(f, f 0 ) = ab2 − 2a(b2 − 2ac) = −a(b2 − 4ac),
and finally disc(f) = b2 − 4ac, which is the familiar formula.
In general, note that if n 6= 0 in A (for instance if A has char-
acteristic 0), the discriminant is a homogeneous polynomial of
degree 2n − 2 in the coefficients of f.

Proposition 4.2.3. Let f ∈ A[x], say

f(x) = an xn + an−1 xn−1 · · · + a0 ,

with an 6= 0, and assume that f 0 has degree n − 1. Let α1 , . . . , αn

be the roots of f in a splitting field of F(A), listed with multiplicities.
Then Y
disc(f) = an−2n (αi − αj )2 .
i<j

Proof. We can enlarge A and assume that the roots of f belong to

A. So we can factor f as

f(x) = an (x − α1 ) · · · (x − αn ).

The usual rules for differentiation yield

X
n
f 0 (x) = an (x − α1 ) · · · (x\
− αi ) · · · (x − αn ),
i=1

where the term (x\

− αi ) is omitted. Corollary 4.1.9 then gives the
thesis.

127
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 computational methods

n
Note that the sign (−1)( 2 ) in the definition of the discriminant
is exactly the sign needed to reorder the terms in the last part of
the proof.
Corollary 4.2.4. Under the same hypothesis, f has repeated roots if,
and only if, disc(f) = 0.
We next discuss a generalization of the discriminant in the con-
text of field theory. For the rest of the section, we fix a finite
separable extension of fields L/K, and denote by K a fixed al-
gebraic closure of K, which is determined up to isomorphism.
Recall from Section A.7 that in this situation we have the trace
homomorphism TrL/K : L → K, and the norm NL/K : L∗ → K∗ .
Looking at L as a vector space over K, we can consider the dual
space L∨ . For any α ∈ L we get a linear functional
Tα : L → K
defined by Tα (β) = TrL/K (αβ). This gives a K-linear map

T : L → L∨
sending α to Tα . By Proposition A.7.6, T is an isomorphism. In-
deed,
ker T = {α ∈ L | TrL/K (αβ) = 0 for all β ∈ L} = 0.
In other words, the K-bilinear pairing on L defined by (α, β) →
TrL/K (αβ) is nondegenerate.
In particular, given a basis u1 , . . . , un of L as a K-vector space,
we can find a dual basis v1 , . . . , vn , such that
TrL/K (ui vj ) = δij .
Definition 4.2.5. We shall say that the two bases u1 , . . . , un and
v1 , . . . , vn are dual with respect to the trace.
We are now ready to study the promised generalization of dis-
criminant.
Definition 4.2.6. Let L/K be a finite separable field extension of
degree n, and let σ1 , . . . , σn be the embeddings of L into K. Given
a1 , . . . , an ∈ L, consider the matrix A = (σi (aj ))ij . The discrimi-
nant of a1 , . . . , an is defined by
disc({a1 , . . . , an }) := (det(A))2 .

128
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 4.2 discriminants

Remark 4.2.7. Because of the square, the discriminant does not

depend on the ordering of the ai . Still, note that it depends on
the fields L and K, that are now fixed, even though the notation
omits this dependence.
There is another convenient formula for the discriminant. Just
note that det(A)2 = det(AT · A), and the matrix AT · A has entries
X
n
(AT · A)ij = σk (ai )σk (aj ) = TrL/K (ai aj ).
k=1

So we have the following expression:

disc({a1 , . . . an }) = det(TrL/K (ai aj ))ij . (4.2.1)
Remark 4.2.8. The last displayed formula shows in particular
that disc({a1 , . . . an }) ∈ K.
Proposition 4.2.9. The discriminant disc({a1 , . . . an }) is 0 if, and
only if, a1 , . . . , an are linearly dependent over K.
Proof. Assume that
c1 a1 + · · · + cn an = 0
for some nontrivial choice of c1 , . . . , cn ∈ K. Then we have
c1 σi (a1 ) + · · · + cn σi (an ) = 0
for every i, so the vector (c1 , . . . , cn ) lies in the kernel of the
matrix A = (σi (aj ))ij . It follows that

disc({a1 , . . . , an }) = (det(A))2 = 0.
Vice versa, assume that disc({a1 , . . . , an }) = 0. Using the ex-
pression (4.2.1) we see that the rows of the matrix (TrL/K (ai aj ))ij
are linearly dependent, hence for some choice of c1 , . . . , cn ∈ K
we have
Tai (c1 a1 + · · · + cn an ) = 0
for all i. If a1 , . . . , an are linearly independent, the functionals
Tai span the whole L∨ , hence
c1 a1 + · · · + cn an = 0,
contradiction.

129
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 computational methods

Proposition 4.2.10. Let a1 , . . . , an ∈ L and let b1 , . . . , bn ∈ L be

linear combinations with integer coefficients of the ai , say ~b = M~ a,
where M is a matrix with integer coefficients. Then

disc({b1 , . . . , bn }) = (det M)2 disc({a1 , . . . , an }).

Proof. For any embedding σ of L in the algebraic closure of K we

have σ(~b) = Mσ(~a), hence

disc({b1 , . . . , bn }) = det(σi bj )2 = (det M)2 (σi aj )2 =

= (det M)2 disc({a1 , . . . , an }).

We can now make the link with our previous notion of discrim-
inant. Take an element α ∈ K of degree n, and let L = K(α). So L
is spanned as a K-vector space by the powers 1, α, . . . , αn−1 .

Proposition 4.2.11. Let α be algebraic over K with minimal polyno-

mial fα of degree n, and take L = K(α). Then

disc(fα ) = disc({1, α, . . . , αn−1 }).

Proof. Let αi = σi (α) be the conjugates of α, so that

fα (x) = (x − α1 ) · · · (x − αn ).

Then the matrix

(σi (αj ))ij = (αji )ij
is in Vandermonde form, and its determinant is
Y
(αi − αj ).
i<j

The square of this is by definition disc({1, α, . . . , αn−1 }), so the

thesis follows from Proposition 4.2.3.

Corollary 4.2.12. With notations as above, we have

n
disc({1, α, . . . , αn−1 }) = (−1)( 2 ) NL/K (fα
0
(α)).

130
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 4.3 gröbner bases

Proof. Indeed this is the same of

n
disc(fα ) = (−1)( 2 ) R(fα , fα
0
),
and by Corollary 4.1.9 we have
Y
n
0 0 0
R(fα , fα )= fα (αi ) = NL/K (fα (α)).
i=1

We will come back to discriminants in the context of number

fields in Section 6.1.

4.3 gröbner bases

In this section we work in a polynomial ring A = k[x1 , . . . , xn ],
where k is a field. We want to solve some algorithmic problems
in A. The most basic problem is ideal membership: given a ∈ A
and an ideal I = (f1 , . . . , fk ), how do we test if a ∈ I?
The answer is easy if n = 1. In this case A is a principal ideal
domain, and we can find a generator f of I using the Euclidean
algorithm. Having done that, we just perform the division of a
by f and check if the remainder is 0. Alternatively, if we do not
want to rely on the fact that A is a principal ideal domain, we can
perform the division of a by f1 , obtaining
a = q1 f1 + r1 ,
then divide r1 by f2 and so on. If the last remainder rk is 0, a ∈ I,
and conversely.
This idea seems good to be generalized to a higher number of
variables, but we run into two problems. The first one, which
is easy to solve, is that there is no obvious way to perform the
division algorithm between polynomials in more than one vari-
able. The first step of the division algorithm is tho perform the
division between the highest order monomials, and when there
is more than one variable there is no single choice of what highest
order monomial means.
This is easily repaired with the following
Definition 4.3.1. A monomial order is a total order 4 on the set of
monomials in the indeterminates x1 , . . . , xn satisfying:

131
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 computational methods

1. if xα , xβ and xγ are monomials and xα 4 xβ , then xα+γ 4

xβ+γ ;
2. there exists no infinite descending chain

xα1  xα2  · · ·  xαr  · · · .

Property 2 just states that 4 is a well-ordering, so every non-

empty set of monomials admits a minimum. Property 1 is called
multiplicativity.
Remark 4.3.2. There is a natural bijective correspondence be-
tween the set of monomials and Nn , sending xα to the multiin-
dex α, and correspodingly we can give a definition of a monomial
order on Nn . Multiplication of monomials becomes addition of
vectors in Nn , and we shall switch between the two points of
view according to convenience.
Remark 4.3.3. A simple consequence of the axioms is that for
every α ∈ Nn we have xα < 1. Indeed if 1  xα , we get by
multiplicativity the infinite descending sequence

1  xα  x2α  · · · .

Again by multiplicativity, it follows that if xα divides xβ , then

xα 4 xβ .
In other words, every monomial order is a total order which
refines the partial order given by divisibility of monomials. In
particular the standard order is the only monomial order in one
variable.
We now give some examples of monomial orders. These are
better described as orders on Nn .
Example 4.3.4. (a) The first example is the lexicographic order LEX.
In order to compare two multiindices α = (α1 , . . . , αn ) and
β = (β1 , . . . , βn ) one first compares α1 with β1 . If α1 < β1
one puts α ≺ β, and conversely. If α1 = β1 , we compare α2
with β2 , and proceed until we find a different entry (unless
of course all entries coincide and so α = β).
In other words α  β if and only if the first nonzero entry
of α − β is positive. From this characterization, property 1

132
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 4.3 gröbner bases

follows immediately. Indeed summing γ to both α and β

does not change their difference. Property 2 follows easily by
induction on the number of variables.

(b) The graded lexicographic order DEGLEX is similar to LEX, but

to compare α and β we first compare their total degree. Here
the degree of α = (α1 , . . . , αn ) is

deg(α) := α1 + · · · + αn .

If deg(α) > deg(β) we declare α  β. For monomial of equal

total degree, we use the lexicographic ordering.

(c) Finally, we shall consider the graded reverse lexicographic order

DEGREVLEX. As with DEGLEX, we first compare the total
degree of α and β. If the degree is the same, we use the
reverse lexicographic order: in this order α  β if and only if
the last nonzero entry of α − β is negative.
The only thing which is not obvious is property 2, which is
proved as follows. In any descending chain, the total degree
is decreasing. Since this is a natural number, it must eventu-
ally stabilize. But there are only finitely many monomials of a
given total degree, hence we cannot have infinite descending
chains.

Remark 4.3.5. All algorithms that we consider require the choice

of a monomial order, and for most of them any one will do. Still,
the performance of the algorithm may depend on the order cho-
sen; this is why we choose not to pick any particular one from
the very beginning.

For the rest of the section, let us fix a monomial order 4.

Definition 4.3.6. Let f ∈ k[x1 , . . . , xn ]. The momomial of f which

is maximum with respect to 4 is called the leading monomial and
it is denoted by LM(f). Its coefficient in f is called the leading
coefficient of f and it is denoted by LC(f). The term LC(f)LM(f)
is called the leading term of f and it is denoted by LT(f).
If LM(f) = xα we define the multidegree of f as

mdeg(f) := α.

133
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 computational methods

We now describe the division algorithm between two polyno-

mials f and g. Our aim is to write

f = q · g + r,

where no monomial of r is divisible by LM(g).

Start from the leading monomial of f. If LM(f) is divisible by
LM(g), say
LM(f) = xα LM(g),
put the term
LC(f) α
h= x
LC(g)
inside q, and replace f by f − h · g.
If instead LM(f) is not divisible by LM(g), we just move the
leading term of f to r. Either way, the multidegree of f has low-
ered. We repeat the process until we reach the point where f = 0,
which will happen because 4 is a monomial order and the se-
quence of multidegrees is decreasing.
We can extend this process to define the division of f by a finite
set of polynomials g1 , . . . , gn . This is done by dividing f by g1 ,
then the remainder of this by g2 and so on.

Remark 4.3.7. The result of the division actually depends on the

monomial order we have chosen.

This process is best explained by way of an example.

Example 4.3.8. Take the lexicographic order on k[x, y] and con-

sider the division of f = x2 y + xy2 + y2 by f1 = xy − 1 and
f2 = y2 − 1. Dividing f by f1 we get f = (x + y)f1 + x + y2 + y.
Next, we put the x term into the remainder and divide y2 + y by
f2 to obtain
f = (x + y)f1 + f2 + (x + y + 1).
We cannot do any more steps since no monomial in x + y + 1 is
divisible by the leading term of either f1 or f2 .

Having discussed the issue of the monomial order, we can

now see a second, more serious, problem. Namely the ring
k[x1 , . . . , xn ] is not a principal ideal domain, so we cannot hope
to find a single generator for an ideal. Let’s take an ideal I =

134
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 4.3 gröbner bases

(f1 , . . . , fk ) and assume we want to check whether g ∈ I. The

naive idea would be to compute the successive divisions of g by
f1 , . . . , fk . Namely, we perform the first division to obtain
g = q1 f1 + r1 ,
then divide r1 by f2 obtaining
r1 = q2 f2 + r2
and so on. At the last step we are left with a remainder rk , which
we can consider as the remainder of the division of g by the
system of polynomials f1 , . . . , fk .
The problem is that this remainder is not determined by the
polynomials alone, but depends on the order we put on the fi .
It may even happen that for some ordering the remainder is 0,
thereby showing that g ∈ I, but for another ordering it is not (see
Exercise 6).
Example 4.3.9. We perform the same computation of Example
4.3.8 but this time we invert the order and we divide by f2 first.
Since LC(f) is not divisible LC(f2 ), x2 y goes in the remainder. By
dividing what remains, we get f = (x + 1)f2 + x2 y + x + 1. But
then x2 y is multiple of LC(f1 ) and the other terms have too small
degree, so the final result is
f = xf1 + (x + 1)f2 + 2x + 1.
To start solving this issue, we need some information about
monomial ideals. We recall from Exercise 11 in chapter 1 the
following
Definition 4.3.10. The ideal I ⊂ k[x1 , . . . , xn ] is called monomial
if it is generated by monomials.
In the same exercise, you proved that if f belongs to a mono-
mial ideal, then each of its monomials does. If I ⊂ k[x1 , . . . , xn ]
is any ideal, we denote by LT(I) the monomial ideal generated by
LT(f) for f ∈ I.
Remark 4.3.11. If I is an ideal, LT(I) is finitely generated by
Hilbert’s basis theorem 2.3.3. Hence it is generated by finitely
many monomials, and each of these monomials is multiple of
LT(f) for some f ∈ I. We deduce that there exist f1 , . . . , fm ∈ I
such that LT(f1 ), . . . , LT(fm ) generate LT(I).

135
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 computational methods

Now, it is not in general the case that any set of generators

f1 , . . . , fm of I will do.
Definition 4.3.12. A set of elements g1 , . . . , gm of I ⊂ k[x1 , . . . , xn ]
is called a Gröbner basis of I (with respect to a given momonial or-
der) if LT(g1 ), . . . , LT(gm ) generate LT(I).
The relevance of Gröbner bases for our task is given by the
following result.
Proposition 4.3.13. Let g1 , . . . , gm be a Gröbner basis for I and take
any f ∈ k[x1 , . . . , xn ]. Then there exists a unique r ∈ k[x1 , . . . , xn ]
such that
i) No monomial of r is divisible by any of LT(g1 ), . . . , LT(gm ).
ii) We can write f = g + r with g ∈ (g1 , . . . , gm ).
Proof. Existence is clear by the division algorithm. For the unique-
ness, assume f = g + r = g 0 + r 0 , with g, g 0 ∈ I and r, r 0 as
stated. Then r − r 0 ∈ I, so in particular LT(r − r 0 ) ∈ LT(I).
Since g1 , . . . , gm is a Gröbner basis, this is excluded by i) unless
r = r 0.
In particular, we see that the remainder of the division of f
by g1 , . . . , gm does not the depend on the order in which we
perform the division if g1 , . . . , gm is a Gröbner basis – although
the other coefficients in the division might change.
Corollary 4.3.14. Let g1 , . . . , gm be a Gröbner basis for I and take
any f ∈ k[x1 , . . . , xn ]. Then f ∈ I if, and only if, the remainder of f by
g1 , . . . , gm is 0.
Remark 4.3.15. This in particular implies that g1 , . . . , gm gener-
ate the ideal I, a fact that we did not assume in the definition of
a Gröbner basis.
This solves the problem of determining if f ∈ I in an algorith-
mic way, provided we can find a Gröbner basis for I. Now, such a
basis certainly exists by Remark 4.3.11. Before asking how to find
one such basis algorithmically, we look at a concrete example.
Example 4.3.16. (a) The set {xy − 1, y2 − 1} in k[x, y] is not a Gröb-
ner basis for the LEX order, since we showed in the previous
examples that the remainder is not unique.

136
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 4.3 gröbner bases

(b) We take k[x, y, z], again with the LEX order. The set

G = {g1 , g2 } = {x + y2 − 2yz + 3z − z5 , y + 3z2 − z3 }

is a Gröbner basis. In fact, let I = (G), take f ∈ LT(I) and

assume that LT(f) does not belong to the ideal generated by
LT(g1 ) = x and LT(g2 ) = y. Then LT(f) is a monomial in z
and by the property of lexicographic ordering f is a polyno-
mial in z alone.
On the other hand, f = f1 g1 + f2 g2 , hence it vanishes if we
specialize y = −3z2 + z3 and x = −y2 + 2yz − 3z + z5 . Since
f does not involve either x or y, there is nothing to specialize,
hence f = 0 identically. In fact, all we used was that we could
solve for y in g2 and then solve for x in g1 .

We are now left with the task of producing a Gröbner ba-

sis for an ideal I ⊂ k[x1 , . . . , xn ] given a set of generators F =
{f1 , . . . , fm }. If F is not already a Gröbner basis, there exists f ∈ I
such that LT(f) does not belong to (LT(f1 ), . . . , LT(fm )).
If we have a way of producing such f algorithmically, we can
add f to F and ask again if we have a Gröbner basis. If not
we can enlarge F again, and so on. The process must end since
k[x1 , . . . , xn ] is Noetherian, leaving us with a Gröbner basis.
Hence, we are reduced to the following problem: given a set
f1 , . . . , fm ∈ k[x1 , . . . , xn ], either show that it is a Gröbner basis, or
produce a polynomial that shows that it is not.
Let f, g be two polynomials, and consider a linear combination
h = af + bg with a, b ∈ k[x1 , . . . , xn ]. If the highest terms of af
and bg do not cancel out, LT(h) ∈ (LT(f), LT(g)). So we start by
looking for cases where a cancellation occurs.

Definition 4.3.17. The S-polynomial of f and g, denoted S(f, g),

is given by
m m
S(f, g) = f− g,
LT(f) LT(g)
where m is the mininum common multiple of LM(f) and LM(g)
(which makes sense as both are monomials).

By construction, S(f, g) has degree less than m, because of can-

cellation.

137
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 computational methods

Lemma 4.3.18. Let f1 , . . . , fm be polynomials of the same degree (mean-

ing that LM(f1 ) = · · · = LM(fm )). If a linear combination

X
m
h= ci fi
i=1

with ci ∈ k has a smaller degree, then h is a linear combination of the

S(fi , fj ) with coefficients in k.

Proof. Let ai = LC(fi ) and pi = fi /ai . The S-polynomials are

simply S(fi , fj ) = fi /ai − fj /aj = pi − pj , so that the expression
for h becomes

h = c1 a1 (p1 − p2 ) + (c1 a1 + c2 a2 )(p2 − p3 ) + · · · +

+ (c1 a1 + · · · + cm−1 am−1 )(pm−1 − pm ) +
+ (c1 a1 + · · · + cm am )pm .
P
Our hypothesis amounts to ci ai = 0, hence we have the de-
sired expression for h.

We can now state the main technical result of this section.

Theorem 4.3.19 (Buchberger’s criterion). Let I ⊂ k[x1 , . . . , xn ] be

an ideal with generators F = {f1 , . . . , fm }. Then F is a Gröbner basis if
and only if for each i, j, the remainder of S(fi , fj ) divided by F is 0.

Note that in the theorem the order in which we use the fi to

perform the division is arbitrary, and in fact any ordering will
do.

Proof. Let h ∈ I so that

X
m
h= fi gi . (4.3.1)
i=1

Let α be the (multi)degree of the highest monomial appearing in

any of the fi gi . If h has itself degree α, its leading monomials
lies in (LT(f1 ), . . . , LT(fm )). By contradiction, assume that we
find h such that this is not the case, so cancellation happens in
the highest degree.

138
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 4.3 gröbner bases

Let S be the set of indices i for which fi gi has degree α. Then

the previous lemma ensures that
X X
fi gi = cij S(fi gi , fj gj )
i∈S i,j∈S

for some constants cij . Note that S(fi gi , fj gj ) is itself a linear

combination of fi and fj with coefficients in k[x1 , . . . , xn ], but
has degree lower than α. Hence we can rewrite (4.3.1) as

X
m
h= fi gi0 ,
i=1

but this time the maximum degree of any combination fi gi0 is

less than α.
We can repeat the process at will, constructing an infinite de-
scending sequence of degrees, thereby contradicting the defini-
tion of monomial order.

As a consequence of Buchberger’s criterion, we finally have

Buchberger’s algorithm to produce a Gröbner basis for an ideal I
starting from a set of generators F = {f1 , . . . , fn }.

1. For each pair of polynomials fi , fj , we consider the S-poly-

nomial S(fi , fj ) and perform the division with respect to
F.

2. If for some pair i, j the remainder is not 0, we add S(fi , fj )

to the set and start again from 1.

3. Otherwise F is a Gröbner basis by Buchberger’s criterion.

If we want to check if a given polynomial f lies in I we add a

final step:

4. Once we have a Gröbner basis G, perform the division of f

by G. If the remainder is 0, then f ∈ I, and conversely.

Remark 4.3.20. As a corollary, we also obtain an algorithm for

ideal inclusion or equality. In fact, say I = (f1 , . . . , fn ) and J =
(g1 , . . . , gm ). Then to check if I ⊂ J it is enough to check whether
each fi ∈ J.

139
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 computational methods

4.4 more algorithmic operations

In the previous section, we have learned to solve a few problems
algorithmically. We work inside k[x1 , . . . , xn ] and we assume that
ideals are given by a finite set of generators. By computing Gröb-
ner bases with respect to any monomial order, we are able to

1. determine if a polynomial f belongs to an ideal I, or

2. for two ideal I and J, determine whether I ⊂ J, and hence

also detect equality.

It is also trivial to compute generators for I ∪ J and I · J – hence

our next problem is to compute generators of I ∩ J.
To do this, we first give a small introduction to elimination the-
ory. As we have discussed in Section 4.1, the core idea of elim-
ination theory is to solve a system of polynomial equations by
trying to eliminate one variable at a time. The main result that
links it to Gröbner bases is the following:

Theorem 4.4.1 (Elimination theorem). Let I ⊂ k[x1 , . . . , xn ] and

G a Gröbner basis for I with respect to LEX. Then, for every i 6 n,
G ∩ k[xi , . . . , xn ] is a Gröbner basis for I ∩ k[xi , . . . , xn ].

In particular, we can find generators for I ∩ k[xi , . . . , xn ] by tak-

ing the elements of G that do not involve the first i − 1 variables.

Proof. Let f ∈ I be a polynomial that does not involve the first i −

1 variables. The leading monomial LM(f) is divisible by LM(g)
for some g ∈ G, and clearly LM(g) does not involve the first i − 1
variables. Since this is the leading monomial for LEX, all other
monomials have the same property, hence g ∈ G ∩ k[xi , . . . , xn ].

Example 4.4.2. [Boo08] Consider the system of equations


 2
x + y + z = 1
x + y2 + z = 1


x + y + z2 = 1

140
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 4.4 more algorithmic operations

To solve this system, we take the ideal I = (f1 , f2 , f3 ) where

f1 (x, y, z) = x2 + y + z − 1
f2 (x, y, z) = x + y2 + z − 1
f3 (x, y, z) = x + y + z2 − 1

The three elements are not a Gröbner basis, but we can apply
Buchberger’s algorithm to find one. A Gröbner basis for I is
given by

g1 (x, y, z) = x + y + z2 − 1
g2 (x, y, z) = y2 − y − z 2 + z
g3 (x, y, z) = 2yz2 + z4 − z2
g4 (x, y, z) = z2 (z − 1)2 (z2 + 2z − 1)

The last equation can be √solved for z giving the four solutions
z = 0, z = 1 and z = −1 ± 2. We can then substitute back z in
g2 or g3 to solve for y and finally back into g1 to solve for x. For
instance, when z = 0, we recover y = 0 or y = 1 from g2 , and
accordingly x = 1 or x = 0 from g1 . Work out the other solutions!

Theorem 4.4.1 guarantees that we can always find a way to

eliminate variables in a system of equations, and then we can
backtrack to solve for one variable at a time.
As we promised, another application for this result is the com-
putation of generators for I ∩ J. We use the following

Proposition 4.4.3. Let I and J be ideals of k[x1 , . . . , xn ]. Inside the

ring k[t, x1 , . . . , xn ] consider the ideal L = t · I + (1 − t) · J. Then

L ∩ k[x1 , . . . , xn ] = I ∩ J.

Proof. If f ∈ I ∩ J we can write f = tf + (1 − t)f, which shows one

inclusion. For the other, let g ∈ k[x1 , . . . , xn ] and assume that

g = ta + (1 − t)b

for some a ∈ I, b ∈ J. Then writing g = t(a − b) + b, we see that

a = b, so they both lie in I ∩ J.

141
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 computational methods

Using this result, we now have an algorithm to compute gen-

erators for I ∩ J, since it is easy to compute generators for t · I +
(1 − t) · J and then we can apply Theorem 4.4.1 to eliminate the t
variable.
A similar reduction exists for the radical of an ideal.

Proposition
√ 4.4.4. Let k be a field, I ⊂ k[x1 , . . . , xn ] an ideal. Then
f ∈ I if, and only if, 1 belongs to the ideal generated by I and 1 − tf
in the ring k[x1 , . . . , xn , t].

Proof. Let A = k[x1 , . . . , xn ], and

√ consider the multiplicative set
S ⊂ A generated by f. Then f ∈ I if and only if I becomes trivial
in S−1 A. But there is an isomorphism

φ : A[t]/(1 − tf) → S−1 A,

t → 1/f

hence this is equivalent to the condition that I and 1 − tf generate

the whole A[t].

Using this result, we get another algorithmic

√ application of
Gröbner bases: to check whether f ∈ I, we only need to test
whether 1 ∈ (I, 1 − tf), which is something we can easily do by
computing a Gröbner basis for the latter ideal.
We end with another example that shows the far reaching ap-
plicability of Gröbner bases.

Example 4.4.5. [Mir11] Many problems in Euclidean geometry

can be reduced to systems of polynomial equations by choosing
coordinates. These, in turn, can be solved using elimination the-
ory as we outlined above. This is the observation at the heart of
automatic theorem proving in Eucliden geometry.
For a concrete example, consider the circle theorem of Apollo-
nius: let ABC be a triangle right in A. The midpoints of the three sides
and the foot of the altitude drawn from A to BC all lie on one circle.

142
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 4.4 more algorithmic operations

C
H
M2 M3

B
A M1

To prove this, we place A in the origin and choose coordinates

(u1 , 0) for B and (0, u2 ) for C. The midpoints are M1 = (x1 , 0),
M2 = (0, x2 ) and M3 = (x3 , x4 ). We also let H = (x5 , x6 ) be the
foot of the altitude and O = (x7 , x8 ) be the center of the circle
through the midpoints.
The construction gives a system of equations. First we have
2x1 = u1 and three other equations for the midpoints. Then, the
condition AH ⊥ BC becomes x5 u1 = x6 u2 , and the fact that H ∈
BC becomes x5 u2 + x6 u1 = u1 u2 . Finally, we get two quadratic
equations for M1 O = M2 O and M1 O = M3 O.
The whole system is


 2x1 − u1 = 0



 2x2 − u2 = 0





 2x3 − u1 = 0

2x − u = 0
4 2

 x u − x6 u2 = 0


5 1




x u
5 2 + x6 u1 − u1 u2 = 0


 2 2 2 2
(x1 − x7 ) + x8 − x7 − (x8 − x2 ) = 0


(x1 − x7 )2 + x28 − (x7 − x3 )2 − (x8 − x4 )2 = 0.

The thesis is given by the equation

(x5 − x7 )2 + (x6 − x8 )2 − (x1 − x7 )2 − x28 = 0.

Using Gröbner bases, one can show that the latter polynomial
lies in the ideal generated by the polynomials appearing in the
system. This can be automated using the algorithms of this chap-
ter, and this technique allows one to transform statements in Eu-
clidean geometry into algorithmic problems.

143
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 computational methods

4.5 exercises
1. Let k be an algebraically closed field, and I ⊂ k[x1 , . . . , xn ] a
radical ideal, say I = (f1 , . . . , fk ). For each pair i 6= j, where 1 6
i 6 k, consider the resultant ri,j := R(fi , fj ), where the resultant is
taken with respect to the variable x1 , so that ri,j ∈ k[x2 , . . . , xn ].
Assume that each p fi has positive degree in x1 . Show that I ∩
k[x2 , . . . , xn ] = (ri,j ).

2. Try to apply resultants to find solutions to a generic system of

polynomial equations. What difficulties do you find? Work out a
concrete case by using resultants to solve the system in Example
4.4.2.

3. A Gröbner basis G = {g1 , . . . , gr } is called reduced if each gi

is monic (that is, LC(gi ) = 1) and for each i, LM(gi ) does not
divide any monomial of any gj for j 6= i.
Show that given a Gröbner basis for an ideal I, one can con-
struct explicitly a reduced one. Moreover, show that a reduced
Gröbner basis for I is unique (up to order). Derive a simple algo-
rithm to test ideal equality in polynomial rings.

4. Let I ⊂ k[x, y] be the ideal generated by xy − x and x2 − y.

Compute a Gröbner basis for I with respect to LEX.

5. Let I ⊂ k[x, y] be the ideal generated by x2 and xy + y2 . Com-

pute a Gröbner basis for I with respect to LEX.

6. Inside k[x, y], consider the polynomial f = x2 y3 − 2xy2 and the

ideal I = (f1 , f2 ), where f1 = x2 y − 2x and f2 = y3 + 4. Using
the LEX order, perform the division of f by f1 , then f2 . Then, do
the same computation using f2 , then f1 . Show that in one case,
the remainder is 0, while in the other case it is nonzero.

7 ([RZ10]). Consider the polynomials f(x, y) = (y2 − y)x2 + x

and g(x, y) = (y2 − y)x + y in k[x, y]. Compute the resultant with
respect to the x variable, R(f, g) = y2 (y − 1)2 . Letting I = (f, g),
use Gröbner bases to compute that I ∩ k[y] = (y). What happens
when you use resultants to solve the system given by f and g?
Explain the discrepancy between the two techniques that we have
introduced for the elimination problem.

144
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 4.5 exercises

8. Let w ∈ Rn be a vector such that the components wi are

linearly independent over Q. Given α, β ∈ Nn , define α ≺w β if
w · α < w · β, where · denotes scalar product. Show that ≺w is a
monomial order, called a weight order.

9. Let w ∈ Rj , and consider the order defined as follows. Given

α, β ∈ Nn , define α ≺w β if w · (α1 , . . . , αj ) < w · (β1 , · · · , βj ),
where · denotes scalar product. If the two products are equal,
fallback to comparing α and β by DEGREVLEX. This is called an
elimination order for {1, . . . , j}. Show that it has the property that
if a variable x1 , . . . , xj does not appear in LM(f), then it does not
appear in f at all.

10. Let ≺w be the elimination order defined in the previous ex-

ercise. Show that it can be used to compute the elimination ideal
for the first j variables. Namely, let I ⊂ k[x1 , . . . , xn ] and G a
Gröbner basis for I with respect to ≺w . Then G ∩ k[xj+1 , . . . , xn ]
is a Gröbner basis for I ∩ k[xj+1 , . . . , xn ].

11. Consider a finite undirected graph G where each pair of ver-

tices is connected by at most one edge. For each vertex vi intro-
duce a variable xi and consider the system of equations given by
x3i = 1 (for each i) and x2i + xi xj + x2j = 0 whenever vi and vj are
adiacent.
Show that G can be colored with 3 colors (that is, each vertex
can be assigned one of 3 colors such that adiacent vertices get
different colors) if, and only if, the system has a solution. Derive
an algorithm to test whether a graph is 3-colorable.

12. Verify that the proof of correctness and termination of Buch-

berger’s algorithm does not require the full strength of Hilbert’s
basis theorem, but only the special case that states that monomial
ideals in k[x1 , . . . , xn ] are finitely generated.

13. The following result is known as Dickson’s lemma. Consider

the partial order on Nn defined by a 4 b if and only if ai 6 bi
for all i = 1, . . . , n. Then every nonempty subset S ⊂ Nn has
finitely many (at least one) minimal elements. Prove Dickson’s
lemma directly, and then show that it is equivalent to the form
of Hilbert’s basis theorem for monomial ideals required in the
previous exercise.

145
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 computational methods

In order to implement the algorithms described in this chapter,

one has to be able to implement operations on the base field k. A
common choice is to take the finite field Fq over q = pd elements.
This can be described as the quotient Fp [x]/(f), where f is an
irreducible polynomial of degree d over Fp , so one can perform
computations in k = Fq by finding one such polynomial f, and
then using the division algorithm in Fp [x]. This inspires one
to find algorithms that generate an irreducible polynomial of a
given degree d over Fp . The next exercises, up to Exercise 15
discuss this problem.
14. Let p be a prime, Fp = Z/pZ, and take d > 2. Compute the
proportion of polynomials of degree d in Fp that are irreducible.
Namely, the number of monic polynomials of degree d is pd , and
the number of irreducible ones is asymptotic to pd /d as d → ∞.
15. By the previous exercise, a viable – even if a little costly – strat-
egy to generate an irreducible polynomial of degree d is to gener-
ate random polynomials and test whether they are irreducible. In
this exercise, we are going to provide such a test of irreducibility.
Namely, let f ∈ Fp [x] be a polynomial of degree d. Show that
r
if f is reducible, then gcd(f, xp − x) is nontrivial for some r 6
deg f/2. Hence a test for irreducibility is just to compute all such
gcd and check if any of them is not a constant.
16. For a polynomial f ∈ k[x1 , . . . , xn ] define the Netwon’s polytope
of f as the convex hull (in Rn ) of points α ∈ Nn such that xα is
a nonzero monomial of f.
Show that if m = xα is the leading monomial of f for any
monomial order, then α is extremal in the Netwon polytope of f
(that is, if we omit α in the above construction, we get a strictly
smaller convex hull).
17. Find a solution in C3 to the system

 2 2
xy − z − z = 0
x2 y − y = 0

 2
y − z2 = 0.
18. Gröbner bases can be used to prove Heron’s formula for the
area of a triangle
s2 = p(p − a)(p − b)(p − c), (4.5.1)

146
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 4.5 exercises

where s is the area, a, b, c are the side lengths and p = a+b+c

2 is
the semiperimeter.
To prove it, choose a reference system such that the vertices are
in (0, 0), (0, a) and (x, y). By the Pythagorean theorem, we have
the relations

b2 = x2 + y2
c2 = (a − x)2 + y2

and the area satisfies 2s = a · y. Let I ⊂ R[s, a, b, c, x, y] be the

ideal generated by these relations. Use elimination theory to find
the intersection I ∩ R[s, a, b, c], and show that it is generated by
the single equation 4.5.1.
19. Let A = k[x1 , . . . , xn ]. Follow the various connections in this
chapter and write an explicit algorithm√ to test whether a polyno-
mial f ∈ A belongs to the radical I of an ideal I = (g1 , . . . , gk ) ⊂
A.
20. Let A = k[x1 , . . . , xn ] and B = k[y1 , . . . , ym ], and let f : A → B
be a homomorphism. Let fi := f(xi ) ∈ B, and define the ideal

If := (x1 − f1 , . . . , xn − fn ) ∈ k[x1 , . . . , xn , y1 , . . . , ym ].

Show that ker f = If ∩ A. Use this result and the elimination

theorem to deduce an algorithm to compute ker f.
21. A polynomial f ∈ k[x1 , . . . , xn ] is called symmetric if it is in-
variant with respect to the action of the symmetric group Sn that
permutes the indeterminates x1 , . . . , xn . The elementary symmetric
polynomials are the polynomials s1 , . . . , sn defined by
X
sk = xi1 · · · xik .
i1 <i2 <···<ik

Of course, these polynomials are in fact symmetric.

Prove that every symmetric polynomial is a polynomial in the
elementary symmetric polynomials s1 , . . . , sn – in other words

k[x1 , . . . , xn ]Sn = k[s1 , . . . , sn ].

In general, it rarely happens that the invariant ring for the action
of a group has this simple structure, see [Dol03]. (Use the order

147
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 computational methods

LEX and do induction on the multidegree. Prove that the multi-

degree of a symmetric polynomial has the form α = (α1 , . . . , αn ),
where α1 > α2 > · · · > αn .)

22. Let I ⊂ k[x1 , . . . , xn ] be an ideal generated by binomials.

Show that there exists a Gröbner basis for I which consists of
binomials (follow the steps of Buchberger’s algorithm). In fact,
even a reduced Gröbner basis for I (see Exercise 3) is composed by
binomials. For much more about binomial ideals, see [HHO18].

23. The result of the preceeding exercise does not hold for poly-
nomials composed of three terms. In fact, take the ideal I =
(x1 + x2 + x3 , x1 + x4 + x5 ) in the ring k[x1 , . . . , x5 ]. Show that
a reduced Gröbner basis for I with respect to DEGREVLEX con-
tains the polynomial x2 + x3 − x4 − x5 .

148
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 5
INTEGRAL DEPENDENCE

In this chapter we study the so called integral extensions: given

an inclusion of rings A ⊂ B we want to understand what hap-
pens for elements of B which satisfy a monic polynomial equa-
tion with coefficients in A. This is the obvious generalization of
the concept of algebraic elements from field theory (see Section
A.1). Many results from field theory can be transferred to this
more general situation more or less literally, and we do so in the
first section. For example we prove that the set of x ∈ B that
satisfy a monic polynomial equation with coefficients in A is a
subring of B, called the integral closure of A. We also prove the
going up and going down theorems, which describe the behavior
of prime ideals in integral extensions, and then prove the famous
Noether normalization lemma, which gives a structure theorem for
integral extensions for k-algebras. Finally, we see how this theory
allows us to construct many important examples of Dedekind do-
mains.

5.1 integral extensions

We want to mimic the theory of algebraic elements for fields in
the more general context of rings.

Definition 5.1.1. Let A ⊂ B be an extension of rings. We say that

an element b ∈ B is integral over A if it satisfies an equation

bn + an−1 bn−1 + · · · + a0 = 0, (5.1.1)

where ai ∈ A.

149
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 integral dependence

Definition 5.1.2. The subset of B of integral elements over A is

called the integral closure of A inside B. If the integral closure
amounts just to A, we say that A is integrally closed in B. If it is
the whole of B, we say that B is integral over A.

Remark 5.1.3. We note that the Equation (5.1.1) that we ask b to

satisfy is monic; this is irrelevant when working with fields since
we can always divide everything by the leading coefficient.

We shall see soon that the integral closure of A is a subring

of B as expected. Explicitly we have to see that if a, b ∈ B are
integral over A, then a + b and ab are integral too. In order to
do this we take a roundabout way: we do not exhibit a monic
polynomial satisfied by a + b or ab; rather we find an equivalent
condition for an element to be integral.
To state the result in full generality we need a

Definition 5.1.4. Let M be an A-module. We say that M is faithful

if the induced homomorphism A → End(M) is injective. Equiva-
lently the ideal
\
Ann(M) := Ann(m)
m∈M

is 0.

Example 5.1.5. (a) If A ⊂ B are rings, B is a faithful A-module.

Indeed assume a · b = 0 for all b ∈ B; taking b = 1 we find
a = 0.

(b) If I is a nonzero ideal of A, the inclusion I ⊂ Ann(A/I) shows

that A/I is not a faithful A-module.

(c) The infinite direct sum

p prime Z/(p)
L
M :=

is a faithful Z-module, but any finitely generated submodule

of M is not faithful.

Theorem 5.1.6. Let A ⊂ B be rings, b ∈ B. The following conditions

are equivalent:

i) b is integral over A;

150
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 5.1 integral extensions

ii) the ring A[b] is finitely generated as an A-module;

iii) there exists a ring C containing b, with A ⊂ C ⊂ B, such that C
is finitely generated as an A-module;
iv) there exists a faithful A[b]-module M which is finitely generated
as an A-module.
Note that (part of) this criterion is completely analogous to the
usual statement for fields: an element b is algebraic over the field
k if and only if k[b]/k is a finite extension (Remark A.1.3).
Proof. If b is integral over A, b satisfies an equation of the form
(5.1.1). We can use the equation to express every power bk for
k > n as a linear combination of 1, b, . . . , bn−1 with coefficients
in A, so i) implies ii).
That ii) implies iii) is obvious. To show that iii) implies iv) it is
enough to take M = C.
Now assume iv). Consider the map

µb : M → M

given by multiplication by b. If we choose a finite set m1 , . . . , mk

of generators of M, we can write

X
k
µ b mi = b · mi = aij mj ,
j=1

for some aij ∈ A. It follows that the matrix

D := b · Ik×k − (aij )

satisfies D · m = 0 for every element

X
n
m= λi mi ∈ C.
i=1

Let D 0 be the matrix of cofactors of D; that is, the (i, j) coeffi-

cient of D 0 is (−1)i+j times the determinant of the minor of D
obtained by removing the i-th column and the j-th row. Then it
is well known from linear algebra that

D 0 · D = det D · I. (5.1.2)

151
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 integral dependence

Hence, from D 0 D · m = 0 we deduce

det D · m = 0

for every m ∈ M. Note that the latter is a scalar equation. Since

M is faithful, we see that det D = 0; now just develop this deter-
minant to find the desired monic equation for b, thereby proving
i).
Remark 5.1.7. Probably Equation (5.1.2) is known to some read-
ers only in the case of matrices with coefficients in a field. Sim-
ilarly to what we have done in Proposition 4.1.8, this particular
case implies that the same equation holds over any ring.
First, note that if A is an integral domain with quotient field k,
the result for k implies that for A trivially. In particular Equation
(5.1.2) holds in the ring Z[xij ], where xij are indeterminates, for
i, j = 1, · · · , k. Now for any ring A and for any matrix D = (dij )
with coefficients in A we can find a homomorphism

Z[xij ] → A

sending xij to dij , and the result follows by specialization.

The method of proof that we have used is known as the deter-
minant trick. Notice that it also yields a more explicit form of
Nakayama’s lemma. More explicitly, let M be a finitely gener-
ated module over A – say m1 , . . . , mn are the generators. If I is
an ideal such that I · M = M, we can use the above identity to
find an explicit element x ∈ I such that 1 + x ∈ Ann M.
Theorem 5.1.8 (Nakayama’s lemma, explicit form). Let M be a
module over the ring A, generated by m1 , . . . , mn . Let I be an ideal
such that I · M = M, and write
X
n
mi = aij mj .
j=1

Let A be the matrix with coefficients (aij ), and y = det(id −A). Then
y ∈ 1 + I and y · M = 0.
The standard form of Nakayama’s lemma follows immediately:
if I ⊂ J(A), 1 + x is a unit for every x ∈ I, hence M must be 0.

152
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 5.1 integral extensions

Proof. By hypothesis, the matrix B = id −A annihilates the vec-

tor (m1 , . . . , mn ). By (5.1.2), det B · id also annihilates this vector,
hence det B · m = 0 for all m ∈ M. All summands in the expan-
sion of the determinant belong to I, except for the product of n
factors 1.
Theorem 5.1.6 has some important corollaries.
Corollary 5.1.9. 1 Let A ⊂ B be rings. The integral closure of A in B
is a subring of B.
Proof. Let a, b ∈ B be integral over A. Then the ring C = A[a, b]
is finitely generated as an A-module; since a + b and a · b belong
to C we see that both are integral over A.
Corollary 5.1.10. Let f : A → B be a morphism of rings. If B is
finitely generated as an A-algebra and it is integral over A, it is finitely
generated as an A-module.
Proof. Write
B = A[b1 , . . . , bn ].
Each of the bi is integral over A, so the claim follows by Theorem
5.1.6 and induction.
Corollary 5.1.11. Let A ⊂ B ⊂ C be rings. If B is integral over A and
C is integral over B, C is integral over A.
Proof. Assume c ∈ C satisfies an equation
cn + bn−1 cn−1 + · · · + b0 = 0,
for some bi ∈ B. The ring
B 0 := A[b1 , . . . , bn−1 ]
is finitely generated as an algebra an integral over A, hence it is
finitely generated as an A-module. Moreover B 0 [c] is finitely gen-
erated as a B 0 -module. It follows that B 0 [c] is finitely generated
as an A-module, so c is integral over A.
1 Indeed this is the origin of the word ring. Dedekind observed that he could obtain
interesting sets of numbers, closed under addition and multiplication, of the form
Z[α], where α is integral over Z. Once he had the first powers 1, α, . . . , αn−1 ,
higher powers could be expressed recursively using an equation for α. In the
important case where αn = 1, he just had to cycle through the powers; hence he
decided to use the expression ring of numbers for such a set, and the name “ring”
has then stuck.

153
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 integral dependence

This easily implies that the integral closure is idempotent.

Corollary 5.1.12. Let A ⊂ B be rings. If C is the integral closure of A
in B, C is integrally closed in B.
Definition 5.1.13. We say that A is integrally closed (or normal) if
it is so in the total ring of fractions F(A).
The simplest example is the following.
Proposition 5.1.14. A unique factorization domain is integrally closed.
Proof. Assume x = s/t satisfies (5.1.1). We can choose s, t ∈ A
without common factors. Assume t ∈ / A∗ ; clearing denominators
we get
sn + an−1 sn−1 t + · · · + a0 tn = 0.
Hence t divides sn , contradiction.
Example 5.1.15. (a) Let A be an integral domain with field of
fractions k, and let k 0 /k be an extension of fields. Then the
integral closure of A inside k 0 is integrally closed. This is
often applied with A = Z, to produce a number of rings of
interest in arithmetics.
Often it is difficult to determine exactly the integral closure,
but we will see that this can be done in some cases. The case
of quadratic extensions of Q is Example 6.1.2, while the case
of cyclotomic extensions is treated in Section 6.8.
(b) By way of example, we can check that Z[i] is integral over Z.
Indeed the element z = a + ib satisfies
z2 − 2az + (a2 + b2 ) = 0.
Since Z[i] is a unique factorization domain, it is integrally
closed, so it must be the integral closure of Z inside Q(i).
(c) The quotient of an integrally closed ring need not be inte-
grally closed. For instance take A = k[x, y], which is a unique
factorization domain, hence integrally closed. Consider the
ideal I = (x2 − y3 ); then A/I is not integrally closed.
Indeed
y
f=
x
satisfies f2 − x = 0, but f ∈
/ A/I.

154
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 5.1 integral extensions

Remark 5.1.16. In spite of the previous example, in some sense

integral closure behaves well with respect to quotients.
More precisely, let A ⊂ B be rings, I ⊂ B. If B is integral over
A, B/I is integral over A/A ∩ I. This is immediately checked, by
reducing a monic equation for b ∈ B modulo I.
For the localization we have:
Proposition 5.1.17. Let A ⊂ B be rings, S ⊂ A a multiplicative set,
and let A be the integral closure of A in B. Then S−1 A is the integral
closure of S−1 A in S−1 B.
In particular, if B is integral over A, S−1 B is integral over S−1 A.
Proof. Let b ∈ A, s ∈ S. Since b is integral over A, b satisfies an
equation
bn + an−1 bn−1 + · · · + a0 = 0,
whence  n
an−1 b n−1
 
b a
+ + · · · + n0 = 0,
s s s s
which shows that b/s is integral over S−1 A.
Vicecersa assume b/s is integral over S−1 A. A monic equation
for b/s can be written in the form
 n  n−1
b a b a
+ n−1 +···+ 0 = 0
s t s t
for some t ∈ S (after multiplying denominators). Then

(bt)n + an−1 st · (bt)n−1 + · · · + a0 sn tn−1 = 0,

so bt ∈ A. Hence
b bt
= ∈ S−1 A.
s st
Corollary 5.1.18. Let A be an integral domain. A is integrally closed
if, and only if, every localization AP at prime ideals is.
Proof. Let k be the field of fractions of A, A the integral closure of
A inside k. Let ι : A ,→ A be the inclusion; for each prime P ⊂ A
let S = A \ P and consider the map

ιP : AP ,→ S−1 A

155
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 integral dependence

induced in the localization. By the proposition S−1 A is the inte-

gral closure of AP in k.
By Corollary 1.6.20, ι is surjective if, and only if, each ιP is,
hence the thesis.
As another application of this concepts, we can give an alterna-
tive characterization of Dedekind rings.
Theorem 5.1.19. Let A be an integral domain. A is a Dedekind ring
if, and only if, it satisfies the following conditions:
i) A is Noetherian;
ii) A is integrally closed;
iii) every nonzero prime ideal of A is maximal.
Proof. First assume that A is a Dedekind ring. A is Noetherian
by definition, and satisfies iii) by Remark 3.4.5. We only need to
show that A is integrally closed, but that is identical to Proposi-
tion 5.1.14, using unique factorization for ideals.
For the converse, assume A satisfies the conditions of the The-
orem, and let P be a prime of A. Then AP is again Noether-
ian and integrally closed, and moreover is local with maximal
ideal M = PAP . By primary decomposition, every nonzero ideal
I ⊂ AP is the intersection of primary ideals; since M is the
only nonzero prime of AP , each component √ must be M-primary.
Hence I is M-primary itself, in particular I = M.
First we show that M is principal. Start from any nonzero a ∈
M; by Proposition 2.3.10 there is some power Mn ⊂ (a). Take n
minimal, so Mn−1 6⊂ (a), and take some element b ∈ Mn−1 \ (a).
Let x := a/b; we claim that M = (x). Note that by construction

1 1
M ⊂ Mn ⊂ A,
x a
since every element of Mn is multiple of A. On the other hand
assume
1
M ⊂ M;
x
since M is a faithful A[1/x]-module, it follows that x is integral
over A by Theorem 5.1.6. Since A is integrally closed, 1/x =
b/a ∈ A, which we have excluded.

156
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 5.2 going up and down

Hence
1
M = A,
x
which means M = (x). Now we exploit that the choice of b above
was arbitrary in Mn−1 \ (a).
Taking b = xn−1 shows that a/xn−1 is another generator x 0
of M, say x 0 = xu for some u ∈ A∗ . In particular (a) = (xn ), so
every principal ideal is a power of M.
Since every element of A is a power of x up to a unit, every
ideal is generated by powers of x, so it must be a power of M.
Corollary 5.1.20. A unique factorization domain A is a principal ideal
domain if, and only if, every prime P ⊂ A is maximal.
Proof. One implication is Proposition 2.1.4.
Vice versa, let A be unique factorization domain; then A is
integrally closed by Proposition 5.1.14, and we are assuming that
every prime is maximal. If we show that A is Noetherian, it is a
Dedekind ring, and so we are done by Proposition 3.4.9.
The first part of the proof of Proposition 3.4.9 carries over, and
tells us that prime ideals of A are principal. By Exercise 8 of
Chapter 2, this is enough to show that A is Noetherian.

5.2 going up and down

In this section we consider the following situation. Let A ⊂ B be
an integral extension; what is the relation between prime ideals
in A and B? It turns out that there is a beautiful description
which relates chains of prime ideals in the two rings. This is due
to Cohen and Seidenberg ([CS46]).
Let P ⊂ A and Q ⊂ B be primes. If P = Q ∩ A, we shall say
that Q is over P, or vice versa P is under Q. The first case that we
study is that of a single prime.
So assume P ⊂ A is a given prime; we want to find whether
there is any prime of B over it. Localizing at P we can assume
that P is actually maximal, and by taking a quotient we can even
assume that P = 0 and A is a field. In this case we have:
Proposition 5.2.1. Let A ⊂ B be an integral extension of integral
domains. Then A is a field if, and only if, B is.

157
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 integral dependence

Proof. Assume A is a field, and take b ∈ B, b 6= 0. Then b satisfies

an equation
bn + an−1 bn−1 + · · · + a0 = 0,
and we can assume a0 6= 0. Then
1 1
= − (bn−1 + an−1 bn−2 + · · · + a1 ) ∈ B.
b a0
Conversely, assume B is a field, and take a nonzero a ∈ A.
Then 1/a ∈ B, so it satisfies an equation
 n  n−1
1 1
+ an−1 + · · · + a0 = 0,
a a
whence
1
= −(an−1 · a + · · · + a0 · an−1 ) ∈ A.
a
We now have a series of formal consequences.
Corollary 5.2.2. Let A ⊂ B be an integral extension, Q a prime of B
over P, that is, Q ∩ A = P. Then P is maximal if, and only if, Q is.
Corollary 5.2.3. Let A ⊂ B be an integral extension, P a prime of A.
Then there exists a prime Q of B over P.
Proof. Let S = A \ P; by Proposition 5.1.17 we know that

AP = S−1 A ⊂ S−1 B
is integral. Let M ⊂ S−1 B be any maximal ideal. Then M ∩ A
is maximal, so it must be PAP . We conclude by the bijective cor-
respondence between prime ideals in the localization (Corollary
1.6.9).
Corollary 5.2.4. Let A ⊂ B be an integral extension, P a prime of A
and assume Q ⊂ Q 0 are primes of B over P. Then Q = Q 0 .
Proof. By localization we can assume that P is maximal. Then Q
and Q 0 must be both maximal, so they are equal.
Remark 5.2.5. The above results tell us that we can always lift a
prime P of A, and that there cannot be two lifts one inside the
other. It is by no means true that the lift is unique, though. For
instance the primes Q1 = (2 + i) and Q2 = (2 − i) of Z[i] both
satisfy Qi ∩ Z = (5).

158
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 5.2 going up and down

We can do a little better and lift chains of prime ideals.

Theorem 5.2.6 (Going up). Let A ⊂ B be an integral extension. Let
P1 ( · · · ( Pn be a chain of primes of A and Q1 ( · · · ( Qm (m 6 n)
a chain of primes of B such that Qi is over Pi . Then we can extend this
to a chain Q1 ( · · · ( Qn of primes of B with the same property.
Proof. We can forget the first m − 1 primes, so assume m = 1.
Moreover by induction we only need to do the step n = 2. Finally
we can quotient by P1 and Q1 , and assume that both are 0. By
Corollary 5.2.3 we can lift P2 , and we are done.
Remark 5.2.7. Producing a chain in A from a chain in B is easy.
Indeed assume Q1 ( · · · ( Qn is a chain of prime ideals of B, and
let Pi = Qi ∩ A. Then the Pi are again prime, and they are distinct
by Corollary 5.2.4, so we have the desired chain P1 ( · · · ( Pn .
Example 5.2.8. (a) Consider the inclusion k ⊂ k[x]/(x2 ); this is
an integral extension since x satisfies x2 = 0. Note that both
0 and (x) restrict to 0 on k, so the hypothesis that B is an
integral domain is necessary in Proposition 5.2.1.
(b) The same example shows that over the prime 0 we have both
0 and (x), so Corollary 5.2.4 does not extend to ideal which
are not prime, not even primary is enough. Indeed in the
ring B = k[x]/(x2 ) the ideal (x) is prime and 0 is primary, as
every divisor of 0 in B is nilpotent.
(c) The going up property fails for non-prime ideals. To see this,
consider the integral extension k[x2 , x3 ] ⊂ k[x], and take the
ideal I = (x2 ) ⊂ k[x2 , x3 ]. Every ideal J ⊂ k[x] that contains I
must contain x3 as well, but x3 ∈ / I, hence there is no ideal of
k[x] that lies over I.
(d) For extensions which are not integral, there is no hope to have
any result as the above. This will be more apparent after we
discuss dimension.
For now just consider the inclusion k[x] ⊂ k[x, y]. For every
polynomial f = f(y) we have (f) ∩ k[x] = (0), in particular
both 0 and (y) are primes above 0.
(e) Similarly consider the extension k[x] ⊂ k[x, 1/x], which is not
integral. Since x becomes invertible in k[x, 1/x], there is no
ideal of k[x, 1/x], prime or not, over the prime ideal (x).

159
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 integral dependence

There is a similar result which allows us to extend chains in

the other direction, but that requires more work. First we have to
generalize our definition of integral dependence.
Definition 5.2.9. Let A ⊂ B be an inclusion of rings, I ⊂ A an
ideal. We say that an element b ∈ B is integral over I if it satisfies
a monic equation

bn + an−1 bn−1 + · · · + a0 = 0, (5.2.1)

where ai ∈ I. The set of all such elements b is called the integral

closure of I in B.
As one may expect, this integral closure turns out to be an
ideal, and it is even expressible in terms of concepts we already
know.
Proposition 5.2.10. Let A ⊂ B be rings, A the integral closure of A
√ B, and let I ⊂ A be an ideal. Then the integral closure of I in B is
in
I · A.
Proof. Every element which is integral over I is certainly in A,
so to simplify notation we can assume A = B. We also denote
J := I · B. √
If b ∈ B satisfies (5.2.1),
√ bn ∈ J, hence b ∈ J.
Vice versa, let b ∈ J, so

b n = ai b i + · · · + ak b k

for some n > 0 and elements ai ∈ I and bi ∈ B. The ring

B 0 = A[b1 , . . . , bk ] is a finitely generated A-module, since all bi
are integral over A.
We can now apply the determinant trick, using the fact that
xn B 0 ⊂ I · B 0 , to obtain a monic equation for xn whose non-
leading coefficients are all in I.
Corollary 5.2.11. Let A ⊂ B be integral domains and assume A is
integrally closed. Denote k = F(A), and take an element b ∈ B integral
over an ideal I ⊂ A. If we write the minimal polynomial of b over k as

f(x) = xn + an−1 xn−1 + · · · + a0 ,

√
then all ai ∈ I.

160
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 5.2 going up and down

Proof. Let k 0 be the field of fractions of B; we can assume that

k 0 contains all the conjugates bi of b. The coefficients of f are
polynomials in the bi (the elementary symmetric functions if the
extension k 0 /k is separable).
Each bi is integral over I, since it satisfies the same equation
as b, so the coefficients of f are integral over I. By the previous
proposition and the √ assumption that A is integrally closed, these
coefficients are in I.
We need a last
Lemma 5.2.12. Let f : A → B be a ring homomorphism, and let P ⊂ A
be a prime. There there exists a prime Q ⊂ B such that f−1 (Q) = P if,
and only if, f−1 (f(P) · B) = P.
Proof. The condition is clearly necessary.
So assume f−1 (f(P) · B) = P. Let S = A \ P; by localizing at S
we can assume that A is local with maximal ideal P.
In this case let Q any maximal ideal containing f(P) · B; then
f−1 (Q) ⊃ P, so we must have equality by maximality.
Theorem 5.2.13 (Going down). Let A ⊂ B be an integral extension
of integral domains, and assume that A is integrally closed.
Let P1 ) · · · ) Pn be a chain of primes of A and Q1 ) · · · ) Qm
(m 6 n) a chain of primes of B such that Qi is over Pi . Then we can
extend this to a chain Q1 ) · · · ) Qn of primes of B with the same
property.
Proof. As in the proof of going up, we can assume m = 1, n = 2.
By Lemma 5.2.12, it is enough to show that

P2 BQ1 ∩ A = P2 .

Take an element
c
∈ P2 BQ1 , b=
s
with c ∈ P2 B and s ∈ B \ Q1 . By Proposition 5.2.10, c is integral
over P2 , so by Corollary 5.2.11 the minimal polynomial of c over
k := F(A) is

f(x) = xn + an−1 xn−1 + · · · + a0 ,

with ai ∈ P2 .

161
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 integral dependence

If moreover b ∈ A we find that a minimal equation for s = c/b

over k is
a a0
sn + n−1 sn−1 + · · · + n = 0.
b b
Again by Corollary 5.2.11 we see that the coefficients of this
equation are in A. If b ∈
/ P2 , all coefficients must lie in P2 , so
sn ∈ P2 · B.
But then
sn ∈ P1 · B ⊂ Q1 ,
which contradicts our choice of s. So b ∈ P2 and we are done.

5.3 noether normalization

In this section we prove the celebrated Noether normalization
lemma, which is a structure result for integral extensions of alge-
bras over a field.
Recall that an extension of fields can always be factored into
a completely transcendental part, followed by an algebraic exten-
sion. Namely, if K ⊂ L is an inclusion of fields, and L is finitely
generated over K, there exists an intermediate field T ⊂ L which
is isomorphic to K(x1 , . . . , xr ) as K-algebras, and such that L is
an algebraic extension of T . Something similar holds for algebras
over a field.

Theorem 5.3.1 (Noether normalization lemma). Let k be a field and

A a finitely generated k-algebra. Then there exist n > 0 algebraically
independent elements x1 , . . . , xn ∈ A such that A is finitely generated
as a module over k[x1 , . . . , xn ].

Proof. Let y1 , . . . , ym be a minimal set of generators of A. We

argue by induction over m, the base case m = 0 being trivial.
If the yi are already algebraically independent, there is noth-
ing to prove, hence assume that there exists a polynomial f over k
such that f(y1 , . . . , ym ) = 0. We can try to alter the set of genera-
tors and use the relation given by f to write one of the generators
in terms of the other ones.
We perform the change of variable
i−1
zi = yi − yr1

162
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 5.3 noether normalization

for i > 2 (r to be chosen later). This is an invertible change of

variable which has the effect of increasing the degree of y1 inside
f.
More precisely, in the equation

m−1
f(y1 , z2 + yr1 , . . . , zm + yr1 )

we can look at the degree of y1 in the various monomials. For a

monomial of multidegree (α1 , . . . , αm ) (in the yi ), the degree of
y1 after the change of variables is

α1 + α2 r + · · · + αm rm−1 .

For r big enough, we see that there is a single monomial which

will be of maximum degree for y1 such that the zi do not ap-
pear in this monomial. This means that the transformed equation
shows that y1 is integral over B := k[z2 , . . . , zm ].
By induction, B already has the desired form of an integral
extension of a polynomial ring, and the thesis follows.

Remark 5.3.2. When k is infinite, one can use a linear change of

variables in the above proof. The only things that is needed is that
in the resulting polynomial relation, y1 appears with maximum
degree in a monomial that contains no other indeterminates.
Consider a change of variable of the form

z i = yi − λ i y 1 .

This change of variables preserves the total degree of monomi-

als, so consider the homogeneous part of maximum degree of f,
call it F. We want to ensure that in F there appears a monomial
containing only y1 . Let t be the degree of F.
The coefficient of yt1 in

F(y1 , z2 + λ2 y1 , . . . , zm + λm y1 )

is a nonzero polynomial in λ2 , . . . , λm . Since k is infinite, there

exists a choice of λ2 , . . . , λm that makes this coefficient not vanish,
which is enough to show that y1 is integral over k[z2 , . . . , zm ].

163
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 integral dependence

5.4 integral extensions of dedekind

rings
The theory developed so far allows us to give some important
examples of Dedekind rings. The main tool which we shall use
is the following
Theorem 5.4.1. Let A be a Dedekind ring with field of fractions K, and
let L be a finite separable extension of K. Let B be the integral closure of
A in L. Then B is itself a Dedekind ring.
A few remarks are in order before proving the theorem. Re-
call from Section A.7 that in this situation we have an additive
homomorphism
Tr := TrL/K : L → K
called the trace. Now assume b ∈ B, and let σ : L → K be an em-
bedding; then the conjugate σ(b) is again integral over A, having
the same minimal polynomial. It follows that the trace Tr(b) is in-
tegral over A, and since A is integrally closed we find Tr(b) ∈ A.
So we get by restriction an additive homomorphism
Tr : B → A.
Similarly we can see that N(B) ⊂ A.
We isolate the main step, which is of independent interest.
Proposition 5.4.2. Let A be a Noetherian integral domain with field of
fractions K, and let L be a finite separable extension of K. Let B be the
integral closure of A in L. Then B is a finitely generated A-module; in
particular it is itself Noetherian.
Proof. Let n = [L : K] be the degree of the extension. First, observe
that B contains a free rank n A-module. Namely, L is generated
over K by a single element α, and after multipliying by a suitable
element of K we can assume that α ∈ B. So one can take the
subring A(α) ⊂ B, which as an A-module is just
D E
A(α) = 1, α, . . . , αn−1 .
A
For any such submodule M ⊂ B with basis u1 , . . . , un , con-
sider the dual basis v1 , . . . , vn with respect to the trace (Definition
4.2.5), and let
N := hv1 , . . . , vn iA .

164
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 5.4 integral extensions of dedekind rings

We claim that B ⊂ N; since N is a Noetherian A-module, it will

follow that B is finitely generated.
To see the claim, first note that since {v1 , . . . , vn } is a basis of L
over K, any element b ∈ B can be written as

b = a1 v1 + · · · + an vn

for some a1 , . . . , an ∈ K. Multiplying by ui and taking the trace

we find
ai = Tr(bui ) ∈ A,
since bui ∈ B.

Proof of Theorem 5.4.1. We use the characterization of Dedekind

rings from Theorem 5.1.19. It is clear that B is integrally closed.
Moreover there cannot be a strict inclusion P ( Q of nonzero
primes of B, by Corollary 5.2.4 and the fact that every nonzero
prime of A is maximal. Finally, B is Noetherian by Proposition
5.4.2.

The result applies in particular when A is a principal ideal

domain, so we can form a lot of new Dedekind rings. A special
case is obtained when A = Z and L is a number field, that is, a
finite extension of Q.

Definition 5.4.3. Let K be a finite extension of Q. Then K is called

a number field. The integral closure of Z inside K is called its ring
of integers, and it is sometimes denoted OK . For brevity, we will
also call it a number ring, although this terminology is not entirely
standard.

This provides a lot of examples of Dedekind rings which are

of great interest in arithmetics – one of the early successes of
commutative algebra was developing the unique factorization for
ideals in integral extensions of Z. We will study these rings in
greater detail in Chapter 6.
These are not the only examples, though: in Exercise 17 of
Chapter 6 you will see an example of a Dedekind ring which is
not the integral closure of a principal ideal domain in its fraction
field.
In fact, Theorem 5.4.1 is true even in the inseparable case. In
this situation, though, Proposition 5.4.2 fails: the integral closure

165
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 integral dependence

of a Dedekind ring in a finite extension of its fraction field can fail

to be finite (Exercise 19). We will thus pursue a different strategy
that avoids the construction of the dual basis.

Theorem 5.4.4 (Krull-Akizuki, [Aki35]). Let A be an integral do-

main with fraction field K, B a ring such that A ⊂ B ( K. Assume
that
i) A is Noetherian

ii) every nonzero prime of A is maximal.

Then i) and ii) hold for B as well.
In the terminology that we will introduce in Chapter 9, condi-
tions i) and ii) say that A is a 1-dimensional Noetherian ring.

Proof. Let a ∈ A be a nonzero element. Then A/aA is Noether-

ian, and every prime of this ring is maximal. By Theorem 2.4.11,
A/aA is Artinian. We claim that B/aB is Artinian as well.
Given this claim, we can easily conclude the proof. In fact, let
I ⊂ B be any nonzero ideal. Then I ∩ A 6= 0 (why?), and we
can choose a nonzero a ∈ I ∩ A. Then the image of I in B/aB is
finitely generated, so I is finitely generated as well, which proves
that B is Noetherian. If moreover I is prime, its image in B/aB is
maximal, hence I is maximal as well.
To prove the claim, consider the ideals

In := an B ∩ A + aA

of A. Let In be the image of In in A/aA. This is a decreasing

sequence of ideals, hence it stabilizes. By the correspondence of
ideals, the same must hold for {In }, so there is a a minimum m
such that In = Im for all n > m. We want to derive from this
that
am B ⊂ am+1 B + A. (5.4.1)
Both members of this relation are A-submodules of K, hence the
inclusion holds if and only if it holds between their localizations
at any maximal ideal of A. This allows us to prove (5.4.1) under
the additional hypothesis that A is local. In this case, let M be its
maximal ideal. We can also assume that a ∈ M, otherwise it is a
unit and the inclusion is trivial.

166
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 5.4 integral extensions of dedekind rings

Take any nonzero element x = b/c ∈ B, where b, c ∈ A. Each

ideal in a primary decomposition for (c) is M-primary, hence
Mn ⊂ cA for some n > m (why?). But then an+1 x ∈ A, so

an+1 x ∈ an+1 B ∩ A + aA = In+1 = In+2 = an+2 B ∩ A + aA.

We can simplify this to

an x ∈ an+1 B + A. (5.4.2)

In fact, this holds for n = m. To see this, take n minimal such

that (5.4.2) holds. If n > m, then

an x ∈ (an+1 B + A) ∩ an B = an+1 B + (an B ∩ A)

⊂ an+1 B + (an+1 B ∩ A) + aA = an+1 B + aA,

so an−1 x ∈ an B + A, which contradicts the fact that n is mini-

mal. Hence, (5.4.2) holds with n = m, and finally since x ∈ B is
arbitrary, this implies (5.4.1).
Knowing this, we can finally prove our claim that B/aB is Ar-
tinian. In fact,

B ∼ am B am+1 B + A ∼ A
= m+1 ⊂ = m+1 ,
aB a B am+1 B a B∩A
which proves that B/aB is an A/aA-module of finite length, hence
it is Artinian as an A/aA-module, and a fortiori as a B/aB-module.

Corollary 5.4.5. Let A be an integral domain with fraction field K, L/K

a finite field extension and B the integral closure of A in L. Assume that
i) A is Noetherian
ii) every nonzero prime of A is maximal.
Then i) and ii) hold for B as well. In particular, if A is a Dedekind
domain, B is also a Dedekind domain.
Proof. Take b1 , . . . , bn ∈ B that generate L over K. Then the ring
C := A[b1 , . . . , bn ] is finitely generated over A, hence Noetherian.
Also, by Theorem 5.2.6, every nonzero prime of C is maximal.
Since C ⊂ B, the thesis follows from Krull-Akizuki theorem.

167
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 integral dependence

5.5 exercises
1. Let A be a ring, A its integral closure inside the total ring of
fractions F(A). Prove that
[
A= (I :F(A) I).
I fractional
ideal of A

2. Prove the following strenghtening of Gauss’ lemma. Let A ⊂ B

be integral domains, f ∈ A[x] a monic polynomial, and assume
we have a factorization f = gh, where g, h ∈ B[x] are monic. Then
the coefficients of g and h are integral over A. Can you prove the
same result without assuming that A and B are integral domains?

3. Let A be an integrally closed integral domain, f ∈ A[x] an

irreducible monic polynomial. Then f is prime.

4. Let k be a field, B = k[x, y], and define

A := {f ∈ B | f(0, 0) = f(1, 1)}.

Consider the ideal Q2 = (x − 1, y − 1) ⊂ B, P2 = Q2 ∩ A and

P1 = (x) ∩ A. Show that P1 ⊂ P2 , but there is no prime Q1 ⊂ Q2
that lies over P1 , hence going-down fails for the inclusion A ⊂ B.

5. Compute the integral closure of A := k[x, y]/(y2 − x3 − x2 ),

where k is a field. (Notice that y/x is integral over A.)

6. Let B be the integral closure of A := k[x, y, z]/(y2 − x3 − x2 ),

where k is a field. Show that the inclusion A ⊂ B does not satisfy
the going-down property.

7. Let A be a ring with an action of a finite group G by ring

homomorphisms. Denoting

AG := {a ∈ A | g · a = a for all g ∈ G}

the invariant ring, prove that A is integral over AG . In particular,

if A is a finitely generated k-algebra, AG is finitely generated as
a k-algebra as well. In this case, we conclude that AG is Noether-
ian.

168
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 5.5 exercises

The following exercises, up to Exercise 13, give some properties

of the conductor ideal, a tool which is often used to study the
properties of rings that are not integrally closed. In particular,
we will use it to extend the property of ideal factorization from
Dedekind rings to a slightly more general class of rings, with
some restrictions.
8. Let A ⊂ B be rings. The conductor ideal of A in B is defined as

f(B/A) := AnnA (B/A) = {a ∈ A | aB ⊂ A}.

Show that f(B/A) is an ideal both of A and B. Moreover, if I

is any ideal of B such that I ⊂ A, then I ⊂ f(B/A). Deduce that
f(B/A) is the largest common ideal of A and B.
9. Let A ⊂ B be rings, S ⊂ A a multiplicative set. Show that

S−1 f(B/A) ⊂ f(S−1 B/S−1 A).

If moreover B is finitely generated as an A-algebra and integral

over A, then equality holds.
10. Let A be an integral domain with field of fractions k, and take
a ring B such that A ⊂ B ⊂ k. Then, as A-modules,
∼ HomA (B, A).
f(B/A) =

11. Let A be a Noetherian domain and assume that every nonzero

prime ideal of A is maximal. Let B be the integral closure of A
– by Krull-Akizuki, this is a Dedekind ring. Let I ⊂ B be an
ideal that is relatively prime to the conductor f(B/A). Show that
I ∩ A is also relatively to f(B/A) as ideals of A. Moreover, the
homomorphism A/I ∩ A → B/I is an isomorphism. Deduce that
I is prime if and only if I ∩ A is prime.
12. Let A ⊂ B be as in Exercise 11. Let I ⊂ A be an ideal that is
relatively prime to the conductor f(B/A). Show that I · B is also
relatively to f(B/A) as ideals of B. Moreover, the homomorphism
A/I → B/IB is an isomorphism. Deduce that I is prime if and
only if I · B is prime.
13. Let A ⊂ B be as in Exercise 11. Show that there is a bijection
between ideals of A relatively prime to f(B/A) and ideals of B

169
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 integral dependence

relatively prime to f(B/A), and that this correspondence is mul-

tiplicative and preserves the property of being prime. Deduce
from the fact that B is a Dedekind domain a factorization theo-
rem for A. Namely, ideals of A that are relatively prime to the
conductor have the property of unique factorization into primes.

14. Use Noether normalization lemma to prove this variant of

generic freeness (see Corollary 2.5.18): let A be a Noetherian inte-
gral domain, B a finitely generated A-algebra; then there exists a
nonzero f ∈ A such that Bf is a free Af -module.

15. [Huc76, Prop 3.1] Let A be a Noetherian local ring, P ⊂ A a

prime and I a P-primary ideal different from P itself. Let B := A/I
and C the integral closure of B inside its total ring of fractions.
Prove that B is Noetherian, but C is not. (Let b ∈ B be any
element that is not a unit or a zero divisor. The nilradical N(C)
satisfies bN(C) = N(C).)

In fact, integral closure does not preserve the property of being

Noetherian even for integral domains – see [Nag62, Appendix,
Example E.5].
The following exercises, up to 19, give (a variant of) an exam-
ple from [Kap70, Theorem 100] of an integral extension of Dede-
kind rings which is not finite. Our presentation here comes from
[Knaa].

16. Let k be a field. Show that k((t)) := F(k[[t]]) is transcendental

over k(t).

17. Let k be a field and a ∈ k[[x]] an element transcendental over

k(x). Show that the ring A := k[[x]] ∩ k(x, a) is a Dedekind ring
and in particular is integrally closed.

18. Let k be a field of characteristic p, a ∈ k[[x]] an element tran-

scendental over k(x), and let b = ap . Define the rings A :=
k[[x]] ∩ k(x, a) and B := k[[x]] ∩ k(x, b). Show that A is the inte-
gral closure of B inside F(A).

19. Let A, B be the rings of the previous exercise. Show that A

is not a finitely generated B-module. (For any n ∈ N, b can be
written as b = fpn +x
pn b , where f ∈ k[x] and b ∈ k[[x]]. Then
n n n
1/p
yn has a unique p-th root satisfying yn = x−n (−fn + a). If A

170
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 5.5 exercises

was finitely generated, one could find c ∈ A such that cB ⊂ A[a].

Show that this implies that xn divides c in A, for all n, and find
a contradiction.)

20. Prove the following strenghtening of the Krull-Akizuki the-

orem. Let A be a reduced Noetherian ring, T its total ring of
fractions, and B a ring such that A ⊂ B ( T . Assume that A is
Noetherian and every nonzero prime ideal of A is maximal; then
the same holds for B. (Consider the minimal primes of A and,
for each such prime P, the map B → T → F(A/P).)
21. Let A be an integral domain, B its integral closure inside
the algebraic closure of k := F(A). Prove that for any primes
P1 , . . . , Pn ⊂ B, the sum P1 + · · · + Pn is either the whole of B or
a prime ideal. This is a result of Artin from [Art71]. (Reduce to
two primes P1 and P2 . Assume xy ∈ P1 + P2 and define z = y − x,
so that x2 + zx = a1 + a2 , with ai ∈ Pi . Let t be a solution to
t2 + zt = a1 , and argue that either t ∈ P1 or t + z ∈ P1 . Find a
way to factor a2 , and check that in all cases either x ∈ P1 + P2 or
y ∈ P1 + P2 . This proof comes from [HH92].)

171
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 6
L AT T I C E M E T H O D S

This chapter is somewhat more specialized to the cases of inter-

est in algebraic number theory. We are interested in rings that
obtained by taking the integral closure of Z inside a finite exten-
sion K of Q. As an additive group, such a ring A is isomorphic to
Zn , where n is the degree of K over Q. Ideals can then be viewed
as sublattices inside Zn , and this allows us to study them by ge-
ometric techniques. In particular, the norm of an ideal I – which
measures how much I is sparser than A – will be a fundamen-
tal technical tool. While most of the chapter is devoted to this
setting, some parts are valid over more general Dedekind rings.
In the first section, we study the properties of the integral clo-
sure of Z inside number fields; in particular their structure as
an additive group. We also investigate what happens to prime
ideals under extension. The norm of an ideal is the tool that we
use here. We also generalize these results to Dedekind rings with
a different technique. In the next section we study the case of an
integral extension of Dedekind domains when the corresponding
field extension is Galois.
We then go back to the special case of number fields, where we
use more geometric techniques to prove bounds on the norm of
ideals, from which we derive important results about the finitess
of the ideal class group and the finite generation of the unit group.
Finally, we specialize these concepts to the interesting case of
rings of integers inside cyclotomic extensions of Q. This allows
us to present a slick proof of the celebrated quadratic reciprocity
formula of Gauss.
This chapter, together with the material needed from the pre-
vious ones (in particular the factorization theorem for Dedekind
domains from Chapter 3), gives a quick introduction to the ideas

173
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 lattice methods

and methods from algebraic number theory. It can be comple-

mented with some parts of Chapter 7, which introduces the com-
pletion of rings and defines the so-called p-adic rings, giving a
local analog of the theory developed here. For more about num-
ber fields, [Mar77] is an excellent reference.

6.1 additive structure of number rings

Let K be a number field, A its ring of integers. By Theorem 5.4.1,
A is a Dedekind ring. In particular, we can apply the results of
Section 3.4, and factor every ideal I as a product of prime powers
I = P1e1 · · · Prer in a unique way.
As a Z-module, the structure of A is pretty simple.

Proposition 6.1.1. Let n = [K : Q]. Then A is a Z-module of rank n.

Proof. By clearing denominators, we can find a Q-basis for K

given by elements a1 , . . . , an ∈ A. Let b1 , . . . , bn be the dual
basis with respect to the trace. Then for every a ∈ A we can
write
X
n
a= ci b i
i=1

for some c1 , . . . , cn ∈ Q. But then ci = Tr (a · ai ) ∈ Z by duality.

This shows that we have inclusions

Za1 + · · · + Zan ⊂ A ⊂ Zb1 + · · · + Zbn .

The latter inclusion implies that A is a free Z-module of rank at

most n, and the former shows that the rank is exactly n.
√
Example 6.1.2. Let K = Q( m) be a quadratic number field; we
can assume that m is squarefree (why?). Let us compute a basis
for its ring of integers. √
A number α = x + y m ∈ K is an algebraic integer if and only
if Tr(α) = 2x and N(α) = x2 − my2 are in Z. Let us write x = a/2
and y = p/q, where a, p, q ∈ Z and p and q are relatively prime.
To have N(α) ∈ Z is equivalent to

a2 q2 − 4mp2 ≡ 0 (mod 4q2 ).

174
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 6.1 additive structure of number rings

This implies that q2 divides 4m, and since m is squarefree, q

divides 2. So we can take y = b/2, with b ∈ Z.
With this notation, α is integral over Z if and only if

a2 ≡ mb2 (mod 4).

Since m is squarefree, m ≡ 1, 2 or 3 modulo 4.

(a) If m ≡ 2,√
3 (mod 4), a and b must both be even. In√this case,
α = s + t m, with s, t ∈ Z, so a basis for OK is {1, m}

(b) If m ≡ 1 (mod 4), a and b need just have the same parity. In
√
this case, a basis for OK is 1, 1+2 m .

Generalizing the example, a natural question is to find a basis

for the ring A of integers in K. In particular, one may ask whether
it is possible to find some a ∈ A such that A = Z[a]. This is not
true in general, but one can get bounds such as the following.

Proposition 6.1.3. Let a1 , . . . , an ∈ A be a basis of K over Q, d =

disc({a1 , . . . , an }). Then dA ⊂ Z[a1 , . . . , an ].

Recall that the discriminant was defined in 4.2.6.

~ = M~b for some

Proof. Let b1 , . . . , bn be a basis of A, so that a
matrix M with integer coefficients. Let B = Z[a1 , . . . , an ]; then
A/B is a finite abelian group of cardinality equal to |det M| by
1.5.9.
By Proposition 4.2.10,

d = disc({a1 , . . . , an }) = m2 disc({b1 , . . . , bn }),

where m = det M. Since m(A/B) = 0, we have mA ⊂ B, and the

result follows since m|d.

Definition 6.1.4. If A is the ring of integers in a number field K,

we define its discriminant disc(A) := disc({a1 , . . . , an }), where
{a1 , . . . , an } is any basis of A over Z. This is well-defined by
4.2.10, since any other basis can be obtained by a change of basis
for an integer matrix with determinant ±1. We sometimes denote
disc(K) := disc(A).

175
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 lattice methods

The discriminant has a natural geometric interpretation. First,

∼ Zn of finite rank
recall that a lattice is a free abelian group L =
with an embedding i : L ,→ R such that the image of a Z-basis
n

for L is a basis of Rn as a vector space. This is equivalent

to saying that i(L) is discrete in the Euclidean topology (Exer-
Pn 8). Given a basis g1 , . . . , gn of L, the linear combinations
cise
j=1 aj i(gj ) with 0 6 aj < 1 form a parallelogram, called a fun-
damental domain for the lattice. The figure shows an example of
2-dimensional lattice, with generators v1 and v2 , and two choices
of a fundamental domain.

v1
v2

Definition 6.1.5. Let i : L ,→ Rn be a lattice, D ⊂ Rn a funda-

mental domain with respect to the basis {g1 , . . . , gn }. The volume
of D, denoted vol(D), is just its Lebesgue measure as a subset of
Rn . By elementary measure theory, vol(D) = |det M|, where M
is a matrix having as columns the vectors {i(g1 ), . . . , i(gn )}.
The volume of L, denoted vol(L) is the volume of one of its fun-
damental domain. To see that this is well-defined, notice that if
we choose a different basis, the matrix for the change of basis has
determinant ±1, hence the volume of the fundamental domain
does not change.
Let K be a number field with degree n over Q, A its number
ring. We have seen that – as a group – A = ∼ Zn . We can in fact
view A as a lattice inside R . To do so, let K = Q(α), and say f is
n

the minimal polynomial of α over Q. Then f has n roots, of which

– say – there are r real ones and s pairs of conjugate nonreal ones,
so that r + 2s = n. K admits n distinct embeddings into C, each
one determined by the image of α. For r such embeddings σ we
have σ(K) ⊂ R, so in fact the numbers r and s are independent
of α.

176
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 6.1 additive structure of number rings

Definition 6.1.6. The pair (r, s) is called the signature of the num-
ber field K.
The set of embeddings will be
{σ1 , . . . , σr , τ1 , τ1 , . . . , τs , τs },
where the σi are real. Together, they determine a homomorphism
σ : K → Rr × Cs (6.1.1)
given by σ = (σ1 , . . . , σr , τ1 , . . . , τs ). As a real vector space, we
can identify Rr × Cs with Rn , and look at the image of A via σ.
Proposition 6.1.7. σ(A) is a lattice inside Rr × Cs , having volume
|disc(A)|.
−s
p
2
Proof. The first assertion follows from the second one: as a group,
A= ∼ Zn , and if σ(A) did not generate Rr × Cs as a real vector
space, its volume would be 0.
Letting {α1 , . . . , αn } be a Z-basis of A, the image of A is gener-
ated by the columns of the matrix M given by
σ1 (α1 ) · · · σr (α1 ) <τ1 (α1 ) =τ1 (α1 ) · · · =τs (α1 )
 
 .. .. .. .. ..
.

 . . . . .
σ1 (αn ) ··· σr (αn ) <τ1 (αn ) =τ1 (αn ) · · · =τs (αn )
The determinant of M is, up to sign, the volume of a funda-
mental domain. To compute it, we use column operation to re-
place the column <τk (→ −
α ) with τk (→
−
α ). Then, by multiplying the
next column by 2i and subtraction, we can turn next column into
τk ( →
−
α ).
The square of the determinant of the matrix thus obtained is
disc(A), and our columnp operations have multiplied det M by
(2i) , hence |det M| = 2
s −s |disc(A)|.
There is an alternative way to look at the embedding σ in
(6.1.1). Namely, let KR = K ⊗Q R. Since K is a Q-vector space
of dimension n, KR is a real vector space of the same dimension.
More precisely, let K = Q(α), and f the minimal polynomial of α.
Then K = ∼ Q[x]/(f(x)), so

∼ R[x] .
KR =
(f(x))

177
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 lattice methods

By factoring f over R[x] as a product of r linear and s quadratic

factors, we obtain an isomorphism R[x]/(f) = ∼ Rr × Cs , and this
gives another definition of the embedding
K ,→ KR = ∼ Rr × Cs . (6.1.2)
It is a simple verification that the two embeddings agree, for a
suitable choice of the ordering of the factors of f (Exercise 11).
The discriminant can be used to bound the ring of integers of
a composite field.
Proposition 6.1.8. Let K, L be two number fields with m = [K :
Q], n = [L : Q] and assume that [KL : Q] = m · n. Let A be the
ring of integers of K, B that of L and C that of KL. Setting d =
(disc(A), disc(B)), we have
1
A · B. C⊂
d
Proof. Clearly, A · B ⊂ C. Let {a1 , . . . , am } be an integer basis of A
and {b1 , . . . , bn } an integer basis of B. The set {ai · bj } is a Q-basis
of KL.
C contains the product A · B as a subgroup of finite index, since
both have rank mn. If c is this index, c · C ⊂ A · B. So, for t ∈ C
we can write
Xm X n
xij
t= ai b j
c
i=1 j=1
for some xij ∈ Z. We can assume that the GCD of all xij and c
is 1.
Consider an embedding σr : KL → C fixing L. There are m of
those, and their restrictions to K remain distinct. Hence
Xm X n
xij
σr (t) = σr (ai )bj =
c
i=1 j=1
 
Xm Xn
xij X
m
=  bj  σr (ai ) = yi σr (ai ).
c
i=1 j=1 i=1

The terms yi can be seen as the solution of a linear system

defined by the matrix (σr (ai ))r,i and vector (σ(t))r . By Cramer’s
rule,
det Mi d
yi = = i,
dR dA

178
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 6.2 prime extensions in number rings

where dA = disc({a1 , . . . , am }) = disc(A), and Mi is some cofac-

tor matrix. In particular, di ∈ C, so dA yi ∈ C. Since yi ∈ L, we
d x
have dA yi ∈ B, which can be expanded to Ac ij ∈ Z.
So c divides dA xij for all i, j. Since c was chosen to be coprime
with the xij , it follows that c divides dA . Symmetrically, c divides
dB , hence c|d, which is the thesis.

6.2 prime extensions in number rings

Let A ⊂ B be an inclusion of Dedekind domains, and let K be the
fraction field of A and L that of B. Assume that B is integral over
A. A key example is when K and L are number fields, with rings
of integers A and B respectively.
We want to relate the factorization for ideals in A to that in B.
Given an ideal I ⊂ A, we can consider the extension I · A and its
factorization. We start with the case of prime ideals.
Remark 6.2.1. Let P ⊂ A be a prime ideal. The ideal P · B is
not trivial. In fact, since P is not the whole of A, there exists
γ ∈ K \ A such that γ · P ⊂ A (because P−1 is not the unit in the
class group). Assume that we can write

1 = p1 b1 + · · · + pk bk

for some choice of p1 , . . . , pk ∈ P and b1 , . . . , bk ∈ B.

Multiplying by γ we get that γ ∈ B is integral over A. But we
know that γ ∈ K, so we get a contradiction because A is integrally
closed.
Since P · B is not trivial, there exists a factorization of the form

P · B = Qe11 · · · Qer r (6.2.1)

for some prime ideals Qi of B. The condition that Q is a prime

over P in B can be expressed in several equivalent ways.
Proposition 6.2.2. Let A ⊂ B be an integral extension of Dedekind
rings, P a prime ideal of A and Q a prime ideal of B. The following
conditions are equivalent:
i) Q divides P · B

179
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 lattice methods

ii) Q ⊃ P · B

iii) Q ⊃ P

iv) Q ∩ A = P.

Proof. The equivalence of the first three points follows immedi-

ately by the unique factorization for ideals in B. For the last
point, it is clear that if Q ∩ A ⊃ P, we must have equality because
P is maximal.

Definition 6.2.3. Let A ⊂ B be an integral extension of Dedekind

rings, P a prime ideal of A and Q a prime ideal of B over P.

1. The degree of the field extension A/P ⊂ B/Q is called the

degree of inertia of Q over P, and is denoted f(Q | P).

2. The exponent ei in the factorization (6.2.1) is called the ram-

ification index of Qi over P, and is denoted e(Qi | P).

3. If e(Qi | P) = 1 for all Qi above P, we say that P is unramified

in B; otherwise it is ramified.

So far we have been general, but now we restrict again to the

case of rings of integers in number fields. In this case we can
prove that the degree of inertia is always finite.

Remark 6.2.4. Let A be the integer ring in the number field K,

I ⊂ A any ideal. If x is any element of I, we have inclusions
x · A ⊂ I ⊂ A, which shows that, as an additive group, I is free
of the same rank as A. In particular, if P ⊂ A is prime, A/P is a
finite field, and the degree of inertia is always finite for the case
of number rings extensons.

Definition 6.2.5. Let I ⊂ A be an ideal in a number ring. The in-

dex A : I as additive groups – which is the same as the cardinality
of A/I – is called the norm of I and denoted kIk.

Remark 6.2.6. By Proposition 6.1.7, it follows immediately that I

embeds into Rr × Cs as a lattice of volume 2−s |disc(A)| · kIk,
p

where (r, s) is the signature of the number field.

The connection with the norm of elements is given by the fol-

lowing

180
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 6.2 prime extensions in number rings

Proposition 6.2.7. If I = (x) is principal, kIk = |N(x)|, where we

simply denote N = NK/Q .
Proof. [Con, Theorem 7.2] Since I is a subgroup of A of the same
rank, by Theorem 1.5.6 we can find a Z-basis of A, say {r1 , . . . , rn },
such that {a1 r1 , . . . , an rn } is a Z-basis of I, for some integers
{a1 , . . . , an }. Another basis is of course {xr1 , . . . , xrn }, hence the
linear application on K which sends ai ri to xri has determinant
±1.
On the other hand, we have a commutative diagram of linear
maps on K given by
ri → ai ri
K →K
ri →
ai ri → xri
→ ↓ xr
i
K
Since the determinant of the vertical map is ±1, the two other
determinants are the same, up to sign. But the one above is kIk,
and the one below is N(x).
Remark 6.2.8. It follows that for any ideal I and x ∈ I, kIk divides
N(x), since (x) is a subgroup of I.
As it happens for the norm of numbers, the norm of ideals is
multiplicative.
Proposition 6.2.9. Let I, J be ideals of the number ring A. Then kI ·
Jk = kIkkJk.
Proof. We use unique factorization and prove the equivalent fact
that if I = P1m1 · · · Pkmk , then kIk = kP1 km1 · · · kPk kmk .
First, let us treat the case of coprime ideals. If I = S · T , with
S + T = (1), we have
A ∼ A A
= ×
I S T
by the Chinese remainder theorem, hence kIk = kSkkT k.
This leaves with the case of a prime power, where we prove
that kPm k = kPkm by induction. Assume it holds for m and
choose any α ∈ Pm \ Pm+1 . Consider the composition
α· m
A →P → Pm /Pm+1 ,
x → α·x

181
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 lattice methods

call it φ. Trivially, P ⊂ ker φ, and vice versa if x ∈ ker φ, we must

have x ∈ P by unique factorization.
Since P = ker φ, we can descend φ to an injective map

ψ : A/P → Pm /Pm+1 .

Moreover, the image of φ is generated by α and Pm+1 . By unique

factorization, this is the gcd of Pm+1 and (α), which is Pm . This
proves that φ is surjective, hence ψ is an isomorphism.
It follows that A/P = ∼ Pm /Pm+1 , hence kPm+1 k = kPm kkPk as
desired.
We can use what we have proved so far about norms of ideals
to get a special case of an important result.
Theorem 6.2.10. Let L be a number field with ring of integers B, p ∈ Z
a prime, n = [L : Q]. Assume p · B = Qei 1 · · · Qer r . Then

X
r
n= ei fi ,
i=1

where fi := f(Qi | p).

We will generalize this in a moment to any extension of num-
ber fields, but this result will be needed as a starting point.
Proof. We compare the norms in the factorization of p · B. For
one, it is clear that kp · Bk = pn , since B is free of rank n.
On the other hand, kQi k = #(B/Qi ), and B/Qi is a field exten-
sion of Z/pZ of degree fi . Hence
P
kp · Bk = kQ1 ke1 · kQr ker = (pf1 )e1 · · · (pfr )er = p ei fi
,

and the thesis follows by comparison.

To study the more general case, we want to compare how the
norm of ideals behaves under extension. First, we need a
Lemma 6.2.11. Let K ⊂ L be number fields with integer rings A and
B respectively. Let P ⊂ A be a prime ideal, and Q ⊂ B a prime over P
– this gives an inclusion of fields A/P ⊂ B/Q. Then

[B/Q : A/P] 6 [L : K].

182
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 6.2 prime extensions in number rings

Proof. Let n = [L : K] and take elements b1 , . . . , bn+1 ∈ B/Q.

Then for the elements bi ∈ B we have a linear dependence rela-
tion
a1 b1 + · · · + an+1 bn+1 = 0,
where can take ai ∈ A by clearing denominators. This gives a
similar relation for the bi over A/P, unless all ai ∈ P. If this is
the case, it is enough to find α ∈ K such that αai ∈ A for all i,
but at least one αai ∈ / P and then multiply the equation by α.
Let I = (a1 , . . . , an ) and choose an inverse ideal J of A, so that
I · J = (γ). By unique factorization, I · J 6⊂ (γ) · P, so we find β ∈ J
such that βI 6⊂ (γ) · P. Then it is easy to check that α := β γ is the
desired element.
Proposition 6.2.12. Let K ⊂ L be number fields with integer rings A
and B respectively. Let I ⊂ A be an ideal. Then kI · Bk = kIkn , where
n = [L : K].
Proof. By the multiplicativity property 6.2.9, it is enough to prove
the assertion for a prime ideal P. First, we consider the case
where K = Q; in this case, P = (p). If {b1 , . . . , bn } is a basis
of B as a Z-module, then a basis of P · B is {pb1 , · · · , pbn }, so
kP · Bk = pn .
For the general case, consider the inclusion Q ⊂ K ⊂ L, and
let (p) = P ∩ Z, so that we have also inclusions (p) ⊂ P ⊂ P · B.
Factor (p) · A = P1e1 · · · Prer , with inertia degrees f1 , . . . , fr , and
P = Pi for some i. By Theorem 6.2.10, we get
P
pm = k(p) · Ak = p ei fi
,

where m = [K : Q].
For B, using 6.2.9 again, we get

pmn = k(p) · Bk = kP1 Bke1 · · · kPr Bker . (6.2.2)

Moreover, kPi · Bk = kPi kni for some ni 6 n by Lemma 6.2.11.

Combining with (6.2.2), we get
P
pmn = kP1 ke1 n1 · · · kPr ker nr = p e i f i ni
.

Since all ni 6 n, we must have equality for all i, hence kPi · Bk =

kPi kn .

183
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 lattice methods

This allows us to extend 6.2.10 to the general case of extensions

of number fields.

Theorem 6.2.13 (Inertia-ramification formula for number rings).

Let K ⊂ L be number fields with ring of integers A and B respectively.
Let P ⊂ A be a prime ideal, n = [L : K]. Assume P · B = Qei 1 · · · Qer r .
Then
Xr
n= ei fi ,
i=1

where fi := f(Qi | P).

Proof. By Propositions 6.2.12 and 6.2.9 we get

kPkn = kP · Bk = kQ1 ke1 · · · kQr ker =

P
= kPke1 f1 · · · kPker fr = kPk ei fi
,

and the result follows by comparing exponents.

6.3 prime extensions in dedekind rings

We want to extend the results of the previous section to the more
general case of Dedekind rings. The situtation is more subtle
because we cannot assume the result of Section 6.1 – in particular
that the rings involved are additively free Z-modules. For this
reason, we will not use the ideal norm, but we will prove the
analogue of Theorem 6.2.13 with a different approach.
Let A be a Dedekind ring with fraction field K, L a finite sepa-
rable extension of fields, and let B be integral closure of A inside
L. By Proposition 5.4.2, we know that B is finitely generated as
an A-module, and in fact it is itself a Dedekind ring by Theorem
5.4.1.
Given a prime ideal P ⊂ A, we want to study the factorization
of P · B as an ideal of B. Recall that Definition 6.2.3 is valid for
general Dedekind rings, so we can talk about the index of ram-
ification and the degree of inertia of a prime ideal Q ⊂ B over
P.

184
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 6.3 prime extensions in dedekind rings

Proposition 6.3.1. Let K ⊂ L be a finite separable extension of fields,

A ⊂ K a Dedekind ring with fraction field K, and B ⊂ L its integral
closure. For a prime P ⊂ A we have

B
dimA/P = [L : K],
P·B
where dimA/P denotes dimension as a vector space over the field A/P.

Proof. Consider the multiplicative set S = A \ P, and let A 0 =

S−1 A = AP and B 0 = S−1 B. Then, we have isomorphisms A/P = ∼
0 0 ∼ 0 0
A /PA and B/P = B /PB , so we can assume that A is local.
In this case, all ideals are powers of P, which then must be prin-
cipal. It follows that in fact, A is a principal ideal domain(actually,
A is a discrete valuation ring, which we will encounter in Section
7.3). From the structure of finitely generated modules over prin-
cipal ideal domains and the fact that B has no torsion, it follows
that B is actually a free A-module. In this case, the thesis is
clear.
We can use this fact to generalize Theorem 6.2.13.
Theorem 6.3.2 (Inertia-ramification formula for Dedekind rings).
Let K ⊂ L be a finite separable extension of fields, A ⊂ K a Dedekind
ring and B ⊂ L its integral closure. Let P ⊂ A be a prime ideal,
n = [L : K]. Assume P · B = Qei 1 · · · Qer r . Then

X
r
n= e i fi ,
i=1

where fi := f(Qi | P).

Proof. By the Chinese remainder theorem, we have

B ∼ Y B
r
= .
P·B Qei i
i=1

Qki B
Moreover, dimA/P = dimA/P for all k (why?), hence
Qk+1
i
Qi

B B
dimA/P = ei dimA/P .
Qei i Qi

185
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 lattice methods

By Proposition 6.3.1 we have

B X
r
B X
r
[L : K] = dimA/P = ei dimA/P = ei fi .
P·B Qi
i=1 i=1

6.4 galois extensions of dedekind rings

Keeping the setting of the previous section, we investigate what
happens when the extension K ⊂ L is Galois. In this case, the
Galois group Gal(L/K) acts on B – because it preserves integral
elements – hence it acts on ideals of B. In particular, given a
prime ideal P ⊂ A, Gal(L/K) must permute the prime ideals
Qi ⊂ B that are over P.

Proposition 6.4.1. Let K ⊂ L be a finite Galois extension of fields,

A ⊂ K a Dedekind ring with fraction field K, and B ⊂ L its integral
closure. Let P ⊂ A be a prime ideal, and Q1 , . . . , Qr the primes of B
over P. Then the action of Gal(L/K) over {Q1 , . . . , Qr } is transitive.

Proof. Take any two primes over P, say Q1 and Q2 . Assuming

that there is no σ ∈ G := Gal(L/K) such that σ(Q1 ) = Q2 , we can
solve the system

x≡0 mod Q2
x≡1 mod σ(Q1 ) for all σ ∈ G.
Q
Let y = NL/K (x) = σ∈G σ(x). On one hand, x itself is a factor
in this product, hence y ∈ Q2 . On the other hand, x ∈ / σ(Q1 ) for
every σ ∈ G, that is, no factor lies in Q1 . Since Q1 is prime,
y∈/ Q1 . But this is a contradiction, since y ∈ A ∩ Q2 = P.

From the proposition, it follows that e = e(Qi |P) and f =

f(Qi |P) do not depend on i, hence the inertia-ramification for-
mula simplifies to

n = [L : K] = r · e · f. (6.4.1)

186
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 6.4 galois extensions of dedekind rings

Definition 6.4.2. Let K ⊂ L a finite Galois extension with Galois

group G, A ⊂ K a Dedekind ring with fraction field K, B ⊂ L its
integral closure. Let P ⊂ A be a prime ideal, and Q a prime of B
over P.

1. The decomposition group of Q over P, denoted D(Q|P) is

D(Q|P) = {σ ∈ G | σ(Q) = Q},

that is, the stabilizer of Q in G.

2. The inertia group of Q over P, denoted E(Q|P) is

E(Q|P) = {σ ∈ G | σ(x) ≡ x (mod Q) for all x ∈ B}.

Remark 6.4.3. Since D(Q|P) is the stabilizer of Q, and the action

is transitive, we have [G : D(Q|P)] = r. Moreover, all subgroups
D(Qi |P) are conjugate with each other: if τ(Qi ) = Qj ,

D(Qj |P) = τD(Qi |P)τ−1 .

Remark 6.4.4. It is immediate to check that E(Q|P) is a subgroup

of D(Q|P). In fact, it is normal subgroup. To see this, notice that
an automorphism σ ∈ D(Q|P) induces a commutative diagram

σ
B →B

↓ ↓
σ
B/Q → B/Q,

thereby inducing a homomorphism φ : D → Aut(B/Q).

B A
In fact, A/P is a subfield of B/Q, and σ ∈ Gal( Q / P ). The ker-
nel of φ is exactly E(Q|P), which must then be normal in D(Q|P).
B A
Moreover, since the cardinality of Gal( Q / P ) is f(Q|P), we
have [D(Q|P) : E(Q|P)] 6 f.

187
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 lattice methods

By the Galois correspondence (Section A.5), we obtain a tower

of extensions that sit between K and L:
L {e}

LE E

LD D
r

K G
where D = D(Q|P), E = E(Q|P), and on the vertical lines we
denote the degree of the field extensions.
Proposition 6.4.5. In the above setting, we have [L : LE ] = e and
[LE : LD ] = f.
Proof. Let us complete the diagram by showing also the integral
closure of A in LE and LD , as well as the restrictions of Q to these
rings:
Q B L {e}

QE BE LE E
6f

QD BD LD D
r
P A K G.
We know some numbers in this diagram: [LD : K] = r and
[LE : LD ] 6 f. We will compute some others and use the inertia-
ramification (6.4.1) formula to conclude.
First, since D = Gal(L/LD ), and the Galois group is transitive
on primes over QD , we deduce that Q is the only prime of B over
QD . By (6.4.1) we see that
e(Q|P) · f(Q|P) = e(Q|QD ) · f(Q|QD ),

188
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 6.4 galois extensions of dedekind rings

and since there are obvious inequalities, we conclude that in fact

e(Q|QD ) = e(Q|P) and f(Q|QD ) = f(Q|P).
The next step is to show that f(Q|QE ) = 1, that is,
B ∼ BE
= .
Q QE
Given α ∈ B, the polynomial
Y
g(x) = (x − σ(α))
σ∈E

is E-invariant, hence it belongs to BE [x]. This gives a quotient

polynomial g(x) ∈ BE /QE [x].
By definition, σ ∈ E implies that σ(α) = α in B/Q, hence g(x) is
a power of (x − α). Since B/Q is a separable extension of BE /QE ,
the minimal polynomial of α over BE /QE must be x − α – that is,
α ∈ BE /QE . Since α was arbitrary, B/Q is the trivial extension of
BE /QE , as claimed.
At this point, since inertia degrees are clearly multiplicative
in extensions, we find that f(QE |QD ) = f(Q|QD ) = f. By the
inequality we already know we conlude that [LE : LD ] = f and
now we can fill in all remaining degrees in the diagram, which
becomes
Q B L {e}

f=r=1 e
QE BE LE E
e=r=1 f

QD BD LD D
e=f=1 r
P A K G.

The fields LD and LE , that we have defined in terms of sub-

groups of the Galois groups, enjoy an intrinsic characterization.
To see this, let F be any intermediate field, K ⊂ F ⊂ L. If BF
is the integral closure of A in F, and QF = Q ∩ BF , by the defi-
nitions it follows immediately that D(Q|QF ) = D(Q|P) ∩ H and

189
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 lattice methods

E(Q|QF ) = E(Q|P) ∩ H, where H = Gal(L/F). Since the intersec-

tion of Galois groups corresponds to the composite of the relative
fields, we get the following diagram of field extensions:

L e0
e
FLE
f0
LE
f FLD
r0
LD
r F

K
where r 0 , f 0 and e 0 are the number of factors, degree of inertia
and ramification index of QF in B.

Proposition 6.4.6. Fix a finite Galois extension K ⊂ L with Galois

group G, A ⊂ K a Dedekind ring with fraction field K, B ⊂ L its
integral closure. Given a prime P ⊂ A and a prime Q ⊂ B over it,
let LD and LE be the subfields of L associated to D(Q|P) and E(Q|P)
respectively.

i) LD is the biggest subfield F ⊂ L for which e(QF |P) = 1 and

f(QF |P) = 1.

ii) LD is the smallest subfield F ⊂ L such that Q is the only prime

over QF .

iii) LE is the biggest subfield F ⊂ L for which e(QF |P) = 1.

iv) LE is the smallest subfield F ⊂ L such that Q is totally ramified

over QF , that is, QF · B = Q[L:F] .

Proof. i) In the diagram above, assume e = e 0 = 1 and f = f 0 =

1. Then [L : FLD ] = [L : F], which means F ⊂ LD .

ii) Let H = Gal(L/F). If Q is the only prime over QF , every

σ ∈ H preserves Q, which means H ⊂ D(Q|P) and the thesis
follows from the Galois correspondence.

iii) In the diagram above, assume e = e 0 = 1. Then [L : FLE ] =

[L : F], which means F ⊂ LE .

190
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 6.5 discriminant and ramification

iv) If Q is totally ramified over QF we must have f 0 = r 0 = 1,

which means FLE = F, or LE ⊂ F.

Corollary 6.4.7. Let A be a Dedekind ring with fraction field K, and

L1 , L2 two finite separable extensions of K. Let P be a prime of A. If
P splits completely (resp. is unramified) in L1 and L2 , then the same
holds in L1 L2 .

Proof. Fix a finite Galois extension L of K containing both L1 and

L2 . The hypothesis is equivalent to L1 , L2 ⊂ LD(Qi |P) for each
prime Qi over P (resp. L1 , L2 ⊂ LE ), hence the same holds for the
composite field L1 L2 .

Since the normal closure of a field L over K is the composite of

σ(L) for all embeddings of L into K, we also get

Corollary 6.4.8. Let A be a Dedekind ring with fraction field K, and

L a finite separable extensions of K. Let P be a prime of A. If A splits
completely (resp. is unramified) in L, then the same holds in the normal
closure of L.

6.5 discriminant and ramification

As it turns out, the discriminant of a number field is strictly re-
lated to the properties of ramification of primes in that field. In
fact, we have the following result:

Theorem 6.5.1. Let K be a number field with ring of integers A. The

prime number p is ramified in A if, and only if, it divides disc(A).

We will see in Corollary 6.7.4 that the discriminant of a number

field is always bigger than 1, so ramified primes always exist.
Since the proof is quite long, we split it in two.

Proof of forward implication. Assume p is ramified, so say p · A =

P1e1 · · · Prer with e1 > 1. We can then write p · A = P1 · I for some
ideal I divisible by all primes over p.
Fix an integer basis {α1 , . . . , αn } of A, and choose α ∈ I \ p · A.
Write
α = m1 α1 + · · · + mn αn ,

191
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 lattice methods

where p does not divide all m1 – say m1 is not multiple of p. The

subring A 0 with basis {α, α2 , . . . , αn } satisfies

disc(A 0 ) = m21 disc(A),

so if we can prove that p divides disc(A 0 ), we are done.

Now let L be the normal closure of K, with ring of integers B.
Since α ∈ I, it is inside every prime of A over p, hence inside
every prime of B over p. The discriminant of A 0 is

disc(A 0 ) = (det(σi (αj )))2

where σ1 , . . . , σn are the complex embeddings of K, and we have

replaced α1 with α. Since α belongs to every prime Q of B over
p, the same holds for each of its conjugates σi (α). It follows that
disc(A 0 ) belongs to every prime Q of B over p, and since it is an
integer number, disc(A 0 ) ∈ Q ∩ Z = (p).
Proof of backward implication. We use the characterization of dis-
criminant given by (4.2.1), so given a basis {α1 , . . . , αn } of A we
have
disc(A) = det(Tr(αi αj ))ij .
The matrix (Tr(αi αj ))ij has integer coefficients, and p dividing
its determinant implies that its rows are linearly dependent in
Z/pZ. So there exist integer numbers m1 , . . . , mn such that
X
n
(Tr(αi α1 ), . . . , Tr(αi αn )) ≡ 0 mod p.
i=1

In turn, this implies that Tr(αx) ≡ 0 mod p for all x ∈ A, where

X
n
α := mi αi .
i=1

Notice that by construction α ∈ / pA, since the linear combination

is not trivial.
Now assume by contradiction that p = P1 · · · Pr is unramified.
Since α ∈ / pA, there is some i such that α ∈/ Pi . If L is the normal
closure of K, with ring of integers B, and Q is a prime of B above
Pi , we have α ∈ / Q, but by Proposition A.7.5 Tr(αx) ∈ Q for all
x ∈ B.

192
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 6.6 computing prime factorizations

Factor Pi B = Q1 · · · Qr , with Q = Q1 , and choose β ∈ B satis-

fying β ∈
/ Q and β ∈ Qi for all i > 2. Our choice of α guarantees
that X
σ(αβy) = Tr(αβy) ∈ Q
σ∈Gal(L/Q)

for all y ∈ B. On the other hand for every σ ∈ Gal(L/Q) such

that σ ∈/ D(Q|Pi ), we have σ−1 (Q) = Qi for some i > 2, hence
σ(αβy) ∈ Q for all y ∈ B. By difference, we find
X
σ(αβy) ∈ Q (6.5.1)
σ∈D(Q|Pi )

for all y ∈ B.
Since (p) is unramified, we have an isomorphism D(Q|Pi ) = ∼
Gal(B/Q/A/Pi ), and then (6.5.1) implies that αβ = 0 in B/Q.
This means that αβ ∈ Q, which is a contradiction since neither α
nor β belong to Q.

Example 6.5.2. Continuing

√ Example 6.1.2, let us compute the dis-
criminant of K = Q( m), with √m squarefree. When m ≡ 1
(mod 4), a basis for OK is just {1, m}, hence
  2
1 1
disc(K) = det √ √ = m,
m − m

so the only ramified primes are divisors of m.

When m ≡ 2, 3 (mod 4),
!!2
1√ 1√
disc(K) = det 1+ m 1− m = 4m,
2 2

so the ramified primes are divisors of m, as well as 2.

6.6 computing prime factorizations

At this point, we know a great deal about factorization of primes
in number rings. It remains to find a way to actually compute
them.

193
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 lattice methods

In order to do this, let K ⊂ L be number fields, with integer

rings A and B respectively. Take any α ∈ B such that L = K(α).
In general, A[α] is a subring of B, but since they have the same
rank as Z-modules (why?), the index m := [B : A[α]] is finite. We
will be able to provide a factorization in B for all primes P ⊂ A
that are coprime with m.

Theorem 6.6.1 (Kummer). Let K ⊂ L be number fields, with integer

rings A and B respectively. Choose some α ∈ B such that L = K(α);
let g(x) ∈ A[x] be its minimal polynomial over K and denote m := [B :
A[α]].
Let P ⊂ A be a prime such that P ∩ Z = (p), for some p that does
not divide m. Factor

g(x) = g1 e1 (x) · · · gr er (x) in A/P[x].

Then P · B = Qe11 · · · Qer r , where Qi := (P, gi (α)) (as an ideal of B).

A/P[x]
Proof. Since A/P is a field, and gi is irreducible in A/P[x],
(gi (x))
is again a field, hence the ideal (P, gi (x)) is maximal inside A[x].
Now consider the composition

f → f(α)
ψi : A[x] → A[α] →B → B/Qi .

The evaluation map is clearly surjective. The map A[α] → B/Qi

is surjective as well. In fact, A[α] and Qi are both subgroups of
full rank in B – the first has index m, while the second has index
a power of p by Proposition 6.2.12. It follows that they must
generate B, hence ψi is surjective. Since both P and gi are in its
kernel, we derive a surjective homomorphism

A[x] B
→ . (6.6.1)
(P, gi (x)) Qi

Since the former is a field, B/Qi is a field or trivial, which means

that Qi is either prime or the whole B.
Now let us show that Qi + Qj = B for i 6= j. To see this, since
gi and gj are coprime in A/P[x], we can write

h i gi + h j gj ≡ 1 mod P

194
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 6.6 computing prime factorizations

for some hi , hj ∈ A[x]. Evaluating at α, we get to write

hi (α)gi (α) + hj (α)gj (α) + f(α) = 1

for some f(x) ∈ P[x], which shows that P, gi (α) and gj (α) gener-
ate B, as desired.
We now prove that P · B divides Qe11 · · · Qer r , or equivalently

Qe11 · · · Qer r ⊂ P · B.

In fact, Qei i ⊂ (P, gi (α)ei ), hence

Qe11 · · · Qer r ⊂ (P, g1 (α)e1 · · · gr (α)er ).

But g1 (α)e1 · · · gr (α)er = g(α) + r(α) = r(α) for some r(x) ∈ P[x],
hence the ideal simplifies to P · B.
It follows that P · B = Qd 1 dr
1 · · · Qr for some di 6 ei (omit Qi if
it is the whole B). We want to prove we have equality for each i.
By Theorem 6.2.13 we get that

X
r
n := [L : K] = d i fi ,
i=1

where fi = f(Qi |P) = deg gi , the last equality in virtue of the iso-
morphism (6.6.1) (we have omitted terms where B/Qi is trivial).
On the other hand,
X
r
n = deg g = deg g = e i fi .
i=1

This means that we have di = ei for all i, and no term can be

omitted, proving the theorem.
Example 6.6.2.
√ By Example 6.5.2, we know the ramified primes
of K = Q( m), with m squarefree. With Kummer’s theorem, we
can be more explicit. Let √ us assume that m ≡ 2, 3 (mod 4), so
that a basis for OK is {1, m}. √
The minimal polynomial for m is just f(x) = x2 − m. Let us
reduce this modulo a prime p.
(a) If p√divides m, f(x) = x2 , and by Kummer’s theorem pOK =
(p, m)2 .

195
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 lattice methods

(b) If p = 2, f(x) = x2 of (x + 1)2 according to whether m ≡

2 or 3 (mod 4), and correspondingly
 √
(2, m)2
2OK = √
(2, m + 1)2

(c) If p does not divide m and is not 2, f(x) has distinct roots
modulo p. √If m is a square
√ modulo p, say m ≡ c2 , then
pOK = (p, m − c)(p, m + c). Otherwise, f is irreducible
modulo p, and in this case pOK = (p) is irreducible as well.
The case where m ≡ 1 (mod 4) is Exercise 1.

6.7 geometry of ideal lattices

The results that we have proved so far in the chapter are mostly
valid over general Dedekind rings, although in the case of num-
ber rings we have been able to give more explicit proofs based
on the properties of the discriminant and the ideal norms. In this
section, we make more explicit use of the embedding of a num-
ber ring as a lattice in Rr × Cs to derive some finiteness results
that are actually specific to the case of number fields. Namely, if
A is the ring of integers in a number field K, the multiplicative
group A∗ if finitely generated, and the class group G(A) is finite.
Remark 6.7.1. The two groups are related by a simple exact se-
quence. Namely, given a nonzero a ∈ A, factor the ideal (a) as a
product of primes
(a) = P1e1 · · · Prer .
This defines a monoid homomorphism A \ 0 → P Z, where the
L
sum ranges over all primes P of A, which sends a to a vector
with componentL ei at Pi . By adding inverses, we get a homomor-
phism φ : K∗ → P Z, and by definition we have ker φ = A∗ and
coker φ = G(A). Putting all together we get an exact sequence

→ A∗ → K∗ → P Z → G(A) → 0.
L
0

The main technical tool of this section is the following geo-

metric theorem of Minkowski. We will use this result – together

196
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 6.7 geometry of ideal lattices

with the computation of the volume of the embedding of a num-

ber ring – to construct nontrivial integer algebraic numbers with
some control on their size.

Theorem 6.7.2 (Minkowski’s convex body theorem). Let L ⊂ Rn

be a lattice and let B ⊂ Rn be a convex, Lebesgue measurable subset,
symmetric with respect to the origin. If vol(B) > 2n vol(L), then there
exists a nonzero λ ∈ L ∩ B. If moreover B is compact, the same holds
under the weaker assumption vol(B) > 2n vol(L).

Proof. Let D be a fundamental domain for L, so that

◦
[
Rn = (λ + D),
λ∈L

where the dot denotes that the union is disjoint. Denote by 12 B a

rescaling of B by 1/2, so that vol( 12 B) > vol(D). We can write
◦  
1 [ 1
B= B ∩ (λ + D) ,
2 2
λ∈L

hence
  X   X   
B 1 1
vol = vol B ∩ (λ + D) = vol B−λ ∩D .
2 2 2
λ∈L λ∈L

This implies that the sets ( 12 B − λ) ∩ D cannot be disjoint. So

we find λ1 , λ2 such that
   
1 1
B − λ1 ∩ B − λ2 6= ∅,
2 2

that is, there are b1 , b2 ∈ B such that b1 /2 − λ1 = b2 /2 − λ2 .

Then b := (b1 − b2 )/2 = λ2 − λ1 ∈ L, and moreover b ∈ B since
B is convex and symmetric.
When B is compact and vol(B) > 2n vol(L), we can apply the
above results for (1 + )B for all  > 0, and find a nonzero el-
ement in (1 + )B ∩ L. The intersection (1 + )B ∩ L is actually
finite, since B is compact and L is discrete, so the elements we
find for various  will converge to a nonzero element in B ∩ L.

197
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 lattice methods

To apply the above result, let K be a number field, A its ring

of integers, I ⊂ A an ideal. If K admits r real and 2s nonreal
embeddings into C, we have an embedding σ : A → Rr × Cs
which exhibits A as a lattice in Rn , for n = r + 2s. By Proposition
6.1.7 and Remark 6.2.6 we know that
vol(A) = 2−s |disc(A)|
p

vol(I) = 2−s |disc(A)| · kIk

p

for an ideal I ⊂ A. By applying Minkowski’s theorem for various

choices of convex bodies, we can produce nontrivial algebraic
integers satisfying various constraints.
Proposition 6.7.3. Let K be a number field of degree n over Q, A its
ring of integers, I ⊂ A an ideal. Let r, s be the signature of K, so that
n = r + 2s. Then I contains an element x of norm
|N(x)| 6 λkIk,
where  s
4 n! p
λ= |disc(A)|. (6.7.1)
π nn
Proof. Choose the convex body Bt ⊂ Rr × Cs defined by
 
 X
r X
s 
Bt := (y1 , . . . , yr , z1 , . . . , zs ) | |yi | + 2 zj 6 t .
 
i=1 j=1

We choose t to have vol(Bt ) = 2n vol(I), in order to apply Minkowski.

A straightforward but long computation (see Exercise 9) gives
 π s  tn
vol(Bt ) = 2r , (6.7.2)
2 n!
hence we choose t so that
 s 
4
|disc(A)| · kIk.
n
p
t = n! (6.7.3)
π
For this value of t, we find x ∈ I such that σ(x) ∈ Bt . If
σ1 , . . . , σr are the real embeddings and τ1 , . . . , τs are the nonreal
embeddings of K,
Y
r Y
s
2
|N(x)| = |σi (x)| · τj (x) ,
i=1 j=1

198
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 6.7 geometry of ideal lattices

n
so |N(x)| 6 n
t
n by the arithmetic-geometric inequality. Substitut-
ing our choice of tn from (6.7.3) gives the desired result.

Corollary 6.7.4. Let K be a number field of degree n over Q. Then

 n−1
π 3π
|disc(K)| > .
3 4

In particular, |disc(K)| > 1 unless K = Q.

You will prove in Exercise 13 that for any d there exist only
finitely many number fields with discriminant bounded by d.

Proof. Taking I = A in Proposition 6.7.3, we find an x ∈ A with

|N(x)| 6 λ, in particular λ > 1. This simplifies to
 π 2s n2n  π 2n n2n
|disc(A)| > > .
4 (n!)2 4 (n!)2

Calling an the right hand side, we find

1 2n
 
an+1 π 3π
= 1+ > ,
an 4 n 4

and the thesis follows by induction.

Corollary 6.7.5. Let C ∈ G(A) be any class of ideals. Then there exists
I ∈ C such that kIk 6 λ, where λ is defined in (6.7.1)

Proof. Let J ∈ C−1 any integer ideal. By Proposition 6.7.3 we

find x ∈ J with |N(x)| 6 λkJk. Then there exists I ∈ C such that
(x) = I · J. By Proposition 6.2.9 we get

|N(x)| = k(x)k = kIk · kJk,

hence kIk 6 λ.

Proposition 6.7.3 also implies the following fundamental result.

Theorem 6.7.6. Let K be a number field, A its ring of integers. The

class group G(A) is finite.

199
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 lattice methods

Proof. Let C ∈ G(A) be any class of ideals. By Corollary 6.7.5,

there is an integer ideal I ∈ C such that kIk 6 λ. This bounds
the norm of each prime P in the factorization of I by Proposition
6.2.9. So there are finitely many choices for (p) = P ∩ Z, and in
turn finitely many choices for P.

Definition 6.7.7. Let K be a number field with ring of integers

A. The cardinality of G(A) is called the class number of K, and
denoted by h(K).

We will see in the exercises some easy cases in which h(K)

can be computed explicitly. Minkowski’s theorem also implies
another fundamental finiteness result.

Theorem 6.7.8. Let K be a number field of degree n over Q, A its

ring of integers, I ⊂ A an ideal. Let (r, s)be the signature of K, so
that n = r + 2s. The group of units of A is finitely generated – more
precisely,
A∗ =∼ Zρ × U,

where ρ = r + s − 1 and U is a finite group, namely U is the group of

roots of unity in K.

As it turns out, the proof is rather elementary, but the delicate

part is the equality ρ = r + s − 1. We need a lemma, which is
useful in itself:

Lemma 6.7.9. Let A be a number ring, x ∈ A. Then x ∈ A∗ if, and

only if, N(x) = ±1.

Proof. Assume N(x) = ±1, so that x satisfies the polynomial

xn + an−1 xn−1 + · · · + a1 x ± 1 = 0

for some a1 , . . . , an−1 ∈ Z. Then

x · (xn−1 + an−1 xn−2 + · · · + a1 ) = ±1,

hence x is invertible. The other implication is trivial.

Proof of Theorem 6.7.8. Since we work inside the multiplicative group

A∗ , we consider a modification of our standard embedding σ. Let

200
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 6.7 geometry of ideal lattices

σ1 , . . . , σr be the real embeddings of K and τ1 , . . . , τs the nonreal

ones. We define a homomorphism

log σ : A∗ → Rr+s

x → (log(|σ1 (x)|), . . . , log(|τs (x)|))

and notice that N(x) = 1 if, and only if, log σ(x) lies in the hyper-
plane H defined by the equation

X
r X
s
yi + 2 zj = 0.
i=1 j=1

Notice that log σ is a group homomorphism, with kernel


K̃ := ker log σ = x ∈ A∗ | |σi (x)| = τj (x) = 1 for all i, j .

The coefficients of the minimal polynomial of any element x ∈ K̃

are bounded, which implies that K̃ is finite. Since K̃ is a multi-
plicative finite subgroup of C∗ , it must be cyclic – hence K̃ is the
group of roots of unity in K.
It remains to study the image S := log σ(A∗ ) ⊂ H. By the same
reasoning, for each compact W ⊂ H, the intersection S ∩ W is
finite. This implies that S is a discrete subgroup of H, hence it is
a free abelian group of rank ρ 6 r + s − 1. We conclude that we
have an exact sequence

0 →U → A∗ → Zρ → 0,

which splits, and this implies the theorem, save for the equality
ρ = r + s − 1.
To conclude, we must prove that S generates H. Given any
λ1 , . . . , λr , µ1 , . . . , µs > 0, consider the convex body Bλ,µ ⊂ Rr ×
Cs defined by

Bλ,µ = {(y1 , . . . , yr , z1 , . . . , zs ) | |yi | 6 λi , zj 6 µj }.

Then
Y
r Y
s
vol(Bλ,µ ) = (2λi ) (πµ2j ).
i=1 j=1

201
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 lattice methods

In order to apply Minkowski, we choose the λi and µj so that

1 p
λ1 . . . λr µ21 . . . µ2s = T > 2n |disc(A)|. (6.7.4)
(2π)s

Then we have

|disc(A)| = 2n vol(A).
p
vol(Bλ,µ ) > 2n−s

By Theorem 6.7.2 we find xλ,µ ∈ R, x 6= 0, such that σ(x) ∈ Bλ,µ .

Moreover, N(xλ,µ ) 6 T by construction. Notice that

λi
6 σi (xλ,µ ) 6 λi ,
T
and similarly
µj
6 τj (xλ,µ ) 6 µj .
T
By taking logarithms and rearranging things a bit, these inequal-
ities become

0 6 log λi − log σi (xλ,µ ) 6 logT

(6.7.5)
0 6 log µj − log τj (xλ,µ ) 6 logT .

The projection on the first r + s − 1 coordinated is an isomor-

phism between H and Rr+s−1 . If we are able to show that there
is no nontrivial linear form on H that vanishes on S, we are done.
Now take any linear form over Rr+s−1 , say

X
r X
s−1
F(y1 , . . . , yr , z1 , . . . , zs−1 ) = ci yi + dj zj .
i=1 j=1

By multiplying (6.7.5) by ci and dj and summing, we get

X
r X
s−1
06 ci log λi + dj log µj − F(log σ(xλ,µ ))
i=1 j=1
  (6.7.6)
X
r X
s−1
6 |ci | + dj  T .
i=1 j=1

202
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 6.8 cyclotomic rings

P P
|ci | + T . Given h ∈ N, choose λi and


Take any B > dj
µj to satisfy

X
r X
s−1
ci log λi + dj log µj = 2Bh,
i=1 j=1

apart from µs , which is chosen to satisfy (6.7.4). The correspond-

ing xλ,µ is denoted xh . Equation (6.7.6) simplifies to

|F(log σ(xh )) − 2Bh| < B,

so
(2h − 1)B < F(log σ(xh )) < (2h + 1)B.
This forces the values obtained for xh to be all distinct, as the
choice of h varies. On the other hand, they all satisfy |N(xh )| 6 T .
Since the ideals of limited norm are finite in number, we must
have (xh ) = (xk ) for some h 6= k, so xh = ξxk for some ξ ∈ A∗ .
From the fact that

F(log σ(xh )) 6= F(log σ(xk ))

we find that F(log σ(ξ)) 6= 0, showing that F does not vanish

identically on S and concluding the proof that ρ = r + s − 1.

Another beautiful application of Minkowski’s theorem is in Ex-

ercise 13.

6.8 cyclotomic rings

We now give an example of the previous theory to a special but
very interesting case, namely the cyclotomic fields. Recall that
the cyclotomic polynomials φn (x) are defined inductively by the
rule Y
xn − 1 = φn (x) · φd (x), (6.8.1)
d|n

starting from φ1 (x) = x − 1. They are constructed in such a way

that the roots of φn (x) are exactly the primitive n-th roots of 1.

203
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 lattice methods

Remark 6.8.1. Using (6.8.1) it is immediate to prove by induction

that φn (x) is monic with coefficients in Z. Just note that the
quotient of two monic polynomials in Z[x] is again monic and
with integer coefficients.

Let ζn be a primitive n-th root of unity. All other primitive

roots of unity are powers of ζn , so the field Q(ζn ) is the splitting
field of φn . In particular the field Q(ζn ) is a Galois extension of
Q.

Definition 6.8.2. A field of the form Q(ζn ) is called a cyclotomic

field.

The following result is well known.

Proposition 6.8.3. The polynomials φn (x) are irreducible.

Proof. It is equivalent to say that all primitive n-th roots of unity

are conjugated. For this it is enough to show that if p is a prime
not dividing n, ζn and ζp n are conjugated.
Let f(x) ∈ Z[x] be the minimal polynomial of ζn , and write

φn (x) = f(x)g(x);

by Gauss’ lemma we can take both f, g ∈ Z[x].

Now assume by contradiction that f(ζp p
n ) 6= 0, hence g(ζn ) = 0.
So ζn is also a root of

h(x) := g(xp ),

which means f(x)|g(xp ). Reducing this modulo p we find

f(x)|g(xp ) = g(x)p .

This implies that φn (x) has some repeated root in a finite ex-
tension of Fp . But xn − 1 does not have repeated roots modulo
p, since it is relatively prime to its derivative nxn−1 , contradic-
tion.

Corollary 6.8.4. The Galois group of Q(ζn ) over Q is isomorphic to

(Z/nZ)∗ .

204
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 6.8 cyclotomic rings

Proof. For any a ∈ (Z/nZ)∗ we can define an automorphism τa

of Q(ζn ) fixing Q by the request that

τa (ζn ) = ζa
n.

Such an automorphism exist because ζn and ζa n are conjugated,

and it is clearly uniquely determined since ζn generates the field.
It is clear that
τa ◦ τb = τab ,
and this gives an injective homomorphism

(Z/nZ)∗ → Gal(Q(ζn )/Q).

Since both groups have cardinality φ(n), this is an isomorphism.

We can also compute the ring of integers of Q(ζn ) by using

discriminants.
Lemma 6.8.5. The discriminant disc(Z[ζn ]) divides a power of n.
Proof. Let N = NQ(ζn )/Q be the norm. By 4.2.11 we need to
compute N(φn 0 (ζ )).
n
n
Write x − 1 = φn (x)g(x). Taking derivatives and evaluating
at ζn we get
0
nζ−1 n−1
n = nζn = φn (ζn )g(ζn ).

Taking the norm and using the fact that N(ζn ) = 1 we get
0
N(n) = N(φn (ζn ))N(g(ζn )), (6.8.2)

and the left hand side is a power of n.

Theorem 6.8.6. The ring of integers of Q(ζn ) is Z[ζn ].
Proof. Let A be the ring of integers of Q(ζn ), so that Z[ζn ] ⊂ A.
First, assume that n = pa is the power of a prime p, so that
disc(Z[ζn ]) is a power of p as well by the above lemma. By
Proposition 6.1.3, we have the inclusion

1 1
A⊂ c
Z[ζn ] = c Z[1 − ζn ]
p p

205
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 lattice methods

for some c > 0.

Assume by contradiction that Z[ζn ] ( A, and let x ∈ A not in
Z[ζn ]. Write

m0 + m1 (1 − ζn ) + · · · + mt−1 (1 − ζn )t−1
x=
pc

where t = φ(n). We can change x with x · ps for some s, so we

can assume c = 1. Also we can remove the first terms if p divides
some mi , so assume that

mi (1 − ζn )i + · · · + mt−1 (1 − ζn )t−1
x=
p

with p 6 |mi .
We can also remove higher order terms, but this is trickier.
Since
Yn
φn (x) = (x − ζk
n ),
k=1,(k,p)=1

we have
Y
n
p = φn (1) = (1 − ζk
n ).
k=1,(k,p)=1

Each term 1 − ζk t
n is a multiple of 1 − ζn , hence (1 − ζn ) divides
p
p in Z[ζn ]. Since i < t, this implies that y := (1−ζ )i+1 ∈ Z[ζn ],
n
hence xy ∈ A. By expanding this we get
mi
+ mi+1 + · · · + mt−1 (1 − ζn )t−1−i ∈ A,
1 − ζn
mi
hence 1−ζ n
∈ A, with p 6 |mi .
Write this as mi = (1 − ζn ) · α and take norms to find mri =
N(1 − ζn )N(α) = pN(α), which is a contradiction since p does
not divide mi . This proves the thesis for n a prime power.
For the general case, write n = pa 1 ak
1 · · · pk . We can apply the
above for each field Q(ζpai ). To conclude, use Proposition 6.1.8
i
and the fact that the discriminant of Z[ζpai ] divides a power of
i
pi , so all these discriminants are prime with each other.

206
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 6.8 cyclotomic rings

We now focus in more detail on the ring Z[ζp ], where p is

an odd prime. First, we can compute the discriminant exactly.
Namely, xp − 1 = (x − 1)φp (x), so (6.8.2) simplifies to

pp−1 = N(φp0 (ζp )) N(ζp − 1).

The minimal polynomial of ζp − 1 is φp (x + 1), and its constant

term is p, so N(ζp − 1) = p. Since
p−1
disc(Z[ζp ]) = (−1)( 2 ) N(φ 0 (ζ )),
p p

we get
disc(Z[ζp ]) = ±pp−2 ,
where the sign is + for p ≡ 1 (mod 4) and − otherwise. Since
the discriminant is defined as the square of the √determinant of a
matrix with entries in Z[ζp ], we conclude
√ that ±p ∈ Z[ζp ]. In
particular, this shows the inclusion Q( ±p) ⊂ Q(ζp ).
Let us understand the splitting of a prime q 6= p in Z[ζp ].
By Kummer’s theorem 6.6.1, we can do this by computing the
factorization of φp inside Z/qZ[x].
The finite extensions of Fq = Z/qZ are the fields Fqk , while
the roots of φp are primitive p-th roots of unity. In fact, the roots
0
of φp over Fq are distinct, since φp is not 0. The group F∗qk
is cyclic, so the roots of φp lie in Fqk if and only if qk − 1 is
multiple of p. In this case, since all the roots are inside the same
extension of Fq of degree k, the factorization of φp is

φ p ≡ g1 · · · gr ,
p−1
where r = k . The corresponding factorization of q looks like

qZ[ζp ] = Q1 · · · Qr ,

where e(Qi |q) = 1 and f(Qi |q) = k.

In this case E(Qi |q) is trivial, but D(Qi |q) is not, and corre-
sponds to a subfield Q(ζp )D of Q(ζp ) of degree r = p−1 k over
Q. By Corollary 6.8.4, the Galois group of Q(ζp ) over Q is cyclic,
so there is a subgroup for each d that divides p − 1. Let us de-
note the by F(d) the subfield of Q(ζp ) of degree d over Q. By
comparison, we get the following useful result.

207
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 lattice methods

Theorem 6.8.7. Let p be an odd prime number, q 6= p another prime,

and d a divisor of p − 1. Then q splits completely inside F(d) if and
only if q is a d-th power mod p.

Proof. The group (Z/pZ)∗ is cyclic, so q is a d-th power if and

only if q(p−1)/d ≡ 1 (mod p). The order of q in (Z/pZ)∗ is
exactly f(Qi |q) = k, so this amounts to (p − 1)/d being a multiple
of k. Equivalently, this is the same as r = (p − 1)/k being a
multiple of d, or F(d) being contained in Q(ζp )D . The conclusion
follows by Proposition 6.4.6.
√
In fact, we know that F(2) = Q( ±p), where the sign is pos-
itive if p ≡ 1 (mod 4) and negative otherwise. We can use this
to obtain a slick proof of the celebrated quadratic reciprocity for-
mula. To introduce that, we first need a

Definition 6.8.8. Let a, p ∈ Z with p prime. The Legendre symbol

of a and p is defined by

  
0 if a is multiple of p
a p−1
=a 2 (mod p) = 1 if a is a square (mod b)
p 

−1 otherwise.

It is immediate to check that

    
ab a b
= ,
p p p

which can be used to reduce the computation of a Legendre sym-

bol to the case where a is a prime number. The following famous
result allows us to simplify the computation more.

Theorem 6.8.9 (Quadratic reciprocity theorem). Let p and q be two

different primes. If both are odd, then
   
p ( p−1 )( q−1 ) q
= (−1) 2 2 .
q p

For the prime 2,

 
2 p2 −1
= (−1) 8
p

208
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 6.9 exercises

Proof. Let us consider the odd case, and assume   that  p ≡ 1

(mod 4), in which case we aim to prove that q = q p
p . Let
us consider the field F(2) ⊂ Q(ζp ); as we have verified above this
√
is just Q( p).  
By Theorem 6.8.7, q p = 1 if and only if q splits completely
√
inside Z[ p]. By Kummer’s theorem 6.6.1, this happens if and
only if the polynomial
  x2 − p factorizes in Fq , which happens
p
exactly when q = 1.
The other cases are handled similarly.

6.9 exercises
1. Compute √ the factorization of a prime p in a quadratic number
field K = Q( m), where m ≡ 1 (mod 4) is squarefree.
√
2. Show that Z[ −3] does not have unique √ factorization for ide-
als by considering the ideal I = (2, 1 + −3) and proving that
I2 = 2I, but I 6= (2).
√
3. Show that Z[ 7] is a principal ideal domain, as follows. By
Corollary 6.7.5, we know that every class contains an ideal of
norm at most λ = 2.6457 . . . . It is thus enough to show that
primes of norm at most 2 are principal. If P is such a prime, show
that P ∩ Z is (2). Factor 2 explicitly using Kummer’s theorem.
You can do the same computation for other small quadratic
fields.
√
4. Show in the same way that Z[ −163] is a principal ideal √ do-
main. In fact, it is known that the only quadratic fields Z[ −m],
m > 0 of class number one appear for m = 1, 2, 3, 7, 11, 19, 43, 67
and 163 (Heegner-Stark theorem, [Sta69]).
√
5. Prove that A = Z[ 1+ 2−19 ] is a principal ideal domain, but
is not Euclidean. (For the last part, let α ∈ A be an non-unit of
minimal Euclidean norm – what are the possibilities for A/(α)?)

6. Let α be a solution of f(x) = x3 + x2 − 2x + 8 = 0, and let

K = Q(α) (K is known as the Dedekind field). Compute disc(f) =

209
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 lattice methods

−22 · 503. Let β = 4/α; prove that β is integral, and check that
the discriminant of Z[α, β] is −503. Deduce that Z[α, β] is the
ring of integers of K, and that Z[α] has index 2 inside it.
7. Let K be the Dedekind field from Exercise 6, A its ring of
integers. Show that 2 is unramified in A, and that there exist no
surjective homomorphism from A to one of the finite fields F4
or F8 . Conclude that 2 factors in A as a product of 3 different
primes.
Deduce from this that A is not of the form Z[γ] for any γ ∈ A,
and in fact for any γ ∈ A the index of Z[γ] in A must be even.
Verify the same thing directly by computing disc(Z[γ]) for a
generic γ ∈ Z[α, β] and showing that it must be even.
8. Let L ⊂ Rn be a subgroup which is abstractly isomorphic to
Zn . Show that L is discrete in the Euclidean topology if and only
if L generates Rn as a vector space.
9. Compute the volume of Bt , proving (6.7.2) (use induction on
both r and s).
10. Use Eisenstein’s criterion to give another proof that φp (x) is
irreducible, when p is prime.
11. Check that the two embeddings defined in (6.1.1) and (6.1.2)
agree, when using the same ordering for the factors of f.
12. Finish the proof of Theorem 6.8.9 by analyzing in detail the
other cases.
13. Prove Hermite’s theorem: for any d there exist only finitely
many number fields K with discriminant less than d. (One can
assume that the degree of K over Q is bounded. Choose a convex
body in Rr × Cs where each coordinate is bounded by a small
constant, except for one bounded by a big constant times disc(K).
Use this to get an algebraic integer α in K; prove that in fact
K = Q(α) and use the fact that the coefficients for the minimal
polynomial of α are bounded.)
The following exercises, up to Exercise 16 discuss the Frobe-
nius element of a Galois number field extension.
14. Let L/K be a Galois extension of number fields, P ⊂ OK be a
prime which is not ramified in OL . Let Q be a prime of OL over P.

210
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 6.9 exercises

Show that there is a distinct Frobenius element Φ(Q|P) ∈ Gal(L/K)

such that
Φ(Q|P)(x) ≡ xf (mod Q)
for all x ∈ OL , where f = f(Q|P).

15. Continuing the previous exercise, show that if Q 0 is another

prime over P, say Q 0 = σ(Q) for some σ ∈ Gal(L/K), the Frobe-
nius elements are related by

Φ(Q 0 |P) = σΦ(Q|P)σ−1 .

In particular, if L/K is an abelian extension, Φ(Q|P) = Φ(P) only

depends on P.

16. Let L = Q(ζm ) be a cyclotomic field, where we can take m 6≡ 2

(mod 4). A prime p ∈ Z is unramified in L if and only if p
does not divide m. The Galois group Gal(Q(ζm )/Q) is cyclic of
degree φ(m), and acts by sending ζm to some power ζk m.
Show that the Frobenius element Φ(p) is the element of the
Galois group defined by Φ(p)(ζm ) = ζp m.

17. The following example is taken from [Cla65]. Let K be a

quadratic number field with√ class number bigger then 1 - for
concreteness, take K = Q[ −5]. Choose a prime p that factors as
(p) = Q1 Q2 in OK , where Q1 and Q2 are principal (in our case,
p = 29 will do). Writing Q1 = (q1 ), let S be the multiplicative
system generated by q1 , and A = S−1 OK . Prove that A is not
the integral closure of a principal ideal domain in K. (A cannot
be the integral closure of a subring of Q since Q1 · A is invertible,
but Q2 · A is not. Prove that the class group of A does not become
trivial.)

18. Prove that the equation x2 − 2y2 = 1 has infinitely many

integer solutions. (Notice that x2 − 2y2 is a norm in a suitable
number ring). More generally, show that the same holds for the
equation
x2 − dy2 = 1, (6.9.1)
where d is a square-free integer, that is not congruent to 1 modulo
4. Can you modify your argument to make it work for the case
where d ≡ 1 (mod 4)?. Equation (6.9.1) is know an Pell’s equation.

211
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 lattice methods

19. Give an alternative proof of quadratic reciprocity following

Eisenstein. Let S be the set of even integers s such that 2 6 s 6
p − 1. Denote by r(s) the remainder of the division of qs by p.
Prove that s → (−1)r(s)  (modPp) is a permutation of S, and
 r(s) qs
deduce from this that q s∈S b p c .
p = (−1)
This formula tells us that q is a square modulo p if and only
if the number of points with integer coordinates in a certain tri-
angle is even. Use symmetry to conclude quadratic reciprocity
from this.
This is just another one of a plethora of different proofs that
have been found since Gauss (see for example the book [Bau15]).

The following exercises (up to Exercise 22) lead to a proof of a

special case of Fermat’s last theorem, usually denoted by Case 1
for regular primes: there is no integer solution to ap + bp = cp ,
where p does not divide abc. The case where p divides abc (Case
2) is much harder – see [Tha99]. The general case (when p is not
regular) is the celebrated Wiles and Taylor theorem – see [CSS97].

20. Let p be an integer prime and ζp a primitive pr -th root of

unity, for some r > 1. Show that pZ[ζp ] = (1 − ζp )n , where
n = φ(pr ), and deduce that 1 − ζp is prime.

21. Let p be a prime, u ∈ Z[ζp ] a unit. Prove that u/u is a power

of ζp , where · denotes complex conjugation.

22. Let p > 5 be a prime, and assume that p does not divide
h(Q(ζp )) (p is called a regular prime). Show that there is no in-
teger solution to ap + bp = cp , where p does not divide abc.
(Assuming there is a solution, show that a + ζp b = u · tp for
u, t ∈ Z[ζp ] with u invertible, then use the previous exercise.)

The following exercises (up to Exercise 27) discuss some as-

pects of the geometry of lattices in more detail. For a detailed
treatment of these topics, see [MG02] or [Sim10].

23. Let L be a lattice in Rn , and denote by λ1 (L) the minimum

length of a nonzero vector in L. Prove the result of Minkowski
√ 1
that λ1 (L) < n(vol L) n . This is a theoretical bound, but actu-
ally finding a short vector inside a lattice is a difficult computa-
tional problem, which we tackle in the next exercises.

212
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 6.9 exercises

24. Let L be a lattice in Rn , and {b1 , . . . , bn } a basis of L. De-

note by {b∗1 , . . . , b∗n } the basis of Rn obtained by Gram-Schmidt
orthogonalization, so
X
b∗k = bk − ui,k bi ,
i<k

hbk ,b∗i i
where ui,k = hb∗i ,b∗i i . Prove that λ1 (L) > min kb∗i k.

25. Keep the notation of the previous exercises. We say that the
basis is LLL-reduced if kui,k k 6 21 for all i < k, and

3 ∗ 2
kb∗i+1 + ui,i+1 b∗i k2 > kb k (6.9.2)
4 i

for all i.
The following algorithm was introduced in [LLL82]. It alter-
nates a reduction step and a swap step. The reduction step looks
like an approximate Gram-Schmidt: for all i = 2, . . . , n and for
all j = i − 1, . . . , 1, replace bi by bi − dui,j cbj , where d·c denotes
rounding by the nearest integer. The swap step consists in look-
ing at violations of (6.9.2): if bi and bi+1 violate the condition,
swap them and start again from the beginning.
Prove that the LLL algorithm always terminates and produces
an
Q LLL-reduced basis for a lattice. (The quantity F(b1 , . . . , bn ) :=
2
i kbi k always decreases in the swap step.)

26. With the notation of the previous exercise, prove that if the
basis {b1 , . . . , bn } is LLL-reduced, then

n−1
kb1 k 6 2 2 λ1 (L),

thus obtaining an algorithmic way to produce a not too long vec-

tor in L.

27. The LLL algorithm can be used to find algebraic equations

with small coefficients satisfied by a number, given a good enough
approximation of the number.

213
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 lattice methods

Let α be some algebraic number of degree n, and α some ap-

proximation of α. Choose some big constant K, and consider the
vectors

b0 = (1, . . . , 0, K)
b1 = (0, 1, . . . , 0, Kα)
..
.
bn = (0, 0, . . . , 1, Kαn ).

Let L be the lattice generated by {b0 , . . . , bn }. Using the LLL

algorithm, we can find a short vector v ∈ L. But such vector
must have the form
X
n
!
i
v = a0 , . . . , an , K · ai α .
i=0
P
In order for it to be short, the polynomial n i
i=0 ai α must be
approximately 0, and we can hope that the equation is satisfied
exactly by α.
Try this in practice using α = 1.6180339887 and K = 104 : you
should recover the equation x2 − x + 1 for the golden ratio.

214
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 7
METRIC AND TOPOLOGICAL METHODS

In this chapter we use topological methods to investigate our

rings. We start with the notion of absolute value on a field, which
is analogue to the notion of norm in the theory of Banach spaces,
and we show that the metric completion of a field endowed with
an absolute value is again a field. The important example of the
p-adic absolute values give rise to the so-called p-adic fields.

In the sequel our treatment becomes increasingly algebraic.

First we consider the notion of valuation, which is strictly con-
nected to that of absolute value. We exhibit a general result of
existence for valuations on a field. As a particular case we study
the so-called discrete valuation rings (or DVR), which turn out to
be rings obtained by localization of Dedekind rings.

Finally, in Section 7.5 we show that the metric was not really
necessary. The choice of an ideal I of the ring A determines
a topology on A, and we can define the completion of A with
respect to this topology using the machinery of inverse limits,
which we also introduce. In the rest of the chapter, we study the
properties of this topology and the relation between a ring and
its completion.

Both the metric and the algebraic approach allow us to give a

definition of the p-adic integers Zp . Section 7.7 ends the chapter
by showing yet another way to build Zp , this time using explicit
algebraic formulae. These can be generalized to any ring, giving
rise the construction of rings of Witt vectors.

215
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 metric and topological methods

7.1 absolute values

Definition 7.1.1. Let k be a field. A function
| · |: k → R
is called an absolute value if it satisfies:
1. |a| > 0 and |a| = 0 if and only if a = 0;
2. |ab| = |a| |b|;
3. |a + b| 6 |a| + |b|
for all choices of a, b ∈ k. As usual, condition 3. is called the
triangular inequality.
Remark 7.1.2. This notion is clearly akin to that of a norm in
functional analysis. A norm is required to be homogeneous, and
that makes only sense for real (or complex) vector spaces. By
changing that into condition 2. we obtain a notion which is better
suited to study abstract fields.
We will give examples soon, but before doing that let us intro-
duce an important distiction.
Definition 7.1.3. The absolute value | · | is called nonarchimedean
if it satisfies the stronger inequality

3’. |a + b| 6 max |a| , |b|
for all a, b ∈ k, and archimedean otherwise.
Example 7.1.4. (a) The standard absolute value | · |st on R is an
absolute value in our sense, and it restricts to an absolute
value on every subfield of R.
(b) Similarly we have the Euclidean absolute value | · |st on C.
(c) On every field, we have the trivial absolute value defined by
|a|tr = 1 for all a 6= 0. This is the only absolute value on a
finite field. A quick way to see this is the following.
The multiplicative group F∗q is cyclic; if a is a generator we
must have |an | = |a|n for any absolute value | · |. Taking
n = q − 1 we find |a|q−1 = |1| = 1; since |a| is real and
positive we then have |a| = 1.

216
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 7.1 absolute values

(d) Here comes the real motivation for this notion. Fix a prime
number p ∈ Z; for all a ∈ Z \ {0} denote vp (a) = r if pr
divides a but pr+1 does not. Note that

vp (a · b) = vp (a) + vp (b), (7.1.1)

hence it makes sense to extend this to Q∗ by the rule

a
vp := vp (a) − vp (b),
b
and (7.1.1) still holds. Also note that

vp (a + b) > min vp (a), vp (b) . (7.1.2)

These relations look like the logarithm of those defining an

absolute value. Indeed we can get an absolute value | · |p on
Q defined by
|a|p := p−vp (a) .
Note that | · |p is nonarchimedean.
The function vp is called the p-adic valuation and | · |p is called
the p-adic absolute value. We will come back on valuations in
next section.
(e) By the same token, let A be a Dedekind ring with fraction
field k. Fix a prime P of A; for a ∈ A we let vP (a) be the
exponent of P in the factorization of (a). This satisfies prop-
erties analogue to (7.1.1) and (7.1.2), hence we can extend it
to k∗ by the rule
a
vP := vP (a) − vP (b).
b
Letting q be the norm of P, we define the nonarchimedean
absolute value
|a|P := q−vP (a) .
Again we call vP the P-adic valuation and | · |P the P-adic abso-
lute value.
An absolute value on a field allows us to define a metric on it.
Given an absolute value | · | on the field k define the distance

d(a, b) := |a − b|

217
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 metric and topological methods

for a, b ∈ k. The function d turns k into a metric space; in turn

this induces a topology on k, such that the field operations are
continuous.
It may happen that two different absolute values dominate
each other; in this case they are equivalent for our purposes.
Definition 7.1.5. Let | · |1 , | · |2 be two absolute values on k. We
say that | · |1 is equivalent to | · |2 if there exist a real number e > 0
such that
|a|2 = |a|e1
for all a ∈ k. An equivalence class of absolute values on k is
called a place of k.
It is clear that two equivalent absolute values determine the
same topology on k; it may come as a surprise that the converse
is true.
Proposition 7.1.6. Let | · |1 , | · |2 be two absolute values on k. If the
topologies induced on k by | · |1 and | · |2 are the same, the two absolute
values are equivalent.
Proof. First note that for any absolute | · | and any a ∈ k we have
|a| < 1 if and only if an → 0. Since this is a topological property,
|a|1 < 1 if, and only if, |a|2 < 1. Passing to the reciprocal we find
|a|1 > 1 if, and only if, |a|2 > 1. It follows that for any a ∈ k we
can write
|a|2 = |a|e1
for some e > 0, depending on a.
Now assume

|a|2 = |a|e1 and |b|2 = |b|f1 ;

we shall prove that e = f. If this is not the case, let

log |b|1
c := .
log |a|1

We can assume that e < f and c > 0. Then ce < cf, so we can
find a rational number r/s such that
r
ce < e < cf.
s

218
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 7.1 absolute values

This gives the two inequalities

cs < r and re < csf.
From the first one we derive
|b|s1 = |a|cs r
1 < |a|1 ,

so
ar
> 1.
bs 1
From the second
|a|r2 = |a|re csf
1 < |a|1 = |b|sf s
1 = |b|2 ,

hence
ar
< 1.
bs 2
This is a contradiction.
Remark 7.1.7. The second part of the proof may seem tricky, but
it is simpler than it looks. Once one looks for a number of the
form ar /bs which violates the conclusion of the first part, it is
just a matter of spelling out the needed costraints.
Remark 7.1.8. The function | · |e may fail to be an absolute value,
even if | · | is. For instance if | · |st is the standard absolute value
on R, | · |2st does not satisfy the triangular inequality. Still, | · |e is
an absolute value when | · | is nonarchimedean.
The main reason why we consider absolute values is that we
can use them to costruct new rings or fields by completion of old.
Let k be a field endowed with an absolute value | · |, and let A ⊂ k
a subring. Then the metric completion A b of A is itself a ring,
with the ring operations extended from A by continuity. To make
sense of the preceding sentence, let us recall some definitions.
Definition 7.1.9. Let (X, d) be a metric space. A sequenxe (xn )
of elements of X is called Cauchy if for any  > 0 we can find N
such that
d(xm , xn ) < 
for all m, n > N. The metric space X is called complete if every
Cauchy sequence has a limit in X.

219
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 metric and topological methods

It is trivial to check that, conversely, a converging sequence is

always Cauchy. If (X, d) is any metric space, there is a construc-
tion of a complete metric space (X, b together with an embedding
b d)
X ,→ X b such that d b restricts to d on X and X is dense in X.
b More-
over such X is unique up to isometry, and is called the completion
b
of X. We give the standard construction of X, b leaving uniqueness
to the reader.
Consider the set Y of Cauchy sequences of elements of X. We
call two sequences (xn ) and (yn ) equivalent if

d(xn , yn ) → 0.

We define Xb as the quotient of Y by this equivalence relation. The

equivalence class of a sequence is denoted [xn ]. On X b we define
the distance
b n ], [yn ]) := lim d(xn , yn ).
d([x
n→∞

Such a limit exists in R, since both xn and yn are Cauchy and d

is uniformly continous, so (d(xn , yn )) is a Cauchy sequence in R.
Moreover it is immediate to see that d b does not depend on the
choice of the representatives.
We claim that Xb is a complete metric space with a dense iso-
metric embedding of X. The embedding of X is given by the
function
ι: X → X.b

x → [x]
This is clearly an isometry, and moreover the image is dense.
Indeed for any Cauchy sequence [xn ] ∈ Xb choose N such that
d(xn , xm ) <  for n, m > N. Then

b n ], [xN ]) = lim d(xn , xN ) 6 .

d([x
n→∞

Finally, we check that X b is complete. Let (x(m)) be a Cauchy

sequence in X and write x(m) = [x(m)n ]. One can define the
b
diagonal sequence y by yn = x(n)n . Using the triangular in-
equality, it is easy to check that

lim (x(m)) = [yn ]

m→∞

220
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 7.1 absolute values

in X.
b
We now come back to the setting of absolute values. Fix an
absolute value | · | on the field k. For a subring A ⊂ k let A
b be the
completion. The ring operations on A extend to A, by putting
b

[an ] + [bn ] := [an + bn ]

[an ] · [bn ] := [an · bn ].

It is immediate to check that the sequences (an + bn ) and (an ·

bn ) are indeed Cauchy, and that the operations are well defined.
Note that this is determined by the requirement that the opera-
tions should be continous on the whole A. b The ring axioms for
the operations are valid on A which is dense, hence hold on all
b by continuity. This gives A
A b the structure of a ring.
If moreover A is itself a field – for instance A = k – we also
define
1/[an ] := [1/an ]
for [an ] 6= 0. In this case it is slightly less obvious that 1/an is
Cauchy. Since [an ] 6= 0, an 6→ 0, which implies that |an | >  for
some  > 0 and infinitely many n. Since moreover an is Cauchy,
we have |an | > /2 eventually. For any δ > 0 choose N such that
|an − am | < δ for n, m > N. Then

1 1 am − an 4δ
− = < 2,
an am an am 

proving that the sequence (1/an ) is Cauchy. We deduce that the

completion of a field is itself a field.
Note that the completion b k inherits an absolute value, again
obtained extending by continuity the absolute value | · | on k. We
shall denote this absolute value by the same symbol.

Remark 7.1.10. Once one has constructed b b is simply the clo-

k, A
sure of A inside b
k. So, it is enough to consider completions of
fields.

Example 7.1.11. (a) Consider the Euclidean absolute value | · |st

on Q. The completion of Q is then R. If we complete Q(i)
with respect to the Euclidean absolute value, we obtain C.

221
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 metric and topological methods

(b) The completion of Q with respect to the p-adic absolute value

| · |p is called the field of p-adic numbers and is denoted by Qp .
It contains the ring Zp of p-adic integers as the completion of
Z.
We claim that Zp is local; its maximal ideal is pZp , or equiv-
alently the closure of (p) inside Zp . To see this, we try to
understand what a p-adic integer looks like. Choose any set
D of representatives for Z/(p), for instance

D = {0, 1, . . . , p − 1}.

Let a ∈ Zp ; then a is the limit of a sequence (an ) of integers

which is Cauchy for | · |p . Up to passing to a subsequence,
we can assume that if m, n > N then am − an is divisible by
pN . In particular the class of an modulo pN is independent
of n > N; we shall call αN ∈ D the representative for such
class.
It follows that
X
n
an − αi pi
i=0
Pn
pn+1 , i is equiva-


is divisible by so the sequence i=0 αi p
lent to (an ). We conclude that we can write
∞
X X
n
a= αi pi := lim αi pi . (7.1.3)
n→∞
i=0 i=0

This is completely analogous to writing a real number be-

tween 0 and 1 in terms of powers of 1/10, and we can think
of the coefficients αn as the p-adic digits of a (note that the
argument above shows that the αi are uniquely determined).
The numbers having an expansion with a finite number of
digits are exactly the usual integers.
The function
Zp → Z/(pn )
P
defined by sending a to the finite sum n−1 i
i=0 αi p is easily
seen to be a surjective ring homomorphism extending the
projection from Z. It follows that pn generates a nontrivial
ideal in Zp , and in particular pZp is maximal.

222
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 7.1 absolute values

We can also verify that any p-adic integer a ∈ Zp \ pZp is

invertible. Indeed, consider the element 1/a ∈ Qp . For any
n ∈ N we can use density to find some bn ∈ Q such that

|bn − 1/a|p < p−n ;

in particular, the denominator of bn is not divisible by p and

we can find some cn ∈ Z such that

|cn − bn |p < p−n .

It follows that (cn ) is a sequence of integers which converges

to 1/a in Qp , in particular 1/a ∈ Zp . This completes the
proof of our claim that Zp is local with maximal ideal pZp .
Moreover the universal property of the localization shows
that we have inclusions

Z ⊂ Z(p) ⊂ Zp .

Since Zp has projections to Z/(pn ) for any n but no other

homomorphisms, it is particularly well-suited for studying
congruences modulo powers of p. We shall make this more
precise in Section 7.5, when we treat the completion of a ring
as an inverse limit.

(c) Let k be a number field with ring of integers A. For a prime

P of A consider the absolute value | · |P of k. Then the com-
pletion of k is a field kP , containing the completion AP of
A. By the same arguments as above, AP is a local ring with
maximal ideal PAP , and every element of AP can be written
as
X∞
a= ai ,
i=0

where ai ∈ \ Pi Pi−1 .
The only difference is the lack of a
preferred system of representatives for the digits.

On p-adic numbers, one can mimic Newton’s method to solve

analytic equations. A possible formulation is the following, but
see Theorem 7.6.1 for a generalization.

223
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 metric and topological methods

Theorem 7.1.12 (Hensel’s lemma for p-adics). Let f ∈ Z[x] be a

polynomial, f(x) ∈ Z/pZ[x] its residue modulo the prime p. Assume
0
f has a simple root α ∈ Z/pZ – that is f (α) 6≡ 0. Then α can be lifted
uniquely to a p-adic root of f in Zp .

Proof. The idea is to lift α to roots αk of f modulo pk . By induc-

tion, assume that we find αk ∈ Z such that

f(αk ) ≡ 0 mod pk

and
f 0 (αk ) 6≡ 0 mod p,
the case k = 1 being the hypothesis.
Expand
X
n
f(αk + t · pk ) = ci ti pik
i=0

as a polynomial in t · p. By construction c0 = f(αk ) and c1 =

f 0 (αk ), so modulo pk+1 we have

f(αk + t · pk ) ≡ f(αk ) + f 0 (αk ) · t · pk mod pk+1 .

Since pk divides f(αk ), we can solve for t in Z/pZ by

f(αk )
t=− . (7.1.4)
f 0 (αk )pk

With this choice of t, we get our desired lift αk+1 . The se-
quence {αk } converges in Zp to a root of f that lifts α.

Notice the resemblance of (7.1.4) with the equations used in

the Newton method to approximate roots of real functions.
We shall generalize the completion construction in Section 7.5.
In the final part of this section we classify all places on number
fields. This consists of two, pretty independent, steps: first we
classify places on Q, and then we study the ways one can extend
absolute values from one field to a bigger one. It will turn out
that the only places are those we already met in the examples.

Theorem 7.1.13 (Ostrowski). Every nontrivial absolute value on Q

is equivalent either to the Euclidean absolute value or to a p-adic one.

224
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 7.1 absolute values

Proof. Let | · | be an absolute value on Q. Given m, n ∈ N with

n > 1 we write m in base n as

m = a0 + a1 n + · · · + ar nr

with ai ∈ {0, 1, · · · , n − 1}. By the triangular inequality

X
r
|m| 6 |ai | Ni ,
i=0

where N = max{1, |n|}. Combining this with the easy estimates

log m
r 6 log n and |ai | 6 ai 6 n we find
  log m
log m
|m| 6 (1 + r)nN 6 1 + r
nN log n . (7.1.5)
log n

This is true for all m ∈ N, so we can substitute mk for m in

(7.1.5). After taking the k-th root and letting k → ∞ we find
log m
|m| 6 N log n . (7.1.6)

Two cases now arise

i) Assume |n| > 1 for all n > 2. In this case N = |n| and (7.1.6)
becomes
1 1
|m| log m 6 |n| log n .
By symmetry we must have equality, so there exists a con-
stant c > 0 such that

|n| = clog n = nlog c

for all n > 2. It follows that | · | is equivalent to | · |st .

ii) Otherwise there is some n > 2 such that |n| 6 1. Then N = 1
and from (7.1.6) we find |m| 6 1 for all m ∈ N.
It is immediate to check that the set

{n ∈ Z | |n| < 1}

is a prime ideal, hence it is generated by some prime p ∈ N.

In this case, | · | is equivalent to | · |p .

225
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 metric and topological methods

Finally we discuss how to extend absolute values. The situa-

tion is easier for complete fields, and is completely analogous to
the fact that all norms on Rn are equivalent.

Proposition 7.1.14. Let k be a field complete with respect to the abso-

lute value | · | and let E/k be a finite extension. Then there is at most
one extension of | · | to E.

Proof. We abuse notation and denote by | · | an extension of | · | to

E. Choose a basis e1 , . . . , en of E over k; for v ∈ E we can write

v = v 1 e1 + · · · v n en .

Triangular inequality yields

X
n X
n
|v| 6 |vi | |ei | 6 M |vi | (7.1.7)
i=1 i=1

where M = max{|ei |}.

Note that the function
X
n
kvk := |vi |
i=1

may fail to be an absolute value, but still allows us to define a

distance on E by d(v, w) := kv − wk. Equation (7.1.7) then shows
that | · | is continous with respect to this distance.
The sphere
S := {v ∈ E | kek = 1}
is compact since | · | is complete on k. It follows that | · | assumes
a minimum m on S, so

m kvk 6 |v| 6 M kvk

for v ∈ S; by homogeneity the same inequality holds for all v ∈ E∗ .

In particular | · | and k·k induce the same topology on E. So all
extensions of | · | to E are equivalent.

To generalize Ostrowski’s theorem, let k be a number field with

an absolute value | · |. The restriction of | · | to Q is either | · |p for
some prime p or the standard absolute value.

226
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 7.2 valuations and valuation rings

Let us consider the first case. The completion b k is a finite ex-

tension of Qp – namely, if k = Q(α), then b k = Qp (α). More
explicitly, if g(x) is the minimal polynomial for α, and
g(x) = g1 (x) · · · gd (x)
over Qp , then k b= ∼ Qp [x]/(gi ) for some i. By Proposition 7.1.14,
the absolute value on b k is uniquely determined by its restriction
to Qp .
In fact, we know how to extend the absolute value | · |p from Qp
k. The ring Zp is a Dedekind domain with a single prime ideal
to b
(in the terminology of Section 7.3 it is a DVR), so the integral
closure A of Zp inside b k is a Dedekind ring. If Pb is any prime
of A over pZp , we can define a P-adic absolute value on b
b k which
will extend | · |p . By uniqueness, this must agree with the absolute
value of bk (and in fact, Pb is the only prime above pZp ). It follows
that the absolute value on k is the P-adic absolute value, where
P=P b ∩ Ok .
In a similar way, we can prove that the only archimedean abso-
lute values on k are obtained by embeddings into C. Our conclu-
sion is the following classification.
Theorem 7.1.15 (Ostrowski). Every nontrivial absolute value on a
number field k is equivalent either to the Euclidean absolute value for
some embedding k ,→ C, or to a P-adic one.

7.2 valuations and valuation rings

Recall that we have defined the p-adic absolute value by taking
the exponential of the function vp defined by vp (a) = r if pr
divides a but pr+1 does not. In this section we consider general-
izations of this function.
Notice that we can see vp as a group homomorphism Q∗ → Z,
and that (7.1.2) implies that if vp (a) ∈ N and vp (b) ∈ N, then
vp (a + b) ∈ N.
Definition 7.2.1. An ordered group G (written additively) is an
abelian group endowed with a subset P closed under the group
operation (the positive elements) such that for every g ∈ G, either
g ∈ P or −g ∈ P.

227
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 metric and topological methods

An ordered group has an order compatible with its group struc-

ture. Namely, define g 4 h if h − g ∈ P. The fact that P is closed
under the group operation implies that ≺ is transitive, and the
last requirement in the definition implies that ≺ is a total order.
Example 7.2.2. Taking G = Z and P = N we see that Z is an
ordered group with the usual order.
Definition 7.2.3. A valuation on the field k is a group homomor-
phism v : k∗ → G, where G is an ordered group, such that

v(a + b) < min v(a), v(b) . (7.2.1)

Valuations and absolute values are strictly related concepts, as

shown in the following remark.
Remark 7.2.4. When the target group of v is Z with the standard
ordering, one can define a nonarchimedean absolute value by
taking |x| = av(x) for any real a > 1. In general, there exist valua-
tions with target groups different from Z, as well as archimedean
absolute values, so neither concept is more general than the other.
Associated to a valuation v, there is a ring

A = {a ∈ k|v(a) < 0} ∪ {0}.

Notice that if a ∈ k is a nonzero element, either a ∈ A, or 1/a ∈ A.

We will see that this property alone characterizes rings of this
form, and in fact it is enough to recover the group G and the
valuation from the ring alone.
Definition 7.2.5. Let A be an integral domain with field of frac-
tions k. We say that A is a valuation ring if for all a ∈ k∗ either
a ∈ A or 1/a ∈ A.
The set M of elements a ∈ A such that 1/a ∈/ A is an ideal. In
fact, let x ∈ A and a ∈ M. Then ax ∈ M. If this was not the case,
1/(ax) ∈ A, hence 1/a = x/(ax) ∈ A, contradiction. Similarly
take a, b ∈ M - then we can assume by symmetry that a/b ∈ A,
so  
a
a+b = 1+ ·b ∈ M
b
by the previous point. Since M consist exactly of non-unit ele-
ments, it follows that it is the only maximal ideal. We conclude:

228
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 7.2 valuations and valuation rings

Proposition 7.2.6. Let A be a valuation ring. Then A is local, with

maximal ideal
/ A} = A \ A∗ .
M = {a ∈ A|1/a ∈
Knowing this, we can construct the valuation from a valuation
ring as follows. Let A∗ be the group of units of A. We know that
– if A arises from a valuation v – v(A∗ ) = 0. So we can just take
the abelian group G = k∗ /A∗ , with positive elements P the image
of A. It is clear that P is closed under the group operation, hence
G is an ordered group.
The valuation v is just the projection k∗ → k∗ /A∗ . It remains
to check condition (7.2.1). This amounts to saying that given
a, b ∈ k, either (a + b)/a ∈ A or (a + b)/b ∈ A. But
a+b b a+b a
= 1+ ; = 1+ ,
a a b b
so this follows from the definition of a valuation ring.
We summarize the discussion so far.
Theorem 7.2.7. Let A be a valuation ring with fraction field k. Then
there exists a valuation v on k such that
A = {x ∈ k|v(x) < 0}. (7.2.2)
Vice versa, given a valuation v on a field k, define A by (7.2.2). Then A
is a valuation ring.
In the rest of the section, we make some important connections
between valuation rings and integral closure.
Proposition 7.2.8. Let A be a valuation ring. Then A is integrally
closed.
Proof. Let k be the field of fractions of A and x ∈ k an element
integral over A so that

xn + an−1 xn−1 + · · · a0 = 0
for some a0 , . . . , an−1 ∈ A. Rewrite this as
 a0 
x = − an−1 + · · · + n−1 .
x
Since A is a valuation ring, either x ∈ A or 1/x ∈ A. In the latter
case the above equation shows that x ∈ A as well.

229
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 metric and topological methods

We now turn our attention to a way to actually construct valu-

ation rings for a field k.
Theorem 7.2.9. Let k and K be two fields, with K algebraically closed,
and consider homomorphisms f : A → K, where A is a subring of k.
Assume (f, A) is maximal among such pairs. This means that there
is no pair (g, B) of a ring B ⊂ k containing A and a homomorphism
g : B → K which restricts to f on A. Then A is a valuation ring.
Notice that maximal pairs of this kind always exist thanks to
Zorn’s lemma.
Example 7.2.10. When k = Q and K is the algebraic closure of
Z/pZ, the maximal ring is the localization Z(p) . In fact the pro-
jection Z → Z/pZ can be extended to any element a/b ∈ Q
such that b is not divisible by p. Unsurprisingly, the associated
valuation is the p-adic valuation.
Before getting to the theorem, we need a couple of intermediate
results.
Lemma 7.2.11. Let f : A → K be a maximal homomorphism as in
Theorem 7.2.9. Then A is a local ring with maximal ideal ker f.
Proof. It is enough to prove that if a ∈ A and f(a) 6= 0, f can be
extended to A[1/a]. By maximality, it will follow that 1/a ∈ A.
Consider the homomorphism g : A[x] → K sending x to 1/f(a).
We want to check that this descends to a homomorphism defined
on A[1/a]. In other words, let g : A[x] → A[1/a] be the evaluation
morphism; we need to check that ker g ⊂ ker g.
For this, let p(x) ∈ A[x], and assume that p(1/a) = 0. If

p(x) = p0 + p1 x + · · · + pr xr

with p0 , . . . , pr ∈ A, then
f(p1 ) f(pr )
g(p) = f(p0 ) + +···+ ,
f(a) f(a)r

hence we can multiply by f(a)r to get

f(a)r g(p) = f(ar p(1/a)) = 0.

230
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 7.2 valuations and valuation rings

Lemma 7.2.12. Let f : A → K be a maximal homomorphism as in

Theorem 7.2.9 and a ∈ k a nonzero element. Let M ⊂ A be the maximal
ideal. Then either M[a] ( A[a] or M[1/a] ( A[1/a].
Proof. Assume by contradiction that M[a] generates A[a] and the
same holds for 1/a. Then we have equations

1 = r0 + r1 a + · · · + rm am
1 = s0 + s1 /a + · · · + sn /an ,

where all the ri , sj ∈ M and we can take m, n minimal. By sym-

metry, assume m > n – then we rewrite the second equation as

(1 − s0 )an = s1 an−1 + · · · + sn .

Notice that 1 − s0 is invertible, so we get

an = t1 an−1 + · · · + tn

for some ti ∈ M. This allows us to write am as a combination of

powers of a of lower degree, hence in the first equation m cannot
be minimal.
Proof of Theorem 7.2.9. Let a ∈ k; we want to show that either a or
1/a belongs to A. Let M be the maximal ideal of A. By Lemma
7.2.12 and symmetry, we can assume that M[a] ( A[a], so M[a]
is contained in some maximal ideal M 0 of A[a].
Since M is maximal, M 0 ∩ A = M, and we get an embedding
of fields L = A/M ⊂ L 0 = A[a]/M 0 . Notice that L 0 = L(a), and a
is algebraic over L. Since L comes equipped with an embedding
into K, we can extend this to an embedding of L 0 into K.
By composition, this gives an extension of f to the ring A[a].
By maximality, it follows that a ∈ A.
Corollary 7.2.13. Let k be a field A ⊂ k a ring. The integral closure
of A in k is the intersection of all valuation rings of k which contain A.
Proof. One inclusion is clear by Proposition 7.2.8. For the other
one let a ∈ k and assume that a is not integral over A. In partic-
ular, a is not in B = A[1/a]. This means that 1/a is not invertible
in B, hence it is contained in a maximal ideal M.
Let K be an algebraic closure of B/M. We have a homomor-
phism f : B → K sending 1/a to 0. In particular, f cannot be

231
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 metric and topological methods

extended to B[a]. On the other hand, it can be extended to a

valuation ring C ⊃ B by Zorn’s lemma and Theorem 7.2.9. This
shows that a ∈/ C, which gives the reverse inclusion.

7.3 discrete valuation rings

A discrete valuation ring, or DVR, is a valuation ring A, where the
underlying valuation v has target the ordered group Z. These
rings happen to have a particularly simple structure.
First, we already know that they are local rings with maximal
ideal M = {a ∈ A | v(a) > 1}. More generally, observe that for
every n ∈ N, we can define the ideal In = {a ∈ A | v(a) > n}, so
that I0 = A and I1 = M.
Let a, b ∈ A be two elements such that v(a) = v(b). Then
v(a/b) = 0, so that a/b is an invertible element of A. It follows
that a and b generate the same ideal. Similarly, if v(a) < v(b), b
lies in the ideal generated by a. Hence, we can easily describe all
ideals of A.
Let t be the minimal valuation of a non-invertible element, say
t = v(a). Let b be any other non-invertible element, and t 0 be
the remainder of v(b) by t, so that v(b) = q ∗ t + t 0 . If t 0 6= 0, we
can produce a ring element of valuation strictly between 0 and t,
namely b/aq . It follows that t 0 = 0, hence v(b) is a multiple of
v(a).
By rescaling, we can assume that t = 1. The ideal Ik contains
aq for q > k. It follows that these ideals are all distinct, and in
fact Ik = (ak ) = Mk .
This tells us that all ideals are powers of the maximal ideals,
and all elements are powers of a single generator, up to invertible
elements. We summarize the discussion so far.

Proposition 7.3.1. Let A be a discrete valuation ring. Then A is local

with a principal maximal ideal M, and all its nonzero ideals are powers
of M.

We can characterize discrete valuation rings via the following

result.

232
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 7.3 discrete valuation rings

Proposition 7.3.2. Let A be a local integral domain with maximal ideal

M 6= 0. Assume that A is Noetherian and integrally closed, and that
M is the only nonzero prime ideal. Then A is a discrete valuation ring.

Proof. By Theorem 5.1.19, A is a Dedekind ring. Since M is the

only nonzero prime, all nonzero ideals are powers of M, and
these powers are distinct by unique factorization.
We can then define a valuation on A as follows. Let a ∈ A
be any element; if (a) = Mk , define v(a) = k. It is easy to
check that v is a well-defined function, and we can extend it to
the field of fractions of A by the rule v(a/b) = v(a) − v(b). The
extension becomes a discrete valuation. To check that A is the
valuation ring of v, assume that v(a/b) > 0. Then v(a) > v(b)
hence (a) = Mk , (b) = Mh for some k > h. We conclude that
a/b ∈ A, hence A is the valuation ring of v.

We can reformulate this result by saying that every local Dede-

kind ring is a discrete valuation ring. There is also a converse:

Proposition 7.3.3. A discrete valuation ring is a local Dedekind ring.

Proof. A discrete valuation ring is integrally closed by 7.2.8. More-

over, it is Noetherian and has a single nonzero prime ideal, both
by 7.3.1.

Remark 7.3.4. The localization of a Dedekind ring is again a De-

dekind ring. This is because being Noetherian or being integrally
closed are properties that are preserved under localization (2.2.12
and 5.1.18) and because of the correspondence between prime
ideals in a localization. Thus, the above can be rephrased by say-
ing that a localization of a Dedekind ring is a discrete valuation
ring.
The valuation can also be expressed explicitly. Namely, if P
is a prime ideal of a Dedekind ring A, for every element a ∈ A
we can factorize the ideal (a) = Pt · P1t1 · · · Pktk . The valuation is
obtained by setting v(a) = t and extending by the valuation rule,
hence it essentially encodes the multiplicity of the factor P at a.

233
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 metric and topological methods

7.4 direct and inverse limits

In Section 7.1 we gave the definition of the completion of an inte-
gral domain with respect to an absolute value on its fraction field.
We want generalize this construction to other rings, as well as ex-
press it in algebraic terms. To do so, we introduce the machinery
of limits.

Definition 7.4.1. Let I 6= ∅ be a set with a partial order 6. I is

said to be directed if for any i, j ∈ I we find k ∈ I such that i 6 k
and j 6 k.

Definition 7.4.2. Let {Ai } be a family of groups (or modules over

a fixed ring, or rings) over a directed set of indices I. For every
pair i < j assume given a homomorphism {fij : Ai → Aj }, and
assume that these are compatible, in the sense that for every i <
j < k the diagram

fij
Ai → Aj

fi
k → f jk
←
Ak

commutes. This datum is called a direct system.

Definition 7.4.3. Let {Ai }i∈I , {fij } be a direct system and let A be
a group (module, ring) with given homomorphisms gi : Ai → A,
which are compatible, in the sense that for every i < j the diagram

fij
Ai → Aj

gi
→ gj
A ←
commutes.
A is called the direct limit of the system – denoted lim Ai – if it
−→
enjoys the following universal property. For every other group
(module, ring) B, equipped with compatible homomorphisms
hi : Ai → B, there exists a unique homomorphism h : A → B
such that hi = h ◦ gi for all i.

234
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 7.4 direct and inverse limits

Remark 7.4.4. The direct limit of a system of groups (or modules

or rings) is unique (up to a uniquely determined homomorphism)
due to its universal property. In fact, given two direct limits A
and A 0 (with their homomorphisms), the universal property de-
termines two homomorpshisms A → A 0 and A 0 → A, and again
it follows from the uniqueness requirement of the universal prop-
erty that these morphisms are inverse to each other.

Let us consider how to construct the direct limits of a system

of groups. Assuming it exists, for every element x in some Ai ,
we have its image gi (x) ∈ A. Moreover, if we have fij : Ai → Aj ,
the image gi (x) has to be the same as gj (fij (x)).
Since there are no other conditions, the idea is to construct
the universal object by just putting together these requirements.
Namely, we start from the disjoint union U of all Ai , and we
introduce the equivalence relation ∼ generated by the following:
elements ai ∈ Ai and aj ∈ Aj are equivalent if aj = fij (ai ).
The resulting quotient is equipped with a group operation.
Namely, let ai ∈ Ai and aj ∈ Aj be any two elements of U. Then,
we can find k bigger than both i and j. The images fik (ai ) ∼ ai
and fjk (aj ) ∼ aj belong to the same group Ak , so we can use the
operation there to combine them and get a new element of U.
It is easy to check that this gives a well-defined operation on
the quotient A := U/ ∼ that satisfies the group axioms. If more-
over the Ai have additional structure, such as being modules over
a ring R or rings themselves, the same construction gives to A the
structure of a module or ring.
The natural inclusion followed by the quotient gives maps Ai →
A, and the universal property is easily checked.

Example 7.4.5. (a) All this may seem very abstract, but in fact
it is something that every child learns to do when summing
fractions. Fractions with a fixed denominator d are easy to
sum: they form an additive group Ad which is isomorphic to
Z – for instance 17 + 37 = 47 .
There are homomorphism Aa → Ab given by multiplication
by k whenever b = k · a. Hence we have a direct system
where the index set is N+ and the ordering relation is divisi-
bility.

235
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 metric and topological methods

In order to sum fractions with different denominators, one

has to consider the equivalence relation induced by these
maps. To sum, say, 37 and 25 , one has to consider the equiv-
alent elements 15 14 29
35 and 35 of A35 to get the result 35 . This
is exactly the construction we have outlined above, which in
particular shows that Q is the direct limit of the system of the
Ad .
(b) When the maps of the system are inclusions, the direct limit
is just the union of the Ai . For an example of this, consider
the construction of the algebraic closure of a field k. Finite
extensions can be constructed explicitly as quotients of the
form k[x]/(f), where f is some irreducible polynomial. To get
the algebraic closure, one has to take the direct limit of the
system of finite extensions.
The above examples should make the notion of direct limit
more natural. The notion of inverse limit is obtained by essen-
tially reversing all the arrows. Unfortunately, examples of inverse
limits are not as natural.
Definition 7.4.6. Let {Ai } be a family of groups (or modules over
a fixed ring, or rings) over a directed set of indices I. For every
pair i < j assume given a homomorphism {fij : Aj → Ai }, and
assume that these are compatible. This datum is called an inverse
system.
Definition 7.4.7. Let {Ai }i∈I , {fij } be an inverse system and let A
be a group (module, ring) with given homomorphisms gi : A →
Ai , which are compatible, in the sense that for every i < j the
diagram
A
g j gi

fij →
←
Aj → Ai
is commutative.
A is called the inverse limit of the system – denoted lim Ai
←−
– if it enjoys the following universal property. For every other
group (module over R, ring) B, equipped with compatible homo-
morphisms hi : B → Ai , there exists a unique homomorphism
h : B → A such that hi = gi ◦ h for all i.

236
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 7.4 direct and inverse limits

As in the case of direct limits, there is a form of uniqueness, up

to a uniquely determined isomorphism. To show the existence,
though, we must follow a different way, since the arrows go in the
reverse direction, and there is no way to push elements ai ∈ Ai
to the inverse limit.
Instead, assuming A = lim Ai exists, for each a ∈ A we get a
←−
compatible system of elements ai = gi (a) ∈ Ai . Compatibility
here means that fij (aj ) = ai for i < j.
The idea is to define the inverse limit A as the set of compatible,
or coherent, system of elements {ai ∈ Ai }. The group operation
can be applied separately on each component, and it is easy to
check that this gives a groups structure on A. When the Ai are
modules over a ring, or ring themselves, the inverse limit inherits
the same structure, again by applying operations component by
component.
Given a group B equipped with compatible homomorphisms
hi : B → Ai , for any b ∈ B we get a compatible sequence (gi (b)).
This defines a homomorphism h : B → A such that hi = gi ◦ h
for all i. This is enough to show that A is the inverse limit of the
Ai , as desired.

Example 7.4.8. The most prominent example of inverse limit is

the construction of p-adic numbers. Recall from Section 7.1 that
these can be constructed by considering the p-adic absolute value
on Q and taking the completion Qp . The closure Zp of Z is the
ring of p-adic integers – each such number a ∈ Zp has a series
expansion

∞
X
a= αi pi ,
i=0

which converges in the p-adic topology.

There is a homomorphism Zp → Z/(pn ) defined by sending
P
a to the finite sum n−1 i
i=0 αi p . These homomorphisms make
Zp into the inverse limit of the system of rings {Z/(pn )}n∈N ,
equipped with the natural projections Z/(pn+1 ) → Z/(pn ).

237
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 metric and topological methods

This is easily proved by making use of the explicit description

of the inverse limit. Namely, given a compatible sequence {an ∈
Z/(pn )}, we can write

X
n−1
an = αi,n pi ,
i=0

for some coefficients αi,n that are determined uniquely modulo p.

The compatibility condition ensures that the coefficients αi,n =
αi are in fact independent from n. This allows us to define the
element
∞
X
a= αi pi ∈ Zp .
i=0
This gives a homomorphism from the inverse limit of the sys-
tem to Zp . The inverse homomorphism is guaranteed by the
universal property.
A similar construction shows that for a Dedekind ring A and
a prime ideal P, the ring of P-adic integers AP is the inverse
limit of the system of rings {A/Pi }i∈N , equipped with the natural
projections.
In the next section we are going to generalize this construction
and define the completion of a ring in purely algebraic terms.
Proposition 7.4.9. Let {Ai }, {Bi } and {Ci } be inverse systems over
the same index set I and assume we have exact sequences 0 → Ai →
Bi → Ci for every i ∈ I, compatible with the system maps (a left-exact
sequence of inverse systems). Then we get an exact sequence

0 → lim Ai → lim Bi → lim Ci

←− ←− ←−
of the inverse limits.
If moreover we have a short exact sequence 0 → Ai → Bi → Ci → 0,
the index set is N, and the maps fij for Ai are surjective, the sequence

0 → lim Ai → lim Bi → lim Ci →0

←− ←− ←−
is exact.
Proof. The maps lim Ai → lim Bi and lim Bi → lim Ci are defined
←− ←− ←− ←−
by the universal property, and in fact are just restrictions of the

238
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 7.5 completion of rings and modules

Q Q Q Q
product maps Ai → Bi and Bi → Ci . This already
makes clear that lim Ai → lim Bi is injective, since the product
←− ←−
map is. Moreover, it also implies that the composition lim Ai →
←−
lim Ci is 0.
←−
Let (bi ) be a compatible set that maps Q to 0 in lim C . We can
←− i
then lift the set uniquely to a set (ai ) ∈ Ai , so we only need to
show that this lifted set is itself compatible. To do so, take i > j,
so that we have a map fij : Ai → Aj . The equality fij (ai ) = aj
can be checked by taking the images in Bj , since Aj → Bj is
injective, but there it follows because (bi ) is a compatible system.
For the last assertion, assume exactness at Ci and that maps
Ai → Aj Q are surjective. Given compatible set (ci ), we can lift it
to (bi ) ∈ Bi as above. The elements bi are determined up to
the image of some ai ∈ Ai . Call αi : Ai → Bi the given map. We
need to find some collection ai ∈ Ai such that bi0 := bi + αi (ai )
is a compatible system. Calling gij the maps for Bi , this amounts
to saying that

gij (bj + αj (aj )) = bi + αi (ai ),

or equivalently

gij (bj ) − bi = αi (fij (aj ) − ai ).

The left side is in the image of αi by exactness, call it αi (ai0 ),

hence we need to solve

fij (aj ) = ai0 + ai .

If fij is surjective and the index set is N, this can be solved

inductively.

7.5 completion of rings and modules

In this section, we reformulate the notion of completing a field
(or a subring of it) with respect to an absolute value in terms of
inverse limits. In doing so we will obtain a notion that is useful
in a more general setting.
For a start, we notice that the operation of constructing Cauchy
sequences only requires that we are able to take the difference of

239
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 metric and topological methods

two elements and tell whether an element is close to 0. The fol-

lowing is thus the natural setting to define Cauchy completions:
Definition 7.5.1. Let G be a group endowed with a structure of
topological space. We say that G is a topological group if the group
operations are continous.
In the following, we will only work with abelian topological
groups and use additive notation.
Remark 7.5.2. Since translations in the group are continuous, the
topology is determined by the neighborhoods of 0, and in fact G
is Hausdorff if, and only if, 0 is closed in G.
In general the closure of 0 is a subgroup of G (why?), and the
quotient of G by this subgroup, with the induced topology, is
Hausdorff.
Assuming that G is first countable, that is, each point has a
countable fundamental system of neighborhoods, we can define
convergence in terms of sequences as usual.
Definition 7.5.3. A sequence (gn ) ⊂ G converges to g ∈ G if
for every neighborhood U of g there exists N such that gn ∈ U
for all n > N. A sequence (gn ) is called Cauchy if for every
neighborhood U of 0 there exists N such that gm − gn ∈ U for all
m, n > N.
It is easily seen that each convergent sequence is Cauchy – if
the converse holds, the group is called complete. The methods of
7.1 are easily translated in this more general setting, allowing us
to construct the completion of a topological group G.
Definition 7.5.4. Let G be an abelian, first countable, topological
group. Let Gb be the set of Cauchy sequences in G modulo the
equivalence relation ∼, where we say that (gn ) ∼ (hn ) if (gn −
hn ) converges to 0. Then G
b has a natural structure of topological
group and is called the completion of G.
Remark 7.5.5. There is a natural map G → G b that sends g to the
constant sequence gn = g, but unlike the case of metric spaces,
this map is not necessarily injective. The kernel is the set of el-
ements equivalent to 0, that is, the intersection of all neighbor-
hoods of 0. In particular this map is injective if, and only if, G is
Hausdorff.

240
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 7.5 completion of rings and modules

This is a simple generalization of something we have already

seen: the main twist of this section is to express this is as an
inverse limit.
Proposition 7.5.6. Let G be an abelian topological group, and assume
that 0 inG has a fundamental system of neighborhoods Gn which are
subgroups. Then G b is the inverse limit of the system G/Gn .

Proof. Assume, as we may, that the groups Gn are in fact nested:

G ⊃ G1 ⊃ G2 ⊃ · · · ⊃ Gn ⊃ · · · . The inverse limit of the
inverse system given by G/Gn is explicitly constructed as the set
of coherent sequences xn ∈ G/Gn . If we lift xn to some gn ∈ G,
the coherence condition tells us exactly that the sequence (gn )
is Cauchy. The different choices in lifting xn differ exactly by a
convergent sequence.
Remark 7.5.7. This reconciles the two different constructions of
p-adic numbers that we have seen in 7.1.11 and 7.4.8. Not all
examples of completions are inverse limits, though. In particular
the condition that 0 has a fundamental system of neighborhoods
that are subgroups rules out the case of archimedean absolute
values.
Remark 7.5.8. This proposition opens the way to studying filtra-
tions of groups topologically. Given a filtration on any group G,
say G ⊃ G1 ⊃ · · · ⊃ Gn ⊃ · · ·, we can endow G with the topology
where neighborhoods of 0 are generated by the Gn , and then link
its completion to the inverse limit of the quotients G/Gn .
By Proposition 7.4.9 we immediately obtain
Proposition 7.5.9. Let 0 → H → G → G/H → 0 be an exact se-
quence of topological abelian groups, and assume G has the topology
generated by a filtration {Gn }. If we give H and G/H the induced
topologies, then we have an exact sequence of completions

0 →H
b →G
b [
→ G/H →0

Corollary 7.5.10. Completion is idempotent, that is, G

b=
b ∼ G.
b

Proof. For the inclusion Gn ,→ G we get

0 →G
cn →G
b \n
→ G/G → 0.

241
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 metric and topological methods

But the topology on G/Gn is discrete, so G/Gn equals its com-

pletion, and we deduce

G ∼ G b
= .
Gn G
cn

The thesis follows by passing to inverse limits.

In the rest of the section we will specialize to the case where
G = M is a filtered A-module.
Definition 7.5.11. Let I be an ideal of the ring A, M an A-module
with a filtration M ⊃ M1 ⊃ · · · ⊃ Mn ⊃ · · ·. We say that this is
an I-filtration if I · Mn ⊂ Mn+1 . The I-filtration is called stable if
I · Mn = Mn+1 for n large enough.
Remark 7.5.12. Given two different filtrations {Mi } and {Mi0 },
both of which are I-stable, we have some N such that MN+i ⊂
Mi0 , and vice versa. We say that they have bounded difference. In
particular, they determine the same topology on M. In the partic-
ular case where Mn = In · M, this is called the I-adic topology.
Remark 7.5.13. The most obvious example of a stable I-filtration
is of course Mn = In · M, but things are easier if we keep a little
more generality.
Recall from Section 1.7 that associated to a filtration

M ⊃ M1 ⊃ · · · ⊃ Mn ⊃ · · ·

we have a graded module Gr(M). If this is an I-filtration, Gr(M)

is in fact a module over the graded ring GrI (A).
A related but different construction arises by taking the graded
ring
BI (A) = ∞ n
L
n=0 I
and graded module
L∞
B(M) = n=0 Mn .

In general, these behave worse than the usual associated graded

ring and module, where we take quotients on each homogeneous
component. Still, they are useful for the following result, which
is the cornerstone of the theory that we are going to develop next.

242
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 7.5 completion of rings and modules

Proposition 7.5.14. Let A be a Noetherian ring, M a finitely generated

A-module with an I-filtration {Mn }. Then {Mn } is stable if, and only
if, B(M) is finitely generated over BI (A).
Proof. First, notice that BI (A) is generated by I as an A-algebra.
Since A is Noetherian, I is finitely generated, so BI (A) is itself
Noetherian by Hilbert’sLn basis theorem 2.3.3.
Consider the sum i=0 Mi ⊂ B(M). This is not an BI (A)-
submodule, because it is not closed under multiplication by I.
The generated submodule includes all terms of the form Ik · Mn ,
hence it is
B(M)n := M0 ⊕ · · · Mn ⊕ I · Mn ⊕ I2 · Mn ⊕ · · ·
The union of all B(M)n is B(M). If the chain {B(M)n } stabi-
lizes, B(M) equals a member of this chain, hence it is finitely
generated. Vice versa, if B(M) is finitely generated over BI (A), it
is Noetherian, so this chain must stabilize.
But it is clear from the definition that the chain stabilizes if,
and only if, the filtration {Mn } is I-stable.
Corollary 7.5.15 (Artin-Rees lemma). Let A be a Noetherian ring,
I ⊂ A an ideal and M ⊃ M1 ⊃ · · · ⊃ Mn ⊃ · · · a stable I-filtration
of the A-module M. If M is finitely generated, for each submodule
N ⊂ M, the {N ∩ Mn } form a stable I-filtration of N.
Proof. It is clear that {N ∩ Mn } is an I-filtration, and stability fol-
lows from 7.5.14 and the fact that B(N) ⊂ B(M), hence it is Noe-
therian if B(M) is.
Remark 7.5.16. As a particular case, taking Mn = In · M, we
obtain that N ∩ (In · M) is stable. In particular, it has bounded
difference from In · N, and it induces the I-topology on N.
Corollary 7.5.17. Let A be a Noetherian ring, I ⊂ A an ideal and M1 ,
M2 , M3 finitely generated modules over A. If
0 → M1 → M2 → M3 →0
is exact, then
0 →M
d1 →M
d2 →M
d3 →0
is exact, where M
di is the completion of Mi with respect to the I-adic
topology.

243
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 metric and topological methods

Proof. This is just a rephrasing of 7.5.9 combined with the above

remark that the I-adic topology on M1 agrees with the topology
induced by the inclusion in M2 .

Corollary 7.5.18. Let A be a Noetherian ring, I ⊂ A an ideal. Then,

for all n > 0, the extension In · A
b is the completion of In , and moreover
In /In+1 = ∼ Icn /I[
n+1 .

Proof. This is just the previous corollary in the case where M2 =

A and M1 = In . In this case, M3 = A/In has the discrete topol-
ogy, hence Md3 = ∼ M3 and the thesis follows by a little diagram
chasing. The second assertion follows by considering M1 = In+1
inside M2 = In .

We are now in a position to understand the kernel of the map

c By construction, this is the intersection T Ik · M, but
M → M. k
we have a more explicit description.

Theorem 7.5.19. Let A be a Noetherian ring, I ⊂ A an ideal and M

a finitely generated module over A, endowed with the I-adic topology.
The kernel of M → M c is

{m ∈ M | (1 − x) · m = 0 for some x ∈ I}.

Proof. Let K = k Ik · M be the kernel of M → M.

T c This is the
intersection of all neighborhoods of 0, hence it has the discrete
topology. By the Artin-Rees lemma 7.5.15, this is the I-adic topol-
ogy on K, so it follows that I · K = K. By 5.1.8 we find x ∈ I such
that (1 − x) · K = 0.
For the converse, assumeT(1 − x) · m = 0, that is, m = xm. Then
m = xk m for all k, so m ∈ k Ik · M = K.

Remark 7.5.20. The identity I · K = K may seem trivial, but in

fact it is not, since products and intersections do not commute!
This is the crucial step where Artin-Rees enters the proof.

Given a Noetherian ring A with an ideal I, let A b be its comple-

tion. For an element x ∈ bI, the series 1 + x + · · · + xn + · · · is well
defined and converges in A. b To see this, consider the truncated
Pk
sums sk := i=0 x . If h > k > N, the difference sk − sh ∈ bIN ,
i

244
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 7.5 completion of rings and modules

so that the sequence {sk } defines an element of the inverse limit

b bIk , which is just the I-completion of A,
lim A/ b and Ab is complete.
←− P∞ i
If we denote this limit by s = i=0 x , it follows that s · (1 −
x) = 1 (why?), so that 1 − x is invertible inside A.
b By 1.1.19, bI is
contained in the Jacobson radical of A. We use this fact to give a
b
few examples of completions.

Example 7.5.21. (a) Let A be a local Artinian ring with maximal

ideal M. Then by 2.4.4, Mk = 0 for k big enough, hence A is
complete with respect to the M-adic topology.

(b) Let I ⊂ A an ideal. Then S = 1 + I is a multiplicative set and

we can take the localization S−1 A. Since every element of S
b there is natural map S−1 A → A.
is invertible inside A, b If A is
Noetherian, Theorem 7.5.19 implies that this map is injective,
so that we can see S−1 A as a subring of the completion.

(c) [Ces] Let M ⊂ A be maximal ideal, and assume that A is

Noetherian. For the M completion A b we know by 7.5.18 that
b b ∼
A/M = A/M is a field, so M is maximal. Moreover we have
b
just seen that it is contained inside J(A),
b so A
b is local.

Moreover, consider the completion of the localization AM .

This is just the inverse limit of quotients AM /(MAM )k . Since
localization is exact, this is just the localization (A/Mk )M .
But A/Mk is already local, so this is in fact A/Mk , and the
limit of these is A.
b

It follows that A
b is the completion of AM as well, and in
particular we have again a natural map AM → A.
b

(d) Let k be a field and A = k[x1 , . . . , xn ], with the ideal M =

{f ∈ A | f(0) = 0}. The completion of A can be identified with
the formal power series ring k[[x1 , . . . , xn ]].
To see this, notice that Ik is the ideal of polynomial vanish-
ing at least of order k in 0, so A/Ik can be identified with
the module of polynomials of degree less than k. Each com-
patible sequence of such polynomials defines a unique power
series, showing that Ab=∼ k[[x1 , . . . , xn ]].

245
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 metric and topological methods

Actually the link between power series rings and completions

allows us to prove that completion preserves the Noetherian prop-
erty.
Theorem 7.5.22. Let A be a Noetherian ring, I ⊂ A an ideal. Then
the completion A
b with respect to the I-adic topology is also Noetherian.

Proof. First, we establish that A[[x]], the ring of power series on

A, is Noetherian. This is the content of Exercise 4 in chapter
2. We will not spoil it completely, but the proof is essentially
the same as Theorem 2.3.3 – just replace the ideal generated by
leading coefficients by the ideal generated by the lowest degree
coefficients. The proof goes on mostly unchanged. By induction –
using the fact that A[[x, y]] is isomorphic to A[[x]][[y]] – it follows
that A[[x1 , . . . , xn ]] is Noetherian for all n.
Now let a1 , . . . , an be generators for I. We show that there is a
well-defined surjective homomorphism

eva : A[[x1 , . . . , xn ]] →A
b

xi → ai
defined by evualuating a power series in a1 , . . . , an , which im-
plies that A
b is Noetherian.
To show that eva is well-defined, take any power series p and
let pk be the polynomial obtained by truncating up to degree k.
Then if h, k > N, we have ph (a) − pk (a) ∈ IN , which shows that
the pk (a) converge to a well-defined element in A. b
To show that eva is surjective, recall that any element b ∈ Ab is
defined by a compatible sequence {bk ∈ A/I }. Choose represen-
k

tatives bk of bk . If we let ck := bk+1 − bk and c0 := b0 , then by

compatibility ck ∈ Ik , and
∞
X
b= ck .
k=0

Recursively write ck as a homogeneous polynomial of degree

k in a1 , . . . , an – plus possibly an element of Ik+1 , which we
incorporate into ck+1 . This gives the desired expression of b as a
power series in a1 , . . . , an , showing that eva is indeed surjective.

246
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 7.6 hensel’s lemma

In the end of the section, we derive a few consequences of

the Artin-Rees lemma, and in particular of the strictly related
Theorem 7.5.19. We try to focus on some applications of the
theory developed so far that do not mention the machinery of
completion.
The first result follows from the fact that for an element x in
the Jacobson radical, 1 + x is invertible.
Corollary 7.5.23. Let A be a Noetherian ring, I ⊂ J(A) an ideal and
M a finitely generated A-module. Then ∩n>0 In · M = 0 – in other
words, the I-adic topology is Hausdorff on M.
Specializing to the case where M = A we get the famous
Corollary 7.5.24 (Krull intersection theorem). Let A be a Noether-
ian ring, I ⊂ J(A) an ideal. Then ∩n>0 In = 0.
Remark 7.5.25. In our terminology, a module M over a ring is
complete with respect to the I topology if every Cauchy sequence
converges. Other authors require the more stringent condition
that the map c : M → M c is an isomorphism. The difference
lies in the fact that the kernel of c may be nontrivial, or in other
words, that M may not be Hausdorff. The above results imply
that the two definitions coincide in many cases of interest.
Another application is the following:
Proposition 7.5.26. Let A be a Noetherian ring, P ⊂ A a prime ideal.
Then the kernel of the localization map A → AP is the intersection of
all P-primary ideals.
Proof. The ring AP is local, hence applying the preceding corol-
lary we see that the intersection of all ideals Pn · AP is 0. This is
the same as the intersection of all P-primary ideals of AP , since
every P-primary ideal sits among P and some power Pk .
Pulling back this to A and using 3.2.24, we get the thesis.

7.6 hensel’s lemma

In this section we study various generalizations of Hensel’s lemma
for p-adics (Theorem 7.1.12) to arbitrary complete local rings.

247
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 metric and topological methods

Since in this case there is not a preferred generator for the maxi-
mal ideal, we cannot translate its proof directly. Rather, we prove
the following more general statement.
Theorem 7.6.1 (Hensel’s lemma, strong form). Let A be a Noether-
ian local ring, complete with respect to its maximal ideal M. Denote
k = A/M. Let f ∈ A[x] be a monic polynomial, f(x) ∈ k[x] its residue
modulo M. Assume f factorizes as f = G · H in k[x], where G and H
are coprime. Then we have a factorization f = g · h in A[X], where g
and h are monic of the same degrees of G and H, and g = G and h = H.
Moreover, g and h are uniquely determined.
The following, more familiar form, follows immediately by tak-
ing G(x) = x − α:
Corollary 7.6.2 (Hensel’s lemma, weak form). Let A be a Noether-
ian local ring, complete with respect to its maximal ideal M. Denote
k = A/M. Let f ∈ A[x] be a monic polynomial, f(x) ∈ k[x] its residue
modulo M. Assume f has a simple root α in k. Then α lifts uniquely
to a root of f in A.
Proof. We show by induction that there exist monic polynomials
gk , hk ∈ A[x] such that
f ≡ gk · h k mod Mk [x],
such that gk = G and hk = H. Moreover, such polynomials
are uniquely determined modulo Mk [x]. The case k = 1 is the
hypothesis of the theorem.
Having produced gk and hk we look for polynomials of the
form
gk+1 = gk + s
hk+1 = hk + t,
where deg s < deg gk and deg t < deg hk .
To find the right s and t, consider the difference
d := f − gk hk ∈ Mk [x].
Using the fact that G and H are coprime, we can write 1 ≡
agk + bhk mod M[x] for suitable a, b ∈ A[x]. Multiplying by d
we almost get what we want:
d ≡ (d · a)gk + (d · b)hk mod Mk+1 [x],

248
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 7.6 hensel’s lemma

so if we let g◦k+1 := gk + d · b and h◦k+1 := hk + d · a we have

g◦k+1 ≡ gk mod Mk [x]

h◦k+1 ≡ hk mod Mk [x]

and

g◦k+1 h◦k+1 ≡ gk · hk + d ≡ f mod Mk+1 [x]. (7.6.1)

The only issue with this choice is that d · a and d · b may have
large degree. To remedy this, use the fact that gk is monic to
perform the division of d · b by gk , so that

d · b = lk · gk + s
d · a = mk · hk + t.

With this choice of s and t, the relation (7.6.1) is valid for gk+1
and hk+1 – that is, gk+1 hk+1 ≡ f mod Mk+1 [x]. Moreover, s
has smaller degree than gk , so that gk+1 is monic, and similarly
for hk+1 .
Uniqueness can be proved inductively, in a similar fashion.
Since by construction gk+1 − gk ∈ Mk [x], and all gk have the
same degree, the coefficients of the {gk } converge in A. This de-
fines a limit polynomial g, monic of the same degree as G. Sym-
metrically, we get a limit polynomial h, monic of the same degree
of H.
The equality f = g · h follows by passing to the limit. More
precisely, the coefficients of f − g · h belong to Mk for all k. But
k M is 0 by Krull intersection theorem 7.5.24.
k
T

Local rings that satisfy the conclusion of Hensel’s lemma are

called Hensel rings, and are not necessarily complete (see [Ray70]
for a detailed treatment, or Exercise 23). The following results
have a flavour similar to Hensel’s lemma, in that they allow to lift
a condition on a module of the form M/I to the I-adic completion
of M.
Proposition 7.6.3. Let A be a Noetherian ring, I ⊂ A an ideal, M, N
two A-modules with a map φ : M → N. If the induced map M/I →
N/I is surjective, then the induced map Mc→N b is surjective, where b·
denotes completion with respect to the I-adic topology.

249
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 metric and topological methods

Proof. By induction, the map M/Ik → N/Ik is surjective for all k.

In fact, assume this holds for k, and take n ∈ N/Ik+1 . Then we
can write
Xr
n = φ(m) + aj nj ,
j=1

for some m ∈ M, aj ∈ Ik and nj ∈ N. Since M/I → N/I is

surjective, we can write nj = φ(mj ) + xj , where mj ∈ m and
xj ∈ I · N, so
 
Xr
n ≡ φ m + aj mj  (mod Ik+1 · N),
j=1

which means that M/Ik+1 → N/Ik+1 is surjective as well.

It follows that there are exact sequences
Lk M N
0 → k
→ → →0
I M Ik M Ik N
where Lk = φ−1 (Ik N). By a similar reasoning as above, the
map Lk+1 /Ik+1 M → Lk /Ik M is surjective (check this!). Using
Proposition 7.4.9, we get an exact sequence of the inverse limits,
c→N
so M b is surjective.

Corollary 7.6.4. Let A be a NoetherianTring, complete with respect to

the I-adic topology, M an A-module. If ∞ n
n=1 I · M = 0 and M/IM
is finitely generated over A/I, then M is finitely generated over A.
Proof. Let m1 , . . . , mk ∈ M such that their classes in M/I gen-
erate M/I as an A/I-module, and let N = hm1 , . . . , mk iA . By
the previous Proposition, the map N b →M c is surjective. On the
k
other hand, N is a quotient of A , hence it is complete (why?).
Since ∞ n
T
n=1 I · M = 0, the map M → M is an inclusion, and we
c
conclude that N = ∼ M= ∼ M.
c

7.7 witt vectors

In the course of this chapter, we have seen two different con-
struction of p-adic numbers. The first one starts with the p-adic

250
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 7.7 witt vectors

absolute value on Q and defines Qp as the Cauchy completion

of Q with respect to this absolute value. The p-adic integers Zp
can be seen either as the topological closure of Z inside Qp , or
alternatively as the integral closure of Z in the same field. The
second approach is more algebraic in nature and constructs the
p-adic integers Zp as the inverse limit of the system of quotients
Z/pk Z, and Qp as the fraction field of Zp . Both points of view
are fruitful, and their generalizations lead to the study of general
absolute values and valuations in one direction, and to I-adic
completions in the other one.
In this section, we are going to introduce a third approach to
study p-adic numbers. The idea is that every a ∈ Zp has an
expansion into digits
X∞
a= αi pi ,
i=0

where the coefficients αi are taken in {0, . . . , p − 1}. One should

be able to define p-adic numbers this way, and perform algebraic
operations using this system of digits.
Predictably, this plan is bound fail due to the issue of carries.
This is very similar to trying to define addition or multiplication
of real numbers using digits: the formulas become very compli-
cated, and keeping track of carries in an organized way is tricky.
The insight of Witt [Wit36] was that one can perform this plan and
find simple, universal formulas, provided one choose a smarter set
of representatives for the digits.
Let π : Zp → Z/pZ be the projection. A choice of “digits” es-
sentially amounts to a section of π, that is a function σ : Z/pZ →
Zp such that π ◦ σ = id. Witt noticed that computations are
simplified if one choose a particular σ. Namely, each element
a ∈ Z/pZ satisfies ap = a, so by Hensel’s lemma there exists
a unique section τ : Z/pZ → Zp such that τ(a)p = τ(a) for all
a ∈ Z/pZ. The section τ is called the Teichmüller character and
τ(a) is called the Teichmüller representative of a. Clearly, τ satis-
fies τ(ab) = τ(a)τ(b). One can find suitable formulas for p-adic
numbers when expressed using Teichmüller representatives.
Instead of pursuing this and building p-adic numbers out of
Z/pZ digits, we are going to define a universal construction that
works over any ring. This will lead us to the definition of p-adic
Witt vectors. Our presentation will follow parts of [Haz09], which

251
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 metric and topological methods

has a much more extensive treatment. Witt vectors satisfy many

universal properties, and accordingly one can construct them in
various different ways. For possible alternative approaches, see
[Rab14] or [Hes05].
The following remark is meant to give some motivation for the
construction of the Witt polynomials. If it is more confusing that
inspiring on a first read, one can safely skip it and come back to
reading it after the rest of the section.

Remark 7.7.1. Fix a prime number p. Given a ring A, we are

going to define another ring Wp (A) together with a projection
π : Wp (A) → A. Mimicking what happens for Zp , we want to be
able to write elements of Wp (A) as infinite sums

∞
X
a= σ(ai )pi ,
i=0

for some ai ∈ A, where σ : A → Wp (A) is a section. This repre-

sentation is going to induce a set bijection

λn : An → Wp (A)/pn Wp (A)

given by
X
n−1
λn (a0 , . . . , an ) := σ(ai )pi .
i=0

Having fixed such a bijection, we get an induced ring structure

on An , and the question becomes how to express addition and
multiplication in such coordinates.
Unfortunately, λn depends on the choice of the section σ, so
it is unlikely that we can find nice universal formulas. On the
other hand, in any ring it is true that if a ≡ b (mod p), then
k k
ap ≡ bp (mod pk+1 ) for all k – this can be seen by expanding
the binomial formula. Using this, it is not difficult to slightly
change the defintion of λ in order to make it independent of σ.
Namely,
X
n−1
n−i
µn (a0 , . . . , an ) := σ(ai )p pi
i=0

252
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 7.7 witt vectors

is independent of the choice of σ modulo pn . Turning this on its

head, in the following we are going to consider the polynomials
n n−1
wn (x0 , . . . , xn ) := xp p
0 + px1 + · · · + pn xn

and see if we can use them to define a ring structure on formal

sums of elements in A.
Guided by the above remark, the point of departure for our
construction is a collection of polynomials wn ∈ Z[x0 , x1 , . . . ]
defined by

w0 (x) = x0
w1 (x) = xp
0 + px1
2
w2 (x) = xp p 2
0 + px1 + p x2
.. (7.7.1)
.
n n−1
wn (x) = xp p
0 + px1 + · · · + pn xn
..
.

The definition seems mysterious at this point – the reason for

the introduction of these polynomials is that one is able to find
universal formulas for addition and multiplication of them. The
polynomials {wn } are called p-adic Witt polynomials.
By construction, they satisfy

wn+1 (x) ≡ wn (xp ) (mod p),

where for notational simplicity we denote x = (x0 , . . . , xn ) and

xp = (xp p
0 , . . . , xn ). This simple observation allows us to prove the
key result:
Lemma 7.7.2. Let f ∈ Z[x, y, z]. Then there are uniquely determined
polynomials

fn ∈ Z[x0 , . . . , xn , y0 , . . . , yn , z0 , . . . , zn ]

such that

f(wn (x), wn (y), wn (z)) = wn (f0 (x, y, z), · · · , fn (x, y, z)). (7.7.2)

253
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 metric and topological methods

Thus, every ternary operation over the polynomials wn is just

another polynomial wn evaluated at other polynomials. There
is nothing special in choosing ternary operations here – it is just
that we will not need to apply the lemma to operations of bigger
arity.
Proof. One can choose f0 (x0 , y0 , z0 ) = f(x0 , y0 , z0 ), and then use
equation (7.7.2) to define fn by induction. This can be done be-
cause fn appears with degree 1 in (7.7.2). The only thing that
needs to be proved is that the polynomials {fn } thus defined have
coefficients in Z, since a priori defining fn requires a division by
pn .
This can be proved by induction using the remark above. For
simplicity of notation, we are going to omit underlines in sets of
variables. Assume all polynomials up to fn are integral. Notice
that by construction
wn+1 (x) ≡ wn (xp ) (mod pn+1 ),
and so
f(wn+1 (x), wn+1 (y), wn+1 (z)) ≡
≡ f(wn (xp ), wn (yp ), wn (zp )) ≡ (7.7.3)
p p p p p p n+1
≡ wn (f0 (x , y , z ), · · · , fn (x , y , z )) (mod p ).
At the same time we can expand
n+1 n
wn+1 (f0 , · · · , fn+1 ) = fp
0 + pfp
1 +···+p
n+1
fn+1 . (7.7.4)
If we expand the last line of (7.7.3), and compare it term by
term with (7.7.4), all terms are congruent modulo pn+1 . There
is a last summand in (7.7.4), which is pn+1 fn+1 , so we get that
pn+1 fn+1 ≡ 0 (mod pn+1 ), which proves that fn+1 has coeffi-
cients in Z.
Choosing the polynomial f(x, y) = x + y, one finds polynomi-
als sn (x0 , . . . , xn , y0 , . . . , xn ) that satisfy
wn (x) + wn (y) = wn (s0 , . . . , sn ).
Similarly, by choosing f(x, y) = xy, one finds another set of poly-
nomials mn (x0 , . . . , xn , y0 , . . . , xn ) that satisfy
wn (x) · wn (y) = wn (m0 , . . . , mn ).

254
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 7.7 witt vectors

Definition 7.7.3. Let A be any ring, p a prime number. The ring

of p-adic Witt vectors over A, denoted Wp (A), is the set

Wp (A) = AN = {(a0 , a1, . . . ) | ai ∈ A},

endowed with the operations defined by

(ai ) + (bi ) := (s0 (a, b), s1 (a, b), . . . )

(ai ) · (bi ) := (m0 (a, b), m1 (a, b), . . . ).

Of course, we have to prove that Wp (A) is a ring, so the oper-

ations are associative, multiplication is distributive over the sum
and so on. This is the reason why we have proved Lemma 7.7.2
with 3 variables. For instance, to prove associativity of the sum,
one considers the polynomial

f(x, y, z) = (x + y) + z = x + (y + z).

Uniqueness in Lemma 7.7.2 implies the identity

sn (x0 , . . . , xn , s0 (y, z), . . . , sn (y, z)) =

sn (s0 (x, y), . . . , sn (x, y), z0 , . . . , zn ),

which amounts to the associativity of the sum in Wp (A). All

other ring identities can be proved in the same way. The zero
element of Wp (A) is just (0, 0, . . . ), while the multiplicative iden-
tity is (1, 0, 0, . . . ). One can explicitly work out the polynomials
involved, but the computations rapidly become messy: for in-
stance,

xp p
0 + y0 − (x0 + y0 )
p
s1 (x0 , x1 , y0 , y1 ) = + x1 + y1 .
p

Remark 7.7.4. There are various relations between a ring and

the associated Witt rings. Namely, a homomorphism A → B of
rings induces a homomorphism Wp (A) → Wp (B). Moreover, by
construction, the evaluation map

wn : Wp (A) →A
(a) → wn (a)

255
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 metric and topological methods

is a ring homomorphism such that the diagram

wn
Wp (A) →A

(7.7.5)
↓ ↓
wn
Wp (B) →B

is commutative for each map A → B. Putting together all the

morphisms wn gives a morphism

w : Wp (A) → AN (7.7.6)

which in general is neither injective nor surjective.

Definition 7.7.5. Given an element a ∈ An , the components
wn (a) are called the ghost components of a.
The rings Wp (A) are quite special: unlike most constructions
that we have met until now (quotients, localizations, polynomial
rings, power series, . . . ) they come endowed with maps Wp (A) →
A which have A as target instead of source. The algebraic struc-
ture on Wp (A) is more or less interesting depending on the char-
acteristic of A.
Remark 7.7.6. Assume that p is invertible in A. Then the homo-
morphism w of (7.7.6) is an isomorphism, since one can recur-
sively solve for xn in wn . In this case, Wp (A) is just another
way to write AN and does not carry any interesting information.
This happens when A is an algebra over Q, or when char(A) is a
prime q 6= p.
This implies that the rings Wp (A) are determined by the fact
that the underlying set is AN , the request that the polynomials
wn are homomorphisms, and that the diagram in (7.7.5) com-
mutes. In fact, this determines both the addition and multipli-
cation on Wp (A) when A is a Q-algebra. Every ring A of char-
acteristic 0 embeds into its fraction field, and commutativity of
(7.7.5) implies that the structure of Wp (A) is determined as well.
Finally, every ring is a quotient of Z[{xi }i∈I ] for a suitable index
set I, and since these rings have characteristic 0, another applica-
tion of (7.7.5) is enough to determine the operations on Wp (A)
for any ring A.

256
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 7.7 witt vectors

One can use computations with ghost components to derive

many relations that hold in Wp (A). We summarize some of these,
but first we need a few definitions.
Definition 7.7.7. Given an element a ∈ A, the element t(a) :=
(a, 0, 0, . . . , ) ∈ Wp (A) is called the Teichmüller representative of a.
It satisfies w0 (t(a)) = a.
The Verschiebung operation is just the shift Vp : Wp (A) → Wp (A)
defined by
Vp ((a0 , a1 , . . . )) = (0, a0 , a1 , . . . ).
Proposition 7.7.8. Let A be a ring, p a prime number.
i) Given a ∈ A and b = (b0 , b1 , . . . ) ∈ Wp (A),
2
t(a) · b = (ab0 , ap b1 , ap b2 , . . . ).
In particular the Teichmüller map is multiplicative, that is, given
a, b ∈ A one has t(ab) = t(a)t(b).
ii) The Verschiebung satisfies w0 (Vp ) = 0 and wn (Vp ) = p · wn−1 .
iii) The Verschiebung is additive, namely for a, b ∈ Vp (A)
wn (Vp (a + b)) = wn (Vp (a)) + wn (Vp (b)).

iv) Every a = (a0 , a1 , . . . ) ∈ Wp can be written as a series

∞
X
a= Vpi (t(ai )).
i=0

v) Define polynomials pn (x) by

wn (p0 , p1 , . . . , pn ) = pwn (x).
Then the {pn } are well-defined and pn (x) ≡ xp
n−1 (mod p).
We must be precise about the meaning of the series in iv). The
ring Wp (A) is endowed with ideals
Ik := {(ai ) ∈ Wp | ai = 0 for i < k} = Vk (Wp (A)).
These ideals define a filtration on Wp (A), and the convergence
in iv) has to be meant in the topology induced by this filtration
(this is just fancy way of saying that the tails in that sum begin
with more and more zeros).

257
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 metric and topological methods

Proof. Item i) is equivalent to stating that

n
mn (x0 , 0, . . . , 0, y0 , . . . , yn ) = xp
0 yn .

This can be proved by induction.

Item ii) is an obvious identity between the Witt polynomials.
Property iii) follows from ii) and the fact that wn−1 is additive.
To prove iv), first notice that if a, b ∈ Wp (A) are such that, for
every i, either ai = 0 or bi = 0, then wn (a + b) = wn (a) + wn (b).
With this, the result is clear since

Vpi (t(ai )) = (0, . . . , 0, ai , 0, . . . , ),

where ai is in position i.
Finally, in v) the polynomials pn are well-defined as in the
proof of Lemma 7.7.2, and in fact, pn can be computed by com-
position of the sum polynomials {si } for i 6 n. The equation
pn (x) ≡ xp
n−1 (mod p) can be proved by induction.

At this point, we are going to assume that char A = p. In this

case, these computations simplify quite a bit.

Corollary 7.7.9. Let A be a ring of characteristic p. Then, for a =

(ai ) ∈ Wp (A),
p · a = (0, ap p
0 , a1 , . . . ).
In particular, char Wp (A) = 0 and p = (0, 1, 0, 0, . . . ) in Wp (A).

Proof. This is just a restatement of item v) of Proposition 7.7.8

Example 7.7.10. Let A = Z/pZ. Then w( x) = xp 0 . In this case,

we can explicitly identify Wp (A) with the ring Zp of p-adic in-
tegers. In fact, by Proposition
P 7.7.8 iv), every element of Wp (A)
can be written as a series ∞ V
i=0 p
i (t(a )). By the above corollary,
i
Vp (a) = p · a for a ∈ Wp (A), so every element of Wp (A) admits
a decomposition
X∞
a= pi · t(ai ).
i=0

Moreover, t satisfies

t(α)p = t(αp ) = t(α).

258
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 7.7 witt vectors

for all α ∈ Z/pZ.

On the other hand, using the Teichmüller character τ defined
at the beginning of the section, every element a ∈ Zp can be
written as a series
X∞
a= pi · τ(ai ).
i=0

This gives a natural bijection between Wp (A) and Z/pZ. It is

not a priori clear that this is an isomorphism, but this will follow
from the next two theorems, which imply that there exists – up to
isomorphism – a unique complete DVR of characteristic 0 having
Z/pZ as a residue field, and that is Wp (Z/pZ).

Corollary 7.7.9 allows us to generalize the construction of p-

adic numbers to any perfect field of characteristic p. Recall that
a field k of characteristic p is called perfect if kp = k. This holds,
for instance, if k is finite or if k is algebraically closed.

Theorem 7.7.11. Let k be a perfect field of characteristic p. Then

the ring Wp (k) is a discrete valuation ring of characteristic 0, with
maximal ideal M = I1 = Vp (Wp (k)). Moreover, k = ∼ Wp (k)/M, M
is generated by p, and Wp (k) is complete with respect to the M-adic
topology.

Proof. Corollary 7.7.9 tells us that char(Wp (k)) = 0. Moreover,

since k is perfect, it shows that I1 is generated by p. The natural
map w0 : Wp (k) → k is surjective, and has I1 for kernel, so I1 is
maximal. If we denote M := I1 , the same corollary implies that
Ik = Mk = (pk ).
By Proposition 7.7.8 iv), Wp (A) is complete in the M-adic topol-
ogy. Every element a ∈ Wp (k) with a0 6= 0 is invertible, since
M is maximal. It follows that the only ideals of Wp (k) are the
powers of M, so Wp (k) is a DVR.

Not only the Witt construction allows us to produce a DVR

out of any perfect field k of positive characteristic – the ring thus
constructed enjoys a universal property among all rings having k
as a quotient.

Theorem 7.7.12. Let k be a perfect field of characteristic p, π : A → k

be any surjective ring map, M := ker π. Assume that A is complete in

259
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 metric and topological methods

the M-adic topology and that ∞ i

T
i=0 M = 0. Then there is a unique
map f : Wp (k) → A that makes the diagram
f
Wp (k) →A
w
0 π
→ ↓
k
commutative.
Proof. To prove the existence of f, we define maps
fn : Wp (k)/Mn → A/Mn
and then try to patch them together. First notice that inside A we
have p ∈ M, as char(k) = p. By definition of the Witt polynomi-
als, it follows that given a0 , . . . , an ∈ M we have wn (a0 , . . . , an ) ∈
Mn+1 . Hence, there is a commutative diagram
wn
Wp (A) →A

↓ ↓
gn
Wp (k) → A/Mn+1
where the left hand map is induced by π. The map gn sends M
into M/Mn+1 , hence it induces the desired homomorphism fn .
The construction of this diagram shows that the system of maps
{fn } is compatible, in the sense that the diagram
Wp (k) fn+1 A
→ n+1
Mn+1 M

↓ ↓
Wp (k) fn A
→ n
Mn M
commutes. Hence we can define f(a) as the inverse limit of the
elements fn (a), which exists and is unique since A is complete
and Hausdorff.
To prove uniqueness, notice that any map f as in the thesis
must send M to M, hence Mn to Mn for all n. This implies that
f is the inverse limit of the maps fn defined above.

260
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 7.7 witt vectors

From the two theorems together, it follows:

Corollary 7.7.13. Let k be a perfect field of characteristic p. Then there

exist a complete DVR A with maximal ideal M having k as residue field.
Moreover, A is unique up to a unique isomorphism.

There is a fairly straightforward generalization of p-adic Witt

vectors. The ring Wp (A) encodes information about characteris-
tic p phenomena, but one can put together this information for
various primes at once. To do so, we are going to introduce some
notation, that is in slight conflict with the notation that we have
used so far.

Definition 7.7.14. Given n ∈ Z+ , we define the Witt polynomial

X n
wn (x) = dxdd .
d|n

Notice that wn is a polynomial involving the variables xd for

d|n. When n = pm , we recover the familiar Witt polynomials
that we previously denoted wm , save for the renaming that uses
the variable xpk in place of xk .
We can develop the theory in this more general setting, al-
though we are going to allow ourserlves to proceed a little faster.

Lemma 7.7.15. Let f ∈ Z[x, y, z]. Then there are uniquely determined
polynomials

fn ∈ Z[x0 , . . . , xn , y0 , . . . , yn , z0 , . . . , zn ]

such that

f(wn (x), wn (y), wn (z)) = wn (f1 (x, y, z), · · · , fn (x, y, z)).

Proof. As in the proof of Lemma 7.7.2, we use the equation to

define fn , and the only thing to prove is that fn has integral
coefficients. This is done by induction: for each prime p dividing
n, we can use the inductive hypothesis on n/p to show that the
coefficients of fn , which a priori are in Q, have a nonnegative
p-adic valuation. Since this is true for all p that divide n, fn is in
fact integral.

261
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 metric and topological methods

Using the lemma, one can define sum polynomials sn (x, y) that
satisfy
wn (x) + wn (y) = wn (s1 , . . . , sn ).
and multiplication polynomials mn (x, y) that satisfy

wn (x) · wn (y) = wn (m1 , . . . , mn ).

We say that a subset S ⊂ Z+ is divisor-stable if whenever n ∈ S

and m divides n, m ∈ S as well. Given a divisor-stable set S and
a ring A, we can define the ring of Witt vectors WS (A) as the set
AS = {(ai )i∈S | ai ∈ A}, endowed with the operations defined
by

(ai ) + (bi ) := (si (a, b))

(ai ) · (bi ) := (mi (a, b)).

Using Lemma 7.7.15, we get that these operations give WS (A)

a ring structure. The ring WS (A) comes with homomorphisms
wn : WS → A for all n ∈ S, and putting them together gives a
homomorphism
wS : WS → AS .

Remark 7.7.16. As we have already noticed, in this more gen-

eral setting we have switched to a different notation. If we take
S = {pk } for a fixed prime p, we recover the ring that we had pre-
viously denoted Wp (A), although the components are indexed
(a1 , ap , ap2 , . . . ) in place of (a0 , a1 , a2 , . . . ). One can also choose
the set S = {pk }k6n – the ring WS (A) thus obtained is just
Wp (A)/pn+1 Wp (A).

When taking S = Z+ , we get the so-called ring of big Witt

vectors.

Remark 7.7.17. If T ⊂ S are two divisor-stable sets, there is a

surjective homomorphism WS (A) → WT (A) which just forgets
all coordinates that are not in T . In this way, the ring of big Witt
vectors has all other rings of Witt vectors as quotients.

Just as in Remark 7.7.6, the homomorphism wS : WS → AS

∼ AS .
is bijective when A is a Q-algebra. In this case, WS (A) =

262
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 7.8 exercises

Now if A → B is a ring map, there is an induced homomorphism

WS (A) → WS (B), such that the diagram

wS
WS (A) →A

↓ ↓
wS
WS (B) →B

commutes. This is enough to determine the structure of WS (A)

for all rings A. In fact, every ring A of characteristic 0 embeds
into the Q-algebra F(A), which implies that WS (A) is a a subring
of WS (F(A)) formed by vectors with coefficients in A. Finally,
every ring A is a quotient of a ring of characteristic 0, and com-
mutativity of the diagram implies that this is enough to fix the
ring structure of WS (A).

7.8 exercises
1. Prove that a Noetherian valuation ring is either a field or a
discrete valuation ring.
2. Prove that the family of arithmetic progressions is a basis for a
topology on Z which makes Z into a topological ring. Prove that
arithmetic progressions are both open and closed, and deduce
that Z has infinitely many primes (otherwise {−1, 1} would be
open).
3. Give an example of a ring A, an ideal I, and an A-module M
with two I-filtrations that induce the same topology on M but do
not have bounded difference.
4. Show that the various p-adic norms on Q for different primes
are not equivalent.
5. Give an alternative proof of Ostrowski’s theorem 7.1.15 that
does not use completions, at least for the nonarchimedean case.
Namely, given a nonarchimedean absolute value | · | on the num-
ber field k, show that x → log |x| is a discrete valuation, and use
the results of Section 7.3 to show that | · | is a P-adic absolute
value.

263
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 metric and topological methods

6. Given a point x ∈ Qp and r > 0, define the disc

D(x, r) := {y ∈ Qp | |x − y| < r}.

Show that every point of a disc is a center – that is, if y ∈ D(x, r),
then D(y, r) = D(x, r).
7. Prove that the p-adic expansion (7.1.3) of a number a ∈ Zp is
eventually periodic (that is, αi+t = αi for some fixed t and i big
enough) if and only if a ∈ Q ∩ Zp = Z(p) .
8. Show that in Z5 there is a third root of 3.
9. Let p ∈ Z be a prime and n not divisible by p. Then any
element a ∈ Z which is congruent to 1 modulo p is a n-th root
in Zp .
10. Prove the following stronger version of Hensel’s lemma. Let
f ∈ Zp [x] be a polynomial, a ∈ Zp such that
2
f 0 (a) p
< f 0 (a) p
.

Then there is a unique root α ∈ Zp of f such that |α − a|p <

|f 0 (a)|p .
11. Show that there is a unique extension of the norm |·|p to the
algebraic closure Qp of Qp . On the other hand, for each finite
extension K of Q, there is an extension of |·|p to K for each prime
Qi of OK over p. Why this is not a contradiction?
12. In Remark 7.5.20, we state that products do not commute with
intersections (even finite ones). Motivate this remark, by finding
an explicit example of a ring A with ideals I1 , I2 and I3 such that
I1 · (I2 ∩ I3 ) 6= I1 · I2 ∩ I1 · I3 .
13. [Car] In the ring

Q[x, z, y1 , y2 , . . . ]
A= ,
(x − zy1 , x − z2 y2 , . . . )

consider the ideal I = (z). Show that ∞ k

T
k=1 I = (x) and that
z · (x) 6= (x). This provides a counterexample to Krull intersection
theorem when the ring is not Noetherian.

264
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 7.8 exercises

The following exercises (up to Exercise 20) develop the begin-

ning of the theory of heights on number fields.

14. Let K be a number field. For every equivalence class of abso-

lute values on K, choose a representative normalized as follows.
If v is archimedean, take the restriction of | · |st for a suitable
embedding K ,→ C. Otherwise, choose the normalization of Ex-
ample 7.1.4e. Define (v) = 1 unless v is the restriction of | · |st for
an embedding of K in C such that K 6⊂ R, in which case (v) = 2.
Show that for any x ∈ K we have |x|v = 1 except for a finite
number of v, and that
Y (v)
|x|v = 1.
v

This is know as the product formula for absolute values. (Hint:

show the result on Q first.)

15. Let K be a number field, and normalize absolute values on K

as in Exercise 14. Given x ∈ K define its height
Y
HK (x) := max{|x|v , 1}.
v

Show that HK is well-defined, and that for a reduced rational

number a/b it reduces to HQ (a/b) = max{|a|st , |b|st }.

16. Let HK be the height on a field K, as defined in the previous

exercise. Show that if K ⊂ L is an inclusion of number fields, we
have
HL (x) = HK (x)[L:K]
for all x ∈ K. Hence given an algebraic number x, we can define
the absolute height
1
H(x) := HK (x) [K:Q]
for any number field K containing x.

17. Let H be the absolute height defined in the previous exer-

cise. Show that for an algebraic number x we have the identities
H(xk ) = H(x)k for all k ∈ N. Moreover, if y is another alge-
braic number, show the two inequalities H(xy) 6 H(x)H(y) and
H(x + y) 6 2H(x)H(y).

265
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 metric and topological methods

18. Given a polynomial with algebraic coefficients

f(x) = an xn + · · · + a1 x + a0 ,

define its height by

!1
Y d
H(f) = max{|a0 |v , · · · , |an |v } ,
v

where the product runs over all absolute values of the field K :=
Q(a0 , · · · , an ), and d = [K : Q].
Assume that f factorizes as

f(x) = an (x − α1 ) · · · (x − αn ).

Show that
Y
n
H(f) 6 2n−1 H(αi ).
i=1

19. Use the previous exercise to prove the following theorem of

Northcott: for any choice of constants A, B, the set

{x | H(x) 6 A, [Q(x) : Q] 6 B}

is finite. In particular, every number field contains a finite num-

ber of elements of bounded height. (The coefficients of the mini-
mal polynomial over Q of an element of this set are bounded.)
20. Prove the following theorem of Kronecker: an algebraic num-
ber x satisfies H(x) = 1 if, and only if, it is a root of unity.
21. Let A be a ring, I ⊂ A an ideal and M an A-module. Consider
the completion with respect to the I-adic topology. Show that
b ⊗A M → M.
there is a well-defined map t : A c
Assuming that M is finitely generated, show that t is surjective.
(Write M as a quotient of a free module.)
22. Continuing the previous exercise, assume that M is finitely
generated and A is Noetherian. Show that the map t is an iso-
morphism.
23. Let A = R{{x}} be the ring of convergent power series around
0. Show that A is not complete, but Hensel’s lemma holds for A.

266
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 7.8 exercises

To better understand ideals in valuation rings, we give some

definitions. Let G be an ordered group. A subset S ⊂ G is called
a segment if for any g ∈ S, the set {h ∈ G | −g 6 h 6 g} is
contained in S. A subgroup which is a segment is called isolated.
Given a field k and a valuation v : k∗ → G, we can always assume
that v is surjective, and then we say that G is the value group of v.

24. Let A be a valuation ring with value group G, and call v the
valuation. Show that for any ideal I, the set

GI := v(A \ I) ∪ −v(A \ I)

is a segment of G, and that this induces a bijection between ideals

of A and segments of G. Moreover, this correspondence restricts
to a bijection between prime ideals of A and isolated subgroups
of G.

25. Show that the set of isolated subgroups in an ordered group

is totally ordered with respect to inclusion. Use the previous
exercise to prove that the set of prime ideals in a valuation ring
is totally ordered with respect to inclusion.

26. Given an ordered group G, construct a field k with a valua-

tion v having value group G.

27. Find a valuation on k(x1 , . . . , xn ) having value group Zn

with the lexicographic order, and describe its valuation ring.

28. Let A be a complete DVR, M its maximal ideal, and let

∼ k[[x]], where the
k = A/M. If k has characteristic 0, then A =
isomorphism preserves the valuation.

29. Let {wn } be the p-adic Witt polynomials. Show that there is
a sequence of integral polynomials {fn (x0 , . . . , xn+1 )} such that

wn (f0 , . . . , fn ) = wn+1 .

Show that for each ring A, the map f : Wp (A) → Wp (A) defined
by
f(a0 , a1 , . . . ) = (f0 (a0 , a1 ), f1 (a0 , a1 , a2 ), . . . )
is a homomorphism. The map f is called the Frobenius homomor-
phism of Wp (A).

267
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 metric and topological methods

30. Let A be a ring of characteristic p. Show that the Frobenius

homomorphism f : Wp (A) → Wp (A) defined in the previous ex-
ercise is given by

f(a0 , a1 , . . . ) = (ap p
0 , a1 , . . . ),

justifying its name.

The next two exercises give a beautiful application of Krull’s
intersection theorem.

31. [KO10] Let A be a Noetherian integral domain, I ⊂ A an

a = |A|, b = |A/I|. Show that a 6 bℵ0 . (The ring A
ideal, and letQ
embeds into A/In .)
32. [Bel] One may wonder whether every Noetherian ring is the
quotient of a Noetherian integral domain. Show that this is false
by applying the previous exercise to the ring K × L, where K
is a finite field and L is a field of cardinality bigger than the
continuum.

268
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 8
GEOMETRIC DICTIONARY

In this chapter, we give a geometric perspective to the topics stud-

ied so far. This is useful because some notions such as dimension
are much easier to grasp under the geometric point of view. We
will also see that many of the notions introduced up to this point
are passible of a geometric interpretation.
The first section introduces the basics definitions, like varieties
in affine space, their coordinate rings and the Zariski topology.
Most of the sequel uses this new language to reinterpret old re-
sults and concepts from the geometric point of view.
In particular, in the second section we give some proofs of
the celebrated Nullstellensatz and use it to illustrate the corre-
spondence between the points of an affine variety and the max-
imal ideals of its coordinate ring. We also derive from it the
Ax-Grothendieck theorem on affine polynomial maps. Then we
introduce the local ring at a point and show that this can be ob-
tained by localization, so the terminology finally makes sense.
Next, we show that the local situation is somehow similar to the
graded one, and introduce the correspondence between projec-
tive varieties and graded rings.
Towards the end of the chapter, we introduce two new geomet-
ric concepts which we will investigate from an algebraic point
of view in the next chapters. The first one is the concept of di-
mension. Actually some different definitions are possible; the
equivalence of them, in a more general setting, will be shown on
the next chapter. Second, we introduce the Zariski tangent space,
together with regular and singular points.
After this, one can define curves as varieties of dimension one.
In the last section we show how Dedekind rings arise as coordi-

269
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 geometric dictionary

nate rings of smooth irreducible curves, and use our knowledge

about Dedekind rings to discuss morphisms of curves.

8.1 affine varieties

We consider rings of the form A = k[x1 , . . . , xn ] where k is a field.
We can interpret a polynomial f ∈ A as an equation that defines
a zero locus
V(f) = {(x1 , . . . , xn ) ∈ kn |f(x1 , . . . , xn ) = 0}.
More generally, we can consider the locus defined by the vanish-
ing of more than one equation. Given a set S ⊂ A we define the
locus
V(S) = {(x1 , . . . , xn ) ∈ kn |f(x1 , . . . , xn ) = 0 for all f ∈ S}.
Definition 8.1.1. A locus of the form V(S) for some S ⊂ A is
called an (affine) algebraic variety. When S is empty we obtain the
whole kn , which we will call an affine space and denote by An (k),
or simply An when k is fixed.
Remark 8.1.2. It is immediate to check that V(S) = V(I), where I
is the ideal generated by S, hence it is enough to consider the zero
loci of ideals. Moreover, since ideals in k[x1 , . . . , xn ] are finitely
generated, we see that a finite number of equations always suffice
to define an algebraic variety.
Example 8.1.3. (a) When k = R, n = 2 and I is generated by a
single quadratic polynomial, we obtain the classical example
of plane conics. Notice that it can happen that the polynomial
has no real zeros (such as x2 + y2 + 1) – in such case, the zero
locus is just the empty set.
(b) More generally, the zero locus of a single equation f ∈ A is
called a hypersurface.
(c) The zero locus V(S) of a set S is just the intersection of all
V(f) for f ∈ S.
(d) The union V(I) ∪ V(J) of two algebraic varieties is itself an
algebraic variety, defined by the ideal I · J. It can also be
defined by the ideal I ∩ J.

270
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 8.1 affine varieties

Definition 8.1.4. The last two points show that the sets of the
form V(S) for S ⊂ A are the closed sets for a topology on An (k),
which we will call the Zariski topology. Varieties V ⊂ An (k) in-
herit this topology – their closed sets are the intersections of V
with other affine algebraic varieties.
Notice that this is a very coarse topology – its only open sets
are complements of algebraic varieties – hence there are no small
open sets such as balls in Rn .
We can also go in the other direction: given a set W ⊂ kn we
define
I(W) = {f ∈ A | f(x1 , . . . , xn ) = 0 for all (x1 , . . . , xn ) ∈ W}.
The set I(W) is always an ideal in R – in fact, it is a radical
ideal.
Remark 8.1.5. The correspondences I → V(I) and W → I(W)
reverse inclusions. Moreover, it is easy to check that J ⊂ I(V(J))
and W ⊂ V(I(W)). In fact, if W is itself an algebraic variety, we
have the equality W = V(I(W)) (check it!).
The fact that J is not always the same as I(V(J)) can be easily
seen in the case where k = R and J is generated by f = x2 +
y2 + 1. In this case, V(J) = ∅, hence I(V(J)) = R. We will discuss
in the next section the shape of I(V(J)) in the case where k is
algebraically closed.
Definition 8.1.6. The ring R(V) := A/I(V) is called the coordinate
ring of the affine variety V. Notice that elements of R(V) define
actual polynomial functions V → k.
When V is a single point {x}, I(V) is just the kernel of the eval-
uation function
evx : A →k

f → f(x)
In particular I({x}) is a maximal ideal.
We end this section with a property of the Zariski topology
that shows that it cannot be very fine.
Definition 8.1.7. A topology is called Noetherian if every ascend-
ing chain of open sets (or equivalently a descending chain of
closed sets) is eventually stationary.

271
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 geometric dictionary

By Hilbert’s basis theorem, it is immediate that the Zariski

topology is Noetherian. In fact a descending chain of closed sets

W1 ⊃ W2 ⊃ · · ·

gives rise to an ascending chain of ideals

I(W1 ) ⊂ I(W2 ) ⊂ · · ·

that must stabilize since A is Noetherian. This implies that the

original chain stabilizes, since Wi = V(I(Wi )).
Remark 8.1.8. In particular, this implies that An is compact un-
der the Zariski topology.

8.2 the nullstellensatz

Starting from this section, we assume that k is algebraically closed.
This simplifies the geometry of An (k) considerably.
We start with the question left open in the previous section:
what can we say about an ideal of the form I(V), where V ∈ An
is an algebraic variety? It is easy to check that I(V) is a radical
ideal√(if fr vanishes on V, so does f). Hence if V = V(J), we see
that J ⊂ I(V).
In this section, we will see that the converse holds in the al-
gebraically closed case. We give three (equivalent) forms of the
celebrated Nullstellensatz, also called the Hilbert zeros theorem
[Hil93].
Theorem 8.2.1 (Nullstellensatz, strong form). Assume that k is al-
gebraically
√ closed, and let J ⊂ k[x1 , . . . , xn ] be an ideal. Then I(V(J)) =
J.
Notice that this implies that there is a bijective correspondence
between algebraic varieties inside An and radical ideals inside
k[x1 , . . . , xn ]. There are other formulations of the Nullstellensatz
that are apparently weaker.
Theorem 8.2.2 (Nullstellensatz, first weak form). Assume that k
is algebraically closed, and let J ⊂ k[x1 , . . . , xr ] be an ideal. If V(J) is
empty, then J is the whole ring.

272
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 8.2 the nullstellensatz

Theorem 8.2.3 (Nullstellensatz, second weak form). Assume that

k is algebraically closed. The maximal ideals of k[x1 , . . . , xr ] are exactly
those of the form I({x}) for some point x ∈ kn .
We start by proving the equivalence of the three forms, and
then prove the theorem itself.
Proof of the equivalence. 1. We first see that the strong form im-
plies the first weak form. This is easy: if V(J) is empty,
I(V(J)) = A = k[x1 , . . . , xn ]. By the strong form, this is the
radical of J, hence J itself is the whole ring.
2. Assume that the first weak form holds, and let M ⊂ A be
a maximal ideal. Then V(M) is not empty, so that we find
some point x ∈ V(M). This implies that M ⊂ I({x}), and
we must have equality since M is maximal; this proves the
second weak form.
3. Assume that the second weak form holds, and let J ⊂ A
be an ideal such that V(I) is empty. This means that J is
not contained in I({x}) for any point x. Since this are all the
maximal ideals, J is the whole ring, proving the first weak
form.
4. The trickiest part is proving that the first weak form implies
the strong form. This is done by the so called Rabinow-
itsch trick ([Rab29]). Let J ⊂ k[x1 , . . . , xn ] be an ideal, and
let g be a polynomial vanishing on V(J). Inside the ring
k[x1 , . . . , xn , y], consider the ideal J generated by elements
of J and the single polynomial y · g − 1.
Since g is zero on V(J), V(J) is empty, and by the first weak
form we deduce that J is the whole ring. This implies that
we can find an identity of the form

1 = h0 (y · g − 1) + h1 f1 + · · · hr fr

for some f1 , . . . , fr ∈ J and h0 , . . . , hr ∈ k[x1 , . . . , xn , y].

We can formally substitute y = 1/g in this equation and
then clear the denominators by multipliying both sides by
gt for some t big enough. This leaves the identity

gt = h10 f1 + · · · hr0 fr ,

273
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 geometric dictionary

√
which shows that g ∈ J.

There is another form of the Nullstellensatz that can be stated

for an arbitrary field. It usually goes under the name of Zariski’s
lemma.

Theorem 8.2.4 (Zariski’s lemma). Let A be a finitely generated k-

algebra that is also a field. Then A is a finite extension of k.

Zariski’s lemma implies the Nullstellensatz, as we show here.

Proof of the Nullstellensatz. We show that Zariski’s lemma implies

8.2.3. Assume that k is algebraically closed, and let M be a max-
imal ideal of k[x1 , . . . , xr ]. Then the algebra A := k[x1 , . . . , xr ]/M
is finitely generated over k and it is also a field, hence it is a fi-
nite extension of k. Since k it is algebraically closed, the natural
inclusion k ,→ A is an isomorphism.
If we denote λi ∈ k the image of xi under this isomorphism, it
follows that fi := xi − λi ∈ M. Since the ideal generated by the
elements fi is maximal, it follows that M = (f1 , . . . , fr ).

We give two proofs of Zariski’s lemma.

First proof of Zariski’s lemma. Let A be a finitely generated k-alge-

bra that is also a field. By Noether normalization lemma 5.3.1, we
can write A = k[x1 , . . . , xr ], where x1 , . . . , xm are algebraically in-
dependent, and xm+1 , . . . , xr are integral over k[x1 , . . . , xm ]. The
assertion that we need to prove is that m = 0.
If not, 1/x1 ∈ A, hence it must be integral over k[x1 , . . . , xm ].
This gives a nontrivial polynomial relation between x1 , . . . , xm ,
contradicting the fact that they are algebraically independent.

The second proof will be an immediate consequence of this

more general lemma, which is useful in its own sake.

Lemma 8.2.5. Let A ⊂ B be integral domains and assume that B is

finitely generated over A (as an A-algebra). For all nonzero b ∈ B
there exists a nonzero a ∈ A with the following extension property:
every homomorphism f : A → K, where K is algebraically closed and
f(a) 6= 0, can be extended to a homomorphism g : B → K such that
g(b) 6= 0.

274
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 8.2 the nullstellensatz

Proof. By induction, we can assume that B is generated over A

by a single element x. We now distinguish two cases, based on
whether x is algebraic over (the field of fractions of) A.
If x is transcendental over A, write

b = an xn + · · · + a0

with a0 , . . . , an ∈ A, and choose a := an . Then, for every f : A →

K, we can extend it to B by choosing at will the image y = g(x) ∈
K. The requirement that g(b) 6= 0 becomes

f(an )yn + · · · + f(a0 ) 6= 0,

which can be guaranteed for a suitable choice of y, since K is

infinite and f(an ) 6= 0.
If x is algebraic over A, a little more care is needed since g(x)
cannot be chosen arbitrarily. In this case, x satisfies an equation

an xn + · · · + a0 = 0 (8.2.1)

for some a0 , . . . , an ∈ A. Moreover, 1/b is a polynomial in x,

hence it satisfies a similar equation

bm b−m + · · · + b0 = 0 (8.2.2)

for some b0 , . . . , bm ∈ A.
In this case, we choose a = an bm . Given f : A → K such
that f(a) 6= 0, we can extend it to f 0 : A[1/a] → K by declaring
f 0 (1/a) = 1/f(a). Then we can choose a maximal extension, say
h : T → K. By Theorem 7.2.9, T is a valuation ring.
By (8.2.1), x is integral over A[1/a], hence over T . But valuation
rings are integrally closed by 7.2.8, hence x ∈ T , and we have the
desired extension to B.
To check that g(b) 6= 0, it is enough to show that h is defined
on 1/b, and this follows in the same way. Namely, 1/b is integral
over A[1/a] by (8.2.2), so it belongs to T .
Second proof of Zariski’s lemma. With the notation of the above lem-
ma, assume that A and B are fields and choose b = 1. Then the
fact that every homomorphism from A to an algebraically closed
field can be extended to B is equivalent to the fact that B is alge-
braic over A.

275
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 geometric dictionary

In the final part of the section, we investigate the geometric

consequences of the Nullstellensatz. By Theorem 8.2.3, points in
An correspond bijectively with maximal ideals of A.
Let V be an affine variety, and consider the quotient R(V) =
A/I(V). Maximal ideals of R(V) correspond to those maximal
ideals I({x}) of A that contain I(V); this happens exactly when
x ∈ V. We conclude that maximal ideals of R(V) correspond
bijectively to points of V. Moreover, V can be reconstructed as
a topological space knowing the ring R(V) alone: in fact closed
sets are exactly sets of maximal ideals that contain a given ideal
J. This shows that the algebraic side completely governs the ge-
ometry of algebraic varieties.

8.3 the ax-grothendieck theorem

In this section, we temporarily go back to our algebraic setup to
give an application of the Nullstellensatz.

Theorem 8.3.1 (Ax-Grothendieck). Let k be an algebraically closed

field, f : kn → kn a polynomial map. If f is injective, then it is surjec-
tive.

The proof by Ax in [Ax68] is based on model theory – we

sketch it in Exercise 31. We follow a simplification of Grothen-
dieck’s argument ([Gro66, Theorem 10.4.11]).
To begin, we rephrase the conditions of being injective and not
being surjective in more explicit terms.

Lemma 8.3.2. Let k be an algebraically closed field, f : kn → kn an

injective polynomial map. Then there exist polynomials h1 , . . . , hn ∈
k[x1 , . . . , xn , y1 , . . . , yn ] such that

f(x) − f(y) = h1 (x, y)(x1 − y1 ) + · · · + hn (x, y)(xn − yn ). (8.3.1)

Proof. By hypothesis, f(x) − f(y) only vanishes on the diagonal

of kn × kn . The radical ideal defining the diagonal is ((x1 −
y1 ), . . . , (xn − yn )), so this follows by the Nullstellensatz.

Another application of the Nullstellensatz immediately trans-

lates the condition of not being surjective.

276
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 8.4 morphisms

Lemma 8.3.3. Let k be an algebraically closed field, f : kn → kn a non-

surjective polynomial map. Then there exists y ∈ kn and polynomials
h1 , . . . , hn ∈ k[x1 , . . . , xn ] such that

(f1 (x) − y1 ) · h1 (x) + · · · + (fn (x) − yn ) · hn (x) = 1. (8.3.2)

Notice that the Ax-Grothendieck theorem also holds when k is

a finite field, trivially since any injection from a finite set to itself
is also surjective. We give the proof by reducing to this case.
Proof of Ax-Grothendieck theorem. By contradiction, assume that f
is injective but not surjective, hence (8.3.1) and (8.3.2) hold. Con-
sider the ring A generated over Z (if k has characteristic 0) or
Z/pZ (if k has characteristic p) by all coefficients of all poly-
nomials involved in these equations. This implies that f is still
injective and not surjective on A.
Choose a maximal ideal M that does not contains all coeffi-
cients, so that equations (8.3.1) and (8.3.2) remain nontrivial over
k 0 := A/M. The map f defines a map on k 0n which is still in-
jective and not surjective. Notice that even if k has characteristic
0, k 0 cannot contain Q, hence in any case it has positive charac-
teristic p. By Lemma 8.2.5, k 0 is an algebraic extension of Z/pZ,
hence it is itself a finite field. But this is a contradiction, since
an injective self map over k 0n , where k 0 is finite, is also surjec-
tive.

8.4 morphisms
In this section we introduce morphisms between affine varieties.
Let V ⊂ An (k) be an affine variety, and consider the ring R(V) =
k[x1 , . . . , xn ]/I(V). Any element f ∈ k[x1 , . . . , xn ] defines a poly-
nomial function An (k) → A1 (k), where we identify A1 (k) with
k. This can be restricted to a function V → A1 (k), and two poly-
nomials f and g will define the same function on V if and only
if f − g ∈ I(V). This allows us to identify R(V) with the ring of
polynomial functions V → A1 (k). Putting together more than
one function, we give the following
Definition 8.4.1. Let V ⊂ An (k) be an affine variety. An affine
morphism V → Am (k) is a function defined by the value of m

277
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 geometric dictionary

elements f1 , . . . , fm ∈ R(V). If W ⊂ Am (k) is another affine

variety, a morphism V → W is just a morphism V → Am (k)
whose image lies inside W.
Using the Nullstellensatz, we can give a precise picture of the
relation between the algebraic and geometric side (assuming, as
we will do from now on, that we work on an algebraically closed
field k). First, we have seen that to any affine variety V, we can
associate a finitely generated, reduced ring R(V). The ring R(V)
is reduced because the ideal I(V) is radical.
In the other direction, let A be any finitely generated, reduced
k-algebra. Since A is finitely generated as a k-algebra, there exists
a surjection f : k[x1 , . . . , xn ] → A. The ideal I = ker f is radical,
since A is reduced. Attached to I there is the affine variety V :=
V(I) ⊂ An (k), and the Nullstellensatz implies that I = I(V). It
follows that in fact, A = ∼ R(V) as k-algebras. This construction
gives us a way to go from rings to varieties and vice versa.
This duality also extends to morphisms. Let V ⊂ An (k) and
W ⊂ Am (k) be two affine varieties. If f : V → W is a morphism,
there is a natural composition map
f∗ : R(W) → R(V),
g → g◦f
and f∗ is clearly a homomorphism of rings.
Vice versa, take any homomorphism t : R(W) → R(V). The
coordinate functions x1 , . . . , xm belong to R(W), and this gives
us m elements f1 , . . . , fm ∈ R(V) defined as fi = t(xi ). Moreover,
for all g ∈ I(W) we have

g(f1 (x), . . . , fm (x)) = 0

for all x ∈ V, just because t is well-defined (do you see this?).

This means that
(f1 (x), . . . , fm (x)) ∈ W
for all x ∈ V, and this defines a morphism V → W.
In other words, studying the algebra of reduced, finitely gener-
ated k-algebras and their homomorphisms is completely equiv-
alent (up to inverting the direction of maps) to studying the ge-
ometry of affine varieties over k. Under this correspondence, re-
duced ideals of rings become closed set of varieties under the

278
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 8.5 local rings and completions revisited

Zariski topology. For those who know the terminology, we have

realized an equivalence between the category of reduced, finitely
generated k-algebras and the opposite category of affine varieties
over k.

8.5 local rings and completions re-

visited
In this section we can give geometric interpretations to the oper-
ations of localization and completion of a ring.
Let V ⊂ An (k) be an affine variety, where k is an algebraically
closed field. Section 8.2 showed that there is a bijective correspon-
dence between points of V and maximal ideals of the ring R(V) –
namely the point x ∈ V corresponds to the ideal Mx of algebraic
functions that vanish at x.
Assume that V is irreducible, so that R(V) is an integral do-
main, and let k(V) be the fraction field of R(V). Elements of k(V)
can be seen as rational functions defined on V. An element of
k(V) takes the form f/g for f, g ∈ R(V), hence it defines a func-
tion Ug → k on the Zariski open set

Ug = {p ∈ V|g(p) 6= 0}.

The localization of R(V) at Mx is the subset of k(V) consisting

of those rational functions whose denominator does not vanish
on x. In other words, a function t = f/g belongs to the localiza-
tion if, and only if, x ∈ Ug . This entails that t is actually defined
on a neighborhood of x.
The ring R(V)Mx is then ring of algebraic rational functions
defined in a neighborhood of x, which gives meaning to the name
localization.
More generally, let P ⊂ R(V) be a prime ideal, and V(P) ⊂ V
its zero locus. The ring R(V)P is the ring of rational functions on
V defined in a neighborhood of V(P).
The completion of a ring gives rise to an even more local zoom
around the point x. The issue is that the Zariski topology is not
very fine. Zariski closed sets in V are actually subvarieties of V –
or symmetrically, Zariski open subsets cannot be too small. In the

279
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 geometric dictionary

case where k = C, complex algebraic varieties are also endowed

with the Euclidean topology, which contains many more open
sets - for instance Euclidean balls. We want to somehow capture
the notion of a function defined in a small neighborhood of x ∈ V,
but the Zariski topology does not have small neighborhoods!
The answer comes by analogy with complex geometry. Com-
plex differentiable functions are locally expressible as a conver-
gent sum of a power series. In our algebraic setting, we do not
have a suitable notion of convergence, but we do away with it by
just considering the ring of all algebraic power series.
Let us first consider the case where V = An (k) is just the
affine space. We might just as well choose the point x = 0. Then
the ring of algebraic functions on V is just k[x1 , . . . , xn ], and its
completion at the ideal Mx is the power series ring k[[x1 , . . . , xn ]],
as we have seen in Example 7.5.21d.
Notice that in this case we have natural inclusions

k[x1 , . . . , xn ] ⊂ k[x1 , . . . , xn ]Mx ⊂ k[[x1 , . . . , xn ]].

This is because a polynomial g such that g(0) 6= 0 is invertible in

k[[x1 , . . . , xn ]], hence a rational function f/g can be expanded as
a well defined power series
This set of inclusions can be interpreted as taking smaller and
smaller neighborhoods of x. In other words, function defined
globally are a subset of functions defined in a neighborhood of x,
which are themselves a subset of functions defined only formally
in an infinitesimal neighborhood of x.
In the general case, Example 7.5.21c shows that for a ring R(V),
the localization at the maximal ideal Mx of a point sits between
R(V) and its completion at Mx . Again, in this case, we interpret
[ as the ring of “analytic functions” defined around the point
R(V)
x.

8.6 graded rings and projective vari-

eties
In this section, we mimic what we have done for affine varieties
to define varieties in projective space.

280
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 8.6 graded rings and projective varieties

Definition 8.6.1. Let k be a field. The n-dimensional projective

space over k is the set of lines through the origin in An (k). It can
be identified with the quotient

Pn (k) := (kn+1 \ 0)/ ∼,

where ∼ is the equivalence relation given by x ∼ y if x = t · y for
some t ∈ k.
We will soon give Pn (k) the structure of a variety. To this end,
notice that if f ∈ k[x0 , . . . , xn ] is a polynomial and p ∈ Pn (k), the
value of f(p) is not well-defined, since it depends on a represen-
tative for p in kn+1 . But if f is homogeneous, it is well-defined
whether f(p) = 0 or not.
Hence, we can define a zero locus
V(f) = {p ∈ Pn (k)|f(p) = 0}.
More generally, we can consider the locus defined by a homoge-
neous ideal I ⊂ k[x1 , . . . , xn+1 ] as
V(I) = {p ∈ Pn (k)|f(p) = 0 for all homogeneous f ∈ I}.
Definition 8.6.2. A locus of the form V(I) for some homogeneous
ideal I is called a (projective) algebraic variety. When I = 0, we
obtain the whole Pn (k), which is a projective variety itself. We
will sometimes denote Pn = P( k) when k is implied.
Remark 8.6.3. As in the affine case, every algebraic variety is
defined by a finite number of equations, since k[x0 , . . . , xn ] is
Noetherian. Moreover, mimicking the affine case, we can define
the Zariski topology on Pn (k) whose closed sets are projective va-
rieties.
The projective space is covered by affine charts. Namely, for
every i = 0, . . . , n, there is a subset Ci ⊂ Pk (n) defined by the
equation xi 6= 0. For each element (x0 , . . . , xn ) ∈ Ci , there is
unique representative having xi = 1, hence Ci can be naturally
identified with An (k) by taking such representative and omitting
the i-th coordinate.
Remark 8.6.4. The sets Ci are open in the Zariski topology and
form an open covering of Pn (k). If V ⊂ Pn (k) is a projective
variety, this restricts to an open covering of V by affine varieties.

281
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 geometric dictionary

We now give some examples of projective varieties.

Example 8.6.5. (a) For every affine variety V ⊂ An (k), we can
construct a projective variety by homogeneizing its equations.
Namely, if f(x1 , . . . , xn ) is any polynomial of degree t van-
ishing on V, we can construct a homogeneous polynomial
fh (x0 , . . . , xn ) of degree t such that

fh (1, x1 , . . . , xn ) = f(x1 , . . . , xn ).

It is sufficient to take any monomial in f of degree d 6 t and

multiply it by xt−d0 .
If we do that for all f ∈ I(V), we get a projective variety
V ⊂ Pn (k) such that V ∩ C0 = V. V is called the projective
closure of V.
(b) As in the affine case, the zero locus of a single equation f is
called a hypersurface.
(c) The zero locus V(I) of a homogeneous ideal I is just the inter-
section of all V(f) for all f generating I.
(d) The union V(I) ∪ V(J) of two projective algebraic varieties is
itself an algebraic variety, defined by the ideal V(I · J).
(e) Let Mm,n (k) be the vector space of m × n matrices with co-
efficients in k. For each r < min{m, n}, the set Dr of matrices
having rank at most r is an affine variety. In fact, the condi-
tion of having rank at most r is equivalent to the vanishing
of all (r + 1) × (r + 1) minor determinants. Since these equa-
tions are homogeneous, they also define a variety Dr in the
corresponding projective space. These are called determinan-
tal varieties.
Let us denote A = k[x0 , . . . , xn ]. As in the affine case, attached
a projective variety V there is an ideal

I(V) = {f ∈ A | f(p) = 0 for all p ∈ V},

and a ring R(V) = A/I(V). Unlike the affine case, I(V) is ho-
mogeneous, hence R(V) is graded. Also, unlike the affine case,
elements of R(V) do not correspond to polynomial functions de-
fined on V.

282
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 8.6 graded rings and projective varieties

You will prove in Exercise 2 that when the field k is alge-

braically closed there is version of the Nullstellensatz adapated
to the projective case. Hence, while the algebra of finitely gener-
ated algebras over k is reflected in the geometry of affine varieties,
graded finitely generated algebras are best seen as counterparts
of projective varieties. Let us see what it means for morphisms.

Definition 8.6.6. Let V ⊂ Pn and W ⊂ Pm be two projective

varieties. A projective morphism f : V → W is a function which
locally is an affine morphism. This means that for each point
x ∈ V we can find affine charts Ci ⊂ Pn and Cj0 ⊂ Pm such that
x ∈ Ci , f(x) ∈ Cj0 and the affine map

f |V∩Ci : V ∩ Ci → W ∩ Cj0

is a morphism of affine varieties.

As in the affine case, a morphism V → W gives a homomor-

phism R(W) → R(V) between the graded rings, although the
elements of this rings cannot be interpreted as functions.

Remark 8.6.7. It is easy to generalize Definition 8.6.6 to speak of

morphisms V → W, where V is affine and W projective. Similarly,
one can also consider the case where V is projective and W is
affine, altough it turns out that in this case the only morphisms
are constant.

Remark 8.6.8. Definition 8.6.6 is pretty cumbersome to work with.

Actually, a morphism V → Pm , where V is projective, is given by
m + 1 elements f0 , . . . , fm ∈ R(V) of the same degree, which do not
vanish simultaneously on V.
At least one direction is easy to see: if f0 , . . . , fm ∈ R(V) is such
a collection of polynomials, the map f : V → Pm given by

f(x) = (f0 (x), . . . , fm (x))

is at least well-defined (why?). It is also clear that f restricts to a

polynomial map on each affine chart, hence it defines a projective
morphism.

Example 8.6.9. Let vn be the map P1 → Pn defined by the ho-

n−1
mogeneous polynomials xn 0 , x0 x1 , . . . , xn
1 . These polynomials

283
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 geometric dictionary

never vanish together on P1 , hence the map vn is a projective

morphism, called the Veronese map.
The image vn (P1 ) is a curve inside Pn , called a rational normal
curve. It is easy to see that its ideal is generated by equations of
the form
ya yb = yc yd
whenever a + b = c + d, where y0 , . . . , yn are projective coor-
dinates on Pn . Another way to describe these equations is by
requiring that the matrix
 
y0 y1 ··· yn−1
y1 y2 ··· yn

has rank 1.

8.7 a new idea: the dimension

In the previous sections, we have established a dictionary that
allows us to translate algebraic concepts in a geometric frame-
work. Under this correspondence, we have associated (reduced)
k-algebras with affine varieties, graded k-algebras with projec-
tive varieties and (radical) ideals with subvarieties. Moreover, we
have seen how localization and completion correspond to the ge-
ometric operations of considering smaller charts around a point.
In this section, we are going to introduce a new concept from
a geometric point of view. It will be the task for the next chapter
to give the full picture from the algebraic side. Actually, we will
state four different definitions of dimension of an algebraic vari-
ety, and give some motivation why each one of those is at least
plausible. In this chapter we deal with finitely generated reduced
algebras over an algebraically closed field, while in the next one,
we are going to generalize these definitions to rings more general
than that. In any case, the definitions that we are going to give
will rely on the intuition gained here. The next chapter will show
that these definitions actually agree and give the same number
whenever it makes sense.
For simplicity, we state our definitions for affine varieties, al-
though it is easy to adapt them for projective varieties with mini-

284
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 8.7 a new idea: the dimension

mal modifications. Before proceeding, we need some remark on

the irreducible components of varieties.

Definition 8.7.1. Let V be a topological space. We say that V is

irreducible if V cannot be written as a union of two proper closed
subset. If V is an affine or projective variety, we call it irreducible
when it is so with respect to the Zariski topology.

Remark 8.7.2. Let V ⊂ An be an affine variety, and assume that

I(V) is prime. Then V is irreducible by Proposition 1.1.20 ii).

Remark 8.7.3. Any algebraic variety can be decomposed uniquely

as a finite union of irreducible varieties, which are called its irre-
ducible components. For an affine variety, this is the geometric
translation of the existence and uniqueness of the primary de-
composition.
In fact, let V be an affine variety, and decompose the ideal
I(V) as the intersection√Q1 ∩ · · · ∩ Qr , where the ideals Qi are
primary. Letting Pi = Qi , we can write V as the union of the
varieties Vi := V(Pi ), where Pi is a minimal prime of I(V), and
the components Vi are irreducible by the previous remark.
The same result for projective varieties follows easily by con-
sidering its intersections with the affine charts Ci .

Remark 8.7.4. In fact, by the above decomposition it follows that

the affine variety V is irreducible if and only if I(V) is prime.

It is easy to see that the dimension is not a well-defined con-

cept for reducible variety. In fact, different components can have
different dimensions – for instance, the equations xy = 0 and
xz = 0 define the union of a line y = z = 0 and a plane x = 0 in
A3 .
There are two ways around this issue. The first one is to only
define dimension for irreducible varieties, and extend it to irre-
ducible varieties by taking the maximum across the components.
The other one is to define the concept of dimension of a variety
around a point, with the implication that this number can depend
on the point. Given a variety V and a point p ∈ V defined by the
ideal M, we are then lead to work with the local ring R(V)M and
to define dimension for such rings. We will follow both ways in
our attempts to define dimension.

285
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 geometric dictionary

First approach. Our first attempt at defining dimension is

based on the idea that if there is an inclusion of irreducible va-
rieties V ⊂ W, the dimension of W should be strictly greater
than that of V.
Definition 8.7.5. Let V be an irreducible algebraic variety. The
Krull dimension of V is the maximum integer n such that there
exists a chain
V0 ⊂ V1 ⊂ · · · ⊂ Vn = V (8.7.1)
of irreducible varieties.
The intution behind this definition is that the existence of a
chain like (8.7.1) should imply that V has dimension at least n.
Conversely, if V has dimension n, there should always be a way
to obtain lower dimensional subvarieties of any intermediate di-
mension, for instance by taking the intersection with a suitable
linear subspace.
In the ring R(V), the chain (8.7.1) corresponds to a chain of
prime ideals Pn ⊂ Pn−1 ⊂ · · · ⊂ P0 , where Pi is the ideal defin-
ing Vi . This is the notion that we will generalize to arbitrary
rings.
Second approach. Let us assume that An is an object of dimen-
sion n. If we consider a variety defined by polynomials f1 , . . . , fr ,
we would like the dimension to drop by r, provided the equa-
tions fi = 0 are in some sense independent. A measure of this is
given by algebraic independence: that there exists no polynomial
identity satistied by f1 , . . . , fr .
To make this precise, let us take an irreducible variety V, so
that the ring R(V) is an integral domain. Its quotient field k(V)
can be considered as the field of rational functions on V. Each
time we add an independent polynomial relation, we impose an
algebraic relation between elements of k(V), thereby decreasing
its transcendence degree by 1.
Definition 8.7.6. Let V be an irreducible affine variety. The tran-
scendence dimension of V is the transcendence degree of the field
k(V) over k.
This definition makes sense for algebraic varieties, but is hard
to generalize over arbitrary rings, since it exploits the fact that
R(V) is an algebra over a field.

286
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 8.7 a new idea: the dimension

Third approach. This is in some sense complementary to the

previous one. Transcendence dimension was an attempt to de-
fine the dimension of V ⊂ An as n − r, where r is the number of
independent equations needed to define V. Dually, we can con-
sider the minimal number of equations needed in R(V) to isolate
a point. Since this can – a priori – change from point to point, we
work inside a localization R(V)M , where M is the ideal of a point
p ∈ V.
Definition 8.7.7. Let V be an affine variety, p ∈ V a point defined
by the ideal M. The Chevalley dimension of V at p is the minimal
number of elements x1 , . . . , xd ∈ R(V)M that satisfy the equality
(x1 , . . . , xd ) = M.
p

Intuitively, to isolate the point p in V we need as many equa-

tions in R(V) as the dimension of V around p. Notice that here
we do not require V to be irreducible – but we will see that, when
this is the case, the Chevalley dimension does not depend on the
point and agrees with the previous definitions.
Fourth approach. Our last definition of dimension is the most
sophisticated and least intuitive one.
To begin, let us consider the case of An , and let us take a
point, say O ∈ An . Let M ⊂ k[x1 , . . . , xn ] be the ideal of poly-
nomials vanishing at O. The local ring around O is the ring
A = k[x1 , . . . , xn ]M of rational function whose denominator is
defined at O.
A way to extract the numerical invariant n from the ring A is
to consider the space As of functions f ∈ R vanishing of order s at
O. The quotient As /As+1 has a basis formed by all monomials
of degree s in n variables, and so its dimension is of order sn .
This allows us to define the dimension of An around O as the
order of growth of this dimension.
Now let V be a variety, p ∈ V. If we had something like a
local chart around p, we could try to define the dimension of V
around p by following the same approach: take the ideal M of p
and consider the dimension of Ms /Ms+1 and its order of growth.
Unfortunately, we do not work in the differentiable setting, and
this implies that we do not have this luxury.
On the other hand, we can hope to have something of the sort
when we pass to the completion with respect to the M topol-
ogy (this is not quite true, but it will turn out to be true for

287
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 geometric dictionary

most points). And we know by Corollary 7.5.18 that passing to

the completion should not change the dimension of Ms /Ms+1 .
Hence, we are somewhat justified to give the following defini-
tion.

Definition 8.7.8. Let V be an affine variety, p ∈ V a point defined

by the ideal M. The Poincaré dimension (or Samuel dimension) of
V at p is the order of growth of the function n(s), which is the
dimension of the space Ms /Ms+1 of polynomial functions van-
ishing of order s at p.

At this point, it is not even clear that this is well-defined. We

will make this definition precise in Section 9.2. There, we will
generalize this definition to arbitrary local Noetherian rings, and
show that it agrees with our previous definitions.

8.8 the zariski tangent space

Borrowing some ideas from differential geometry, we now want
to define the tangent space at point of an algebraic variety. We
will use the setting of smooth manifolds (say Ck , C∞ or analytic)
as a source of inspiration – the reader that is not familiar with
them can consult any standard source, such as [Hir97], but in any
case our final definitions will be independent of this discussion.
In the differentiable (or holomorphic) setting, there are many
ways to introduce the tangent space. For instance, a standard
way to introduce the tangent space in a point p to a smooth
manifold M is to consider classes of equivalence of smooth arcs
c : (−1, 1) → M such that c(0) = p, modulo the equivalence rela-
tion that identifies two such arcs if their diffentials in 0 agree, in
a suitable local chart around p.
In the algebraic setting we are more constrained: for instance,
given a point p on an algebraic variety V, it may very well happen
that the only morphism c : A1 → V such that c(0) = p is constant.
A second complication that we face is the fact that some of the
varieties we consider will not be smooth, and we would like to
be able to define the tangent space even in singular points.
To work around these issues, we start from the affine case. Let
V ⊂ An (k) be an affine variety, p ∈ V a point, I ⊂ k[x1 , . . . , xn ]

288
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 8.8 the zariski tangent space

the ideal defining V. Each polynomial f ∈ k[x0 , . . . , xn ] has a

formal differential given by the vector
 
∂f ∂f
dfp = (p), . . . , (p) .
∂x1 ∂xn

By analogy with the implicit function theorem, one possible

definition of the tangent space would be

Tp V = {v ∈ kn | dfp · v = 0 for all f ∈ I} . (8.8.1)

This is a well-defined subspace of kn , and using the Leibniz

rule it is easy to check that if I = (f1 , . . . , fr ), then

Tp V = {v ∈ kn | d(fi )p · v = 0 for i = 1, . . . , r} .

In other words, Tp V is the kernel of the matrix of the partial

derivatives  ∂f ∂f1

1
(p) · · · (p)
 ∂x1. ∂xn
.. 
 .
 . . .

∂fr ∂fr
∂x1 (p) · · · ∂xn (p)

The only disadvantage of this definition is that Tp V is not in-

trinsic, but is defined in terms of the ambient embedding. To
recover from this, we again go by analogy with the differentiable
case.
Given a smooth manifold M and a point p ∈ M, one can define
the cotangent space Tp? M := (Tp M)? . To any smooth function f
defined in a neighborhood of p, one can then attach a differential
form dfp , which is an element of Tp? M. Let Op M be the ring of
smooth functions defined in a neighborhood of p. Then Op M is
a local ring with maximal ideal

Mp = {f ∈ Op M | f(p) = 0},

(why?) and there is a natural surjective map

dp : Mp → Tp∗ M.
f → dfp

289
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 geometric dictionary

By the Leibniz rule, it follows that dp is identically 0 on M2p , and

in fact it is a simple verification that dp induces an isomorphism
of real vector spaces
Mp ∼ ∗
= Tp M.
M2p

This looks like something we are able to translate to the alge-

braic setting. Let V be an affine variety over the field k, p ∈ V a
point defined by the maximal ideal Mp ⊂ R(V). Let A = R(V)Mp
be the local ring of V around p, having maximal ideal Mp =
Mp · A. The set Mp /M2p has the structure of vector space over
the field A/Mp = ∼ k.

Definition 8.8.1. The vector space Mp /M2p is called the Zariski

cotangent space of V at p, denoted by Tp∗ M. Its dual (Mp /M2p )∗ is
called the Zariski tangent space of V at p, denoted by Tp M.

Notice that Mp is finitely generated, hence Tp M is a finite di-

mensional vector space. To reconcile this intrinsic definition with
the previous one, we use the following result.

Proposition 8.8.2. Let V ⊂ An (k) an affine variety, p ∈ V, and T the

extrinsic tangent space at p, defined by (8.8.1). Then there is a natural
duality between T and Mp /M2p .

Proof. Let I be the ideal of V, R(V) = k[x1 , . . . , xn ]/I and Mp ⊂

R(V) the ideal defining p. Given a polynomial f ∈ k[x1 , . . . , xn ]
such that f(p) = 0, and a vector v ∈ kn we can compute the
number
hf, vi := dfp · v.
By definition, hf, vi = 0 for v ∈ T and f ∈ I, hence this descends
to a k-bilinear function on Mp × T . By the Leibniz rule, hf, ·i is
identically 0 if f ∈ M2p , hence this gives a bilinear form

Mp
b: × T → k.
M2p

We claim that b is a perfect pairing. One direction is easy: f

b(·, v) = 0 identically, then v is in the kernel of dp (xi − pi ) for
each coordinate function xi , which implies that v = 0.

290
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 8.8 the zariski tangent space

Vice versa, take f ∈ Mp such that b(f, ·) = 0 identically. We can

lift f to a polynomial f, which we can expand in the form
X
n
f= ai (xi − pi ) + g1 ,
i=1

where g1 vanishes of order 2 at p. The condition that b(f, v) = 0

for all v ∈ T implies that there is a polynomial h ∈ I such that
X
n
h= ai (xi − pi ) + g2 ,
i=1

where g2 vanishes of order 2 at p. It follows that

f = f − h = g1 − g2 ∈ M2p ,
so b is a perfect pairing of finite-dimensional k-vector spaces.
The isomorphism between Mp /M2p and Mp /M2p allows us to
conclude.
Remark 8.8.3. We have defined the Zariski tangent space Tp V
only when V is affine. Since Tp V is defined only in terms of
the local ring R(V)p , we can extend the definition to projective
varieties. Namely, if V is a projective variety and p ∈ V, we
can take one of the standard affine charts Ci containing p, and
consider the affine variety V ∩ Ci . The local ring of V ∩ Ci around
p does not depend on i, up to a natural isomorphism (prove this!),
hence there is a well-defined Zariski tangent space Tp V even in
the projective case.
Remark 8.8.4. Using the definition of Chevalley dimension 8.7.7
and Nakayama’s lemma, it follows immediately that
dim Tp V > dim V.
We will make this precise and prove this in Section 10.1.
In general, we expect by analogy with the smooth case that the
two dimensions agree. We isolate this condition with a
Definition 8.8.5. Let V be an algebraic variety, p a point of V. If
dim Tp V = dim V, we say that V is regular at p, otherwise that it
is singular. We say that V is regular (or smooth) if it is regular at
each of its points, singular otherwise.

291
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 geometric dictionary

It is now time to look at a few examples.

Example 8.8.6. (a) The surface S defined by

f(x, y, z) = xy − z2 − 1

is smooth. In fact, the differential is

df(x,y,z) = (y, x, −2z),

which never vanishes on a point p ∈ S. It follows that Tp S is

everywhere 2-dimensional, and dim S = 2.

(b) Let V = V(f) ⊂ An be an affine hypersurface and take a

point p ∈ V. The condition that V is regular at p amounts to
saying that not all partial derivatives of f vanish at p. When
p = 0, this means that the linear part of f is not zero. Hence,
we can picture any hypersurface singularity by taking a suit-
able polynomial whose lowest homogeneous component has
degree at least 2.

(c) Consider the variety V formed by two lines in A2 (k) meeting

at the origin – for concreteness take the x and y axes. The
ideal defining the union of the two lines is I = (xy), so V is
singular at the origin.

(d) With the same notation, consider now the curve defined by
the equation y2 = x2 + x3 . From the analytic point of view,
the singularity looks the same as the previous one: for small
x, x3 becomes negligible, and the locus looks like y2 = x2 ,
which is the union of the two diagonals. This kind of singu-
larity is called a node.

(e) Consider the curve given by the equation y2 = x3 . In this

case, from the analytic point of view the locus has a double
tangent line given by y2 = 0. This kind of singularity is called
a cusp

The following figure depicts the planar singularities that we

have described above.

292
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 8.9 curves and dedekind rings

y2 = x2 + x3 y2 = x3

8.9 curves and dedekind rings

Having gained some intuition about dimension and smoothness,
we now turn to the simplest varieties, apart from points: those of
dimension 1, which we will call curves. Let C ⊂ An be a regular,
irreducible variety of Krull dimension 1. This implies that its
only Zariski-closed subsets are finite collections of points.
The ring R(C) is an integral domain, because C is irreducible.
Moreover, it is Noetherian, and the fact that C is a curve says that
the nonzero prime ideals of R(C) are maximal. As we will see in
Chapter 10, the condition that C is regular ensures that R(C) is
integrally closed. By Theorem 5.1.19, we conclude that R(C) is a
Dedekind ring.
This means that in our dictionary, Dedekind rings become the
algebraic counterpart to smooth, irreducible curves. Let x ∈ C be
any point, corresponding to the maximal ideal M ⊂ R(C). By our
definition of Dedekind rings, the localization Rx = R(C)M is a
discrete valuation ring. In fact, let f ∈ M \ M2 , that is, a function
vanishing at x of order 1 – f is called a uniformizer at x.
Since the only ideals of Rx are powers of M, every other func-
tion g ∈ Rx can be written as g = u · fk for some u invertible
in Rx – that is – not vanishing at x. This shows that elements of
Rx are determined, up to invertible elements, by their order of
vanishing at x.
In the infinitesimal case, the same geometric picture holds as
in the local case. Namely, let R cx be the completion of Rx at M.
As we have discussed before, elements of R cx can be interpreted
as germs of functions in infinitesimal neighborhoods of x. By
Corollary 7.5.18, the only ideals of R cx are powers of M, b hence

293
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 geometric dictionary

every element g ∈ R cx can be written as g = u · fk , where u ∈ R cx

is unit and f ∈ M b \M 2
b is a uniformizer.
We now turn to morphisms. Let f : C → D be a nonconstant
morphism of affine smooth irreducible curves. There is a pull-
back map on the rings f∗ : R(D) → R(C). We claim that f∗ is
injective. In fact, let s ∈ R(D) such that s ◦ f = 0 identically. This
means that s vanishes on f(C), and if s is not 0, this implies that
f(C) is a finite set of points. By irreducibility, it follows that f is
constant on C.
The inclusion f∗ : R(D) ,→ R(C) gives R(C) the structure of a
R(D) algebra, and clearly R(C) is finitely generated as algebra.
We will consider what happens when we assume in addition that
R(C) is an integral extension of R(D). This happens at least in two
interesting cases.

Example 8.9.1. (a) Let C be any smooth irreducible curve. By

Noether normalization lemma 5.3.1, we can write R(C) as
integral extension of k[x1 , . . . , xd ] for some d – and in fact,
d = 1 for dimensional reasons. This gives us a morphism
C → A1 with the desired properties.

(b) Another case where one can prove that f∗ is an integral exten-
sion is when f is the restriction of a morphism of projective
curves. This is a nontrivial result, and follows for instance by
the Stein factorization – see [Har77, Corollary III.11.5].

In any case, assume that R(C) is integral over R(D), and let
k(C) be the fraction field of R(C). Since we assumed that R(C) is
integrally closed, R(C) is the integral closure of R(D) in k(C). In
this situation, we can apply the results of Section 6.3. Let y ∈ D
be a point, corresponding to a maximal ideal P of R(D). We can
then factorize
P · R(C) = Qe11 · · · Qer r ,
where Q1 , . . . , Qr are the primes over P. Since we assumed that
k is algebraically closed, the quotient fields R(C)/Qi and R(D)/P
are all isomorphic to k, hence f(Qi |P) = 1 for all i.
Let xi ∈ C be the point defined by the ideal Qi . These are
exactly the preimages of y. The ramification index ei = e(Qi |P)
expresses in a precise way the multiplicity of xi in f−1 (y). In
fact, let g ∈ P \ P2 be a uniformizer at y. We can pull back g

294
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 8.10 exercises

to a function f∗ (g) = g ◦ f ∈ R(C), and it is easy to verify that

f∗ (g) ∈ Qei i \ Qei i +1 , so f∗ (g) vanishes of order ei at xi .
In this setting, Theorem 6.3.2 simplifies to
X
r
e(Qi |P) = n,
i=1

where n is the degree of the field extension [k(C) : k(D)]. We

can interpret this as saying that the number of preimages of y is
constant, when they are counted with multiplicities.
The map f can be seen as the algebraic analogue of a branched
covering in topology or complex analysis. Unramified primes
P (if any) correspond to points which have exactly n preimages
– these are the points where the covering is unramified. Over
the other points, we can have multiple branches of the covering
converging into one, hence we need a local notion of multiplicity
to keep the number of preimages constant.

8.10 exercises
1. Prove the following geometric version of Noether normaliza-
tion lemma: any affine variety V over a field k admits a surjective
map onto an affine space An (k) such that R(V) is an integral ex-
tension of k[x1 , . . . , xn ]. If moreover the field is infinite, a generic
projection to an affine subspace of suitable dimension will do.
2. Prove the projective version of the Nullstellensatz: given a
projective variety V over an algebraically closed field, points of
V are in bijective correspondence with maximal homogeneous
ideals of R(V) different from the irrelevant ideal R(V)+ , by the corre-
spondence
p → Mp = {f ∈ R(V) | f(p) = 0}.
3. Prove the claim of Remark 8.6.8.
4. Let f : P1 (k) → P2 (k) be the map given by x30 , x0 x21 , x31 (a pro-
jection of the rational normal curve). Show that the image of f is
given by the equation y0 y22 = y31 . In the affine chart C0 , this is
simply y2 = z3 – draw it when k = R, and check that it has a
cusp at the origin.

295
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 geometric dictionary

The following exercises, up to Exercise 8, describe a result

called the combinatorial Nullstellensatz, by Noga Alon ([Alo99]),
and some applications of it. This is a variant of the standard
Nullstellensatz: on the one hand, it works on any field, and does
not require taking radicals; on the other hand it only concerns
ideals generated by polynomials of a certain special form.

5. Let k be a field, p ∈ k[x1 , . . . , xn ] a polynomial of degree ti in

the variable xi . Let Si ⊂ k be a subset of size ti + 1. If p vanishes
on S1 × · · · × Sn , then p = 0.

6. Use the previous exercise to prove the combinatorial Nullstel-

lensatz. Let k be a field, p ∈ k[x1 , . . . , xn ] a polynomial and
S1 , . . . , Sn ⊂ k be finite, nonempty subsets. Define polynomials
Y
gi (xi ) = (xi − s).
s∈Si

If p vanishes on S1 × · · · × Sn , then there are polynomials hi ∈

k[x1 , . . . , xn ], satisfying deg hi 6 deg f − deg gi , such that

p = g1 h1 + · · · + gn h n .

7. Use the previous exercise to prove the following result. Let k

be a field, p ∈ k[x1 , . . . , xn ] a polynomial of total degree t, say
by a monomial xt11 · · · xtnn . Let Si ⊂ k be a subset of size at least
ti + 1. Then there exists s ∈ S1 × · · · × Sn such that p(s) 6= 0.

8. Use the previous exercise to prove the Cauchy-Davenport theo-

rem: let A, B be nonempty subsets of Z/pZ, for some prime p.
Then
|A + B| > min{p, |A| + |B| − 1}.

9. Let A be the coordinate ring of an affine variety, M a finitely

generated module over A. For concreteness, think of the case
where M is the module of sections of a vector bundle over a sub-
variety. Give a geometric interpretation of the associated primes
of M, of its support, and a plausible explanation why the mini-
mal primes of the two sets are the same.

10. Let k be an uncountable, algebraically closed field. Give an

alternative proof of the Nullstellensatz in the form of Theorem

296
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 8.10 exercises

8.2.3, as follows. Let M be a maximal ideal of k[x1 , . . . , xn ] and

consider the field k 0 := k[x1 , . . . , xn ]/M. Assume that k 0 contains
an element t transcendental over k and show that the elements
1/(t − x) for x ∈ k are linearly independent. Derive a contradic-
tion and conclude that k 0 = k, from which Theorem 8.2.3 follows.

11. To generalize the result of the previous exercise to any alge-

braically closed field, argue as in [CL]. Namely, given a field k,
consider the quotient of kN by a maximal ideal M that contains
the ideal

I := (xi ) ∈ kN | xi = 0 except for a finite number of i .

The field k 0 thus obtained is called an ultrapower of k. Show that

k 0 is uncountable and algebraically closed, hence the previous
exercise applies. Deduce the Nullstellensatz for k.

12. Generalize the Veronese map to a map

vd : Pn → PN ,

where N + 1 is the number of monomial of degree d in n + 1

variables. Show that the image of vd is a projective variety.

13. Show that there is well-defined map

sm,n : Pm × Pn → P(m+1)(n+1)−1

which sends a pair ((xi ), (yj )) to the point having as coordinates

all possible pairwise products (xi yj ). Moreover, sm,n is injective.
This is called the Segre map.

14. Show that the image of the Segre map defined in Exercise 13
is a subvariety of P(m+1)(n+1)−1 . Since sm,n is injective, this
allows us to give the structure of a projective variety to a product
of projective spaces, and by extension to a product of projective
varieties.

15. Show that the restriction of the Segre map sn,n defined in Ex-
ercise 13 to the diagonal ∆ ⊂ Pn × Pn agrees with the Veronese
map v2 defined in Exercise 12, after having identified ∆ with Pn
in the natural way.

297
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 geometric dictionary

16. Give a coordinate-free description of the Segre map defined

in Exercise 13, as a map P(V) × P(W) → P(V ⊗ W).

17. Let p ∈ Pn , and H ⊂ Pn a hyperplane not meeting p. Define

the projection from p as the map πp : Pn \ {p} → H which sends a
point q to the intersection of the line pq and H:

πp (q) = pq ∩ H.

Show that in suitable coordinates, this is just the map sending

(x0 , . . . , xn ) to (x0 , . . . , xn−1 ). Deduce that if V ⊂ Pn is a subva-
riety not meeting p, the map πp : V → H is an algebraic map.

18. Let V ⊂ Pn be a projective variety, p ∈

/ V a point and consider
the projection map πp : V → H of the previous exercise. Show
that the image πp (V) is a projective subvariety. (For a point q ∈ H
show, using coordinates, that the line l = pq meets V if and only
if every pair of polynomials f, g ∈ I(V) have a common zero on l,
then use resultants.)

19. A quasiprojective variety is a Zariski open set in a projective

variety. Show that affine and projective varieties are quasiprojec-
tive, and that a quasiprojective is covered by open sets that are
affine varieties. Use the last fact to give a definition of morphism
between quasiprojective varieties, by reducing to the affine case.

20. Let V, W be quasiprojective varieties, as in the previous exer-

cise, and assume that V is irreducible. A rational map f : V 99K W
is an equivalence class of maps U → W, where U ⊂ V is open
and dense, and two maps are equivalent if they agree on the in-
tersection of their domains. Show that this notion is well-defined
and that every rational map is defined on a maximal open set.
Can you give examples of rational maps that are not morphisms?

21. Let V be an irreducible affine variety. Show that the set of

rational maps V 99K A1 (see Exercise 20) can be identified with
the fraction field of the coordinate ring R(V).

The following exercises, up to Exercise 26, define the Grass-

mann varieties, or Grassmannians, which are one of the most im-
portant family of projective varieties, and study their basic prop-
erties.

298
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 8.10 exercises

22. Let V be an n-dimensional vector space over a field k. For

m 6 n, define the set

G(m, V) := {H ⊂ V subspace of dimension m}.

Given a subspace H ∈ G(m, V), choose a basis v1 , . . . , vm of H.

Show that the tensor

v1 ∧ · · · ∧ vm ∈ Λm V

is determined, up to scalar multiples, by H only, hence it is well-

defined a map
ψ : G(m, V) → P(Λm V),
which is called the Plücker map. Show that the Plücker map is
injective. Finally, identify G(1, V) with P(V).
23. Let V be an n-dimensional vector space, m 6 n and let φ ∈
Λm V. Show that φ is the product of elements of V if and only if
the rank of the linear map

mφ : V → Λm+1 V
v → φ∧v
is at most n − m. Deduce that the image of G(m, V) via the
Plücker map is a closed subvariety of P(Λm V), in fact a deter-
minantal variety. By the Plücker embedding, the Grassmannians
take the structure of a projective variety.
24. Let k be a field not of characteristic 2, V a finite-dimensional
vector space over k. Let φ ∈ Λ2 V. Show that φ is the product
of two vectors of V if and only if φ ∧ φ = 0. Deduce that the
Grassmannian G(2, V) is defined by quadratic equations.
25. Let V be a finite dimensional vector space and consider the
Grassmannian G(m, V) – this can be identified with the set of m −
1-dimensional projective subspaces of P(V). Define the universal
family

U(m, V) := {(H, p) | p ∈ H} ⊂ G(m, V) × P(V).

Show that a point (H, p) ∈ U(m, V) if and only if

v1 ∧ · · · ∧ vm ∧ w = 0,

299
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 geometric dictionary

where v1 , . . . , vm span H and w spans p. Deduce that U(m, V)

is a closed subvariety of G(m, V) × P(V) (the latter inherits the
structure of variety via the Plücker and Segre embeddings).
26. The universal family U(m, V) described in the previous exer-
cise has two natural morphisms given by projections, U(m, V) →
G(m, V) and U(m, V) → P(V). Describe the fibers of these mor-
phisms. We have not studied the dimension formally yet, but
what do you think is the dimension of U(m, V)? And the dimen-
sion of G(m, V)?
27. Consider the subset
B0 (An ) := {(x, y) | xi yj = xj yi } ⊂ An × Pn−1 .

The product An × Pn−1 is a quasiprojective variety (Exercise

19) via the Segre embedding (Exercise 13). Show that B0 (An ) is
a closed subvariety of An × Pn−1 , hence it is itself a quasiprojec-
tive variety. B0 (An ) is called the blow-up1 of An in the origin.
28. Let B0 (An ) be the blow-up of An in the origin, defined in
the previous exercise, and consider the morphisms given by pro-
jection π1 : B0 (An ) → An and π2 : B0 (An ) → Pn−1 . Show that
π1 is bijective outside the origin, hence it admits an inverse ratio-
nal map (Exercise 20) An 99K B0 (An ). What is the fiber of π1
over the origin? Show that the fibers of π2 are all lines – can you
link this to the universal family of Exercise 25?
For the next exercises, we need some definitions. Let V be an
affine (resp. projective) variety of dimension k. Say V ⊂ An
(resp. V ⊂ Pn ) so that its codimension is n − k. We say that
V is a complete intersection if the ideal I(V) is generated by n − k
elements (resp n − k homogeneous elements). We say that V
is a set-theoretic complete intersection if there exists an ideal (resp.
a homogeneous ideal) I generated by n − k elements such that
V(I) = V. Clearly a complete intersection is also a set-theoretic
complete intersection, in general the second condition is weaker.
Even if we do not have a formal definition of dimension, for the
exercises assume that the rational normal curve has dimension 1,
while points have dimension 0.
1 Actually, the term blow-up is kind of a mis-translation. The original italian term
is scoppiamento, which can be translated as blow-up, but in its original form is
more akin to “decoupling”.

300
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 8.10 exercises

29. Let V ⊂ An be a finite set of points. Show that V is a set-

theoretic complete intersection. Find an example of a finite set in
A2 that is not a complete intersection.

30. Let c : P1 → P3 be the rational normal curve; its image C is

called the twisted cubic. Find explicit equations for C and show
that C is not a complete intersection, but it is a set-theoretic com-
plete intersection. Also, the intersection of C with each affine
chart of P3 is an affine curve that is a complete intersection. (To
prove that C is a set-theoretic complete intersection, start from 3
equations for C and combine two of them with a suitable linear
combination with coefficients in k[x0 , . . . , x3 ].)
31. This exercise requires some model theory. Give an alternative
proof of the Ax-Grothendieck theorem as follows. First, prove it
for finite fields, and deduce that the result holds for the algebraic
closure of a finite field. To get the proof in general, consider the
axioms for an algebraically closed field of characteristic p on the
language {0, 1, +, ·} of rings. Up to isomorphism, there is at most
one algebraically closed field of given characteristic and cardi-
nality. From the Löwenheim–Skolem theorem, it follows that the
theory of algebraically closed fields of characteristic p is complete.
Since the Ax-Grothendieck theorem is a first order statement, use
Gödel’s completeness theorem to prove that it holds for all alge-
braically closed fields of characteristic p. By a similar reasoning,
derive the result in characteristic 0.

301
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 9
DIMENSION THEORY

In this chapter, we revisit the notion of dimension, that was in-

troduced from a geometric point of view in Section 8.7. Here,
our focus becomes more algebraic, since we want to extend these
ideas to rings that do not appear necessarily as coordinate rings
of algebraic varieties. In particular, in the earlier chapter we were
concerned with finitely generated algebras over an algebraically
closed field, while many rings of interest do not even contain a
field.

We will start with some basic definitions of dimension of rings,

which we soon generalize to modules, since it easier to develop
the dimension theory of rings and modules at the same time.
We then go on to prove the main results about Hilbert functions,
which will be the basic technical tool to prove that all definitions
that we give are equivalent. Once we have proved this impor-
tant result, we introduce the notion of height of ideals, which is
a slight generalization of Krull dimension, and use the theory
developed so far to prove the famous Krull Hauptidealsatz. In
the rest of the chapter, we study the nonlocal case. We derive
the main properties of Krull dimension, and its behaviour under
quotients, polynomial extensions and integral extensions, as well
as investigate the dimension of graded rings. A good reference
for the topics of this chapter and the next one is [Ser00].

303
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 dimension theory

9.1 dimension of rings and modules

Definition 9.1.1. Let A be a ring. The Krull dimension of A, de-
noted dim A, is the maximum length of a chain of prime ideals
P0 ⊂ P1 ⊂ · · · ⊂ Pn
contained in A.
This is of course a direct generalization of Definition 8.7.5.
Remark 9.1.2. Since every prime ideal contains the nilradical, it
is clear that dim A = dim A/N(A).
We have already met some examples of rings of small Krull
dimension.
Example 9.1.3. (a) A ring A has dimension 0 if and only if ev-
ery prime ideal is maximal. This holds for Artinian rings
by Corollary 2.4.2. The converse holds assuming that A is
Noetherian, by Theorem 2.4.11
(b) Let A be an integral domain, so that 0 is prime. Then dim A =
1 if and only if every prime other than 0 is maximal. This
holds for Dedekind rings by Theorem 5.1.19.
The next definition of dimension will only make sense in cer-
tain cases. We introduce the class of rings where the theory works
nicely.
Definition 9.1.4. Let A be a ring. We say that a is semilocal if A
has only finitely many maximal ideals.
This class includes local rings, as well as Artinian rings, by
Proposition 2.4.3.
Definition 9.1.5. Let A be a semilocal ring with maximal ideals
M1 , . . . , Mn , and let
M = M1 ∩ · · · ∩ Mn
be its Jacobson radical. The minimum number d (if any) for
which there exist a1 , . . . , ad ∈ M such that the length
 
A
`
(a1 , · · · , ad )

304
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 9.1 dimension of rings and modules

is finite is called the Chevalley dimension of A, and denoted (tem-

porarily) by δ(A).

This definition is in fact temporary, since we will see that this

number agrees with dim A.

Remark 9.1.6. Let J = (a1 , · · · , ad ). Then `(A/J) < ∞ implies

that the sequence M J+J must be stationary, hence Mk ⊂ J ⊂ M.
k

If A is in fact local, this implies that J is M-primary by Propo-

sition 3.2.6. Vice versa, if A is Noetherian and local with Mk ⊂
J ⊂ M, A/J has finite length by Corollary 2.5.17.
It follows that for a local Noetherian ring A, δA is the minimum
number of elements a1 , . . . , ad ∈ M such that Mk ⊂ (a1 , . . . , ad )
for some k > 1, or in other words, the minimum number of
generators of an M-primary ideal.

The above remark should make clear the connection with Defi-
nition 8.7.7. We extrapolate the condition, since it is useful termi-
nology.

Definition 9.1.7. Let A be a local ring with maximal ideal M.

Elements a1 , . . . , ad ∈ M are called a system of parameters for A if
the ideal (a1 , . . . , ad ) is M-primary, and moreover d is minimal,
that is, d = δ(A).

We can generalize the above definitions to modules, in such

a way that the theorem on dimension will be valid in this more
general setting.

Definition 9.1.8. Let A be a ring, M an A-module. The Krull

dimension of M is defined as dim M := dim A/ Ann(M).

Remark 9.1.9. When talking about a module of the form M =

A/I, there may be an ambiguity in the definition of dim(M), since
it can be considered as a module over A or as a ring. In fact, there
is no ambiguity, as the two definitions agree. This follows with a
moment’s thought from the fact that Ann(A/I) = I.

Similarly we can generalize the Chevalley dimension.

Definition 9.1.10. Let A be a semilocal ring, and let

M = M1 ∩ · · · ∩ Mn

305
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 dimension theory

be its Jacobson radical. If M is an A-module, we define the Cheval-

ley dimension of M – denoted δM – as the least number d such
that there exist a1 , . . . , ad ∈ M such that the length
 
M
`
a1 M + · · · + ad M
is finite.
Remark 9.1.11. If A is Noetherian and M is finitely generated,
then δM is finite. In fact, if a1 , . . . , ad is any system of parameters
of A and I = (a1 , . . . , ad ), A/I is Artinian, hence M/I · M has
finite length.

9.2 hilbert functions

In order to make Definition 8.7.8 precise, and generalize it to a
wider class of rings, we need to understand the growth of the size
of the homogeneous components of graded rings and modules.
L∞
Remark 9.2.1. L∞ Let A = i=0 An be a graded Noetherian ring,
and M = i=0 Mn a finitely generated graded A-module. Then
each Mn is finitely generated as an A0 -module.
In fact, let x1 , . . . , xs be homogeneous generators of A as an
A0 -algebra (these exist by Proposition 1.7.15). Let m1 , . . . , mt
be homogeneous generators of M as an A-module, of degrees
d1 , . . . , dt . Then every m ∈ Mn can be written as
m = y1 m 1 + · · · + yt m t ,
where yi ∈ An−di . Each yi is a polynomial in the xj , hence Mn
is generated by the products xa1 as
1 · · · xs mi having total degree n.
Assume now that A0 is in fact Artinian. Then each compo-
nent Mn has finite length `(Mn ) and we can put these lengths
together in a generating function.
L∞
Definition 9.2.2. Let AL= i=0 An be a graded ring with A0
∞
Artinian, and let M = i=0 Mn be a finitely generated graded
A-module. The Hilbert-Poincaré series of M is the formal series
∞
X
P(M, t) = `(Mn )tn ∈ Z[[t]].
n=0

306
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 9.2 hilbert functions

The starting point is the following observation, that was proved

by Hilbert in the polynomial setting, and later extended by Serre
in this generality.
Proposition 9.2.3 (Hilbert, Serre). Let A be a graded ring, and as-
sume that A0 is Artinian, and let M be a finitely generated graded
A-module. The Hilbert-Poincaré series P(M, t) is rational, more pre-
cisely
f(t)
P(M, t) = Qs ki
,
i=1 (1 − t )
where f ∈ Z[t] is a polynomial, and A = A0 [x1 , . . . , xs ] with xi
homogeneous of degree ki .
Proof. By induction on s. When s = 0, we have An = 0 for n > 1.
Since M is finitely generated over A, we must have Mn = 0 for
all large n, hence P(M, t) is in fact a polynomial.
For the inductive step, consider the multiplication map

fs : M → M,
m → xs · m
which gives rise to an exact sequence

0 → Kn → Mn → Mn+ks → Ln+ks → 0,

where K and L are the kernel and cokernel respectively.

Both K and L are finitely generated graded A-modules, and
they are annihilated by xs , hence they are finitely generated over
A 0 = A[x1 , . . . , xs−1 ]. The additivity of the length gives

`(Kn ) − `(Mn ) + `(Mn+ks ) − `(Ln+ks ) = 0,

and multiplying by tn+ks and summing over n gives

tks P(K, t) − tks P(M, t) + P(M, t) − P(L, t) = g(t),

where g(t) ∈ Z[t] is a polynomial. We can obtain

1 
ks

P(M, t) = −t P(K, t) + P(L, t) + g(t)
(1 − tks )

and we conclude by the inductive hypothesis.

307
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 dimension theory

Remark 9.2.4. The above proof goes unchanged – without assum-

ing that A0 is Artinian – for every integer-valued function λ(M)
over finitely generated A0 -modules having the property of being
additive over exact sequences.

Definition 9.2.5. Under the hypothesis of the above Proposition,

the order of pole of P(M, t) at t = 1 is called the Poincaré dimen-
sion (or Samuel dimension) of M and denoted d(M).

Corollary 9.2.6. Let A, M as in Proposition 9.2.3 and assume that

each xi is homogeneous of degree 1. Then `(Mn ) = p(n) for all n large
enough, where p(n) ∈ Q[n] is a polynomial function of degree d − 1,
and d is the Poincaré dimension of M.

Proof. Under the hypothesis we have

f(t) g(t)
P(M, t) = s
=
(1 − t) (1 − t)d

6 0. The generalized binomial

after simplification, so that g(1) =
theorem allows us to write
X∞  
1 d+n−1 n
= t ,
(1 − t)d d−1
n=0

with the convention that −1 m

−1 = 1 and −1 = 0 for m > 0.
Writing g(t) = a0 + a1 t + · · · + ar tr , we get

X
r 
d+n−k−1

`(Mn ) = ak
d−1
k=0

for all n such that d + n − r − 1 > 0, and this is a polynomial in

n of degree at most d. To check that the degree is exactly d, we
compute the first coefficient, which is

X
r
1 g(1)
ak = 6= 0.
(d − 1)! (d − 1)!
k=0

308
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 9.2 hilbert functions

Example 9.2.7. Let A = A0 [x1 , . . . , xs ], where the xi are indeter-

minates. Then An is generated by all monomials xa 1 as
1 · · · xs such
that a1 + · · · as = n. Letting `0 = `(A0 ), we have
n+s−1
∼ A( s−1 ) ,
An = 0

hence `(An ) = `0 n+s−1

s−1 .

Let now A be a semilocal ring, with Jacobson radical M, and

let M be a finitely generated A-module. Any ideal I such that
Mk ⊂ I ⊂ M is called a definition ideal of A. In this case, we can
consider the graded ring A 0 = GrI (A) and the graded A 0 -module
M 0 = GrI (M).
Since A00 = A/I is Artinian, the above results apply. In partic-
ular, if I = (x1 , . . . , xs ), we can write A 0 = A00 [ξ1 , . . . , ξs ], where
ξi is the class of xi in A 0 .
Definition 9.2.8. The Hilbert function of M at I is

X
n
0
χIM (n) := `(Mn ) = `(M/In M),
i=0

By Corollary 9.2.6, χIM (n) agrees for all large n with a rational
polynomial of degree d, where d is the Poincaré dimension of M.
This polynomial is called the Hilbert polynomial of M at I.
Remark 9.2.9. As in Remark 9.1.9, one may wonder whether the
definition of Poincaré dimension is ambiguous for a module of
the form M = A/I, which can be seen as a module over A or as a
ring. Since d(M) is the degree of the Hilbert polynomial, which
can be computed as `(M/In M) for n large, and since this length
does not depend on whether we see M as a module over A or
over itself, the definition of Poincaré dimension agrees in the two
cases.
Proposition 9.2.10. The degree of χIM (n) does not depend on the
choice of I.
Proof. If J is another definition ideal, we have Ia ⊂ J and Jb ⊂ I
for suitable integers a, b. It follows that

χIM (a(n + 1) − 1) > χJM (n)

309
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 dimension theory

and conversely

χJM (b(n + 1) − 1) > χIM (n).

This implies that the two polynomials χIM and χJM have the same
order of growth.
The following result has a similar proof, but in fact it is much
more subtle due to the use of the Artin-Rees lemma.
Proposition 9.2.11. Let A be a semilocal Noetherian ring, I a defini-
tion ideal of A, and let

0 → M0 →M → M 00 →0

be a short exact sequence of finitely generated A-modules. Then χIM −

χIM 00 and χIM 0 have the same degree and first coefficient.
Proof. It is easy to check that

M 00 M0
     
M
` n = ` n 00 + ` n ,
I M I M I M ∩ M0

so χIM = χIM 00 + ψ(n) for

M0
 
ψ(n) := ` .
I M ∩ M0
n

By Lemma 7.5.15, we know that In M ∩ M 0 and In M 0 have

bounded difference, hence ψ is a polynomial of the same degree
and first coefficient as χIM 0 .

9.3 the main theorem on dimension

In this section, we are going to prove that all definitions that we
have given for dimension actually agree with each other.
Theorem 9.3.1 (Dimension theorem). Let A be a semilocal Noether-
ian ring, M a finitely generated A-module. Then

dim(M) = δ(M) = d(M).

310
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 9.3 the main theorem on dimension

Hence, after this section, we will not need to distinguish be-

tween Krull, Chevalley and Poincaré dimension, and we will just
talk about the dimension of a module. We will prove this the-
orem as a consequence of three inequalities. The first one, we
prove separately in the ring case.

Proof that d(A) > dim(A). By induction on d = d(A). Let M be

the Jacobson radical of A. When d = 0, χM n
A (n) = `(A/M ) is
eventually constant, hence Mn = Mn+1 for n big enough. By
Nakayama’s lemma 1.3.19, Mn = 0. If P ⊂ A is a prime ideal,
since 0 ⊂ P, it follows that M ⊂ P. Writing M = M1 ∩ · · · ∩ Mk ,
where the Mi are maximal, we get that P ⊃ Mi for some i by
Proposition 1.1.20, so dim(A) = 0.
For the induction step, we can assume that dim(A) > 0. Let

P0 ( P1 ( · · · ( Pr

be a chain of primes. Choose x ∈ P1 \ P0 and consider the exact

sequence

A ·x A A
0 → → → → 0.
P0 P0 (x) + P0

Denote B = A/((x) + P0 ). Proposition 9.2.11 implies that d(B) <

d(A), hence we can apply the inductive hypothesis to get

dim(B) 6 d(B) 6 d(A) − 1.

But in B we have the chain

P1 ( P2 ( · · · ( Pr ,

which implies that r − 1 6 d(A) − 1, that is, r 6 d(A).

Since d(A) is finite, we can already state a corollary of the first

part of the proof, which we will use in the sequel.

Corollary 9.3.2. Let A be a semilocal Noetherian ring. Then dim(A)

is finite – in particular, every descending chain of prime ideals is
stationary.

311
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 dimension theory

Remark 9.3.3. The Krull dimension is defined for arbitrary rings,

and the above Corollary is not valid for a general ring, not even a
Noetherian one. The first example of a Noetherian ring of infinite
Krull dimension was given in [Nag62, Appendix, example E1],
as follows.

Example 9.3.4 (Nagata). Take integers mi such that the differ-

ences mi+1 − mi are increasing. In the ring A := k[x1 , x2 , . . . ]
take the prime ideals Pi := (xmi +i , . . . , xmi+1 ). Let S be the com-
plement of the union of the Pi . Then the localization S−1 A is Noe-
therian and has infinite dimension. You will prove both claims in
Exercises 18-22.

Having proved the first inequality for rings, we consider the

module case.

Proof that d(M) > dim(M). By Theorem 2.5.16, we find a chain

0 ( M1 ( M2 ( · · · ( Mq = M, (9.3.1)

where Mi+1 /Mi = ∼ A/Pi , and Pi ∈ Ass(M) is an associated

prime of M. We will prove the result by linking the dimension of
M and that of the rings A/Pi .
To get there, let

0 → M0 →M → M 00 →0

be an exact sequence of A-modules. By Proposition 3.3.3, if P ⊃

Ann(M) is a prime ideal of A, then either P ⊃ Ann(M 0 ) or P ⊃
Ann(M 00 ). Hence if we have a chain of primes,

Ann(M) ⊂ P0 ( P1 ( · · · ( Pr ,

we have either Ann(M 0 ) ⊂ P0 or Ann(M 00 ) ⊂ P0 . In conclusion,

dim(M) = max{dim(M 0 ), dim(M 00 )}.

A similar reasoning can be done for the Poincaré dimension. By

Proposition 9.2.11, χM − χM 00 has the same leading coefficient as
χM 0 , hence either χM 0 or χM 00 has the same degree as χM , which
implies
d(M) = max{d(M 0 ), d(M 00 )}.

312
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 9.3 the main theorem on dimension

Using (9.3.1), we obtain

dim(M) = max {dim(A/Pi )}

d(M) = max {d(A/Pi )} ,

and we can conclude since the A/Pi are rings.

This is probably the hardest inequality, the next ones don’t

need to treat the ring case separately.

Proof that dim(M) > δ(M). Again, we use induction on dim(M).

When dim(M) = 0, we have dim(A/ Ann(M)) = 0, which means
that every prime in that ring is in fact maximal. This implies that
A/ Ann(M) is Artinian, and since M is finitely generated over
A/ Ann(M), it is an Artinian and Noetherian module. Hence M
has finite length, and δ(M) = 0.
For the inductive step, consider a maximal chain of prime ide-
als containing Ann(M). This means that the smallest one is a min-
imal prime of M. By primary decomposition for modules, M has
finitely many minimal primes, say P1 , . . . , Pt . Since dim(M) > 0,
these are not maximal.
We claim that this allows us to find some elements x ∈ M such
that x ∈/ Pj for any j, where M is the Jacobson radical. In fact,
write M = M1 ∩ · · · ∩ Mk . By Proposition 1.1.20,

Mi 6⊂ P1 ∪ · · · ∪ Pt ,

so we can find yi ∈ Mi such that yi ∈/ Pj for any j. Then, taking

x := y1 · · · yk will do.
If we take M1 = M/xM, then Ann(M1 ) ⊃ (x) + Ann(M).
In particular, no minimal prime Pj contains Ann(M1 ), and a
chain for M1 must be shorter than one for M. This implies that
dim(M1 ) < dim(M). Also, by construction δ(M) 6 δ(M1 ) + 1,
since M1 is just the quotient of M by an element of M. By induc-
tion,
δ(M) 6 δ(M1 ) + 1 6 dim(M1 ) + 1 6 dim(M).

We finally turn to the last inequality in the theorem.

313
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 dimension theory

Proof that δ(M) > d(M). Again, we do induction on δ(M). When

δ(M) = 0, M has finite length. In this case, the polynomial
 
M
χM (n) = ` 6 `(M)
Mn M
is bounded, hence it must have degree 0, so d(M) = 0.
For the inductive step, assume δ(M) = s, and take x1 , . . . , xs ∈
M such that  
M
` < ∞.
x1 M + · · · + xs M
For i = 1, . . . , s, call Mi = M/(x1 M + · · · + xi M). It is a simple
verification that δ(Mi ) = s − i.
We are going to relate χM and χM1 . To do this, observe that
M1 ∼ M
= n ,
Mn M1 M M + x1 M
so
       
M1 M M x1 M
` = ` = ` − ` .
Mn M1 Mn M + x1 M Mn M Mn M ∩ x1 M
On the other hand, we have a surjective homomorphism
·x1 x1 M
M → x1 M → ,
M M ∩ x1 M
n

whose kernel is (Mn : x1 ). Hence

x1 M M
= ,
Mn M ∩ x1 M (Mn : x1 )
and      
M1 M M
` =` −` .
Mn M1 Mn M n
(M : x1 )
Finally, the inclusion Mn−1 ⊂ (Mn : x1 ) allows us to simplify
this to
     
M1 M M
` > ` − ` .
Mn M1 Mn M Mn−1 M
This just reads χM1 (n) > χM (n) − χM (n − 1), hence χM1 has
smaller degree than χM . By induction,
d(M) 6 d(M1 ) + 1 6 δ(M1 ) + 1 = δ(M).

314
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 9.4 height

9.4 height
Let A be a Noetherian ring, P ⊂ A a prime ideal. By Corollary
9.3.2, every descending chain of prime ideals

P = Pn ⊃ Pn−1 ⊃ · · · ⊃ P0 (9.4.1)

must terminate, and in fact has length bounded by dim AP , which

is finite.

Definition 9.4.1. The maximum length of a chain as in (9.4.1) is

called the height of the prime ideal P, denoted ht(P). By Proposi-
tion 1.6.6, we have ht(P) = dim AP .

The height of a prime ideal is a more refined invariant than the

dimension, in the sense that dim A = supP⊂Aprime ht(P). It turns
out that this is not only finite, but we can give a precise bound in
terms of the number of generators.

Theorem 9.4.2. Let A be a Noetherian ring, I ⊂ A an ideal and P a

minimal prime of I. If I = (a1 , . . . , ar ), then ht(P) 6 r.

Proof. First, notice that the only prime ideal containing IAP is
PAP . In fact, let Q be a prime of A such that IAP ⊂ QAP . This
means that I ⊂ Q ⊂ P, and since P is minimal, we must have
Q = P. √
In particular, IAP = PAP , which is maximal, so IAP is PAP -
primary. This means that IAP is a definition ideal of AP , so
dim AP = dAP 6 r.

The theorem was firsdt discovered in the following restricted

form as Krull’s Hauptidealsatz or Krull’s principal ideal theo-
rem.

Corollary 9.4.3 (Krull). Let A be a Noetherian ring, I ⊂ A a principal

ideal, say I = (x) where x is not a unit or a divisor of zero. Let P be a
minimal prime of I. Then ht(P) = 1.

Proof. The fact that ht(P) 6 1 is a special case of the theorem.

Assume that ht(P) = 0. If I ( A, then P is a minimal prime of 0,
and by Corollary 3.2.22, the union of the primes belonging to 0 is
the set of zero divisors.

315
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 dimension theory

Remark 9.4.4. The assumption that the ring is Noetherian is fun-

damental for Krull’s Hauptidealsatz – see Exercise 11 for a coun-
terexample in the non-Noetherian case.
We can also state a partial converse. Taken together, these re-
sult give a strong relationship between the height of ideals and
the number of generators.
Proposition 9.4.5. Let P be a prime of height r in the Noetherian ring
A.
i) P is the minimal prime of an ideal I = (a1 , . . . , ar ).
ii) If b1 , . . . , bs ∈ P, then ht(P/(b1 , . . . , bs )) > r − s.
iii) If a1 , . . . , ar are as in i and s 6 r, then ht(P/(a1 , . . . , as )) =
r − s.
Proof.
i) AP is a local ring of dimension r, hence it has a PAP -primary
ideal generated by r elements. Call it IAP = (α1 , . . . , αr ).
Write αi = a si , where ai ∈ P. Then P is a minimal prime of
i

I = (a1 , . . . , ar ).
ii) Let A := A/(b1 , . . . , bs ), P := P/(b1 , . . . , bs ), and denote
t = ht(P). By the first point, P is the minimal prime of an
ideal I = (c1 , . . . , ct ). Then P is a minimal prime of I :=
(b1 , . . . , bs , c1 , . . . , ct ), and by Theorem 9.4.2 ht(P) 6 s + t.
iii) One inequality is the previous point. The converse comes
from the fact that P/(a1 , . . . , as ) is a minimal prime of the
ideal (as+1 , . . . , ar ) in A/(a1 , . . . , as ).

The form in which we state Krull’s principal ideal theorem

may look a little awkward. One may wonder whether a minimal
prime of a principal ideal is itself principal, allowing to simplify
the statement. In fact, this is true only in a very special case.
Proposition 9.4.6. Let A be a ring. The following are equivalent:
i) for every principal ideal I ⊂ A and minimal prime P of I, P is also
principal

316
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 9.5 properties of dimension

ii) A is a unique factorization domain.

Proof. Assume that A is a unique factorization domain, take a ∈
A and let I = (a). Factorize a = pe11 · · · per r . If P is a minimal
prime of I, P must contain one of the pi , hence P = (pi ) by
minimality.
For the converse, assume i). If Q ⊂ A is a prime ideal, take any
a ∈ Q and let P be a minimal prime of A. By the hypothesis, P
is principal, and we conclude the proof by Kaplansky’s theorem
3.1.20.

9.5 properties of dimension

In this section, we take a closer look to the dimension of rings,
and in particular we study how the dimension changes under
various operations.
Remark 9.5.1. In a sense, we have already discussed the case of
localization by a prime ideal. By definition, dim AP = ht(P), and
we have already the Hauptidealsatz and its converse that allow
us to control the height of ideals.
The first thing we look at is the behaviour of dimension under
quotients. It is not always well-behaved, but there are simple
cases where we can control the drop in dimension.
Proposition 9.5.2. Let A be a local Noetherian ring with maximal ideal
M, x ∈ M, and assume that x is not a zero divisor. Then dim(A/(x)) =
dim A − 1.
Proof. We have the exact sequence

0 → (x) →A → A/(x) →0

and we know by Proposition 9.2.11 that χA − χA/(x) and χ(x)

have the same first coefficient.
On the other hand, since x is not a zero divisor, A and (x)
are isomorphic as A-modules, which implies that deg χA/(x) <
deg χA , so dA/(x) < dA.
The reverse inequality is clear by using the Chevalley dimen-
sion: if a1 , . . . , ad ∈ A and (a1 , . . . , ad ) is M/(x)-primary in

317
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 dimension theory

A/(x), then (x, a1 , . . . , ad ) is M-primary in A, hence δ(A) 6

δ(A/(x)) + 1.
Another useful property is that the dimension does not change
under completion, as the geometric intuition suggests.
Proposition 9.5.3. Let A, M be a local Noetherian ring, A
b its comple-
tion in the M-adic topology. Then dim A = dim A.
b

Proof. This follows at once from Corollary 7.5.18, since A and A

b
have the same associated graded ring, hence the same Hilbert
polynomial.
Next, we take on studying the dimension of polynomial rings.
For that, we need a lemma, which is an useful result of its own.
Lemma 9.5.4. Let f : A → B be a homomorphism of Noetherian rings.
Take a prime Q ⊂ B and let P = f−1 Q be the contraction. Then
BQ
ht(Q) 6 ht(P) + dim .
PBQ
Proof. By localization, we can assume that A is local with maxi-
mal ideal P and B is local with maximal ideal Q, in which case
the thesis reads
B
dim B 6 dim A + dim .
PB
Let r = dim A, and s = dim B/(PB). Then by Proposition 9.4.5,
P is minimal over an ideal I = (a1 , . . . , ar ) and Q/(PB) is minimal
over an ideal (b1 , . . . , bs ). The quotient Q/(PB + (b1 , . . . , bs )) is
both maximal and minimal inside B/(PB + ((b1 , . . . , bs )), hence
it is nilpotent (why?). It follows that
Qn ⊂ PB + (b1 , . . . , bs )
for n big enough. By a similar reasoning,
Pm ⊂ (a1 , . . . , ar )
for m big enough, and so
Qm+n ⊂ J := (f(a1 ), . . . , f(ar ), b1 , . . . , bs ).
This implies that Q is a minimal prime of J, and the thesis follows
by Theorem 9.4.2.

318
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 9.5 properties of dimension

This allows us to study the relation between a ring and its

polynomial extension. For Noetherian rings, we have a very neat
result.
Proposition 9.5.5. If A is a Noetherian ring, dim A[x] = dim A + 1.
Proof. The inequality dim A[x] > dim A + 1 is easy. If

P0 ( P1 ( · · · ( Pd

is a chain of primes in A, we get ideals Pi [x] ⊂ A[x]. These

are primes, since A/Pi [x] is a domain, hence dim A[x] > dim A.
Moreover, we have a strict inequality, since Pd [x] is not maximal,
as A/Pd [x] is not a field.
For the converse, let Q be a maximal ideal of A[x], and P =
Q ∩ A. By Lemma 9.5.4,

A[x]Q
ht(Q) 6 ht(P) + dim ,
PA[x]Q

hence it is enough to prove that the last term is at most 1. Now

 
A
[x] = k[x]Q ,
P Q

where k is the fraction field of A/P. But dim k[x] = 1, and local-
ization at Q can only lower the dimension.
The same proof also gives the following result (Exercise 7).
Proposition 9.5.6. If A is Noetherian ring, dim A[[x]] = dim A + 1.
Corollary 9.5.7. If k is a field, dim k[x1 , . . . , xn ] = n.
Proposition 9.5.5 does not hold for rings that are not Noether-
ian, but still we can state a partial result.
Proposition 9.5.8. Let A be a ring. Then

dim A + 1 6 dim A[x] 6 2 dim A + 1.

Proof. The first inequality follows as in Proposition 9.5.5, as its

proof does not use the Noetherian hypothesis. The second one
follows at once if we prove that there do not exist three prime

319
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 dimension theory

ideals Q1 ( Q2 ( Q3 ⊂ A[x] having the same restriction to A. In

fact, any chain of prime ideals in A[x] can have at most twice the
length of a chain in A.
To prove the claim, assume

Q1 ∩ A = Q2 ∩ A = Q3 ∩ A = P.

By taking the quotient at P, we can assume that A is domain and

P = 0. Localizing at Q3 we get a chain of three prime ideals
inside A[x]Q3 . This ring is a further localization of k[x], where
k = F(A) is the fraction field of A. This is a contradiction, since
dim k[x] = 1.

In fact, this is not a limitation of our techniques: all intermedi-

ate dimensions can actually appear, as shown in [Sei53]. We only
give a simple example.

Example 9.5.9. [Sla] Let t, y be two indeterminates, k a field, and

take the ring

A := {f(y) ∈ k(t)[[y]] | f(0) ∈ k}.

The ideal
P := {f(y) ∈ k(t)[[y]] | f(0) = 0} ⊂ A
is prime, and A/P = ∼ k by evaluation, so in fact P is maximal. In
fact, P is the only nontrivial prime ideal of A (why?), so dim A =
1.
On the other hand, dim A[x] = 3. To see this, consider the map

evt : A[x] → k(t)[[y]].

p(x) → p(t)

Since, k(t)[[y]] is a domain, Q = ker evt is a prime ideal, and

clearly yx − yt ∈ Q, so Q 6= 0. On the other hand, Q ⊂ P[x], and
the inclusion is strict since y ∈ P, but y ∈
/ Q. This show that we
have the chain of 4 prime ideals

0 ( Q ( P[x] ( M,

where M is any maximal ideal containing P[x]. Hence, dim A[x] >
3, and the reverse inequality is Proposition 9.5.8.

320
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 9.6 dimension of graded rings

Now that the picture for polynomial extensions is clear, we

turn to the opposite case: the dimension of integral extensions.
Here things are much simpler. In fact, the going-up theorem
5.2.6 immediately gives:

Proposition 9.5.10. Let A ⊂ B be rings, with B integral over A. Then

dim A = dim B.

Putting the two results together, we are finally able to conclude

that our definition of transcendence dimension 8.7.6 also agrees
with the other definitions we have given. The precise statement
is the following.

Theorem 9.5.11. Let k be a field and A a finitely generated k-algebra

which is an integral domain. Let K = F(A) be the fraction field of A.
Then the transcendence degree of K over k equals dim A.

Proof. Let s be the transcendence degree of K over k. By Noether

normalization lemma 5.3.1, we can find algebraically indepen-
dent a1 , . . . , as ∈ A such that A is integral over k[a1 , . . . , as ].
But dim k[a1 , . . . , as ] = s, and integral extensions preserve the
dimension.

9.6 dimension of graded rings

If we track our definition of Poincaré dimension, it starts by con-
sidering the lengths of homogeneous components in a graded
ring A, where A0 is Artinian. In turn, we were lead to consider
the special case of the graded ring associated to a local Noether-
ian ring.
Now that the picture for local Noetherian rings is complete
thanks to Theorem 9.3.1, it makes sense to go back to investigate
graded rings. In particular, one may wonder whether the dimen-
sion of graded rings can be expressed in terms of its Hilbert poly-
nomial, and whether the dimension of a local Noetherian ring
and its associated graded ring are related. As it turns out, both
things are true.
First, we establish a few results that allow us to only consider
homogeneous ideals when dealing with the Krull dimension of
graded rings.

321
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 dimension theory

Proposition 9.6.1. Let A be a graded ring, I ⊂ A a homogeneous ideal,

and P a minimal prime of I. Then P is homogeneous.

Proof. We can write P = Ann(x) for some x ∈ A/I. Since A/I

is a graded ring, we can decompose x = x0 + x1 + · · · + xd into
its homogeneous components. Choose any a ∈ P, and again
decompose a = a0 + a1 + · · · + ar . The condition ax = 0 becomes

a0 x0 = 0
a1 x0 + a0 x1 = 0
a2 x0 + a1 x1 + a0 x2 = 0
···

By multiplying each of these in turn by a0 , we recursively get

ak+1
0 xk= 0. In particular, ad+1
0 ∈ Ann(x) = P, so in fact a1 +
· · · + ar∈ P. We can then repeat the reasoning for all ai to find
that all of them lie in P.

Proposition 9.6.2. Let A be a graded ring, and P ⊂ A a homogeneous

prime of height r. Then, there exists a chain

P0 ( P1 ( · · · ( Pr = P

composed by homogeneous primes.

Proof. Let P0 be the smallest element in a maximal chain of primes

ending at P. Then P0 is a minimal prime of 0, and by Proposition
9.6.1 it is homogeneous. By taking the quotient with respect to P0 ,
we can prove the result assuming that A is an integral domain.
Take any homogeneous a ∈ P, a 6= 0. By Proposition 9.4.5,
ht(P/a) > r − 1. Since A is an integral domain, any maximal
chain ending in P has to start from 0, so in fact ht(P/a) = r − 1.
This gives us a chain

P1 ( · · · ( Pr = P,

where P1 is a minimal prime over a. By Proposition 9.6.1, P1 is

homogeneous and we can then finish the proof by induction.

Having established these basic results, we turn to the relation-

ship between a local ring and its associated graded ring.

322
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 9.6 dimension of graded rings

Theorem 9.6.3. Let k be a field, and consider a graded ring A =

k[x1 , . . . , xr ], where the xi have degree 1. Let M = (x1 , . . . , xr ), and
consider the localization AM , with maximal ideal M = MAM .

i) Let
φ(n) = `(An ) (for n large enough)
be the Hilbert polynomial of A and
 
AM
χ(n) = ` (for n large enough)
Mn

be the Hilbert polynomial of AM . Then φ(n) = χ(n) − χ(n − 1).

ii) The dimensions are equal, in fact

dim A = dim AM = deg φ + 1.

iii) As graded rings,

∼ GrM (AM ).
A=

Proof. i) There is an isomorphism

∼ M ,
n n
∼ M
An = =
Mn+1 Mn+1
hence
Mn
 
φ(n) = `(An ) = ` = χ(n) − χ(n − 1).
Mn+1

ii) By the previous point,

deg φ + 1 = deg χ = dim AM .

On the other hand, since AM is local, dim AM = ht(M) =

ht(M). The problem is, a priori A may have other maximal
ideals of greater height.
First, assume that A is an integral domain, and let t be its
transcendence degree over k. We know that t = dim A by
Theorem 9.5.11. It is not restrictive to assume that in fact
x1 , . . . , xt are algebraically independent over k. Then, the

323
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 dimension theory

monomials of a given degree n in x1 , . . . , xt are linearly in-

dependent over k. This implies
 
n+t−1
`(An ) > ,
t−1

hence deg φ > t − 1 = dim A − 1, which is what we need.

In the general case, take any maximal chain of primes in A,
terminating at P0 . Then P0 is a minimal prime of A, and by
Proposition 9.6.1 it is homogeneous. Then the inequalities

A M
dim A = dim = ht 6 ht M 6 dim A
P0 P0
show that dim A = ht M = dim AM .
iii) This follows from by verifying that the group isomorphism

∼
A=
L ∼ L Mn+1
n
∼ GrM (AM )
n An = M
=

is in fact a ring homomorphism.

Corollary 9.6.4. Let A be a local Noetherian ring with maximal ideal

M, k = A/M. Then

dim A = dim GrM (A).

Proof. Just apply the previous theorem by taking as xi the images

in M/M2 of a finite set of generators of M.

9.7 exercises
1. Let k be a field and A = k[x1 , . . . , xs ]. Let f ∈ A be a polyno-
mial of degree d. Prove that for n large enough we have
   
n+s n−d+s
`(An ) = dim An = − ,
s s

and that this is a polynomial of degree s − 1, with leading coeffi-

d
cient (s−1)! .

324
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 9.7 exercises

2. Find an explicit expression for the Hilbert-Poincaré series of

the ring k[x1 , x22 , . . . , xn
n ] (as a module over itself) as a rational
function.
The following exercises, up to Exercise 6, follow [Car80] to give
an alternative proof of Krull’s principal ideal theorem that does
not rely on the theory developed in this chapter, and especially
avoids the usage of the Artin-Rees lemma.
3. Let A be an integral domain, a, b ∈ A nonzero. Show that
∼ ((b) : (a)), and that this
there is an isomorphism ((a) : (b)) =
induces an isomorphism
((a) : (b)) ∼ ((b) : (a))
= .
(a) (b)
4. Let A be an integral domain, a, b ∈ A nonzero. Show that
there is an isomorphism
A ∼ (a, b) .
=
((a) : (b)) (a)
5. Let A be a Noetherian local domain with maximal ideal M,
and assume that there exists x ∈ A such that (x) is M-primary.
Prove that ht(M) = 1. (Assume 0 ( P ( M with P prime and
take a nonzero z ∈ P. The chain ((z) : (xn )) is eventually station-
ary. Use the previous two exercises to deduce that `(A/(xn , z)) is
eventually constant, and so (xn , z) is also stationary. From this,
find the contradiction that x ∈ P.)
6. Reduce Krull’s principal ideal theorem to the local version
proved in the previous exercise.
7. Prove in detail Proposition 9.5.6 – in particular you will need
to prove that dim k[[x]] = 1.
8. Give an alternative proof of Proposition 9.5.6 by reducing to
the local case and then showing that for a Noetherian local ring A,
the ring A[[x]] is also local and the Chevalley dimension satisfies
δ(A[[x]]) = δ(A) + 1.
9. Show that R is not the field of fractions of a Noetherian ring
other than itself. (By localization at a suitable prime and the
Krull-Akizuki theorem, you can assume that this ring is integrally
closed.)

325
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 dimension theory

10. Let An be the valuation ring with value group (Zn , LEX)
constructed in Exercise 27 of Chapter 7. Show that dim An = n,
and in fact all prime ideals of An fit into a chain

P0 ( P1 ( · · · ( Pn .

11. With the notation of the previous exercise, find a principal

ideal I ⊂ An and a minimal prime P of I having ht(P) = n. This
shows a counterexample for Krull’s Hauptidealsatz 9.4.3 when
the ring is not Noetherian. Why this does not contradict Proposi-
tion 2.2.12?

12. Let A be Noetherian local ring of dimension 2. Show that

A admits infinitely many primes of height 1. Deduce the more
general case: if A is a Noetherian ring of finite dimension d, and
there exist finitely many primes of height k, then either k = 0 or
k = d.

13. Let A be a valuation ring. Using the previous exercise and

Exercise 25 in Chapter 7, show that if dim A > 2, then A is not
Noetherian

14. Lest the previous exercise seduces you into thinking that val-
uation rings of dimension 1 are Noetherian, here is a counterex-
ample from [Knab]. Consider the rings
k
k[x] ⊂ k[x1/2 ] ⊂ · · · ⊂ k[x1/2 ] ⊂ · · ·

Abstractly, each of these rings is isomorphic to k[x], hence if we

k k
let Pk = (x1/2 ) ⊂ k[x1/2 ], the localization
k
Ak := k[x1/2 ]Pk

there are natural inclusions Ak ⊂ Ak+1 and

is a DVR. Show that S
that the union A := ∞ k=0 Ak is a valuation ring having a value
group isomorphic to the dyadic rationals
a
∈ Q | a ∈ Z, b ∈ N .
2k
Conclude that A is a non-Noetherian valuation domain of dimen-
sion 1.

326
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 9.7 exercises

15. Let A be a Noetherian ring, and consider a chain of prime

ideals
P0 ( P1 ( · · · ( Pn .
Given any a ∈ Pn , show that we can find a different chain
P00 ( P10 ( · · · ( Pn−1
0
( Pn
ending at Pn , such that a ∈ P1
16. Let S be the multiplicative set generated by x and the cy-
clotomic polynomials. Prove that S−1 Z[x] is a principal ideal
domain.
17. Let A be a principal ideal domain, so dim A = 1. Show that a
maximal prime ideal of A[x] can have height at most 2 by Krull’s
Hauptidealsatz, and conclude that dim A[x] = 2. This gives a
simple proof of Proposition 9.5.5 for the case of a PID.
The next Exercises, up to 22, go in detail in Nagata’s example
of a Noetherian ring of infinite dimension. We use the notation of
Example 9.3.4. Our presentation is taken from [gne]. Recall the
notation: A = k[x1 , x2 , . . . ], mi is an increasing sequence of inte-
gers such that
S mi+1 − mi is increasing, Pi = (xmi +1 , . . . , xmi+1 )
and S = A \ i Pi .
18. Prove that ht(Pi ) = mi+1 − mi , and deduce that S−1 A has
infinite dimension.
19. Prove that the ideals of the form S−1 Pi are maximal in S−1 A,
and that every nonzero a ∈ S−1 A is contained in finitely many
ideals of the form S−1 Pi .
S
20. Let I ⊂ A an ideal. Show that if I ⊂ i Pi , then I ⊂ Pi for
some i. Deduce that the ideals of the form S−1 Pi are the only
maximal ideals in S−1 A. (Find a way to avoid cancellation of
monomials)
21. Prove that the localizations (S−1 A)S−1 Pi are Noetherian.
22. Let A be a ring, and assume that for each maximal ideal
M ⊂ A, the localization AM is Noetherian. Assume further that
each nonzero a ∈ A is contained in finitely many maximal ide-
als. Show that A is Noetherian. Use this criterion, together with
the previous exercises, to show that the ring S−1 A in Nagata’s
example is Noetherian.

327
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 dimension theory

23. Use the converse Hauptidealsatz Proposition 9.4.5 to give an

alternative proof of Proposition 3.4.9.

24. Let A be a Noetherian ring, B = A[x1 , . . . , xr ], Q a prime of B

and P = Q ∩ A. Prove that
BQ
ht(Q) = ht(P) + dim .
PBQ

25. Prove Nagata’s height formula: let A ⊂ B be Noetherian in-

tegral domains, with B finitely generated as an A-algebra, Q a
prime of B and P = Q ∩ A. Let k(P) = AP /PAP be the residue
field at P and k(Q) = BQ /QBQ that at Q. Then

ht(Q) 6 ht(P) + trdegF(A) F(B) − trdegk(P) k(Q),

where trdeg denotes the transcendence degree of a field exten-

sion. (By induction, one can assume that B is generated over A
by a single element b. Distinguish the cases where b is transcen-
dental or algebraic over A, and use the previous exercise.)

328
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 10
LOCAL STRUCTURE

In this chapter, we investigate in more detail the local structure

of rings – in particular, we study the condition of regularity and
the related notion of multiplicity. Regular rings are the algebraic
counterpart to smooth algebraic varieties. Smoothness for an al-
gebraic variety is measured by looking at the dimension of its
Zariski tangent space – this is at least the dimension of the va-
riety, and equality happens in the smooth case. By Nakayama’s
lemma, the dimension of the Zariski tangent space at a point
x ∈ V is the same as the minimal number of generators of the
maximal ideal of the local ring R(V)x . This notion can readily be
generalized to a Noetherian local ring, giving rise to the concept
of regular ring.
In the first section, we study the elementary properties of reg-
ular rings – we show that from an analytic point of view they
are all very similar, and that a regular ring is necessarily integral,
which translates to the fact that the union of two algebraic vari-
eties is singular along the intersection. Many more results are
known for regular local rings – in particular they are unique fac-
torization domains – but we are only able to prove the simplest
of them, since the most important results require homological
techniques and will be proved in the sequel of this book.
Next, we define the notion of multiplicity of a local ring. This
is in some sense a measure of how much the ring fails to be reg-
ular, or how complex is a singularity. In fact, regular rings have
multiplicity 1, and the converse is true under some additional hy-
pothesis. The latter implication is not easy, though, and a good
part of the chapter builds enough theory to prove at least a spe-
cial case.

329
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 local structure

We can also define the multiplicity of a finitely generated mod-

ule, although the ring case is the most interesting one. The mul-
tiplicity of a local ring is related the degree of the associated
graded ring, via the tangent cone construction.
In the next sections, we develop various results around mul-
tiplicity, in particular the very useful additivity formula. We also
study the behavior of the multiplicity of rings of the form A/(a),
where A is a fixed local ring and a ∈ A varies. This is a way to
express the notion of order of vanishing of the element a, and in
some cases can be used to define a valuation on the ring A.
We end the chapter with the famous structure therem of Co-
hen, that gives a precise description of complete local Noetherian
rings.

10.1 regular rings

Let A be a local Noetherian ring with maximal ideal M. From
the previous chapter we know that A has finite dimension d, and
there is an M-primary ideal Q generated by elements a1 , . . . , ad .
In general, though, we cannot just take Q = M.

Definition 10.1.1. Let A, M be a local Noetherian ring, a1 , . . . , ad

a system of parameters. We say that a1 , . . . , ad is a regular system
of parameters if (a1 , . . . , ad ) = M. If M admits a regular system
of parameters, then we say that A is regular.

Definition 10.1.2. Let A be a Noetherian ring. We say that A is

regular if AP is a regular local ring for all primes P ⊂ A.

Remark 10.1.3. In the definition we require that A is Noetherian,

so that by Corollary 9.3.2 we know that AP has finite dimension
for all primes P.

Remark 10.1.4. Let A, M be a local Noetherian ring. The mini-

mum number of generators of M is called the embedding dimen-
sion of A, denoted embdim A. By definition, we have

embdim A > dim A,

and A is regular exactly when the above is an equality.

330
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 10.1 regular rings

Before giving examples, we can look at the definition from a

different angle.
Remark 10.1.5. Let a1 , . . . , ad be any system of parameters for
A, M. The quotient module M/M2 is a vector space over k :=
A/M. By Nakayama’s lemma 1.3.19, a1 , . . . , ad is regular if and
only if the images a1 , . . . , ad are linearly independent over k.
The above remark makes the geometric meaning of regularity
more apparent. As discussed in Section 8.8, in the geometric
case this notion correspond to a nonsingular variety. To be more
specific, take a variety V and a point p ∈ V, corresponding to a
maximal ideal M ⊂ R(V). Assume that V has dimension d in p –
then elements a1 , . . . , ad ∈ R(V)M form a regular system of pa-
rameters if and only if their linear components are independent
over k. This can be guaranteed exactly when the Zariski tangent
space Tp V has dimension d. In general, the dimension of Tp V is
the embedding dimension of the ring R(V)M .
The name embedding dimension comes from the remark that
V cannot be embedded in An for n < embdim R(V)M , since its
tangent space is a subspace of kn , so the embedding dimension
is a lower bound for the dimension of an affine space in which
V can be embedded. Notice that in any case this bound is not at
all sharp, as the example of a regular variety should immediately
show.
Example 10.1.6. (a) Let A be a local ring of dimension 0. Then
A is regular if and only if it is a field.
(b) More generally, let k be a field, A = k[[x1 , . . . , xd ]]. Then A
is local with maximal ideal (x1 , . . . , xd ). Since dim A = d, A
is regular of dimension d. This is our prototypical example:
we will prove with Cohen’s theorem that the completion of
a regular local rings has this form whenever it has the same
characteristic as its residue field.
(c) Let A, M be a regular local ring of dimension d. Then the
completion Ab has dimension d, and generators for M map to
generators of M.
b This implies that A
b is regular as well.

(d) Let A, M be a regular local ring of dimension 1. Then M has

a single generator a. It turns out (we will not prove this here)

331
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 local structure

that a regular local ring is a UFD, and in particular integrally

closed. By Proposition 7.3.2, it follows that A is a DVR. You
will prove this directly in Exercise 1.
(e) Vice versa, every DVR has dimension 1, and its maximal ideal
is principal by Proposition 7.3.1. Hence local regular rings of
dimension 1 are the same as discrete valuation rings. This
gives a lot of examples of regular rings, such as the p-adic
integers Zp .
(f) For a non-geometric example, take A = Z[x] and the maximal
ideal M = (p, x) for a prime p. Then AM has dimension 2,
and its maximal ideal is generated by two elements, so AM
is a regular local ring.
Remark 10.1.5 helps understanding when a quotient of a regu-
lar ring remains regular.
Proposition 10.1.7. Let A be a regular local ring, with maximal ideal
M. Elements a1 , . . . , ai are a subset of a regular system of parameters
if and only if A/(a1 , . . . , ai ) is regular.
Proof. Let I = (a1 , . . . , ai ). Notice that dim A/I = d − i by Propo-
sition 9.4.5. Assume that a1 , . . . , ad is a regular system of param-
eters. Then ai+1 , . . . , ad are a regular system of parameters for
the quotient A/I, because their images remain linearly indepen-
dent over k = A/M. Vice versa, any regular system of parameters
for A/I lifts to a regular system of parameters for A together with
a1 , . . . , ai .
From an analytic point of view, regular local rings have a very
simple structure. This is expected, since they correspond to non-
singular points on a variety. In differential geometry, the inverse
function theorem ensures that the nonsingular points of the zero
locus of finitely many C∞ functions admits a local chart, hence
the local structure around all nonsingular points looks the same.
We have something like this in the algebraic setting.
Proposition 10.1.8. Let A, M be a regular local ring of dimension d.
Then, as graded rings,
∼ k[x1 , . . . , xd ],
GrM (A) =
where k = A/M.

332
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 10.1 regular rings

Proof. Let a1 , . . . , ad be a regular system of parameters. There is

a surjective homomorphism of graded rings φ : k[x1 , . . . , xd ] →
GrM (A), so
GrM (A) =∼ k[x1 , . . . , xd ] .
I
for some homogeneous ideal I. Assuming I 6= 0, take a homoge-
neous f ∈ I, say of degree r. Then every multiple of f lies in I, so
we can bound the length
   
n+d−1 n+d−r−1
`(GrM (A)n ) 6 −
d−1 d−1
for n big enough. But the right hand side is a polynomial of
degree d − 1, while Theorem 9.6.3 guarantees that dim GrM (A) =
d.
This result has an important corollary.
Theorem 10.1.9. A regular local ring is an integral domain.
Proof. Let A, M be a regular local ring, and takeTsome nonzero
a, b ∈ A. By Krull’s intersection theorem 7.5.24, n∈N Mn = 0,
hence a ∈ Mr \ Mr+1 and b ∈ Ms \ Ms+1 for some r, s ∈ N. This
implies that a ∈ GrM (A)r and b ∈ GrM (A)s are not zero. By the
above Proposition, GrM (A) is a domain – in particular ab 6= 0,
which implies that ab 6= 0.
In the nonlocal case, we cannot hope to have such a result: a
variety made by multiple smooth, not intersecting components
has a coordinate ring that is regular but not integral. This is
essentially the only thing that can go wrong:
Proposition 10.1.10. Let A be a regular ring. Then A is a finite direct
sum of integral domains.
Proof. Let P1 , . . . , Pr be the minimal primes of A, so that
√
N(A) = 0 = P1 ∩ · · · ∩ Pr .

Since A is Noetherian, N(A) is finitely generated, so N(A)n = 0

for n big enough.
Let M be a maximal ideal of A. Then AM is a regular local
ring, in particular it is a domain. The only minimal prime of AM

333
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 local structure

is 0, hence M contains exactly one of the Pi by Corollary 1.6.9. It

follows that the primes P1 , . . . , Pr are all coprime. By the Chinese
remainder theorem,

∼ A ⊕···⊕ A .
A=
P1n Prn

Let Ai := A/Pin , and take a divisor of zero a ∈ Ai . By con-

struction, a is nilpotent in Ai . For every maximal ideal M ⊂ A,
the image of a inside AM is 0, since AM is a domain. But then
Ann(a) is not contained in any maximal ideal, so a = 0. It fol-
lows that each Ai is an integral domain, and A is a finite sum of
integral domains.
We have another way to express the fact that the analytic struc-
ture of a regular ring is especially simple.
Lemma 10.1.11. Let A be a local Noetherian ring of dimension d, with
a system of parameters a1 , . . . , ad , and let Q = (a1 , . . . , ad ). Let f ∈
A[x1 , . . . , xd ] be a homogeneous polynomial of degree s, and assume
that f(a1 , . . . , ad ) ∈ Qs+1 . Then f ∈ M[x1 , . . . , xd ].
Proof. Evaluation at a1 , . . . , ad gives a surjective homomorphism

A eva L Qn
[x1 , . . . , xd ] → GQ (A) = Qn+1
.
Q

By hypothesis, eva (f) = 0, where f is the class of f modulo Q. If

f∈/ M[x1 , . . . , xd ], f has an invertible coefficient, hence f is not a
zero divisor. By Proposition 9.5.2,
A/Q[x]
dim GQ (A) 6 dim = d − 1,
(f)
which is a contradiction since dim GQ (A) = d.
The following theorem expresses the fact that a system of local
coordinates for a regular ring is given by analytically indepen-
dent parameters. It is just a special case of the above lemma.
Theorem 10.1.12. Let A be a regular local ring of dimension d, with
a regular system of parameters a1 , . . . , ad . Then the ai are analyt-
ically independent, that is, if f ∈ A[x1 , . . . , xd ] is a homogeneous
polynomial such that f(a1 , . . . , ad ) = 0, then f ∈ M[x1 , . . . , xd ].

334
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 10.1 regular rings

When the ring contains a field, this result has a simpler state-
ment.

Corollary 10.1.13. Let A, M be a regular local ring of dimension d,

with a regular system of parameters a1 , . . . , ad . Assume that there is
a field k ⊂ A that maps isomorphically onto A/M. Then the ai are
algebraically independent over k.

Proof. Take a polynomial f ∈ k[x1 , . . . , xd ] such that f(a1 , . . . , ad ) =

0. Let s be the minimal degree of a monomial of f, so that
f = fs + g, where each monomial of g has degree at least s + 1.
Then fs (a1 , . . . , ad ) ∈ Ms+1 , and Lemma 10.1.11 guarantees that
fs ∈ M[x1 , . . . , xd ]. Since fs ∈ k[x1 , . . . , xd ], this means that
fs = 0, hence f = 0.

There are a few results about regular rings that are too impor-
tant to omit, but which we cannot prove with the techniques we
have at hand. The proof these results marks the beginning of the
usage of homological methods in commutative algebra. We plan
to expand on this circle of ideas in a following volume.
The most important result is that Theorem 10.1.9 can be strength-
ened considerably:

Theorem 10.1.14 (Auslander-Buchsbaum). A regular local ring is a

unique factorization domain.

Proof. See [Eis95, Theorem 19.19]

Remark 10.1.15. In particular, this means that regular local rings

are integrally closed. A singular point of variety having an inte-
grally closed local ring is called normal – this is considered a mild
form of singularity.

Example 10.1.16. The planar singularity y2 = x2 + x3 (a node) is

not normal. In fact, this is a rephrasing of the content of Exercise
5 in Chapter 5.

The other important result is

Theorem 10.1.17. Let A be regular local ring, P ⊂ A a prime. Then

the local ring AP is regular as well.

Proof. See [Eis95, Corollary 19.14]

335
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 local structure

A consequence of this fact is that in order for a ring to be

regular it is sufficient that all localizations AM for a maximal
ideal M are regular local rings (that is, it is redundant to ask this
for all prime ideals). Some texts even adopt this as the definition
of regular ring.

10.2 multiplicity and degree

In this section, we study in more detail the leading coefficient
of the Hilbert polynomial, both in the local and in the graded
case. In the local case, this term is an algebraic expression of the
multiplicity of a singularity. In particular, it gives an interesting
invariant for local Noetherian rings that are not regular. In the
graded case, it computes the degree of a projective variety. We
also show how the two notions are related via the tangent cone
construction.
We start with a semilocal Noetherian ring √ A of dimension d.
Let Q be any definition ideal for A, so that Q = J(A), and fix a
finitely generated A-module M. Then we know by Corollary 9.2.6
that for n big enough we can express the length of M/Qn+1 M
with a polynomial
 
M
χQ
M (n) = ` .
Qn+1 M

We can bound the degree of χQ

M (n) since

dim M = dim A/ Ann(M) 6 d.

Hence, we can write

e d
χQ
M (n) = n + ad−1 nd−1 + · · · + a0 (10.2.1)
d!
for some e ∈ N.

Definition 10.2.1. Let A be a semilocal Noetherian ring with

dim A = d, M a finitely generated A-module. The natural num-
ber e in (10.2.1) is called the Hilbert-Samuel multiplicity, or simply
multiplicity, of M at Q, denoted e(Q, M).

336
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 10.2 multiplicity and degree

As a notation, we will simply write e(Q) = e(Q, A). When A

is local with maximal ideal M, we let e(A) = e(M) = e(M, A).
In the geometric case, let V be an affine variety, M ⊂ R(V)
the maximal ideal of a point p ∈ V. We will call e(R(V)M ) the
multiplicity of V at the point p.
Remark 10.2.2. Some authors introduce a notation for all coeffi-
cients of the Hilbert polynomial, and so denote by e0 (Q, M) what
we denote by e(Q, M).
Remark 10.2.3. Let A be a regular local ring. Then Proposition
10.1.8 gives the expression
 
M n+d 1
χA (n) = = nd + · · · ,
d d!

hence e(A) = 1. It follows that the multiplicity is an interesting

invariant of the ring only in the singular case.
The converse is not true without additional assumptions.
Example 10.2.4. Consider the ring A = k[x, y]/(x2 , xy) and the
maximal ideal M = (x, y) ⊂ A. Let B = AM , which is a local ring
with maximal ideal M = M · B.
For n > 2 we have
 
A A
` = dimk n = n + 1,
Mn M

which shows that dim A = 1 and e(A) = 1. On the other hand,

A is not regular, for instance because it is not even an integral
domain (in fact x is nilpotent in A).
We can also consider the M-primary ideal Q = (y). Then, by
the same computation,
 
A
` = n + 1,
Qn

so e(Q, A) = 1 as well.
Remark 10.2.5. In the above example, we may be tempted to de-
scribe A as the local ring in 0 of the variety defined by the ideal
I = (x2 , xy). This is not correct, since I is not radical, and in fact

337
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 local structure

V(I) is just the line x = 0. Talking properly of such singularities

requires the language of schemes, which are geometric objects
that are not fully described by their set of geometric points. In
this language, the zero locus of I would be a line with an embed-
ded double point.
Our counterexample is not integral, but this is not the only
restriction. For an example of an integral domain of multiplicity
1 which is not regular, see [Nag62, Appendix A1, Example 2]. In
general, the appendix of Nagata’s book is a fantastic source of
counterexamples in commutative algebra.
On the other hand, the condition having multiplicity 1 is not
too far from being regular. To state this precisely, we need a

Definition 10.2.6. Let A be a local Noetherian ring, A b its com-

pletion with respect to the topology defined by its maximal ideal.
We say that A is unmixed if for every associated prime P of 0 in A
b
we have
Ab
dim = dim A b = dim A.
P
With this definition, we can state the following multiplicity 1
criterion.
Theorem 10.2.7 (Nagata). Let A be a local Noetherian ring. Then A
is regular if and only if e(A) = 1 and A is unmixed.
The proof of this result is surprisingly subtle, see [Nag62, The-
orem 40.6]. Following [Now97], we will give the proof in the case
where the residue field of A has characteristic 0, as a consequence
of Cohen’s theorem.
We now pass to the graded case, where we can give similar
definitions. Let A be a graded ring of dimension d, and assume
that A0 is Artinian. For a finitely generated graded A-module M
we have the bound

dim M = dim A/ Ann(M) 6 d.

Hence, for n big enough we can write

e d
`(Mn ) = n + ad−1 nd−1 + · · · + a0 (10.2.2)
d!
for some e ∈ N.

338
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 10.2 multiplicity and degree

Definition 10.2.8. Let A be a graded Noetherian ring of dimen-

sion d, M a finitely generated graded A-module. The natural
number e in (10.2.2) is called the (Hilbert-Samuel) multiplicity of
M, denoted e(M).
In particular, we are interested in the multiplicity of a graded
ring as a module over itself. In this case, we will also call e(A)
the degree of A, denoted deg A.
Remark 10.2.9. The notions of multiplicity and degree are strictly
related. In fact, let A, M be a local Noetherian ring of dimension
d. Then by construction the Hilbert polynomials of A, M and
GrM (A) are the same, since

X
n 
Mi

χM
A (n) = ` = χGrM (A) (n).
Mi+1
i=0

In particular
e(A) = deg GrM (A).
To better understand this relation, we consider the geometric
case. Let V ⊂ An (k) be an affine variety, p a point of V defined
by the maximal ideal M ⊂ R(V). To this we associate the local
ring A = R(V)M , with maximal ideal M = M · A. If we translate
the variety so that p = 0, each polynomial f ∈ I(V) has zero
constant term, hence we can write

f = fd + fd+1 + · · · ,

where fk is homogeneous of degree k and d > 0. In particular,

we can consider the homogeneous component of lowest degree,
fd . For the purpose of this section, we denote H(f) := fd .
Definition 10.2.10. Let V ⊂ An (k) be an affine variety with 0 ∈ V.
Let H(I(V)) be the homogeneous ideal generated by all polynomi-
als H(f) for f ∈ I(V). The affine variety C0 V defined by H(I(V))
is called the tangent cone of V in 0. Since H(I(V)) is homogeneous,
it also defines a projective variety in Pn−1 (k) called the projective
tangent cone of V in 0, and denoted PC0 V.
Remark 10.2.11. Assume that I(V) = (f1 , . . . , fr ). Then

H(I(V)) = (H(f1 ), . . . , H(fr )).

339
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 local structure

This is especially useful when V is regular in 0. In this case,

we can choose a regular system of parameters f1 , . . . , fr . Then,
each fi has a nonzero linear term, and the tangent cone is exactly
the Zariski tangent space to V in 0. But when V is singular, the
tangent cone contains strictly more information.
By construction, the ring associated to the projective tangent
cone PC0 V is exactly the associated graded ring to the local ring
A, M. In particular, we can translate Remark 10.2.9 as follows:

Proposition 10.2.12. Let V ⊂ An (k) be an affine variety with 0 ∈ V.

Then the multiplicity of V in 0 is equal to the degree of the projective
variety PC0 V.
Notice that we have defined the notion of tangent cone in the
point 0, but using a suitable translation this concept readily gen-
eralizes to other points in V.
Example 10.2.13. Let C be the node defined by the equation y2 =
x2 + x3 . The tangent cone is defined by the equation y2 = x2 , so
it is the union of two lines. This expresses the fact that – while C
itself is irreducible – locally there are two different branches.

The notion of multiplicity has a much simpler interpretation

for singularities of a hypersurface, as the degree of vanishing
of a singular polynomial. The following Proposition makes this
precise.
Proposition 10.2.14. Let k be a field and V ⊂ An (k) an affine hyper-
surface, given by the equation f = 0. Assume that 0 ∈ V, which means
that f(0) = 0, and write f as a sum of homogeneous components

f = fd + fd+1 + · · · ,

where deg fk = k and fd 6= 0. Let A be the local ring of V in 0. Then

e(A) = d.
Proof. Let M = I(0) ⊂ k[x1 , . . . , xn ] be the ideal of the point 0,
and let I = (f), M 0 = M/I, so that A = R(V)M 0 . To compute the
multiplicity, we have to evaluate the length of A/M 0k .
Let Bk := k[x1 , . . . , xn ]/(f, Mk ). First, we claim that
∼ Bk .
A/M 0k =

340
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 10.2 multiplicity and degree

In fact, since quotients and localizations commute, A/M 0k is a

localization of Bk at the multiplicative set S constisting of images
of polynomials with nonzero constant term. But such polyno-
mials are already invertible in Bk . In fact, take a polynomial
g ∈ k[x1 , . . . , xn ] with g(0) 6= 0. Multiplication by g is an injec-
tive map Bk → Bk , and since Bk is a finite-dimensional k-vector
space, it must be surjective as well. This means that g has an
inverse in Bk , proving the claim.
To compute dim Bk , for k > d, we look at the map

k[x1 , . . . , xn ] k[x1 , . . . , xn ]
µf : →
M k−d Mk
given by multiplication by f. Since fd 6= 0, µf is injective, giving
an exact sequence

k[x1 , . . . , xn ] k[x1 , . . . , xn ]
0 → → → Bk →0
Mk−d Mk
This allows us to compute
   
A n+k n+k−d
dim 0k = dim Bk = − =
M n n
(10.2.3)
d
= kn−1 + lower order terms,
(n − 1)!

which implies that e(A) = d.

With exactly the same proof we get an analogous result for the
graded case:

Proposition 10.2.15. Let k be a field and V ⊂ Pn (k) a projective

hypersurface, given by the equation f = 0. Let A be the ring associated
to V. Then deg A = deg f.

In fact, the notions of multiplicity and degree were originally

understood in simple cases such as this, and the definition with
the Hilbert polynomial was introduced later [Sam49].

Remark 10.2.16. The concepts of multiplicity and degree are both

fundamental in intersection theory. To understand what this is
about, we recall some notions on the topology of manifolds.

341
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 local structure

Let M be a compact, oriented smooth manifold of real dimen-

sion d. The cup product gives a ring structure on the graded
sum of the cohomology groups H∗ (M, A), for all rings A. To a
compact oriented submanifold S ⊂ M one can associate a funda-
mental class c(S) ∈ H∗ (M, Z) using Poincaré duality. If S, T ⊂ M
are two such submanifolds, one can then compute the product
c(S) · c(T ) ∈ H∗ (M, Z).
Let s = dim S, t = dim T . Assuming s + t = d, and that S and
T are transverse, the product
∼Z
c(S) · c(T ) ∈ Hd (M, Z) =

computes the number of points of intersection between S and T ,

counted with sign. For a point p ∈ S ∩ T , the sign is positive
when the natural isomorphism
∼ Tp M
Tp S ⊕ T p T =

preserves the orientation on the tangent space, negative other-

wise. In the case where S and T are not transverse, one can de-
form S in its tubular neighborhood to a submanifold S 0 having
c(S 0 ) = c(S), in such a way that S 0 and T are transverse, and then
the product c(S 0 ) · c(T ) has this geometric interpretation as the
number of signed intersections.
One would like to be able to obtain a similar theory for pro-
jective algebraic varieties, but there are some subtleties. First, for
fields other than R or C, there is not an obvious replacement
for singular cohomology. In any case, one would like to be able
to compute products even inside singular varieties, where some-
thing like Poincaré duality cannot be expected to hold. Third,
since subvarieties are defined algebraically, there is no obvious
way to deform them to obtain tranversality, as one can do in the
differentiable case.
It turns out that one can develop intersection theory in this
algebraic setting, but some care has to be taken. In particular,
the notion of multiplicity is fundamental in computing products
in cases where the intersections cannot be made transverse. The
degree of a projective variety V ⊂ Pk , instead, measures the num-
ber of intersections (counted with multiplicities) with a generic
linear space of dimension k − dim V. For much more about inter-
section theory, the standard reference is [Ful84].

342
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 10.3 formulas for multiplicity

10.3 formulas for multiplicity

In this section, we are going to investigate some properties of
multiplicity. In particular we will derive some convenient formu-
las to compute the multiplicity in a local ring in terms of mul-
tiplicity in smaller rings such as quotients and localizations. We
start by mentioning some elementary facts, which are immediate.
Proposition 10.3.1. Let A be a semilocal Noetherian ring of dimension
d, Q and Q 0 two ideals of definition for A and M a finitely generated
A-module.
 
i) e(Q, M) = limn→∞ nd!d ` QM nM ;

ii) e(Qr , M) = e(Q, M) · rd

iii) if Q ⊂ Q 0 , e(Q 0 , M) 6 e(Q, M).
Also, we can rephrase Proposition 9.2.11 as follows:
Proposition 10.3.2. Let A be a semilocal Noetherian ring, Q ⊂ A an
ideal of definition. If we have an exact sequence of finitely generated
A-modules
0 → M0 →M → M 00 → 0,
then e(Q, M) = e(Q, M 0 ) + e(Q, M 00 ).
Although we have defined the multiplicity for an A-module
M, in many cases of interest only the multiplicity of A carries
some new information. In particular, we can reduce the com-
putation of the multiplicity of an A-module to that of a ring in
many cases. The following result, known as the additivity formula,
achieves this, and at the same time reduces the computation of
multiplicity to the case of integral domains. Some authors call
this the associativity formula (see for instance [Eis95, Ex. 12.11]),
but following [Lec57] we reserve this name for Theorem 10.3.8.
Proposition 10.3.3 (Additivity formula). Let A be a local Noetherian
ring, Q ⊂ A an ideal of definition, M a finitely generated A-module.
Let d = dim A and let P1 , . . . , Pr be the minimal primes of A for which
dim A/Pi = d. Then
X
r
e(Q, M) = e(Qi , A/Pi )`(MPi ),
i=1

343
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 local structure

Q+Pi
where Qi := Pi .
P
Proof. We use an induction over s(M) := `(MPi ).
When s(M) = 0, we must have MPi = 0 for all i, that is,
Pi ∈/ Supp(M) for all i. But this means that Ann(M) is not con-
tained in any minimal prime Pi of A such that dim A/Pi = d.
It follows that dim(M) = dim(A/ Ann(M)) 6 d − 1, and in this
case e(Q, M) = 0.
For the inductive step, choose a minimal prime P ∈ Supp(M)
that is one of P1 , . . . , Pr , say P = P1 . By Corollary 3.3.14, the
minimal primes of Supp(M) and Ass(M) are the same, hence P
is associated to M. This means that we can find N ⊂ M such that
N= ∼ A/P. By Proposition 10.3.2, we have

e(Q, M) = e(Q, N) + e(Q, M/N). (10.3.1)

Remark that NP = ∼ AP , so `(NP ) = 1, while NP = 0 for

PAP i
i > 2, since these are different minimal primes. This allows us to
compute

X
r X
r
s(M) = `(NPi ) + `((M/N)Pi ) = 1 + s(M/N).
i=1 i=1

By inductive hypothesis,

X
r
e(Q, M/N) = e(Qi , A/Pi )`((M/N)Pi ).
i=1

Moreover, by our choice of N we have e(Q, N) = e(Q, A/P) =

e(Q1 , A/P). Equation (10.3.1) becomes

X
r
e(Q, M) = e(Q1 , A/P) + e(Qi , A/Pi )`((M/N)Pi ).
i=1

∼ MP , while for i = 1 we have

For all i > 2, we have (M/N)Pi = i
`((M/N)P1 ) = `(MP1 ) − 1, and the thesis follows.

Corollary 10.3.4. Under the hypothesis of Proposition 10.3.3, assume

that A is an integral domain. Then e(Q, M) = e(Q) rk(M).

344
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 10.3 formulas for multiplicity

Proof. In this case, the only minimal prime is 0, hence e(Q, M) =

e(Q)`(M0 ) and `(M0 ) = dimk M ⊗ k, where k is the fraction field
of A.

In particular, if A is a regular local ring, the multiplicity just

measures the rank of an A-module, and gives no new informa-
tion.

Remark 10.3.5. Proposition 10.3.3 can also be used taking M =

A. In this case, it reduces the computation of e(Q, A) to that of
e(Q, A/P) for various primes P.
Consider an affine variety V and a point p ∈ V. From a geo-
metric point of view, the additivity formula allows us to compute
the multiplicity of V at p in term of the multiplicity of the com-
ponents of V of maximal dimension passing through p.

Remark 10.3.6. The previous remark does not add anything in

the case where V is irreducible. But even then, the additivity
formula can be useful. In fact, let A be a local Noetherian ring
with maximal ideal M, and Ab its completion in the M-adic topol-
ogy. By Corollary 7.5.18, A and A b have the same associated
graded ring, hence the same Hilbert polynomial. It follows that
b and it can happen that A is an integral domain
e(A) = e(A),
while A is not. Geometrically, this can happen if an irreducible
b
variety has two branches at a point that are analytically separa-
ble.

Example 10.3.7. Let k be a ring of characteristic different from

2. The curve C defined by y2 = x2 + x3 in A2 (k) is irreducible.
√
But in the ring k[[x, y]] one has the factorization y2 = (x 1 + x)2 ,
where
√ 1 1 1
1 + x = 1 + x − x2 + x3 + · · · ,
2 8 16
so the completion of the local ring of C at 0 is not an integral
domain, and Proposition 10.3.3 applies. Again, we see that while
C is irreducible, around 0 we can consider C as composed by two
different components.

Related to the additivity formula, there is the following associa-

tivity formula of Lech from [Lec57], generalizing a previous result
from [Che45].

345
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 local structure

Theorem 10.3.8 (Lech). Let A, M be a local ring, a1 , . . . , ad a system

of parameters for A and Q = (a1 , . . . , ad ). Fix a natural number
m 6 d and let Q1 = (a1 , . . . , am ) and Q2 = (am+1 , . . . , ad ). Then
X  Q1 + P 
e(Q) = e e (Q2 · AP ) ,
P
P

where the sum ranges over the primes P ⊂ A that are minimal over Q2
and that satisfy dim A/P + ht P = dim A.
Notice that the elements of the formula are well defined: for
such a prime P, a1 , . . . , am is a system of parameters for A/P, so
(Q1 + P)/P is M/P-primary, and am+1 /1, . . . , ad /1 is a system of
parameters for AP , so Q2 · AP is P · AP -primary.
Let us set up some terminology. Just for the purpose of this
proof, we will call a prime P ⊂ A balanced if dim A/P + ht P =
dim A, and a chain

Pd ( P1 ( · · · ( P0

compatible with a1 , . . . , ad if for each k = 0, . . . , d we have

(ak+1 , . . . , ad ) ⊂ Pk .

This allows us to state some simple lemmas.

Lemma 10.3.9. Each prime P appearing in a chain compatible with
a1 , . . . , ad is balanced.
Proof. The chain can be split into a chain for A/P and one for
AP .
Lemma 10.3.10. The set of chains compatible with a1 , . . . , ad is finite.
Proof. The condition implies that Pk is a minimal prime of the
ideal (ak+1 , . . . , ad ), and minimal primes of an ideal are finite in
a Noetherian ring.
Lemma 10.3.11. The set of chains compatible with a1 , . . . , ad is not
empty.
Proof. There exists a minimal prime of 0 which is balanced, call it
Pd . Recursively choose Pk as a minimal prime of Pk+1 + (ak+1 )
which is balanced.

346
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 10.3 formulas for multiplicity

Lemma 10.3.12. Each balanced prime Pk such that (ak+1 , . . . , ad ) ⊂

Pk and dim A/Pk = k appears in a chain compatible with a1 , . . . , ad .

Proof. Apply the previous lemma to find a chain for A/Pk and
one for APk , and join them.

We are now ready to prove the associativity formula.

Proof of Theorem 10.3.8. Let Σ be the set of chains compatible with

a1 , . . . , ad , which is finite and nonempty by the previous lem-
mas. A repeated application of the additivity formula (Proposi-
tion 10.3.3) gives
X


e(Q, A) = ` (A/P1 )P0 · · · ` (A/Pd )Pd−1 `(APd ). (10.3.2)
P∈Σ

A similar repeated application, just stopped earlier, gives

X  Q + Pm 

e(Q, A) = e ` ((A/Pm+1 )Pm ) · · · ` (A/Pd )Pd−1 `(APd ).
Pm
P∈Σ

By applying (10.3.2) to APm , we simplify this to

X  Q + Pm 
e(Q, A) = e e(QPm , APm ),
Pm
Pm

where the sum is over all balanced primes that contain (am+1 , . . . , ad )
appearing in a member of Σ. By the previous lemmas, these are
exactly the balanced primes minimal over (am+1 , . . . , ad ), and
we get the thesis.

We conclude this section with two classical result by Samuel.

The first one ([Sam53]) relates length and multiplicity.

Proposition 10.3.13 (Samuel). Let A, M be a local Noetherian ring

of dimension d, Q an ideal of definition generated by the system of
parameters a1 , . . . , ad . Then
 
A
e(Q, A) 6 ` .
Q

347
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 local structure

Proof. Evaluation at a1 , . . . , ad gives a surjective homomorphism

of graded rings
A
[x1 , . . . , xd ] → GrQ (A).
Q
This gives a corresponding inequality between their Hilbert poly-
nomials, and since both rings have dimension d, an inequality
between their first coefficient
 
A
e [x1 , . . . , xd ] > e(GrQ (A)) = e(Q, A).
Q
To conclude, we just note that the Hilbert polynomial of the first
ring is   
A n+d
` .
Q d

The next theorem gives a way to relate multiplicities in integral

extensions. It appears in [ZS76b, Theorem 24, Chapter 10], but
we give a slightly simplified statement.
Theorem 10.3.14. Let A ⊂ B be Noetherian integral domains, and
assume that A is local with maximal ideal M. Let Q ⊂ A be an ideal
of definition, and assume that B is integral over A. Take a primary
decomposition
\r
Q·B = Qi ,
i=1
where Qi is Pi -primary in B. Then the polynomials

[F(B) : F(A)]χQ
A (n) (10.3.3)

and
X
r
[B/Pi : A/M]χQ i
BP (n) (10.3.4)
i
i=1
have the same degree and leading term.
The statement can be simplified when all localizations BPi have
the same dimension as A. In this case, all summands contribute
to the leading term, and we get

348
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 10.3 formulas for multiplicity

Corollary 10.3.15. In the above theorem, assume that ht Pi = ht M

for all primes Pi . Then

X
r
[F(B) : F(A)]e(Q, A) = [B/Pi : A/M]e(Qi , BPi ).
i=1

Remark 10.3.16. If we assume that A is integrally closed, then

we can apply the above corollary. In fact, for every prime Pi
as in Theorem 10.3.14, we have Q ⊂ Pi ∩ A, which implies that
Pi ∩ A = M. In this case, we can apply the going down Theorem
5.2.13 to conclude that ht Pi = ht M.
Remark 10.3.17. If B is local, there is only one prime Pi that
appears in the sum, and the corollary applies again.
Proof of Theorem 10.3.14.
For a given M, we denote `A (M) the length of M as an A-
module and `B (M) the length of M as a B-module. Take n big
enough, and let d = χQ i
BP (n). By Theorem 2.5.16, there is a chain
i
of BPi -modules

Qn
i = Md ( Md−1 ( · · · ( M1 = Pi · BPi

such that `BP (Mk /Mk+1 ) = 1. This is also a chain of B-modules,

i
and their length as B-modules is the same – and in fact it is the
same as dimB/Pi (Mk /Mk+1 ). As A/M-vector spaces, though,
each term Mk /Mk+1 has dimension [B/Pi : A/M]. This implies
that

[B/Pi : A/M]χQ i n
BP (n) = [B/Pi : A/M]d = `A (B/Qi ).
i
Tr
Moreover, ·B = Qn n
i=1 Qi , and by the Chinese remainder
theorem we have
X
r
`A (B/Qn n
i ) = `A (B/Q · B).
i=1

Putting the two equations together, we recognize that the sum

in (10.3.4) is just `A (B/Qn · B). On the other hand, the term
in (10.3.3) amounts to [F(B) : F(A)]`A (A/Qn ), so we need to
compare these two lengths.

349
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 local structure

Let k = [F(B) : F(A)]. If B is a free module over A, its rank

must be k, in which case the equality
`A (B/Qn · B) = k · `A (A/Qn )
holds trivially. In general we cannot assume that, but we can find
b1 , . . . , bk ∈ B that span the field F(B) over F(A), and then one
can fit B in
1
C ⊂ B ⊂ C,
d
where C = hb1 , . . . , bn iA and d ∈ A. We use this to compare
`A (B/Qn · B) with `A (C/Qn · C).
Namely, there is a surjection
C C + Qn · B
→ ,
Qn · C Qn · B
which implies the inequality
C + Qn · B d · B + Qn · B
     
C
`A > `A > `A =
Qn · C Qn · B Qn · B
   
B B
= `A n
− `A .
Q ·B d · B + Qn · B
By a symmetric reasoning,
     
C B C
`A 6 `A + `A .
Qn · C Qn · B d · C + Qn · C
Hence, to understand the difference between `A (B/Qn · B) and
`A (C/Qn · C), we look at the terms
 
B
`A
d · B + Qn · B
and  
C
`A .
d · C + Qn · B
In both cases, these are Hilbert functions for modules over the
ring A/dA, which has dimension strictly less than dim A, so
`A (B/Qn · B) and `A (C/Qn · C) have the same degree and lead-
ing coefficient. Since for C we have
`A (C/Qn · C) = k · `A (A/Qn ),
the theorem is proved.

350
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 10.4 multiplicity and valuations

10.4 multiplicity and valuations

In this section, we investigate the meaning of multiplicity as the
order of vanishing of a suitable element. The prototypical re-
sult in this spirit is Proposition 10.2.14, which links the order of
vanishing of a polynomial f in 0 to the multiplicity of the hyper-
surface defined by f in 0. We are going to generalize greatly this
result, and in doing so state some conditions under which the or-
der of vanishing of an element can be interpreted as a valuation.
Let A be a Noetherian ring, I an ideal. Then by Krull intersec-
tion theorem 7.5.24 we have
∞
\
In = 0,
n=0

hence for each element a ∈ A \ {0} we can find a unique n such

that a ∈ In \ In+1 .
Definition 10.4.1. Let A be a Noetherian ring, I an ideal. If a ∈
In \ In+1 , we denote vI (a) = n. The function vI : A \ {0} → N is
called the order function of I. When A is local with maximal ideal
M, we denote vA = vM .
Now assume that A is local with maximal ideal M. By con-
struction, we have the inequality
vA (a + b) > min{vA (a), vA (b)}, (10.4.1)
which makes vA something similar to a discrete valuation, but
this is not always the case. For one thing, valuations are defined
on a field, and A need not be an integral domain, so it may not
have a fraction field. But even if A is a domain, vA can fail to be
multiplicative, hence it is not always a valuation.
Remark 10.4.2. In fact, we always have the inequality
vA (ab) > vA (a) + vA (b),
but it can be strict. The condition that we always have equal-
ity is equivalent to saying that the associated graded ring A 0 =
GrM (A) is an integral domain. In fact let a ∈ Mr \ Mr+1 and
b ∈ Ms \ Ms+1 . Then a is a nonzero element of Ar0 and b is
nonzero in As0 , so their product is nonzero in Ar+s
0 if and only if
vA (ab) = vA (a) + vA (b).

351
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 local structure

In this section, following [Hor76], we want to relate the order

function of A to the function that measures multiplicity of an
element a ∈ A.
Definition 10.4.3. Let A, M be a Noetherian local ring of dimen-
sion d. Given a ∈ M \ {0}, we define
 
A
µA (a) = e
(a)
if dim A/(a) = d − 1, and µA (a) = ∞ otherwise. We define
µ(a) = 0 for a ∈ A \ M (this is consistent, since in this case A/(a)
is trivial). The function
µA : A \ {0} → N ∪ {∞}
is called the multiplicity function of A.
Remark 10.4.4. By Proposition 9.4.5, dim A/(a) > d − 1, so the
condition µA (a) = ∞ means that dim A/(a) = d. This can only
happen if a is a divisor of 0 by Proposition 9.5.2. Hence, on an
integral domain, the multiplicity function takes values in N.
It is no wonder that the functions vA and µA are related, since
both measure some kind of order of vanishing of an element. In
some cases, we have already established a relation between the
functions vA and µA .
Example 10.4.5. (a) Let A be a DVR. In this case, vA is exactly
the discrete valuation on A. Moreover, a quotient A/(a) has
dimension 0, and if vA (a) = n, then `(A/(a)) = n as well,
since all ideals are powers of the maximal ideal. Hence µA =
vA in this case.
(b) Let k be a field, M ⊂ k[x1 , . . . , xn ] the ideal of 0 and A =
k[x1 , . . . , xn ]M . Then we can rephrase Proposition 10.2.14 by
saying that µA = vA . Moreover it is immediate that vA is
multiplicative by looking at the monomials of lowest total
degre. So we can extend vA to a valuation on k(x1 , . . . , xn ).
In both examples, things behave as good as we can hope: the
functions vA and µA agree and both are valuations. We will
prove in this section that these two phenomena are strictly related.
In fact, while vA satisfies (10.4.1), the multiplicity function µA is
almost always multiplicative.

352
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 10.4 multiplicity and valuations

Proposition 10.4.6. Let A, M be a local Noetherian ring. If µA (a) =

∞ or a is not a divisor of zero, then

µA (ab) = µA (a) + µA (b).

The hypothesis that a is no a divisor of zero is necessary – see

Exercise 8.

Proof. In the case where µA (a) = ∞, the is a minimal prime

P ⊂ A such that a ∈ P and dim A/P = dim A. Hence, ab ∈ P as
well, and µA (ab) = ∞.
So we can assume that µA (a) and µA (b) are both finite. Since a
is not a divisor of 0, multiplication by a induces an isomorphism

A ∼ (a)
= .
(b) (ab)

Using the exact sequence

(a) A A
0 → → → →0
(ab) (ab) (a)

we deduce the equality

     
A A A
` =` +` . (10.4.2)
(ab) (a) (b)

Let d = dim A. We can specialize the additivity formula in

Proposition 10.3.3 to get
  X    
A A AP
e = e ` ,
(a) P a · AP
P

where the sum ranges over all primes P 3 a such that dim A/P =
d − 1. We can also include primes P that do not contain a, since
in that case AP /(a · AP ) is 0. A similar formula holds for b and
ab.
Using additivity of the lengths (10.4.2) in the rings of the form
AP , we obtain the result.

Corollary 10.4.7. Let A, M be a local Noetherian integral domain.

Then the multiplicity function µA is multiplicative.

353
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 local structure

We now investigate under what conditions vA is multiplicative

as well. A simple case is the following.
Proposition 10.4.8. Let A be a regular local ring. Then the order
function vA is a discrete valuation on A.
Remark 10.4.9. Notice that A need not be a discrete valuation
ring – in fact this cannot happen unless dim A = 1. What we
mean is just that vA extends to a valuation with values in Z on
the fraction field F(A). Here we are implicitly using the fact that
A is an integral domain (Theorem 10.1.9).
Proof. By Remark 10.4.2, it is enough to prove that GrM (A) is a
domain, and Proposition 10.1.8 tells us that GrM (A) is isomor-
phic to k[x1 , . . . , xd ], where d = dim A and k = A/M.
In order to connect the functions vA and µA , we start with a
lemma that simplifies the computation of µA .
Lemma 10.4.10. Let A be a semilocal Noetherian ring, Q an ideal of
definition, a ∈ Q. Then
 
Q/(a) A
χQ
A (n) − χ A/(a)
(n) = ` .
(Qn : a)
Proof. This is just a computation:
   
Q Q/(a) A A
χA (n) − χA/(a) (n) = ` −` =
Qn Qn + (a)
 n   
Q + (a) (a)
=` =` =
Qn Qn ∩ (a)
   
(a) A
=` =` .
(a) · (Qn : a) (Qn : a)
The last equality uses the fact that, while multiplication by a
is not necessarily injective, its kernel is contained in (Qn : a)
anyway.
Using this lemma, we see that to control the difference between
e(Q, A) and e(Q/(a), A/(a)) we need to understand the ideal
(Qn : a). Assume that a ∈ Qs . Then we have the inclusion
Qn−s ⊂ (Qn : a). Samuel introduced in [Sam53] the following
definition to capture the case where we are able to control the
behavior of (Qn : a).

354
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 10.4 multiplicity and valuations

Definition 10.4.11. Let A be a semilocal Noetherian ring, Q an

ideal of definition. We say that the element a ∈ Qs is superficial
(of order s) for Q if there exist an integer c such that

(Qn : a) ∩ Qc = Qn−s

for all n large enough.

When A, M is a local ring, we just call the element superficial

of order s if it is so for M. Notice that this implies that a ∈
/ Ms+1 ,
so in fact the order vA (a) = s. This gives a hint that superficial
elements can be used to connect vA to µA . We can specialize the
previous lemma to the case of superficial elements:

Proposition 10.4.12. Let A be a semilocal Noetherian ring, Q an ideal

of definition, a superficial of order s for Q. Then there exists an integer
c such that
Q
χQ Q (a) Q Q Q
A (n) − χA (n − s) 6 χ A (n) 6 χA (n) − χA (n − s) + χA (c)
(a)

for n large enough.

Proof. By definition of superficial element, we get

 n
(Qn : a)
  
(Q : a)
` =` =
Qn−s (Qn : a) ∩ Qc
 c
Q + (Qn : a)
  
A
=` 6 ` ,
Qc Qc

which we can translate to

 
A
06 χQ
A (n − s) − ` n
6 χQ
A (c).
(Q : A)

We can conclude by Lemma 10.4.10.

By comparing just the first coefficients in the Proposition, we

get

Corollary 10.4.13. Let A be a semilocal Noetherian ring of dimension

d > 1, a ∈ Qs superficial of order s for the ideal of definition Q. Then
e(Q/(a), A/(a)) = e(Q, A) · s.

355
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 local structure

Proof. We need to compute the first coefficient of the polynomial

χQ Q
A (n) − χA (n − s). The coefficients in the same degree cancel
each other, so the first nonzero monomial is in degree d − 1 and
is given by

e(Q, A) d e(Q, A)
(n − (n − s)d ) = · dsnd−1 + · · ·
d! d!
giving e(Q/(a), A/(a)) = e(Q, A) · s.

Corollary 10.4.14. Let A, M be a Noetherian local ring. Then A[[x]]

is local as well, and e(A[[x]]) = e(A).

Proof. The ring A[[x]] is clearly local with maximal ideal (M, x).
Moreover, it is easy to check that x is superficial of order 1. If
dim A > 0, then Corollary 10.4.13 applies and gives the thesis. If
dim A = 0, then e(A) = `(A) and the Hilbert polynomial of A[[x]]
is χA[[x]] (n) = `(A) · n.

We can finally prove the result that links vA to µA .

Theorem 10.4.15. Let A, M be a local Noetherian integral domain.

Then vA is a valuation if and only if vA = k · µA , in which case
k = e(A).

Proof. If vA is a multiple of µA , then it is multiplicative by Corol-

lary 10.4.7, hence it is a valuation.
Vice versa, assume that vA is multiplicative. Then the ring
GM (A) is an integral domain, and this implies that every nonzero
element a ∈ A is superficial (this should be clear, but see Lemma
10.5.2). Assuming dim A > 1, by Corollary 10.4.13, we get
 
A
µA (a) = e = vA (a) · e(A).
(a)

For the case where dim A = 1, apply the result to A[[x]]. It is a

simple check that the order function vA[[x]] is also a valuation,
hence vA[[x]] = k · µA[[x]] . For a ∈ A, we have vA[[x]] (a) = vA (a),
and by Corollary 10.4.14 also µA[[x]] (a) = µA (a), so we get the
conclusion.

Putting this together with Proposition 10.4.8, we get

356
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 10.5 superficial elements

Corollary 10.4.16. Let A be a regular local integral domain. Then

vA = µA , and both functions are valuations.
In the case where vA and µA disagree, we cannot expect that
µA is a valuation, but still we have the following result that we
quote without proof (see [Hor76, Theorem 4]).
Theorem 10.4.17. Let A be a local Noetherian integral domain. Then
there exist r discrete valuations v1 , . . . , vr on F(A) and corresponding
integers n1 , . . . , nr such that

µA = n1 v1 + · · · + nr vr .

10.5 superficial elements

Let A be a semilocal ring, Q ⊂ A an ideal of definition. In the
previous section, we defined an element a ∈ Qs to be superficial
of order s if it has the property that

(Qn : a) ∩ Qc = Qn−s

for a fixed c ∈ N and for all n large enough. We used this

to prove Corollary 10.4.13 that states that in this case we have
e(Q/(a), A/(a)) = e(Q, A) · s, provided dim A > 1. This allows
us to prove properties of multiplicity by induction on the dimen-
sion.
In this section, we investigate in more detail this technical tool,
starting from some existence results for superficial elements, and
use it to prove some deeper properties of multiplicity. Most of
this material is taken from [ZS76b]. The main existence result is
Theorem 10.5.1. Let A be a semilocal Noetherian ring, Q ⊂ A an
ideal of definition. Then there exists a superficial element of order s for
some s > 1.
In order to prove it, we rephrase the condition of being super-
ficial.
Lemma 10.5.2. Let A be a semilocal Noetherian ring, Q ⊂ A an
ideal of definition, A 0 = GQ (A) the associated graded ring. Given
an element a ∈ Qs , let a ∈ As0 be its image. Then a is superficial of
0
order s if and only if AnnA 0 (a) ⊂ A<c for some c ∈ N.

357
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 local structure

Proof. This is just a matter of spelling out the definitions.

To say that a is not superficial (with constant c) means that we
can find b ∈ Qc \ Qn−s such that ab ∈ Qn . Rephrasing in terms
of the order function vQ of Q, this means that c 6 vQ (b) < n − s
and vQ (ab) > n.
To say that AnnA 0 (a) 6⊂ A<c 0 means that we can find b with
0
vQ (b) > c such that b ∈ Av (b) satisfies ab = 0, which is the
Q
same as ab ∈ Qs+vQ (b)+1 .
The two conditions are now seen to be equivalent.

Proof of Theorem 10.5.1.

Let A 0 = GrQ (A) be the associated graded ring. By the lemma,
we need to understand divisors of 0 in A 0 , so we look at the
associated primes of 0.
Let A+ 0 be the irrelevant ideal, and partition the associated

primes P1 , . . . , Pr of 0 in A 0 , so that the last r − h contain A+

and the first h do not. We claim that we can find a homogeneous
element a ∈ A+ 0 such that a ∈ / Pi for i = 1, . . . , h.
To prove this, choose homogeneous elements bi ∈ A+ 0 \ P and
i
cij ∈ Pj \ Pi for i, j = 1, . . . , h, and then take
Y
ai = b i cij ,
j6=i

so that ai ∈ A+ 0 ,a ∈
i / Pi but at the same time ai ∈ Pj for all j 6= i.
Finally, letting
Xh
a= ad
i
i

i=1

for suitable di such that a is homogeneous, we have the desired

element.
Say a is homogeneous of order s, and let a ∈ Qs be a repre-
sentative of a. We want to show that a is superficial of order s.
Using Lemma 10.5.2, we look at an element b ∈ Ann(a). We can
assume that b is homogeneous.
Consider a primary decomposition of 0
r
\
0= Qi ,
i=1

358
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 10.5 superficial elements

where Qi is Pi -primary. For a suitable exponent c we have

r
\
0 c
(A+ ) ⊂ Qi . (10.5.1)
i=h+1

By our choice of a we have b ∈ Pi for i = 1, · · · , h. If we

assume b ∈ A>c 0 , we find from (10.5.1) that b ∈ P for i = h +
i
1, . . . , r, hence b = 0. This proves that Ann(a) ⊂ A<c 0 , and the

theorem.

Unfortunately, we are not free to choose the order s at will in

the above theorem. This is not a limitation of our approach.

Example 10.5.3. Let k = Z/2Z and consider the ring

k[[x, y]]
A := .
(xy(x + y))

Then A does not have superficial elements of order 1 for its max-
imal ideal M. For if a ∈ A is such an element, then a ∈ M \ M2 .
But there are only 3 such elements: x, y and x + y, and each
of them is annihilated by a homogeneous element of arbitrarily
large order in GrM (A).

We can however guarantee the existence of superficial elements

of any order, assuming the residue field is infinite.

Theorem 10.5.4. Let A, M a Noetherian local ring of positive dimen-

sion, and assume that k = A/M is infinite. Let Q be a M-primary
ideal. Then for any s > 1 there exists a superficial element of order s for
Q. Moreover, we can take such an element outside all minimal primes
of 0 in A.

Proof. As in Theorem 10.5.1, we consider the graded ring A 0 :=

GQ (A) and take a primary decomposition of 0 in A’:
r
\
0= Qi ,
i=1

where Qi is Pi -primary, and the indices have been chosen so that

0 ⊂ P exactly for i = h + 1, . . . , r, for some h 6 r.
A+ i

359
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 local structure

Let D1 , . . . , Dt be the minimal primes of 0 in A. Given a

nonzero element a ∈ A with vQ (a) = s, denote a ∈ As0 its class,
and given an ideal I ⊂ A denote
I := (a | a ∈ I, a 6= 0).
We claim that As0 6⊂ Di for any i = 1, . . . , t and any s ∈ N.
Otherwise, Qs ⊂ Qs+1 + Di , and by Nakayama’s lemma Qs ⊂
Di . Since Di is prime, this implies that M ⊂ Di , which can only
happen if dim A = 0.
Now choose any s > 0. Then As is not contained in any of the
Di , as well as in Pj for j 6 h. Let M = As0 /As+1 0 , which we regard
∼
as a module over A/Q = A /A+ . Every ideal J of A 0 defines a
0 0

submodule J ∩ As0 /As+1 0 , and in particular the ideals Di and Pj

for j 6 h define a collection of submodules Ni , i = 1, . . . , h + t.
By what we have just said, we have strict inclusions Ni ( M.
If we prove that M 6= h+t 0
S
i=1 Ni , then we can find a ∈ As such
that a ∈/ Di for i = 1, . . . , t and a ∈ / Pj for j = 1, . . . , h. As in the
proof of Theorem 10.5.1, this means that a is superficial of order
s, and we are done.
It remains to prove that M 6= Ni . This is clear when Q = M,
S
for in this case M is a vector space over the infinite field k, hence
it cannot be the union of finitely many subspaces. In general,
we reduce to this case as follows. Choose g such that Mg ⊂ Q.
Given a submodule N ( M, we have
M
N+ M ( M,
Q
otherwise, we would have the chain
M M M M2
 
M = N+ M = N+ N+ M = N+ M=
Q Q Q Q
Mg
= ··· = N+ M = N.
Q
Therefore, as vector spaces,
N + MM M
( .
MM MM
In particular, (Ni + MM)/MM is a strict subspace of M/MM,
and the union over all i cannot be the whole space.

360
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 10.5 superficial elements

Superficial elements are the basis for many results on multiplic-

ities, including the proof of the multiplicity 1 criterion (Theorem
10.2.7). We will just give an application in this section.
Proposition 10.5.5. Let A, M be a Noetherian local ring, Q ⊂ A an
ideal of definition, and assume that k = A/M is infinite. Then there
exists an ideal of definition Q 0 ⊂ Q generated by a system of parameters
such that e(Q 0 , A) = e(Q, A).
Proof. Induction on d := dim A. The case d = 0 is trivial, since we
can take Q 0 = 0. The difficult case is d = 1. Choose an element
a ∈ Q which is superficial of order 1 for Q. By definition we have

(Qn : a) ∩ Qc = Qn−1 (10.5.2)

for a fixed c ∈ N. Now apply the Artin-Rees lemma to the ideal
(a). The Q-adic topology on (a) and the topology induced by the
Q-adic topology on A have bounded difference, so

Qn ∩ (a) ⊂ Qn−k · (a) (10.5.3)

for some k ∈ N.
Assume first that a is not a divisor of zero, so that the multipli-
cation map
µa : A → a · A
is injective. In this case, µ−1
a (a · Q
n−k ) = Qn−k , while by defi-
−1
nition µA (Q ∩ (a)) = (Q ∩ (a) : a) = (Qn : a). So, by taking
n n

preimages in (10.5.3), we get

(Qn : a) ⊂ Qn−k .
When n − k > c, we can put this together with (10.5.2) and get

(Qn : a) = Qn−1 .
Using Lemma 10.4.10 we get
Q/(a)
χA/(a) (n) = χQ Q
A (n) − χA (n − 1) = e(Q, A),

since dim A = 1. In particular, dim A/(a) = 0, so the left hand

side is just `(A/(a)). Since a is not a zero divisor,
(an )
   
A
` =` = e(Q, A),
(an+1 ) (a)

361
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 local structure

which proves that e((a), A) = e(Q, A) and we can take Q 0 = (a).

This ends the proof for the case d = 1, assuming a is not a divisor
of 0.
For the general case - still with d = 1 – we first notice that all
chains of primes in A have the form P ⊂ M, where P is a minimal
prime of 0. Choose any superficial element a ∈ Q of order 1 that
lies outside the minimal primes, so that dim A/(a) = 0. If a is
not a divisor of 0 we are done, otherwise we can apply the above
result to A/ Ann(a). To this end, define I := Ann(a), A := A/I
and Q := (Q + I)/I. Then we can compute
 
A
χQ (n) = ` =
A Qn + I
   n 
A Q +I
=` − ` =
Qn Qn
 
I
= χQ
A (n) − ` .
Qn ∩ I

Now, I is a finitely generated module over

T A/(a), which is Artin-
ian, so it has finite length. Moreover, n Qn ∩ I = 0 by Krull’s
intersection theorem, so we have Qn ∩ I = 0 for n big enough. In
this case, we can simplify the last equation to

χQ
A
(n) = χQ
A (n) − `(I).

This implies that dim A = 1 and that e(Q, A) = e(Q, A). The
same reasoning, applied to (a), show that e((a), A) = e((a), A).
Since a is not a divisor of 0 in A, we can apply the first part of
the proof to A and conclude that e((a), A) = e(Q, A), so finally
e((a), A) = e(Q, A) and we can take Q 0 = (a). This ends the
proof for the case d = 1.
For d > 1, we perform an inductive step. Using Theorem 10.5.4,
choose an element a ∈ Q which is superficial of order 1 for Q.
Let A = A/(a) and Q = Q/(a); then e(Q, A) = e(Q, A) and by
induction we find a system of parameters a2 , . . . , ad for A such
that
e((a2 , . . . , ad ), A) = e(Q, A).

362
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 10.6 cohen’s structure theorem

Then {a, a2 , . . . , ad } ⊂ Q is the desired system of parameters for

A. In fact, let Q 0 = (a, a2 , . . . , ad ). Then using Lemma 10.4.10
we deduce that
e(Q 0 , A) 6 e(Q 0 /(a), A) = e(Q, A) = e(Q, A),
and vice versa e(Q 0 , A) > e(Q, A) in any case, so we must have
equality.

10.6 cohen’s structure theorem

In this section, we discuss a famous result of Cohen ([Coh46])
that gives a precise description of complete local rings. In particu-
lar, we will be able to exactly describe the structure of a complete,
regular local ring. Under an additional hypothesis, which we will
describe below, such a ring is isomorphic to a ring of power se-
ries over its residue field. Geometrically, this is essentially an
algebraic version of the inverse function theorem.
In differential geometry, the inverse function theorem describes
the local structure of the zero locus V of k smooth functions
f1 , . . . , fk defined on an open set in Rn . Namely, if p is a point in
V, and the differentials of the functions {fi } are independent in p,
then V admits a local chart φ : Rn−k → V around p, giving to V
the structure of an embedded differentiable manifold.
In the algebraic setting, one does not have the luxury of a local
chart. Still, let V ⊂ An (k) be an affine variety of dimension d, p
a point of V, and A, M the local ring of V in p. If V is regular in
p, we will be able to say that the completion A b of A with respect to
the M-adic topology is isomorphic to k[[x1 , . . . , xd ]]. In a sense,
from an analytic point of view V has something that resembles
a local chart, even though the Zariski topology is too coarse to
express this.
Our treatment follows in part [Katb] and [Mur], but the reader
can also consult tag 0323 in [dJea20]. The equicharacteristic case
is Theorem 7.8 in [Eis95].
In order to state our results we need a
Definition 10.6.1. Let A, M be a local ring with residue field k =
A/M. We say that the field K ⊂ A is a coefficient field if the induced
map K → A/M = k is surjective (hence an isomorphism).

363
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 local structure

This condition cannot always be ensured. For a simple exam-

ple, A may be of characteristic 0, while k may have characteristic
p > 0 (as it happens with the ring Zp ). In this case, a coefficient
field cannot exist.
In fact, a few cases can arise. If char(k) = 0, we must have
char(A) = 0 as well. If instead char(k) = p > 0, then either
char(A) = 0 or char(A) = pn for some n. In fact, A is local and
for every prime q 6= p we have q ∈ / M, so q is invertible in A.

Definition 10.6.2. Let A, M be a local ring with residue field k =

A/M. We say that A is equicharacteristic if char(A) = char(k).

By the above discussion, if A is not equicharacteristic then

char(k) = p > 0 and either char(A) = 0 or char(A) = pn for
some n > 1. If A admits a coefficient field, then A is equicharac-
teristic. To handle the other cases, we need a subtler definition.

Definition 10.6.3. Let A, M be a local ring with residue field k =

A/M. We say that the ring C ⊂ A is a coefficient ring if

1. C is a complete, Hausdorff local ring

2. the ideal C ∩ M is generated by p = char(k)

3. the induced map C → A/M = k is surjective.

Remark 10.6.4. From 3), it follows that the maximal ideal of C is

C ∩ M = p · C. Moreover, since T C is Hausdorff in the topology
defined by its maximal ideal, pk · C = 0. If I ⊂ C is any ideal,
we can find a biggest k such that I ⊂ pk · C. In particular there
exists x ∈ I \ (pk+1 · C), and we can write x = pk · y for some
y ∈ C. But y ∈ / p · C, hence y is invertible in C and I = pk · C. It
follows that C is a principal ideal domain, and all ideals of C are
powers of its maximal ideal.
If moreover, p ∈ / (C ∩ M)2 , it follows by Nakayama’s lemma
k
that all ideals p · C are distinct. In this case, one can define a
valuation v on C \ {0} by declaring that v(x) = n if x ∈ pn · C but
x∈/ pn+1 · C and extend it to the fraction field of C. In this case,
C is a discrete valuation ring.

The following Proposition allows us to get some structure from

the mere existence of a coefficient field or ring.

364
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 10.6 cohen’s structure theorem

Proposition 10.6.5. Let A, M be a Noetherian complete local ring, C ⊂

A a coefficient field or ring. Then there is a surjective map

φ : C[[x1 , . . . , xt ]] → A.

More precisely if M = (a1 , . . . , at ), the map defined by φ(xi ) = ai is

surjective, hence one can take t = embdim A.
Proof. For n > 0, take an element a ∈ Mn , and write
X
a= b i mi ,

where bi ∈ A and mi is a monomial of degree n in a1 , . . . , at . By

definition of a coefficient field or ring, we can write bi = ci + di ,
where ci ∈ C and di ∈ M. Hence we obtain a decomposition
a = c + d, where d ∈ Mn+1 and c is a form of degree n in
a1 , . . . , at with coefficients in C.
Now given any a ∈ A we apply this construction repeatedly,
starting from
a = c1 + d 1 ,
with c1 a linear term in a1 , . . . , at with coefficients in C and d1 ∈
M. We repeat this writing

d 1 = c2 + d 2 ,

and so on. After n steps, we have

a = (c1 + c2 + · + cn ) + dn ,

where the term in parenthesis is a polynomial in a1 , . . . , at with

coefficients in C. Taking the limit (which exists and is unique
since A is complete and Noetherian) we get
∞
X
a= ci ,
i=1

which is a power series in a1 , . . . , at with coefficients in C.

We can define φ by sending xi to ai – then φ is well-defined
since A is complete, and is surjective by the above argument.
In view of this result, it makes sense to ask when a complete
local ring admits a coefficient field or a coefficient ring. This
beautiful result of Cohen gives a pretty complete answer.

365
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 local structure

Theorem 10.6.6 (Cohen). Let A be a complete Noetherian local ring.

i) If A is equicharacteristic, then A admits a coefficient field.

ii) If A is not equicharacteristic, then A admits a coefficient ring that

is the image of a DVR.
Before going into the proof of the theorem, we will state and
prove some important consequences.

Corollary 10.6.7. Let A, M be a complete regular local ring of dimen-

sion d, with residue field k.
i) If A is equicharacteristic, then
∼ k[[x1 , . . . , xd ]].
A=

/ M2 , then
ii) If A is not equicharacteristic, char(k) = p > 0 and p ∈
there exists complete DVR C such that
∼ C[[x1 , . . . , xd−1 ]].
A=

Proof. Let us first consider the equicharacteristic case. By Cohen’s

theorem, there is a coefficient field k ⊂ A. If a1 , . . . , ad is a
regular system of parameters, using Proposition 10.6.5, we find a
surjection
φ : k[[x1 , . . . , xd ]] → A
such that φ(xi ) = ai . It remains to prove that φ is injective.
There are two ways to see this. For one thing, Proposition 10.1.8
implies that φ induces an isomorphism between the associated
graded rings, and then one can use Remark 10.4.2. Alternatively,
one see that
dim k[[x1 , . . . , xd ]] = dim A = d.
If I := ker φ was not trivial, the dimension of A would be less
than d since k[[x1 , . . . , xd ]] is integral, contradiction.
The same proof works in the mixed characteristic case with
slight modifications. In this case, we can take a regular system of
parameters of the form p, a1 , . . . , ad−1 . By Cohen’s theorem, we
find a coefficient ring C, and using Proposition 10.6.5 a surjection

φ : C[[x1 , . . . , xd−1 ]] → A

366
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 10.6 cohen’s structure theorem

such that φ(xi ) = ai . By Remark 10.6.4, C is in fact a DVR, so

the dimension on the two sides are equal, and φ is injective as in
the previous case.

Corollary 10.6.8. Let A be a complete Noetherian local ring. Then A

is the image of a complete regular local ring.

Proof. This follows immediately by Proposition 10.6.5 and Theo-

rem 10.6.6, since the rings k[[x1 , . . . , xd ]], where k is a field, and
C[[x1 , . . . , xd−1 ]], where C is a DVR, are complete and regular
(see Exercise 10).

Corollary 10.6.9. Let A, M be a complete, equicharacteristic local ring

of dimension d, with residue field k. Then A is a finitely generated
module over a subring B ⊂ A such that
∼ k[[x1 , . . . , xd ]].
B=

Proof. Let a1 , . . . , ad√be a system of parameters for A, and Q =

(a1 , . . . , ad ). Since Q = M and A is complete with respect to
M, it is also complete with respect to Q. It follows that it is well-
defined a homomorphism

φ : k[[x1 , . . . , xd ]] → A

such that φ(xi ) = ai . Let B := im φ. Since A/Q is finitely gen-

erated over B/Q, by Corollary 7.6.4 A is finitely generated as a
B-module.
In particular, dim B = d by Proposition 9.5.10. Since the ring
k[[x1 , . . . , xd ]] is an integral domain, each chain of primes ends in
(0), so φ must be injective, otherwise dim B < d. It follows that
B= ∼ k[[x1 , . . . , xd ]] as desired.

We can also derive from Cohen’s theorem a proof of the multi-

plicity 1 criterion of Nagata, albeit with some additional assump-
tions.

Partial Proof of Theorem 10.2.7. Let A be a local ring with maximal

ideal M. If A is regular, then e(A) = 1 by Remark 10.2.3. More-
over, the completion A b with respect to the M-adic topology is
regular as well, hence an integral domain by Theorem 10.1.9. So,
the only associated prime of A b is 0, and A is unmixed.

367
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 local structure

Conversely, assume that e(A) = 1 and A is unmixed. To

prove that A is regular, it is not restrictive to assume that A is
M-adically complete. To prove this implication, we make the ad-
ditional assumption that A is equicharacteristic and the residue
field k = A/M is infinite (for instance, both are true if char k = 0).
Since k is infinite, by Proposition 10.5.5 there is an ideal Q ⊂ M
generated by a system of parameters such that e(Q, A) = 1.
If Q = (a1 , . . . , ad ), where dim A = d, using Corollary 10.6.9,
we know that A is finitely generated over the power series ring
B = k[[a1 , . . . , ad ]]. Using the additivity formula (Proposition
10.3.3) and the hypothesis that A is unmixed, we conclude that A
has only one associated prime of 0. If P is this associated prime,
by primary decomposition it follows that P = N(A). Moreover,
by the additivity formula, `(AP ) = 1, which means that AP is a
field. Since P is nilpotent, this can only happen if P = 0, so in fact
A is an integral domain.
We can now use Corollary 10.3.15 to deduce that A has the
same fraction field as B. Since B is integrally closed and A is an
integral extension of B, we must have A = B, so A is regular.

We now go on in proving Cohen’s theorem. We split the proof

in various cases, which we treat differently. We recall the notation
that A is complete local ring, M its maximal ideal and k = A/M
its residue field.

Proof of Cohen’s theorem when char(A) = char(k) = 0.

By Zorn’s lemma, there exists a maximal subfield K ⊂ A. We
will prove that the induced map K → k is surjective, hence an
isomorphism. Let
π: A → k
the projection, K = π(K) the image of K. If K ( k, we find a ∈ A
such that a = π(a) ∈ / K.
If a is transcendental over K, then a is transcendental over K.
In this case, K[a] ∩ M = 0, so every nonzero element of K[a] has
an inverse in A. This means that K(a) ⊂ A, contradicting the
maximality of K.
If a is algebraic over K, let f be its minimal polynomial over K,
where f ∈ K[x]. Then, since char(k) = 0, a is a simple root of f,
hence it can be lifted to a simple root of f in A by Hensel’s lemma

368
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 10.6 cohen’s structure theorem

7.6.2, call it a 0 . Again we find that K(a 0 ) ⊂ A, contradicting the

maximality of K.

We now pass to the equicharacteristic case where char(A) =

char(k) = p > 0. In this case, the last step of the proof can fail, so
we need a slightly adapted argument. The argument is similar to
the proof of Hensel’s lemma, but we need an additional result.

Lemma 10.6.10. Let A, M be a local ring with residue field k and

assume that char(A) = char(k) = p > 0. If Mp = 0, then A admits a
coefficient field.

Proof. Since char(A) = p, the set Ap := {ap | a ∈ A} is a subring

of A. Moreover let b = ap ∈ Ap be any nonzero element. Then
a∈ / M (as Mp = 0), so a has an inverse in A, and b has one in
A . It follows that Ap ⊂ A is a subfield.
p

By Zorn’s lemma, there exists a maximal subfield K ⊂ A con-

taining Ap . We want to prove that the induced map K → k is
surjective. Let
π: A → k
the projection, K = π(K) the image of K. If this is not the case, we
find a ∈ A such that a = π(a) ∈ / K.
The situation is similar to the previous proof, but this time we
know that ap ∈ K, so the minimal polynomial of a is xp − ap . It
follow that a satisfies the same polynomial over K, hence K(a) is
a subfield of A, contradicting the maximality of K.

We can now easily conclude the proof of Cohen’s theorem in

the equicharacteristic case.

Proof of Cohen’s theorem when char(A) = char(k) = p > 0.

Denote
A A
πn : n+1 → n
M M
the projection. We will recursively find a coefficient field Kn ⊂
A/Mn such that πn (Kn+1 ) = Kn . Then the inverse limit of the
fields {Kn } is the desired coefficient field.
We start with n = 2. In this case, Lemma 10.6.10 applies, as
p > 2, and gives a coefficient field K2 ⊂ A/M2 .
For the induction step, assume that we have found Kn and let
n (Kn ). Also, denote P := ker πn = M /M
B := π−1 n n+1 .

369
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 local structure

Given b ∈ B \ P, let b := πn (b) 6= 0. Since b ∈ Kn , we have

b ∈/ M/Mn , which implies that b ∈ / M/Mn+1 . It follows that
b is invertible in A/M n+1 , and in fact the inverse of b lies in B
(why?). This proves that B is local with maximal ideal P.
We can then apply Lemma 10.6.10 to the ring B having residue
field B/P =∼ Kn , since Pp = 0. This gives us a coefficient field in
B, which we can take as Kn+1 .

For the proof of Cohen’s theorem in the mixed characteristic

case, we are going to rely on the theory of Witt vectors developed
in Section 7.7.

Proof of Cohen’s theorem when char(A) 6= char(k).

First, assume that k is a perfect field, and let p = char k. In
this case, Theorem 7.7.11 tells us that the ring Wp (k) is a DVR,
having maximal ideal M = p · Wp (k), and complete with respect
to the M-adic topology. Moreover, k = ∼ Wp (k)/M.
By Theorem 7.7.12, there exists a homomorphism Wp (k) → A.
The image of Wp (k) inside A is the desired coefficient ring C.
Notice that in this case if char A = 0, the map is injective, hence
C is itself a DVR.
The case where k is not perfect is done by reduction to the
previous case, but we are only giving a brief sketch. The steps
are:

i) Construct a DVR V complete with respect to its maximal

∼ V/M, even when k is not perfect
ideal M and such that k =
per
ii) Consider the perfect closure k of k – this is a construction
similar to the algebraic closure of k, but done by recusively
adding p-th roots of elements of k.

iii) By a similar procedure, starting from A, construct another

complete local ring A
b with residue field b
k.

k) → A
iv) Apply Theorem 7.7.12 to find a lift φ : Wp (b b

v) Prove that V ⊂ Wp (b k), and in fact φ(V) ⊂ A – this is the

desired coefficient ring C.

Some more details can be found in [Katb], or the original paper

[Coh46].

370
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 10.7 exercises

10.7 exercises
1. Prove directly that a regular local ring of dimension 1 is a
discrete valuation ring.
2. Let A be a regular ring of dimension 1. Prove that A is a
Dedekind domain.
3. Prove the properties stated in Proposition 10.3.1.
4. Compute the multiplicity of n lines meeting at the origin in
A2 .
5. Use Example 10.3.7 to compute the multiplicity e(A), where A
is the local ring of the node y2 = x2 + x3 in a simpler way.
6. Verify the final computation (10.2.3) in Proposition 10.2.14.
7. Let A be an integral domain of dimension 1, and assume that
the order function vA is a valuation. Prove that A is a discrete
valuation ring.
8. Let B = k[x, y]/(x2 , xy) and A the localization of B at 0. Com-
pute µA (y) and µA (y2 ) and show that µA (y2 ) 6= 2µA (y) – in
particular the hypothesis that a is a regular element is necessary
in Proposition
9. Let A be a Noetherian ring, a ∈ A a nonzero divisor. Assume
that for a ∈ M \ M2 for every maximal ideal M ⊂ A. If A/(a) is
regular, prove that A is regular.
10. Let A be a regular local ring. Prove that the ring A[[x]] is
regular (use the previous exercise).
11. Let A, M be a complete local ring of dimension d, of mixed
characteristic, and let p = char A/M. Assume that ht(p · A) = 1.
Prove that A is a finitely generated module over a subring B ⊂ A
such that
B= ∼ C[[x1 , . . . , xd−1 ]],
where C is a DVR.
12. Prove Corollary 10.4.14 – that is, e(A[[x]]) = e(A) for a local
Noetherian ring A – by a direct computation (it is easier to write
the Hilbert polynomial of A as a sum of binomial coefficients,
instead of powers of n).

371
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 local structure

The following exercises discuss the notion of reduction of ideals

as a means to compute multiplicities. Given a ring A with two
ideals J ⊂ I, we say that J is a reduction of I if J · In = In+1 for
some n ∈ N.
13. Let A be a semilocal Noetherian ring, J ⊂ I two ideals of
definition. Prove that if J is a reduction of I, then e(J) = e(I).
A sort of converse was proved by Rees in [Ree61]:
Theorem (Rees). Let A, M be a local Noetherian ring. Assume that
A is unmixed, and let J ⊂ I be two M-primary ideals such that e(J) =
e(I). Then J is a reduction of I.
14. Let A be a Noetherian ring, J ⊂ I ⊂ A two ideals. Let B be
the integral closure of A in its total fraction ring. Prove that J is a
reduction of I if and only if I and J have the same integral closure
inside B.
15. Let A = C[[x, y]] and I = (x3 , x2 y, y2 ). Find a reduction of I
and use it to compute e(I).
16. Use Corollary 10.3.15 to give an alternative proof of the inertia-
ramification formula (Theorem 6.3.2).
The following exercises, up to Exercise 21, discuss a structure
result for principal ideal rings, due to Hungerford [Hun68], build-
ing on Cohen’s theorem and [ZS76a, Theorem 33, Part IV]. A prin-
cipal ideal ring is just a ring (not necessarily an integral domain)
whose ideals are all principal. Such a ring is called special if it has
a single prime ideal.
17. Let A be a principal ideal ring, P1 , P2 ⊂ A prime ideals. Prove
that either P1 and P2 are coprime, or P1 ⊂ P2 , or P2 ⊂ P1 .
18. Let A be a principal ideal ring, P1 ⊂ P2 ⊂ A prime ideals. If
Q is a P2 -primary ideal, then P1 ⊂ Q.
19. Let A be a principal ideal ring. Prove that A is a finite direct
sum of principal ideal domains and special principal ideal rings.
(Use primary decomposition for 0 and the previous exercises.)
20. Let A be a special principal ideal ring. Prove that A is an
image of a PID. Deduce the theorem of Hungerford: every prin-
cipal ideal ring is a finite direct sum of images of principal ideal

372
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 10.7 exercises

domains. (A is a complete local ring, so we can apply the struc-

ture theorem of Cohen. If A has a coefficient field, we are done.
Assume that A is the image of C[[x]], where C is a coefficient ring.
Find a quotient of C[[x]] which is a PID and still surjects onto A.)
21. One may want a stronger form of Hungerford’s theorem, but
it is not the case that every principal ideal ring is a quotient of a
principal ideal domain. To see this, take A = R ⊕ Z, and show
that Q, seen as an A-module with a trivial R-action, is not a
direct sum of cyclic A-modules. Conclude by the classification of
modules over a PID. (Compare this with Exercise 32 in Chapter
7.)

373
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 A
FIELDS

In this appendix, we investigate the properties of fields. The

theory has a different flavour from the study of general rings, for
at least two reasons. For one thing, all field homomorphisms are
injective, so the study of morphisms between fields is the same as
the study of field extensions. Second, fields have no proper ideals,
and modules over fields are vector spaces, hence they are fully
characterized by their dimension alone. As a consequence, many
interesting questions of ring theory become moot in this setting.
On the other hand, we will be able to prove much more about
field extensions than about general ring homomorphisms.
This appendix goes through the standard results on field the-
ory, up to the Galois correspondence and the Kummer theory of
Abelian extensions. First, we study algebraic and transcendental
extensions. Next, we specialize to the algebraic setting, and we in-
troduce the problem of separability. This is in order to investigate
the phenomenon – that appears only in positive characteristic –
that irreducible polynomials can have multiple roots. In the next
section, we study normal extensions, which are the maximally
symmetric ones. Extensions that are both normal and separable
admit a Galois correspondence which relates intermediate fields
and subgroups of the automorphism group. We explain the Ga-
lois correspondence both in the finite and infinite case. In the last
section, we specialize the theory to the case of Abelian extensions,
where a more precise description can be obtained.
The appendix covers more than what is needed in the book,
which only uses the notion of transcendence degree, basic results
on separability and the Galois correspondence in the finite case.
Still, many important facts about fields are not covered. In partic-
ular, many classical applications of Galois theory to the feasibility

375
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 fields

of geometric constructions with ruler and compass are only pre-

sented in the exercises, and we just hint at the problem of inverse
Galois theory.

a.1 algebraic elements

Let K be a field. If L ⊃ K is a bigger field and S ⊂ L is any set, we
denote by K(S) the smallest subfield of L containing S – this is the
by S over K. If {Li } is a collection of subfields of L,
field generated Q
we denote by i Li their composite, that is, the smallest subfield
of L containing all the Li . In other words,

Y
!
[
Li = K Li .
i i

The composite of two fields L1 and L2 will be simply denoted

L1 L2 .
As we have remarked, any ring morphisms K → L between
fields is injective. For this reason, we are going to be concerned
with field extensions, that is, given a field K we will investigate the
fields L such that K ⊂ L. We usually denote such an extension by
the notation L/K. The first observation is that in this case L is a
vector space over K, by restriction of the multiplication map.
Let us start with the case where L = K(α) is obtained by adding
a single element to K. The natural evaluation map

evα : K[x] →L

f(x) → f(α)

is injective if and only if α does not satisfy any algebraic equation

with coefficients in K. In this case, we can extend evα to an iso-
morphism K(x) = ∼ L between L and the field of rational functions
in one variable over K. In this case, the structure of L is easy to
understand. We capture this distinction in a

Definition A.1.1. Let K ⊂ L be two fields, α ∈ L. We say that

α is algebraic over K if there exists f ∈ K[x] such that f(α) = 0;
otherwise we say that α is transcendental over K.

376
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 a.1 algebraic elements

If all elements of L are algebraic over K, we say that L is an alge-

braic extension of K, otherwise that L is a transcendental extension
of K.

Let α be an element algebraic over K. Then the set of polyno-

mials that vanish on α is a nonzero ideal of K[x]. Since K[x] is a
UFD, this ideal is generated by a polynomial of minimal degree.

Definition A.1.2. Let α be an element algebraic over K, I the

ideal of K[x] consisting of polynomials vanishing over α. Any
generator of I is called a minimal polynomial of α over K. We
usually normalize such a polynomial by requiring that its leading
coefficient is 1, in which case we speak of the minimal polynomial
of α, and denote it by µα .

Remark A.1.3. Let K ⊂ L be two fields, α ∈ L. Then saying that

α is algebraic over K amounts to a relation of linear dependence
between the powers 1, α, α2 , . . . . Hence, α is algebraic over K if
and only if K(α) is a vector space of finite dimension over K.

To restate this, we introduce another piece of terminology.

Definition A.1.4. Let K ⊂ L be a field extension. Then we say

that L is a finite extension of K if L is finite dimensional as a
vector space over K. The dimension of this vector space is called
the degree of L over K, denoted [L : K].

With this terminology, α is algebraic over K if an only if K(α)/K

is a finite extension.

Remark A.1.5. Let K ⊂ L ⊂ M be three fields. If L/K and M/L are

finite, then M/K is finite as well, and [M : K] = [M : L] · [L : K]. We
say that the degree is multiplicative in towers of field extensions.

The above remark has an important consequence:

Proposition A.1.6. Let L/K be an extension of fields, α, β ∈ L ele-

ments algebraic over K. Then α + β, α · β and α/β (when β 6= 0) are
algebraic over K.

Proof. All such elements are contained in K(α, β), which is a finite
extension of K – since K(α, β)/K(α) and K(α)/K are finite.

377
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 fields

Corollary A.1.7. Let K ⊂ M be fields, and let {Li } be a set of interme-

diate fields. If all Li /K are algebraic extensions, then the composite of
the Li is also algebraic over K
Proof. Every element in the composite is a rational function in
finitely many algebraic elements.
Corollary A.1.8. Let L/K be an extension of fields. The set of elements
of L that are algebraic over K is a subfield of L, called the algebraic
closure of K in L.
Usually, when we talk about the algebraic closure, we do so in
an absolute sense: the algebraic closure of K is a field obtained
from K by the process of adding all possible algebraic elements,
that is, all possible roots of polynomials with coefficients in K. To
make this precise, we need some more language.
Definition A.1.9. Let K be a field. We say that K is algebraically
closed if every polynomial f ∈ K[x] has at least one root in K.
The field L ⊃ K is called an algebraic closure of K if it is alge-
braically closed and algebraic over K.
We will shortly prove existence and uniqueness of the algebraic
closure, but before doing so we first need to understand what it
means to add to K the root of an irreducible polynomial f ∈ K[x].
We need to find a field L ⊃ K such that f admits a root in L.
This is easy, since K[x] is a PID: the ideal generated by f is prime,
because f is irreducible, hence maximal. It follows that K[x]/(f)
is a field, and by construction f has a root in this field (the image
of x). By generalizing this construction we can prove:
Proposition A.1.10. Let K be a field. Then K admits an algebraic
closure.
Proof. Let F be the set of irreducible polynomials of positive de-
gree in K[x], and for each f ∈ F take a formal variable xf . In
the ring K[xf ]f∈F the ideal (xf )f∈F is not the whole ring, hence
it is contained in a maximal ideal M. Then K1 := K[xf ]f∈F /M is
an algebraic extension of K because of Corollary A.1.7, and each
irreducible polynomial in K has a root in K1 .
Now K1 may not be algebraically closed, but we can repeat
the process to produce a field K2 ⊃ K1 such that all irreducible
polynomials with coefficients in K1 have a root in K2 , and so

378
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 a.1 algebraic elements

on. The union of the ascending chain K ⊂ K1 ⊂ K2 ⊂ · · · is an

algebraic closure of K.

The algebraic closure of a field enjoys a uniqueness property,

although non-canonical. We can state a slightly stronger result.

Proposition A.1.11. Let L/K be an algebraic extension, T an alge-

braically closed field with a morphism φ : K → T . Then φ can be
extended to a morphism φL : L → T .

Proof. By applying Zorn’s lemma to pairs (M, φM ) – where K ⊂

M ⊂ L is a field and φM : M → T is a morphism extending φ –
one finds a maximal pair (H, φH ). We claim that H = L. If this
was not the case, take any α ∈ L \ H, and let f be its minimal
polynomial. By the embedding φH , we can see H as a subfield of
T . The morphism φH can be extended to H(α) by sending α to
any root of f inside T (check this!).

Corollary A.1.12. If L1 , L2 are two algebraic closures of the field K,

∼ L2 .
then there is an isomorphism L1 =

Proof. The above proposition allows us to extend the inclusion

K ⊂ L2 to a morphism f : L1 ⊂ L2 . Since L1 is a field, f is injective.
It is also surjective, otherwise the inverse morphism f(L1 ) → L1
could not be extended to the whole L2 .

The isomorphism between different algebraic closures is not

canonical, but usually we are not concerned with this. Most of
the time, we will just stick with one choice for the algebraic clo-
sure of a field K, which we will call the algebraic closure of K and
denote by K.
We now move to the general setting of an extension that may
not necessarily be algebraic. To understand such an extension, it
is useful to factor it in two stages: first a maximally transcenden-
tal extension, and then an algebraic one.

Definition A.1.13. Let L/K be an extension of fields. The set

A ⊂ L is said to be algebraically independent over K if there is no
nonzero polynomial f ∈ K[x1 , . . . , xn ] such that f(a1 , . . . , an ) = 0
for some a1 , . . . , an ∈ A.
A is called a transcendence basis of L over K if it is algebraically
independent and the extension L/K(A) is algebraic.

379
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 fields

Algebraically independent sets are ordered by inclusion, and

by Zorn’s lemma, there is a maximal such set. Clearly, a max-
imal algebraically independent set is a transcendence basis, so
transcendence bases always exist. Notice that if A is an alge-
braically independent set over K, we can form a set of indeter-
minates {xa }a∈A and define a homomorphism K({xa }) → L that
sends xa to a. Hence, every field extension factors as an ex-
tension isomorphic to a rational function field, followed by an
algebraic extension.
What is less obvious is the following result.
Theorem A.1.14. Let L/K be a field extension. Any two transcendence
bases for L/K have the same cardinality.
Definition A.1.15. Given a field extension L/K, the cardinality of
any transcendence bases for it is called the transcendence degree of
L over K, denoted trdegK L.
Proof. Let A, B be two transcendence bases, and assume at first
that at least one of them is finite, say A = {a1 , . . . , ar }. Each b ∈ B
is algebraic over K(A), so we can find a polynomial pb ∈ K(A)[x]
such that pb (b) = 0. Choose an element of A, say a1 . There
exists a b ∈ B such that a1 appears in pb , otherwise all elements
of B would be algebraic over K(a2 , . . . , ar ). Since a1 is algebraic
over K(B), it would follow that a1 is algebraic over K(a2 , . . . , ar ),
a contradiction.
Choose one such element b1 , and let p1 be the associated poly-
nomial relation. Then, we claim that C := {b1 , a2 , . . . , ar } is an-
other transcendence basis. In fact, a1 is algebraic over K(C) (us-
ing p1 ), so L is algebraic over K(C) as well. Moreover, a poly-
nomial relation between elements of C would show that b1 is
algebraic over K(a2 , . . . , ar ), which we have already excluded.
By iterating this procedure, we find elements b1 , . . . , br ∈ B
that constitute a transcendence bases, hence B is finite and |B| = r.
It remains to prove the theorem when both A and B are infinite.
In this case, for each b ∈ B choose a finite subset Ab ⊂ A such
that b is algebraic over K(Ab ), and let
[
A∗ := Ab .
b∈B

Then clearly A∗
is a subset of A of the same cardinality of B.
Moreover, every a ∈ A is algebraic over B, hence over A∗ . Since

380
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 a.2 finite fields

A is algebraically independent, it follows that A = A∗ , and so the

thesis.

This proof should be compared to the result that every two

bases of a vector space have the same cardinality. The basic idea
is similar, but in the vector space case things are made simpler by
the fact that one can use a linear equation to express one variable
in terms of the other ones – something that one cannot do with
polynomial relations of higher degree.

Remark A.1.16. Let K ⊂ L ⊂ M be three fields. If A is a transcen-

dence basis for L/K and B one for M/L, it is immediate to check
that A ∪ B is a transcendence basis for M/K. In particular,

trdegK M = trdegK L + trdegL M.

a.2 finite fields

Let K be a finite field. Then K has positive characteristic p, a
prime, hence it contains a copy of Z/pZ, which we will call the
prime field of K. We can regard K as a vector space over Z/pZ, of
some finite dimension d – in particular K has q := pd elements.
We are now going to turn things around. We fix such prime
power q = pd , and let K be a field with q elements. We will
show that K is uniquely determined up to isomorphism. First, all
elements of K are algebraic over Z/pZ, so we can assume that
K is a subfield of the algebraic closure Z/pZ. We are going to
identify K by looking at the multiplicative group K∗ . This has
q − 1 element, each of which satisfies the equation xq−1 = 1.
Adding 0, we get that all elements of K satisfy the equation

xq = x. (A.2.1)

Since this equation has at most q roots, this determines K as a set,

that is,
K = x ∈ Z/pZ | xq = x . (A.2.2)

This in enough to prove the uniqueness of the field with q ele-

ments.

381
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 fields

Moreover, the set of solutions to (A.2.1) is easily checked to

be closed under sum, multiplication and inversion, since K has
characteristic p. It follows we can use (A.2.2) as a definition of
K, and this is in fact a field. The roots of (A.2.1) are distinct
because the derivative of xq − x in characteristic p is −1, which
never vanishes (see Proposition A.3.3). This implies that (A.2.1)
has exactly q distinct roots. Hence the field K defined by (A.2.1)
has exactly q elements, and we have an existence result.
We summarize the previous discussion:

Theorem A.2.1. Let p be a prime, q = pd . Inside Z/pZ there exists

a unique field with q elements, which is the set of solutions to xq = x.

Since the field with q elements is essentially unique, we are

going to give it a name, and denote it Fq . In particular, Fp is just
Z/pZ, seen as a field.
In the rest of the section, we are going to analyze the structure
of Fq in more detail. First, F∗q is cyclic. Actually, this holds in
slightly greater generality:

Proposition A.2.2. Let K be a field, G ⊂ K∗ a finite multiplicative

subgroup. Then G is cyclic.

Proof. By Corollary 1.5.7, G is a product of finite cyclic groups

∼ G1 × · · · × Gr .
G=

If G was not cyclic, the minimum common multiple of the car-

dinalities of the Gi would be strictly less than their product. If
m is such minimum common multiple, this means that |G| > m,
and all elements g ∈ G satisfy the equation gm = 1. This cannot
happen in a field, since this equation has at most m roots.

In particular, inside Fq , there exists a primitive q − 1-th root of

1, which generates the whole field.

Proposition A.2.3. Let p be a prime, q = pd . Then there exists α ∈

Fq such that α generates F∗q as a multiplicative group – in particular,
Fq = Fp (α).

Since the fields Fpd all live inside Fp , we can also consider
whether there is any inclusion among them.

382
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 a.3 separability

Proposition A.2.4. Let a, b be natural numbers, p a prime. Then

Fpa ⊂ Fpb if and only if a divides b.

Proof. One direction is obvious: if b is multiple of a, then all

solutions of xa = x are also solutions of xb = x.
Vice versa, assume Fpa ⊂ Fpb . The inclusion between the
multiplicative groups implies that pa − 1 divides pb − 1. In other
words,
pb ≡ 1 (mod pa − 1).
But the order of p modulo pa − 1 is clearly a, hence b is multiple
of a.

The fields Fq are also endowed with automorphisms.

Definition A.2.5. Let K be a field of characteristic p. The map

x → xp is a field automorphism of K, which is called the Frobenius
endomorphism of K.

For a general field, the Frobenius endomorphism need not be

surjective: for instance in Fp (x), the indeterminate x is not the
p-th power of any element. But for finite fields, every injective
endomorphism is also surjective, hence an automorphism.
Let φ : Fq → Fq be the Frobenius automorphism on Fq , with
q = pd . Then φd (x) = xq = x for all x ∈ Fq , so φd is the
identity. Also, φc is not the identity for any c < d, since φc only
fixes elements in the field Fpc , which does not contain Fq . We
conclude:

Proposition A.2.6. Let p be a prime, q = pd . The Frobenius element

is an automorphism of Fq , fixing each elements of Fp , and it has order
d.

a.3 separability
Before delving into Galois theory, we need to understand a be-
haviour which is typical of extensions of fields in positive char-
acteristic. In many situations, it would be tempting to assume
that an irreducible polynomial must have distinct roots, but this
is not always the case.

383
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 fields

Example A.3.1. Let K be a field of characteristic p > 0, L = K(t)

for some indeterminate t. The polynomial f(x) = xp − t ∈ L[x] is
irreducible. In fact, if it was reducible we would have a nontrivial
factorization f(x) = a(x, t)b(x, t) for some a, b ∈ K[x, t] by Gauss’
lemma, and this is easily excluded since f has degree 1 in t.
On the other hand, let α ∈ L be any root of f, so that αp = t.
Then
(x − α)p = xp − αp = xp − t = f(x),
since char L = p. It follows that α has multiplicity p as a root of
f, and in particular f does not have distinct roots.
The above is the prototypical example of an inseparable exten-
sion, which we are now going to define. As it turns out, this
is not only a nuisance, but the manifestation of a fundamental
phenomenon typical of characteristic p extensions.
Definition A.3.2. Let K be a field.

1. The irreducible polynomial f ∈ K[x] is called separable if it

has distinct roots in K – that is, it factorizes over K as a
product of distinct linear factors.
2. The element α ∈ K is called separable over K if the minimal
polynomial µα is separable.

3. The algebraic extension L/K is called separable if every ele-

ment of L is separable over K.
In all this cases, we will use the word inseparable to mean not
separable.
The main tool to undestand this notion is the following stan-
dard criterion using derivatives.
Proposition A.3.3. Let K be a field, f ∈ K[x] a polynomial. Then f has
distinct roots in K if and only if f and f 0 do not share common factors.
Proof. If f has a multiple root a, we can write f(x) = (x − a)2 g(x)
over K. Since

f 0 (x) = 2 · (x − a)g(x) + (x − a)2 g 0 (x),

(x − a) is a common factor of f and f 0 .

384
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 a.3 separability

Conversely, if f has distinct roots a1 , . . . , an , we can write

f(x) = λ(x − a1 ) · · · (x − an ),

and then for each i we have

f 0 (ai ) = λ(ai − a1 ) · · · (a\

i − ai ) · · · (ai − an ),

where the term ai − ai is omitted, so f 0 (ai ) 6= 0.

Corollary A.3.4. Let K be a field, f ∈ K[x] an irreducible polynomial.

Then f has distinct roots if and only if f 0 6= 0.

Proof. Since both f and f 0 are defined over K, their greatest com-
mon divisor is also with coefficients in K. Assuming f is irre-
ducible, it cannot share a nontrivial factor with f 0 , unless f 0 =
0.

A simple consequence is that a field of characteristic 0 is always

separable: in fact, the derivative of a nonzero polynomial cannot
be 0. In characteristic p > 0, the derivative of f can be identically
0 exactly when f contains only powers of xp , so we conclude:

Corollary A.3.5. Let K be a field of characteristic p > 0. Then f ∈ K[x]

is inseparable if and only if f(x) = g(xp ) for some other polynomial
g ∈ K[x].

Separable finite extensions are easier to study thanks to the the

primitive element theorem, which ensures that they are generated
by a single element.

Theorem A.3.6. Let L/K be a finite separable extension. Then there

exists α ∈ L such that L = K(α).

Proof. By induction, it is enough to prove that if L = K(α, β), we

can find a single γ ∈ L such that L = K(γ). Let µα , µβ be the
minimal polynomials of α and β.
Take any combination γ = α + tβ for some fixed t ∈ K. The key
observation is that the polynomial f(x) := µα (γ − tx) is defined
over K(γ) and vanishes for x = β. Hence, f and µβ have β as a
common root. If we can ensure that this is the only common root,
then the gcd of f and µβ is x − β, and in particular β ∈ K(γ) (and
so also α ∈ K(γ), which means we are done).

385
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 fields

Say µα , µβ have roots α1 = α, α2 , . . . , αr and β1 = β, β2 , . . . , βs .

These sets of roots are distinct because L/K is separable. Then β
is the only common root of f and µβ , unless

αi = γ − tβj = α + tβ − tβj ,

for some i, j 6= 1, 1. This can be solved for t, giving

α − αj
t=− .
β − βj
In other words, it is enough to choose t outside this finite set
of values to prove the thesis. This prove the theorem assuming
that K is infinite – the finite case has been already proved as
Proposition A.2.3.
In order to measure exactly the failure of an extension to be
separable, we define a separable degree.
Definition A.3.7. Let L/K be an algebraic field extension, K an
algebraic closure of K. The separable degree of L over K, denoted
[L : K]s , is the cardinality of HomK (L, K).
We can assume that in fact L ⊂ K, in which case, this is the
same as the cardinality of the set

φ: L → K | φ = id .
L

Using this observation, it is easy to check that this notion be-

haves nicely in towers of extensions.
Proposition A.3.8. Let K ⊂ L ⊂ M be algebraic extensions of fields.
Then
[M : K]s = [L : K]s · [M : L]s .
Proof. Let σ ∈ HomK (L, K) and τ ∈ HomL (M, K). We can extend
σ to a homomorphism σ : K → K, and form the composition

(σ ◦ τ) ∈ HomK (M, K).

M

It is a simple verification that this gives a bijection

HomK (L, K) × HomL (M, K) → HomK (M, K).

386
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 a.3 separability

The connection between this degree and the notion of separa-

bility is as follows. Let α be algebraic over K, and take L = K(α).
Let µα be the minimal polynomial of α over K, and α1 , . . . , αr
the distinct roots of µα . Then a homomorphism φ : L → K as
K-algebras is uniquely defined by choosing φ(α), which must
be one of the αi . Moreover, each αi appears as φ(α) for a suit-
able homomorphism φ (why?). In other words, [L : K]s is ex-
actly the number of distinct roots of µα . On the other hand,
[L : K] = deg µα is the number of roots of µα , counted with multi-
plicities. We conclude:

Proposition A.3.9. Let L = K(α), with α algebraic over K. Then

[L : K] = [L : K]s if and only if α is separable over K.

In general, all roots of µα have the same multiplicity m, so that

[L : K] = m[L : K]s . We can use this to prove the following impor-
tant corollary, that in particular implies that separability of a field
extension L/K is determined by the separability of generators of L
over K. This makes checking whether an extension is separable
much more manageable.

Corollary A.3.10. Let L/K be a finite algebraic extension, generated

by the finite set {α1 , . . . , αs }. Then the following are equivalent:

i) L is separable over K

ii) Each αi is separable over K

iii) [L : K] = [L : K]s .

Proof. Implications i) to ii) to iii) are immediate, so we need only

to prove that iii) implies i).
To do this, take any α ∈ L, so that [K(α) : K] = m[K(α) : K]s for
some m > 1. Then

[L : K] = [L : K(α)][K(α) : K] = [L : K(α)] · m[K(α) : K]s >

> [L : K(α)]s · m[K(α) : K]s = m[L : K]s = m[L : K],

which implies that m = 1, and so that α is separable over K.

This result is easily generalizable to the infinite case. Since any

element algebraic over K lives in a finite extension of K, we have

387
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 fields

Proposition A.3.11. Let L/K be an algebraic extension, where L =

K(S) for some set S. Then L is separable over K if and only if each s ∈ S
is separable over K.
We can also use the above results to guarantee that separability
is preserved in a tower of extensions.
Proposition A.3.12. Let K ⊂ L ⊂ M be algebraic extensions. Then
M/K is separable if and only if both L/K and M/L are separable.
Proof. One direction is obvious, so assume that L/K and M/L are
separable. If the extensions are finite, the result follows from
Corollary A.3.10 and the multiplicativity of separable degree. In
the infinite case, take any α ∈ M, and let µα be its minimal
polynomial over L. Then we can consider the finite extension F
of K generated by the coefficients of µα . Both extensions F(α)/F
and F/K are finite and separable, hence F(α)/K is separable as
well.
Corollary A.3.13. Let L, M ⊂ K be fields separable over K. Then the
composite L · M = L(M) is itself separable.
The above result allows us to define the biggest separable ex-
tension of a field. In fact, let K be a field, K a fixed algebraic
closure. Then the composite field
sep Y
K := L (A.3.1)
L⊂K separable
sep
is itself separable. In fact, every element of K lives in a finite
product of separable extensions of K, which is itself separable by
repeated applications of Corollary A.3.13.
sep
Definition A.3.14. The field K defined by (A.3.1) is called the
separable closure of K.
Now that we have a pretty clear picture of separable extensions,
we are going to investigate the opposite situation. Let K be a
field of characteristic p > 0, f ∈ K[x] an irreducible polynomial.
By Corollary A.3.5, f is inseparable if and only if we can write
f(x) = g(xp ) for some g ∈ K[x]. If g is itself inseparable, we
can repeat the process – after a finite number of steps we end
r
up writing f(x) = h(xp ) for some r > 0 and some separable
irreducible polynomial h ∈ K[x].

388
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 a.3 separability

Definition A.3.15. Let L/K be a finite extension. We define its

inseparable degree, denoted [L : K]i , as the quotient

[L : K]
[L : K]i := .
[L : K]s

By induction of the number of generators of L over K and the

above discussion, we have:

Proposition A.3.16. Let L/K be a finite extension of characteristic

p > 0. Then [L : K]i is a power of p.

Remark A.3.17. We can rephrase Corollary A.3.10 to conclude

that L/K is separable if an only if [L : K]i = 1. Moreover, if
K ⊂ L ⊂ M are finite extensions of fields, we have

[M : K]i = [L : K]i · [M : L]i

by Proposition A.3.8.

We can mimic Definition A.3.2 to describe the opposite phe-

nomenon.

Definition A.3.18. Let K be a field.

1. The polynomial f ∈ K[x] is called purely inseparable if it has

a single distinct root in K.

2. The element α ∈ K is called purely inseparable over K if the

minimal polynomial µα is purely inseparable.

3. The algebraic extension L/K is called purely inseparable if

every element of L is purely inseparable over K.

By definition, if α ∈ L is purely inseparable over K, µα has a

single distinct root in K, which means that [K(α) : K]s = 1. We
can say a little more:

Proposition A.3.19. Let L/K be an algebraic extension. The following

are equivalent:

i) L/K is purely inseparable.

ii) L = K(S) for a set S of purely inseparable elements.

389
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 fields

iii) [L : K]s = 1
r
iv) For every α ∈ L there exists r such that αp ∈ K.
Proof. That i) implies ii) is obvious. Assume ii) and take any
α ∈ L. Then α ∈ K(S 0 ) for a finite set S 0 ⊂ S, so
[K(α) : K]s 6 [K(S 0 ) : K]s = 1.
Assume by contraction that [L : K]s > 1. Then there is an embed-
ding σ : L → K over K which is not the identity on L. Taking any
α ∈ L such that σ(α) 6= α, we also have [K(α) : K]s > 1, which we
have excluded. Hence, ii) implies iii).
Now assume iii) and let α ∈ L. Then [K(α) : K]s = 1, so the
minimal polynomial µα has a single root over K, which means
r
that µα (x) = xp − c for some c ∈ K, so iv) holds.
Finally, assume iv). Then every α ∈ L satisfies the polynomial
r r
xp − αp ∈ K[x], for some r > 0. The minimal polynomial µα is
a divisor of this polynomial, hence it is purely inseparable.
Remark A.3.20. Clearly, in the finite case, all of the above are also
equivalent to the condition [L : K]i = [L : K].
As a consequence of Proposition A.3.19, we prove the analogue
of Proposition A.3.12.
Proposition A.3.21. Let K ⊂ L ⊂ M be algebraic extensions. Then
M/K is purely inseparable if and only if both L/K and M/L are purely
inseparable.
Proof. Use the fact that L/K is purely inseparable if and only if
[L : K]s = 1, together with the multiplicativity of the separable
degree in towers.
Corollary A.3.22. Let L, M ⊂ K be fields purely inseparable over K.
Then the composite L · M = L(M) is itself purely inseparable.
The above theory allows us to factorize any extensions as a
tower of a separable and a purely inseparable one, thus cleanly
splitting the two phenomena. To see this, take any algebraic ex-
tension L/K, and define the field
sep
(L/K)s := {α ∈ L | α is separable over K} = L ∩ K . (A.3.2)
By Proposition A.3.13, (L/K)s is itself a field, and we can factorize
the extension as K ⊂ (L/K)s ⊂ L.

390
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 a.3 separability

Theorem A.3.23. Let L/K be an algebraic extension. Then (L/K)s is

the only subfield of L such that L/(L/K)s is purely inseparable exten-
sion and (L/K)s /K is separable. Moreover,

[L : K]s = [(L/K)s : K],

and if L/K is finite

[L : K]i = [L : (L/K)s ].
Proof. First, (L/K)s is separable over K, by construction. Take any
α ∈ L and let µα ∈ K[x] be its minimal polynomial. Then we can
r
write µα (x) = h(xp ) for some h ∈ K[x] separable. This means
r
that αp ∈ (L/K)s , and if r > 1, α is purely inseparable over
(L/K)s .
The equality [L : K]s = [(L/K)s : K] then follows from the fact
that [L : (L/K)s ]s = 1, and the claim about [L : K]i follows by
multiplicativity assuming L/K is finite.
Finally, let K 0 ⊂ L be any other field such that K 0 /K is separable
and L/K 0 is purely inseparable. Then K 0 ⊂ (L/K)s , and (L/K)s
is at the same time a separable and a purely inseparable exten-
sion of K 0 (because L/K 0 is purely inseparable, while (L/K)s /K is
separable). It follows that K 0 = (L/K)s .
We end this section by considering fields whose finite exten-
sions are always separable.
Definition A.3.24. Let K be a field. We say that K is perfect if all
finite extensions L/K are separable.
Example A.3.25. (a) Any field of characteristic 0 is perfect.
(b) Any finite field is perfect. In fact, any finite field extension
Fpb /Fpa is generated by the pb -th roots of 1, and the pb -
th cyclotomic polynomial has exactly φ(pb ) = (p − 1)pb−1
distinct roots in Fp .
Remark A.3.26. If K is a perfect field, any algebraic extension of K
is separable, since it is generated by finite, separable extensions.
There is a convenient characterization of perfect fields.
Proposition A.3.27. Let K be a field of characteristic p > 0. Then K
is perfect if and only if Kp = K.

391
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 fields

Proof. Assume that K is perfect and let α ∈ K. Let β ∈ K be any

p-th root of α. Then
(x − β)p = xp − βp = xp − α,
so β is the only root of xp − α. This means that β ∈ K, otherwise
K(β)/K would be inseparable. It follows that α ∈ Kp .
Vice versa, assume that K = Kp , and take any α ∈ K. Then we
r
can write its minimal polynomial as µα (x) = h(xp ) for some h
separable over K. Taking pr -th roots of the coefficients of h, we
r r
can write h(xp ) = g(x)p for some g ∈ K[x], which implies that
pr = 1, since µα is irreducible. It follows that µα = h, so α is
separable, hence all finite extensions of K are separable.
Remark A.3.28. A finite extension of a perfect field is perfect as
well. In fact, let K ⊂ L ⊂ M be finite extensions of fields, with K
perfect. Then M/K is separable, and so M/L is separable as well
(for instance, by computing the separable degree).
With the above criterion, we can rephrase this by saying that if
K is a field of characteristic p such that Kp = K and L is a finite
extension of K, then Lp = L as well, something that is much less
obvious to prove directly.
Using this characterization, we can always enlarge a field to
make it become perfect. Starting from a field K0 = K of charac-
teristic p > 0, we let
K1 = K1/p = K({α ∈ K | αp ∈ K}).
1/p
Iterating this procedure, we define K2 = K1 and so on. The
union
per ∞ [
K := K1/p = Ki (A.3.3)
i>0
per per
satisfies K = (K )p , so it is perfect. In fact, it is the smallest
perfect field containing K.
Definition A.3.29. Let K be a field of characteristic p > 0. The
per
field K defined by (A.3.3) is called the perfect closure of K.
per
By construction, K is the union of a tower of purely insep-
per
arable extensions of K, so the extension K /K is purely insepa-
rable as well. Given an algebraic extension L/K, we define
r per
(L/K)i := {α ∈ L | αp ∈ K for some r} = L ∩ K . (A.3.4)

392
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 a.3 separability

Clearly, (L/K)i is a purely inseparable extension of K. The

analogue of Theorem A.3.23 does not necessarily hold, though.
In other words, L/(L/K)i can fail to be separable.

Example A.3.30. Let k be a field of characteristic p > 2, and

K = k(x, y). Consider the polynomial

f(t) = t2p − xtp − y,

and let α be a root of f in K. We take L := K(α). By construction,

f is not separable, hence L/K is an inseparable extension. On the
other hand, no element of L is purely inseparable over K, that is,
(L/K)i = K. To see this, assume that some element β ∈ L satisfies
βp ∈ K. Write
X
2p−1
β = g(α) = gi αi
i=0

for some g ∈ K[t]. Then

X
2p−1 X
2p−1
p
β = g(α) = p
gp p i
i (α ) = gp i
iγ ,
i=0 i=0

where γ = αp . Using the fact that f(α) = 0, we can simplify this

equation to eliminate all powers γd for d > 2. Namely, γ2 =
xγ + y so we end up with

βp = aαp + b,

with a, b ∈ K. The condition βp ∈ K amounts to a = 0.

We now choose p = 3 and carry out the computation explicitly,
to find

a = g31 + xg32 + (x2 + y)g33 + (x3 + 2y)g34 + (x4 + y2 )g35 = 0.

(A.3.5)
This is a linear equation in the powers g3i with coefficients in
K. We can derive (A.3.5) with respect to either x or y to find
new equations. Using the fact that the derivatives of g3i are 0 in
characteristic 3, these are new equations in the powers g3i . It is
not difficult to five 5 independent equations, which means that
βp ∈ K implies that gi = 0 for all i > 1, or equivalently β ∈ K.

393
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 fields

In fact, we can tell exactly when an algebraic extension admits

such a splitting.
Proposition A.3.31. Let L/K be an algebraic extension. Then L is
separable over (L/K)i if and only if L = (L/K)s (L/K)i .
Proof. Assume that L = (L/K)s (L/K)i . Then L is generated over
(L/K)i by elements that are separable over K, and a fortiori over
(L/K)i . This means that L is a separable extension of (L/K)i .
Vice versa, assume that L/K is separable, and consider the com-
posite M = (L/K)s (L/K)i . Then L is at the same time separable
and purely inseparable over M, hence L = M.

a.4 normal extensions

In this section, we study those field extensions that satisfy as
many symmetries as possible.
Definition A.4.1. Let L/K be an algebraic extension, and regard
L as a subfield of K. We say that L is normal if for every field
morphism σ : L → K such that σ = id we have σL ⊂ L.
K
Remark A.4.2. In fact, in the above definition we can equivalently
say that σ(L) = L. Otherwise, σ(L) = L 0 ( L, and we can extend
σ−1 0 : L 0 → K to the whole of L (by Proposition A.1.11), thereby
L
showing that L is not normal.
Let L/K be a normal extension, α ∈ L and let µα be its minimal
polynomial. Then every other root of µα lies in L. Otherwise
we could define a morphism K(α) → K which sends α to a root
outside L, and extend this to a morphism L → K. This prompts
the next definition.
Definition A.4.3. Let K be a field, F a set of polynomials over K.
The splitting field of F is the smallest subfield of K containing all
roots of all polynomials in F.
The two concepts are clearly linked, and we make this precise
in the next result.
Proposition A.4.4. Let F be a set of polynomials defined over K. Then
the splitting field of F is a normal extension of K. Vice versa, if L/K is

394
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 a.4 normal extensions

a normal extension, there is a set F of polynomials over K such that L is

the splitting field of F.
Before proving this, it is useful to introduce a last piece of
terminology.
Definition A.4.5. Let K be a field α, β ∈ K. We say that α and β
are conjugate if they have the same minimal polynomial. Equiva-
lently, there is a morphism σ : K(α) → K such that σ(α) = β.
Proof. Let L be the splitting field of F, and let S be the set of
roots of all polynomials in F. Then each morphism L → K must
permute the set S, and so preserve L. Vice versa, assume that
L/K is normal. If S is any set of generators of L, all conjugates of
elements in S also lie in L. Hence L is the splitting field of the set
of the minimal polynomials of elements in S.
We now give some examples.
Example A.4.6. (a) Let α be purely inseparable over K. Then
K(α) is the splitting field of µα , hence K(α)/K is normal. By
the same reasoning, any purely inseparable extension is nor-
mal.
(b) Let K be a field of characteristic different from 2, L/K an exten-
sion of degree 2. Then L = K(α) for some α that satisfies √an
equation of degree 2. By the quadratic formula, L = K( ∆)
for some ∆ ∈ K. √ There √ is only one nontrivial morphism
L → K, sending ∆ to − ∆, hence L/K is normal.
(c) Every finite field extension is normal, for cardinality reasons:
Fq is the only subfield of Fq with q elements, so every em-
bedding of Fq into its algebraic closure must actually map
Fq into itself.
√
(d) The√extension√Q( 3 2)/Q√is not normal. In fact, the conjugates
of 3 2 are ζ3 3 2 and ζ23 3 2, where ζ3 is a third root of 1, and
√ √
/ Q( 3 2) since it is not a real number.
ζ3 3 2 ∈
(e) Unlike other concepts we have introduced, normality is not
preserved in towers of extensions. That is, if K ⊂ L ⊂ M
are fields, with M/L and L/K normal, it may be the case that
M/K is not normal. A simple example is the composition of

395
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 fields
√ √
two degree 2 extensions, such as Q ⊂ Q( 2) ⊂ Q( 4 2). The
two extensions of √ degree 2 are normal by b),
√ but the minimal
polynomial
√ of 4
2 over Q is x4 − 2, so 4 2 is conjugate to

i · 4 2, which is not real.

We can single out some easy properties of normal extensions,

whose proof is a simple verification.

Proposition A.4.7. i) Let L/K and M/K be two normal extensions.

Then the composite LM is normal over K.

ii) Let K ⊂ L ⊂ M be extensions. If M/K is normal, then M/L is

normal.

iii) Let L/K be a normal extension. Then (L/K)s is also normal over
K.

With the concept of normal extension we can also revisit Exam-

ple A.3.30.

Example A.4.8. Let L/K the extension of example A.3.30. Notice

that in that example [L : K]s = 2, hence (L/K)s is a normal exten-
sion of K. If (L/K)i was not trivial, it would have degree p over
K, which implies that the composite of (L/K)s and (L/K)i is the
whole L. Since both are normal, L/K would be normal as well, by
the previous proposition. But it is easy to check that L/K is not
normal.
In fact, keeping the notation of that example, let γ = αp be
one solution of t2 − xt − y = 0, and let γ∗ be the other one. Also,
choose α∗ such that αp ∗ = γ∗ . Then we have the equations

αp αp
∗ =x
αp αp
∗ = −y.
√ √
If L/K is normal, we have α∗ ∈ L, √ which means that p x, p y ∈
√
L. But this is impossible, since K( p x, p y)/K is an extension of
degree p2 .

It is not a coincidence that normality can shed a light on Exam-

ple A.3.30. In fact, under normality assumptions, we can prove
an analogue of Proposition A.3.23.

396
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 a.5 the galois correspondence

Proposition A.4.9. Let L/K be a normal extension. Then L is separable

over (L/K)i , and in fact (L/K)i is the only intermediate field K 0 such
that L/K 0 is separable and K 0 /K is purely inseparable.

Proof. Let α ∈ L, and let α1 = α, α2 , . . . , αr be its conjugates over

(L/K)i . Then the polynomial

f(x) = (x − α1 ) · · · (x − αr )

lies in (L/K)i , and so must be the minimal polynomial of α over

(L/K)i . To see this, notice that all embeddings L ,→ K send L into
itself, hence they form a group G = AutK (L). The elements of L
fixed by G are by definition purely inseparable, so the fixed field
of G is (L/K)i . But every elements of G permutes the set {αi },
so the polynomial f is unchanged under the action of G, which
means that f ∈ (L/K)i [x]. Since f is separable by construction, α
is separable over (L/K)i .
To prove the second assertion, every field K 0 purely inseparable
over K is contained in (L/K)i , so assuming L/K 0 separable, we
must have K 0 = (L/K)i .

a.5 the galois correspondence

Let L/K be a finite extension, where we assume that L ⊂ K, a
fixed algebraic closure of K. We have seen that the number of
distinct embeddings L ,→ K that fix K equals the separable degree
[L : K]s . Moreover, for such an embedding σ, we do not have
necessarily σ(L) ⊂ L, unless L/K is a normal extension. If σ(L) ⊂
L, then one has equality σ(L) = L, and σ can be regarded as an
automorphism of L fixing K. Thus, L/K has the maximal possible
number of automorphisms precisely when L/K is separable and
normal.

Definition A.5.1. Let L/K be an algebraic extension. We say that

L/K is Galois if L/K is separable and normal. In this case, the
automorphism group AutK (L) will be called the Galois group of L
over K, and denoted Gal(L/K). If f ∈ K[x] is a separable polyno-
mial, the splitting field of f is a Galois extension, and we define
the Galois group of f as the Galois group of this extension.

397
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 fields

We remark again that if L/K is a Galois extension, every em-

bedding σ : L ,→ K that fixes K actually satisfies σ(L) = L, and so
can be regarded as an element of Gal(L/K).

Remark A.5.2. When L/K is finite and Galois, by the above dis-
cussion the Galois group Gal(L/K) has exactly [L : K] elements.

The Galois group can be used to study subfields of L, through

the Galois correspondence, which relates subgroups of Gal(L/K)
and subfields of L that contain K. In one direction, let L 0 ⊂ L be a
subfield such that K ⊂ L 0 . Then L is Galois over L 0 , and one can
consider the Galois group Gal(L/L 0 ). By construction, this is a
subgroup of Gal(K/L), since the automorphisms of L that fix L 0
a fortiori fix all elements of K.
Going in the other direction, let H < Gal(L/K) be a subgroup.
Then the set

LH := {a ∈ L | σ(a) = a for all σ ∈ H}

is a subfield of L that contains K. The Galois correspondence is

given by the two maps

{ subfields of L containing K} → {subgroups of Gal(L/K)}

L0 → Gal(L/L 0 )

LH ← H.
(A.5.1)
0
There are trivial inclusions H < Gal(L/LH ) and L 0 ⊂ LGal(L/L ) .
In fact, much more is true. We first state the main theorem of Ga-
lois theory for the case of finite extensions.

Theorem A.5.3 (Main theorem of Galois theory, finite case). Let

L/K be a finite Galois extension. Then the maps of Galois correspon-
dence (A.5.1) are inverse to each other.

Proof. Let H < Gal(L/K) be a subgroup, LH its fixed field. Using

Theorem A.3.6, write L = LH (α), and let
Y
f(x) := (x − σ(α)).
σ∈H

398
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 a.5 the galois correspondence

Then f is invariant under H, hence f ∈ LH [x]. Since f has degree

n := |H|, we have [L : LH ] 6 n. But L is a Galois extension of
LH , and its Galois group has exactly [L : LH ] elements. Since
H < Gal(L/LH ), we must have equality, so H = Gal(L/LH ).
In the other direction, let L 0 ⊂ L be an intermediate field, and
let H = Gal(L/L 0 ). Consider the field LH ⊃ L 0 . The extension
L/LH is Galois, and by the first part of the proof its Galois group
is H. This implies that [L : LH ] = [L : L 0 ], which means that
LH = L 0 .

Corollary A.5.4. Let L/K be a finite separable extension. Then there

are only finitely many subfield L 0 ⊂ L containing K.

Proof. Let M be the normal closure of L/K, that is, the composite
of the extensions σ(L) for all embeddings σ : L ,→ K. Then M
is a finite extension of K, and it is normal and separable by con-
struction. The main theorem of Galois theory applied to M/K
shows that the intermediate fields between K and M are in bijec-
tion with the subgroups of Gal(M/K), hence they are finite in
number. A fortiori, this is true of the extension L/K.

Let L/K be a Galois extension, L 0 an intermediate extension

that is also Galois over K. Then one can consider the Galois
group Gal(L 0 /K). Any automorphism of L that fixes K will send
L 0 to itself, hence restriction gives a natural map Gal(L/K) →
Gal(L 0 /K). This map is surjective, since every automorphism of
L 0 can be extended to a map L → K, that in turn will send L to
itself. The main theorem of Galois theory can be strengthened
to also describe this picture, by connecting Galois groups of the
form Gal(L 0 /K) to quotients of Gal(L/K).

Theorem A.5.5. Let L/K be a finite Galois extension. Then a sub-

group H < Gal(L/K) is normal if and only if LH is normal over K.
In this case, H is the kernel of the natural surjective homomorphism
Gal(L/K) → Gal(LH /K).

Proof. Let σ ∈ Gal(L/K), and H 0 = σHσ−1 a conjugate of H.

Then the fixed field of H 0 is exactly σ(LH ), so LH is invariant un-
der Gal(L/K) if and only if H is stable under conjugation, which
proves the first half of the theorem. The kernel of the restric-
tion map Gal(L/K) → Gal(LH /K) consists of those elements of

399
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 fields

Gal(L/K) that fix LH , and so is exactly H by the main theorem of

Galois theory.

The main theorem of Galois theory can be extended to the case

of infinite extensions, but it requires some subtlety. In one di-
rection, the correspondence works flawlessly even in the infinite
case.

Proposition A.5.6. Let L/K be a (possibly infinite) Galois extension,

L 0 ⊂ L an intermediate field, and let H := Gal(L/L 0 ). Then LH = L 0 .

Proof. To simplify the notation, we can assume without loss of

generality that L 0 = K. Let α ∈ L \ K, and let M be the normal
closure of K(α) over K. Then M/K is Galois, so using the Galois
correspondence in the finite case we find σ : M → M such that
σ(α) 6= α. We can extend σ to a map σ : L → K, and since L/K
is Galois, σ ∈ Gal(L/K). It follows that α ∈/ LH , and since α is
H
arbitrary, L = K.

The other half of Galois correspondence, though, cannot be ex-

tended literally, for the following reason. An algebraic extension
is the composite of its finite subextensions. Dually, this should
entail that the Galois group is determined by its finite quotients.
But not all groups have this property. To make this observation
precise, we introduce some topological language.
Let G be a topological group. We recall from Section 7.5 that
this is a group endowed with the structure of a topological space
in such a way that the group operations (multiplication and in-
verse) are continuous. In this case, since translation in the group
are continuous homeomorphisms, the topology is determined by
the set of neighbourhoods of the identity element.

Definition A.5.7. The topological group G is called profinite if

the subgroups of G of finite index form a fundamental system of
neighbourhoods of the identity.

In other words, for every homomorphism φ : G → H, where H

is a finite group, we require that the map φ is continuous, where
H is given the discrete topology. In the language of Section 7.4,
a profinite group is the inverse limit (as topological groups) of a
family of finite groups (Exercise 22).

400
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 a.5 the galois correspondence

Remark A.5.8. A finite profinite group necessarily has the dis-

crete topology, so the notion is only meaningful for infinite groups.
If L/K is a Galois extension, we can always endow Gal(L/K)
with a profinite topology. Namely, let {Li } be the family of finite
Galois extensions of K contained in L. Each element of Gal(L/K)
is determined by its action on the finite extensions Li /K, which
shows that there is a natural injection
Y
Gal(L/K) ,→ Gal(Li /K). (A.5.2)
Li

Inside this product, the image of Gal(L/K) can be identified

as the set of compatible sequences – that is, sequences (σi ) where
σi ∈ Gal(Li /K) such that σi and σj agree on the intersection Li ∩
Lj . This shows that Gal(L/K) = lim Gal(Li /K). If we give each
←−
Gal(Li /K) the discrete topology, then Gal(L/K) can be given the
smallest topology that makes all restriction maps
Gal(L/K) → Gal(Li /K)
continuous. In other words, Gal(L/K) inherits the subspace topol-
ogy from the injection (A.5.2), where the right hand side is en-
dowed with the product topology. This description exhibits the
group Gal(L/K) as the inverse limit (as topological groups) of
a family of discrete finite groups, hence Gal(L/K) is a profinite
group.
We can use this language to make the previous remark precise:
Proposition A.5.9. Let L/K be a Galois extension, H < Gal(L/K) a
subgroup, LH its fixed field. Then Gal(L/LH ) = H is the topological
closure of H.
Proof. Let {Li } be the family of finite Galois extensions of K con-
tained in L, so that L is the composite of the Li . If we let Hi be
the image of H inside Gal(Li /K), then Li ∩ LH = LH i
i , hence
[ H
LH = Li i .
i

Let φi : Gal(L/K) → Gal(Li /K) be the restriction map. Then,

by definition of the profinite topology,
\
H= φ−1
i (Hi ).
i

401
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 fields

On the other hand, H is the biggest subgroup of Gal(L/K) that

fixes all fields LH i
i (here we are using the Galois correspondence
inside Li ). Equivalently, H is the biggest subgroup that fixes LH ,
and this is Gal(L/LH ) by definition.

By putting together Propositions A.5.6 and A.5.9, we obtain the

main theorem of Galois theory in its general form.

Theorem A.5.10 (Main theorem of Galois theory). Let L/K be a

Galois extension. Then the maps of Galois correspondence

{ subfields of L containing K} → {closed subgroups of Gal(L/K)}

L0 → Gal(L/L 0 )

LH ← H.
(A.5.3)
are inverse to each other.

We remark again that in (A.5.3) we are only considering closed

subgroups with respect to the natural profinite topology on the
Galois group. Once the Galois correspondence is established, the
following refinement can be proved exactly as in the finite case.

Theorem A.5.11. Let L/K be a Galois extension. Then a closed sub-

group H < Gal(L/K) is normal if and only if LH is normal over K.
In this case, H is the kernel of the natural surjective homomorphism
Gal(L/K) → Gal(LH /K).

Given this result, all Galois groups of the form Gal(L/K), where
L/K is a Galois extension, can be regarded as quotients of a fixed
group:

Definition A.5.12. Let K be a field. The absolute Galois group of K

sep
is Gal(K /K), endowed with its profinite topology.

The absolute Galois group is the biggest Galois group that can
be taken over K, and encodes in a single (albeit complicated) ob-
ject, the structure of all possible (separable) algebraic extensions
of K.

402
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 a.6 some computations

a.6 some computations

In the previous section, we have exposed the Galois correspon-
dence, without actually computing even a single Galois group.
In this section, we are giving some examples.

Example A.6.1. (a) Let L/K be a finite Galois extension of prime

degree p. Then Gal(L/K) is a group with p elements, which
is necessarily isomorphic to Z/pZ.

(b) Take an extension of finite fields Fpb /Fpa . Then we know

that a divides b by Proposition A.2.4. The extension is nor-
mal because Fpb is the only subfield of Fp of cardinality pb .
Let φ : Fp → Fp be the Frobenius homomorphism given by
φ(x) = xp . Then φa leaves each element of Fpa fixed, hence
it is an element of Gal(Fpb /Fpa ). Letting b = ka, we see that
the power (φa )k = φb acts as the identity on Fpb , while no
smaller power of φa is the identity. It follows that φa gener-
ates a subgroup of Gal(Fpb /Fpa ) of cardinality k. Since the
extension itself has degree k, we conclude that Gal(Fpb /Fpa )
is cyclic, generated by φa .

(c) We can extend the previous example to compute the abso-

lute Galois group Gal(Fp /Fp ). Namely, each finite quotient
Gal(Fpn /Fp ) is isomorphic to Z/nZ, and these form an in-
verse systems with the natural maps Z/nZ → Z/mZ each
time m divides n. The absolute Galois groups is then Z, b the
inverse limit of the finite cyclic groups ordered by divisilibity.
This group naturally contains Z (it is its completion with re-
spect to the topology generated by arithmetic progressions),
and the Frobenius element φ can be identified with 1 under
this map. In particular, the subgroup generated by φ is dense
in Gal(Fp /Fp ).

(d) Let Q(ζm )/Q be a cyclotomic extension, where m is not twice

an odd integer. Then we compute in Proposition 6.8.4 that
Gal(Q(ζm )/Q) is isomorphic to (Z/mZ)∗ , where the ele-
ment a ∈ (Z/mZ)∗ acts by sending ζm to ζa m.

(e) Let L/K be a finite Galois extension of degree n. By the

primitive element theorem, L = K(α) for some α ∈ L. Let

403
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 fields

α1 = α, α2 , . . . , αn be the conjugates of α. Then the Galois

group Gal(L/K) acts on the elements {α1 , α2 , . . . , αn } by per-
mutations. Since an element of Gal(L/K) is determined by
its action on α, it follows that the restriction homomorphism
Gal(L/K) → Sn it injective. This exhibits the Galois group
as a subgroup of a group of permutations.
(f) As a partial converse, let K be a field and let Sn act on
K(x1 , . . . , xn ) by permuting the indeterminates. If we let
L = K(x1 , . . . , xn )Sn be the subfield of symmetric rational
function, then Gal(K(x1 , . . . , xn )/L) = Sn . In fact, in this
case every permutation of {1, . . . , n} defines a field automor-
phism. Define the polynomial
Y
n
f(T ) = (T − xi ) = T n − σ1 T n−1 + · · · ± σn
i=1

over the field K(x1 , . . . , xn ), where the polynomials {σi } (de-

fined by this equality) are the elementary symmetric func-
tions. By constructions, the σi are symmetric, hence f ∈ L[T ].
This proves that K(x1 , . . . , xn ) is the splitting field of f, so the
Galois group of f is Sn .
(g) Let G be any finite group. By letting G act on itself, we find
an injective homomorphism G ,→ Sn into some symmmetric
group. Let L/K be any Galois extension with Galois group
Sn – for instance the one in the previous item. Then G corre-
sponds to a subfield LG ⊂ L such that Gal(L/LG ) = ∼ G. This
allows us to realize every finite group as a Galois group of
some Galois extension. Unfortunately, this does not allow us
to control LG , in particular to construct – for example – Galois
extensions with given Galois group over Q. The problem of
realizing a given finite group as a Galois group over a given
field is called the inverse Galois problem, and is in general open
even over Q [MM18].
To give some more examples, we quote Hilbert’s irreducibility
theorem.
Theorem (Hilbert). Let K be a number field, f[x1 , . . . , xn , y] an ir-
reducible polynomial. Then there exist (a1 , . . . , an ) ∈ Kn such that
f(a1 , . . . , an , y) remains irreducible as a polynomial in y.

404
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 a.7 the trace and norm

In fact, much more is known: the same can be done for finitely
many polynomials at once, and the set of specializations that
leave the polynomials irreducible is “big” in an appropriate sense
(see [Ser88] for the notion of thin set, or [CD16] for quantitative
results).
Example A.6.2. (h) Let K be a number field, and consider the
general polynomial of degree n

Y
n
f(T ) = (T − xi ) = T n − σ1 T n−1 + · · · ± σn ,
i=1

which is defined over K(σ1 , . . . , σn ). By Hilbert’s irreducibil-

ity theorem, we can find (a1 , . . . , an ) ∈ Kn such that g(T ) =
f(a1 , . . . , an )(T ) remains irreducible as a polynomial in T .
This entails that the splitting field of G has degree n! over
K. Since its Galois groups is a subgroup of Sn , we must have
equality, which shows that one can always realize Sn as a
Galois group over K.

a.7 the trace and norm

Let L/K be a finite extension of separable degree n = [L : K]s
and inseparable degree f = [L : K]i . Denote by σ1 , . . . , σn the
distinct embeddings L ,→ K. We define two important group
homomorphisms.
Definition A.7.1. The trace is the additive homomorphism

TrL/K : L → K

defined by

TrL/K (α) = f · (σ1 (α) + · · · + σn (α)).

Similarly we define a multiplicative homomorphism called norm

NL/K : L∗ → K∗

by
NL/K (α) = (σ1 (α) · · · σn (α))f .

405
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 fields

It is not a priori clear that these homomorphisms actually take

value in K. But this follows since the values TrL/K (α) and NL/K (α)
are fixed by every element of the Galois group of the normal clo-
sure of L.

Remark A.7.2. We do not ask in the definition that L is a normal

extension of K, so the individual values σi (α) may lie outside L.

Remark A.7.3. The coefficient f appears in the above formulae

because we want to count each embedding σi with multiplicity f.
This ensures that the number of (repeated) embeddings consid-
ered in these formulae is [L : K].

Example A.7.4. (a) The √ simplest example is a quadratic exten-

sion of K, say L = K( a). In this case the extension is Galois,
√
and √the only nontrivial automoprhism of L over K sends a
to − a. It follows that
√
Tr(x + y a) = 2x
√
N(x + y a) = x2 + ay2 .

(b) More generally assume that L = K(α). Let αi = σi (α) be the

conjugates of α. Then the minimal polynomial of α is

µα (x) = xn + an−1 xn−1 + · · · + a0 = (x − α1 ) · · · (x − αn ).

In particular we see that Tr(α) = −an−1 and N(α) = ±a0

are, up to sign, coefficients of the minimal polynomial of α.

(c) Even more generally, take any α ∈ L, which does not neces-
sarily generate the field. Multiplication by α gives a K-linear
map
mα : L → L,
which sends β to αβ. Let

fα (x) = xn + an−1 xn−1 + · · · + a0

be the characteristic polynomial of mα . Then Tr(α) = −an−1

and N(α) = ±a0 . This relation is the reason why we con-
sidered the factor [L : K]i in the definition of the trace and
norm.

406
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 a.7 the trace and norm

For a tower of extensions, trace and norm behave nicely.

Proposition A.7.5. Let K ⊂ L ⊂ M be extensions of fields. Then

TrM/K = TrL/K ◦ TrM/L

NM/K = NL/K ◦ NM/L .

Proof. Let {σi } be the embeddings of M in K, {τi } be those of M in

L = K (fixing L) and {ηi } those of L in K. Since M is algebraic over
L, we can extend the homomorphisms τi to homomorphisms

M → K,

which we still denote τi .

Note that every composition τi ◦ ηj is an embedding of M into
K, hence it is equal to some σk . Moreover, since

[M : K]s = [M : L]s · [L : K]s ,

the two sets {σi } and {τj ◦ ηk } have the same number of elements.
If we prove that they are equal, the desired relations follow im-
mediately.
So we only need to prove that there are no repetitions in the
set {τj ◦ ηk }. If we have

τj ◦ ηk = τl ◦ ηm

we obtain τ−1 −1 −1
l ◦ τj = ηm ◦ ηk , so ηm ◦ ηk fixes L. It follows
that m = k, and in turn l = j.
The trace can be used as a measure of separability.
Proposition A.7.6. Let L/K be a finite extension. Then L/K is separa-
ble if, and only if, TrL/K is not identically 0.
We only prove the half of this result here; the last part of the
proof will be given as a consequence of Artin’s theorem on the
independence of characters.
Proof of Proposition A.7.6, first half. If L/K is inseparable, we can
factor the extension through F := (L/K)s . The extension L/F is
purely inseparable by Proposition A.3.23. Using the property of
composition of the trace in Proposition A.7.5, it is enough to show

407
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 fields

that TrL/F = 0 – in other words, we only need to consider a purely

inseparable extension. But if α ∈ L is purely inseparable over F,
its minimal polynomial is a divisor of xr − αr for some r > 0,
hence its second coefficient is 0, that is, TrL/F (α) = 0.

a.8 abelian extensions

Let L/K be a Galois extension with Galois group G. According to
the nature of G, we can distinguish some classes of extensions.
Definition A.8.1. Let L/K be a Galois extension with Galois group
G
1. We say that L/K is cyclic if G is (finite) cyclic.
2. We say that L/K is Abelian if G is Abelian.
3. We say that L/K is cyclic if G is solvable.
We recall that a group G is called solvable if there exists a finite
chain of subgroups

{e} = G0 < G1 < · · · < Gr = G,

with Gk normal inside Gk+1 , such that the quotiens Gk+1 /Gk
are Abelian. If G is finite, it is equivalent to require that each
quotient is cyclic (up to refining the chain).
Remark A.8.2. Certain authors call an extension L/K solvable if
there exists a field M ⊃ L such that M/K is a Galois extension
with solvable Galois group. In our terminology, a solvable ex-
tension will always be implicitly Galois (and the same holds for
cyclic and Abelian).
In this section, we want to show that these conditions on the
Galois group have a natural interpretation in terms of extensions.
In particular, we will be able to characterize the properties of
Abelian extensions. The prototypical result is the following.
Proposition A.8.3. Let L/K be a Galois extension of degree d, where
L = K(α) for some α such that αd ∈ K. Assume that char K does not
divide d, and that K contains the d-th roots of 1. Then L/K is cyclic.

408
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 a.8 abelian extensions

Proof. Let ζd ∈ K be a primitive d-th root of 1, and let β = αd ∈

K. Then α is a root of f(x) = xd − β. Since [L : K] = d, f is
irreducible, so α is conjugate to ζd α, which is another root of f.
Let σ ∈ Gal(L/K) such that σ(α) = ζd α. Then σk (α) = ζk d α. In
other words, σ acts by cyclically permuting the conjugates of α.
Since an element of Gal(L/K) is determined by its action on α, it
follows that the Galois group is cyclic, generated by σ.

Actually, a kind of converse holds: a cyclic extension L/K (un-

der suitable conditions) is generated by adding roots of elements
of K. Let us try to invert the argument in the above proof. As-
sume that Gal(L/K) = hσi is cyclic of order d, and that a primi-
tive d-th root of 1, ζd , lies in K. If we are able to find an element
α ∈ L∗ such that σ(α) = ζd α, then

Y
d Y
d
d
α = ζid α = σi (α) = NL/K (α) ∈ K∗ .
i=1 i=1

This suggests that we are able to find d-th roots of elements of K

inside L, provided ζd ∈ K and we are able to write

σ(α)
ζd =
α
for some α ∈ L∗ .

Remark A.8.4. Let L/K be a Galois extension and σ ∈ G =

Gal(L/K). Then for any α ∈ L∗ we have
  Q
σ(α) τ∈G τ(σ(α))
NL/K = Q = 1.
α τ∈G τ(α)

This observation is consistent with our setting, since we as-

sumed that ζd ∈ K, hence NL/K (ζd ) = ζd
d = 1. The key result is
the following:

Theorem A.8.5 (Hilbert’s Theorem 90). Let L/K be a finite cyclic

extension with Galois group G = hσi.

i) If β ∈ L∗ satisfies NL/K (β) = 1, there exists α ∈ L∗ such that

β = σ(α)/α.

409
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 fields

ii) If β ∈ L satisfies TrL/K (β) = 0, there exists α ∈ L such that

β = σ(α) − α.

Before being able to prove Hilbert’s Theorem 90, we need some

terminology. If G is a group and L is a field, a homomorphism
G → L∗ will be called a character of G (with values in L).

Theorem A.8.6 (Artin). Let σ1 , . . . , σn be character of G with values

in L. Then the characters are linearly independent over L.

The following proof is taken from [Kata].

Proof. Let A be the group algebra L[G]. Its elements are formal
finite linear combinations
X
n
ai g i
i=1

with ai ∈ L and gi ∈ G. The sum is made formally, while mul-

tiplication is lifted by linearity from the multiplication of G (all
we need here is that G is a monoid). By construction, A is a
(noncommutative) algebra with unit over L. Every character of
G gives rise to a L-linear homomorphism A → L – we will still
denote these homomorphisms by σ1 , . . . , σn .
We claim that – more generally – such homomorphisms of al-
gebras are linearly independent for every L-algebra A. In fact, let
σ : A → Ln be the map of L-algebras given by σ = (σ1 , . . . , σn ).
We claim that σ is surjective – this is easily done by adapting the
proof of the Chinese remainder theorem to the case of noncom-
mutative rings. In particular, we can choose elements x1 , . . . , xn ∈
A such that σi (xj ) = δij . Assume a linear relation

X
n
ai σi = 0
i=1

with ai ∈ L. Evaluating at xj gives aj = 0, hence the characters

are linearly independent.

With Artin’s theorem on the independence of characters at

hand, we can prove the second half of Proposition A.7.6.

410
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 a.8 abelian extensions

Proof of Proposition A.7.6, second half. Assume that L/K is separa-

ble. The embeddings σi : L → K can be seen as characters of
the group L∗ with values in K. Hence they are linearly indepen-
dent, and in particular the element TrL/K = σ1 + · · · + σn is not
zero.

We can now prove Hilbert’s Theorem 90.

Proof of Theorem A.8.5. To prove i), take α ∈ L∗ with NL/K (α) =

1. Define a function f : G → L given by f(e) = 1, f(σ) = α,
f(σ2 ) = ασ(α) and so on, up to f(σd−1 ) = ασ(α) · · · σd−2 (α).
This is done to ensure the relation f(τ1 τ2 ) = f(τ1 ) · τ1 (f(τ2 )) for
all τ1 , τ2 ∈ G.
By Artin’s theorem, the linear combination
X
χ := f(τ)τ
τ∈G

is not 0, hence we find γ ∈ L such that β := χ(γ) 6= 0. We can

compute
X X f(στ) β
σ(β) = σ(f(τ))σ(τ(γ)) = σ(τ(γ)) = ,
f(σ) f(σ)
τ∈G τ∈G

which we can rewrite as

σ(β)
= 1/f(σ) = 1/α.
β

Since NL/K (α) = 1 if and only if NL/K (1/α) = 1, we have proved

the first part.
The proof of ii) is similar. Define f : G → L by f(e) = 0, f(σ) = α
and so on, up to f(σd ) = α + σ(α) + · · · + σd−2 (α). As above,
consider X
χ =:= f(τ)τ.
τ∈G

This time, just choose any γ ∈ L having TrL/K (γ) 6= 0 (here we

use Proposition A.7.6), and let β := χ(γ)/ TrL/K (γ). A computa-
tion similar as before (do it!) shows that β − σ(β) = f(σ) = α.

We can now prove a converse to Proposition A.8.3:

411
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 fields

Proposition A.8.7. Let L/K be a cyclic extension of degree d. Assume

that char K does not divide d, and that K contains the d-th roots of 1.
Then L = K(α) for some α such that αd ∈ K.

Proof. Let ζd ∈ K be a primitive d-th root of 1. Then NL/K (ζd ) =

1, so by Hilbert’s Theorem 90 we can write ζd = σ(α)/α, where σ
is a generator of Gal(L/K) and α ∈ K. Raising to the d-th power
we deduce that αd is invariant under σ, so αd ∈ K. Moreover, d
is the smallest exponent k such that αk ∈ K, so α has degree d
over K, and it generates the whole of L.

This fact, together with Proposition A.8.3, is enough to derive

the classical results of Galois theory about the solvability of equa-
tions using roots. To state the result precisely, let us first agree
what it means to solve a polynomial with a formula involving
roots.

Definition A.8.8. Let L/K be a finite extension. We say that L is

obtained from K by adding roots if there is a tower of fields

K = L0 ⊂ L1 ⊂ · · · ⊂ Lr = L

such that Li+1 = Li (αi ) for some αi such that αd i

i ∈ Li , for all
i = 0, . . . , r − 1.
The polynomial f ∈ K[x] is said solvable by radicals if the split-
ting field of f is obtained from K by adding roots.

We have already proved that extensions that are obtained by

adding a single root are cyclic (under some additional condi-
tions), so the following should not come unexpected.

Theorem A.8.9. Let L/K be a Galois extension of degree d, and assume

char K = 0 or is a prime not dividing [L : K]. Then L is obtained from
K by adding roots if and only if L/K is a solvable extension.

Proof. Let d = [L : K]. It is not restrictive to assume that K con-

tains the d-th roots of 1. In fact, if this is not the case, one can
consider a primitive d-th root of 1 ζd and apply the result to
the extension L(ζd )/K(ζd ). Since K(ζd )/K is both a cyclic exten-
sion and is obtained from K by adding roots, the result for L/K
follows.

412
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 a.8 abelian extensions

By the Galois correspondence, L/K is solvable if and only if

there is a tower of fields

K = L0 ⊂ L1 ⊂ · · · ⊂ Lr = L

such that each extension Li+1 /Li has cyclic Galois group. Using
Propositions A.8.3 and A.8.7, this is the same as saying that L is
obtained by K adding roots.
Remark A.8.10. Of course, we introduced the terminology in a
way that is backwards with respect to historical usage. First, the
theorem about solvability of equations was proved, then it be-
came customary to call groups obtained as a tower of cyclic ex-
tensions solvable.
Corollary A.8.11 (Abel). Let K be a field of characteristic 0. The
generic equation of degree > 5 does not have a solution in radicals –
that is, the generic polynomial

X
d
f(t) = ai xi
i=0

is not solvable by radicals over K(a0 , . . . , ad ) if d > 5.

Proof. By Theorem A.8.9, this amounts to saying that the Galois
group of f is not solvable. We computed in Example A.6.1 that
this Galois group is the full symmetric group Sd , and this is not
solvable for d > 5 (Exercise 13).
Remark A.8.12. Using Hilbert’s irreducibility theorem, one can
give concrete polynomials over Q of degree 5 that are not solvable
by radicals. This means that not only there is no generic formula
to solve the quintic equation – there are explicit quintic equations
whose roots are not writable in terms of radicals (see Exercise 16
for an explicit example).
Propositions A.8.3 and A.8.7 admit a generalization to Abelian
extensions. In this case, one needs to add simultaneously many
roots of elements of the base field at once. To simplify the state-
ment of the result, we introduce some terminology.
Definition A.8.13. Let L/K be a Galois extension. We say that
L/K is a Kummer extension of exponent n if

413
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 fields

1. The Galois group Gal(L/K) is Abelian and has exponent n,

that is, σn = e for all σ ∈ Gal(L/K)
2. the field K contains a primitive n-th root of 1
3. the characteristic of K does not divide n.
Remark A.8.14. If char K is not multiple of n and K contains a
primitive n-th root of 1, K∗ contains the subgroup Un of n-th
roots of 1. Moreover, if a ∈ K and b, c ∈ K are two elements such
that bn = cn = a, the fields K(b) and K(c) are the same, since b
and c differ
√
by a n-th root of 1. We will simply denote this field
by K n a . Similarly, if ∆ ⊂ K∗ is any set, we will denote by
√ √

K n ∆ the composite of K n a for all a ∈ ∆.

Propositions A.8.3 generalizes as follows.

Proposition A.8.15. Let K be a field such that char K does not divide
n and K contains all n-th roots of 1. Let ∆ ⊂ K∗ and denote L =
√
K n ∆ . Then L/K is a Kummer extension of exponent n.

Proof. By definition, L is the splitting field of a family of separable

polynomials,
√
hence L/K is Galois. For each a ∈ K∗ , the extension
n
K a is cyclic of exponent n, by Proposition A.8.3. Putting
together all restriction homomorphisms gives an embedding
Y √

Gal(L/K) ,→ Gal(K n a /K),
a∈∆

which proves that Gal(L/K) is Abelian of exponent n.

In the other direction, we can generalize Proposition A.8.7.

 Let L/K be a Kummer extension of exponent n.

Proposition A.8.16.
√
Then L = K n
∆ , where ∆ := (L∗ )n ∩ K∗ .
√ 
Proof. By definition, K n ∆ ⊂ L. As we have already observed,
by the Galois correspondence, L/K is the composite of cyclic ex-
tensions. √If
M/K is a cyclic extension with M ⊂ L, we have
M = K n a forsome a ∈ K, by Proposition A.8.7. It follows
√
that M ⊂ K n ∆ , and since L is the composite of all such exten-
√ 
sions, L ⊂ K n ∆ .

414
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 a.8 abelian extensions

The statement of Proposition A.8.16 tells more than just the

converse of Proposition A.8.15. In fact, it tells how to reconstruct
L from the subgroup ∆ := (L∗ )n ∩ K∗ . This hints at a correspon-
dence between subgroups of K∗ and Abelian extensions. In fact,
a famous result of Kummer describes precisely such a link.
Let K be a field such that char K does not divide n and K con-
tains all n-th roots of 1. The Kummer correspondence is given by
the two maps
{L/K Kummer of exponent n} → {subgroups ∆ of K∗ /(K∗ )n }

α (L∗ )n ∩ K∗
L →
(K∗ )n
√  β
n
K ∆ ← ∆.
(A.8.1)
Proposition A.8.16 already proves that β ◦ α is the identity. The
converse is more subtle, and needs a little detour on characters
of finite groups.
Definition A.8.17. Let G be a finite Abelian group. The dual group
of G is defined as
b := Hom(G, C∗ )
G
Remark A.8.18. Let Un < C∗ be the subgroup of n-th roots of 1.
If G has exponent n, then G
b := Hom(G, Un ), since the image of
each element of g by a homomorphism is a n-th root of 1.
As in the case of vector spaces, there is a natural homomor-
phism f from G to its double dual G,
b given by
b

f(g)(α) = α(g)

for g ∈ G and α ∈ G.b The following result, known as Pontryagin

duality, holds in greater generality (see [Fol94, Section 4.3]), but
we will only need the finite case.
Theorem A.8.19 (Pontryagin). Let G be a finite Abelian group. The
natural homomorphism f : G → G
b is an isomorphism.
b

Proof. Since every finite Abelian group is a product of cyclic ones

(for instance by Corollary 1.5.9), it is enough to prove the result

415
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 fields

for cyclic groups. If G = Z/mZ, every character φ of G sends 1

to a m-th root of 1 – call it ω – and moreover φ is determined by
ω, since φ(k) = ωk .
It follows that the dual of Z/mZ can be identified with Um ,
and in particular G and G b are isomorphic, albeit not canonically.
Still, this implies that G and G b have the same cardinality, so it is
b
sufficient to check that f is injective.
If f(g) = e, then φ(g) = 1 for all φ ∈ G. b This means that
every character on G descends to a character on G/ hgi – in other
words there is an injection G b ,→ G/\ hgi. This can only happen
when g = e, for cardinality reasons.
If H < G is a subgroup, we can define its orthogonal

H⊥ := φ ∈ G
b | φ(h) = 1 for all h ∈ H .

Remark A.8.20. There is a natural identification

∼ G/H,
H⊥ = [

since characters in H⊥ descend to the quotient modulo H. With

the natural identification G =∼ G,
b we have the inclusion H ⊂
b
⊥

⊥
H , and by comparing cardinalities we conclude that this is

⊥
in fact an equality H = H⊥ .
With these remarks on group duality out of the way, we can
now state a converse to the stronger statement in Proposition
A.8.16.
Theorem A.8.21 (Kummer). Let K be a field such that char K does not
divide n and K contains a primitive n-th root of 1. The Kummer corre-
spondence in (A.8.1) is a bijection. Moreover, for each ∆ < K∗ /(K∗ )n ,
there is an isomorphism
ψ: ∆ → Homc (Gal(L/K), Un ),
a → χa
√ 
where L = K n ∆ is the field corresponding to ∆, χa is the character
defined by √
σ( n a)
χa (σ) = √ n
,
a

416
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 a.8 abelian extensions

and Homc denotes continuous homomorphism with respect to the nat-

ural profinite topology on Gal(L/K) (on Un we just use the discrete
topology).
Proof. We already know from Proposition A.8.16 that β ◦ α is the
identity. To prove that α ◦ β is the identity requires the second
part of the statement, so we proceed to that first.
Fix a Kummer extension L/K and let ∆ := (L∗ )n ∩ K∗ , G :=
√ First,
Gal(L/K). √ we prove that ψ is well-defined. To start, notice
that σ( n a)/ n a is a n-th root of 1, and does not depend on the
choice of an n-th root of a, as we assumed that Un ⊂ K. It
is immediate that χa is a homomorphism, and it is continuous
since
√ √ √

ker χa = {σ ∈ G | σ( n a) = n a} = Gal(L/K n a )

is a subgroup of finite index. Finally, χa only depends on the

class of a modulo (K∗ )n , so ψ is well-defined. Also,
√ √
ker ψ = {a ∈ ∆ | σ( n a) = n a for all σ ∈ G} = (K∗ )n /(K∗ )n = {1},

so ψ is injective.
To check that ψ is surjective, assume first that L/K is finite.
Given a homomorphism χ : G → Un , as in the proof of Hilbert’s
Theorem 90, we find an element b ∈ K∗ such that
σ(b)
χ(σ) = ,
b
But then bn is fixed by all elements of G, so bn ∈ K.
for all σ ∈ G. √
Writing b = n a for some a ∈ K∗ ∩ (L∗ )n = ∆, we recover that
b = χa .
In the infinite case, let χ : G → Un be a continuous homomor-
phism. Then ker χ has finite index, and there exists a subfield
M ⊂ L such that ker χ = Gal(L/M). By the finite case, we find
a ∈ (L∗ )n ∩ M∗ such that χ = χa . But actually, given σ ∈ G we
have  √ n
σ(a) σ( n a)
= √
n
= 1,
a a
so σ(a) = a. Since this holds for all σ ∈ G, a ∈ (L∗ )n ∩ K∗ = ∆.
At this point, we have the second part of the theorem. To check
that α ◦ β is the identity, fix ∆ < K∗ /(K∗ )n , and let L = β(∆) =

417
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 fields
√ 
K n ∆ and ∆ 0 = α(L) = (L∗ )n ∩ K∗ . Clearly, ∆ ⊂ ∆ 0 , and by
the previous part of the proof, we have an isomorphism
∼ Homc (G, Un ),
∆0 =

where G = Gal(L/K). Our goal is to show that ∆ = ∆ 0 .

Assume first that L/K is finite. In this case, we can identify
Homc (G, Un ) with G.b By Pontryagin duality, it is enough to
check that the orthogonal of ψ(∆) is trivial. But

ψ(∆)⊥ = {σ ∈ G | χa (σ) = 1 for all a ∈ ∆} .

√ √
Saying that χa (σ) = 1 amounts to σ( n a) = n a; if this happens
for all a ∈ ∆, σ is the identity on L. So φ(∆)⊥ is trivial, which
implies that ∆ = ∆ 0 .
In the infinite case, take any ∆∗ < ∆ which is finite. Corre-
spondingly, construct L∗ and ∆∗0 . By the finite case, we have
∆∗ = ∆∗0 . Notice that ∆ is the union of all such finite subgroups
∆∗ , since it has finite exponent. Correspondingly, L is the union
of all such subfields L∗ . It follows that ∆ 0 is the union of all such
∆∗0 , hence ∆ = ∆ 0 .

a.9 exercises
1. Show that the infinite recursion in the proof of Theorem A.1.10
is not necessary: namely, the field K1 constructed in the proof is
already algebraically closed.

The following exercises, up to Exercise 9, apply Galois theory

to the classical problem of geometric constructions with ruler and
compass. Let S be a set of points in the plane. Two distinct
points p, q ∈ S determine the line pq, as well as the circle having
p as center and passing through q. In turn, each pair of these
curves (lines and circles) is either disjoint or determines 1 or 2
intersection points. We can add these new points to the set S and
repeat the construction. We say that a point s is constructible with
ruler and compass from S if it can be obtained by a finite number
of steps as above.

418
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 a.9 exercises

2. Assume we start with S = {(0, 0), (1, 0)}. Prove that a point
s is constructible with ruler and compass if and only if both its
coordinates (x, y) are algebraic of degree 2n for some n.
More generally, if S is a finite set of points – regarded as com-
plex numbers a1 , . . . , ak – prove that s ∈ C is constructible if and
only if it is algebraic of degree 2n for some n over Q(a1 , . . . , an ).

3. Use Exercise 2 to prove that doubling a cube is not doable with

ruler and compass. The problem of doubling the cube consists
of starting from a segment – which is the side of a cube – and
producing a segment which is the side of a cube having twice the
volume.

4. Use Exercise 2 to prove that angle trisection is not doable with

ruler and compass. The problem of trisecting the angle consists
of starting from an angle α (determined by its vertex and two
points on its sides) and constructing the angle α/3.

5. Determine for which n > 3 the regular n-gon is constructible

with ruler and compass. Your answer should predict that the
regular 9-gon is not constructible, but the regular 17-gon is.

The last classical problem with ruler and compass is the squar-
ing of the circle. This requires understanding the algebraic prop-
erties of π. We offer some exercises proving the classical theorem
of Lindemann that π is transcendental. The approach we take
is from [Fil11], where the transcendence of e is proved as well,
using similar methods.

6. Let f be a real polynomial of degree n. Define

Zt
I(t) := et−u f(u)du.
0

Prove the equality

X
n X
n
I(t) = et f(j) (0) − f(j) (t),
j=0 j=0

where f(j) denotes the j-th derivative of f.

419
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 fields

7. Let f be a real polynomial of degree n, say

X
n
f(x) = aj xj ,
j=0

and define the polynomial f by

X
n
f(x) = aj xj .
j=0

If I(t) is the quantity defined in the previous exercise, prove the

bound
|I(t)| 6 |t| e|t| f(|t|).

8. Prove the π is transcendental, as follows. Assume that θ = iπ

is algebraic, and let g be its minimal polynomial, with roots θ =
θ1 , θ2 , . . . , θr . Denote by b the leading coefficient of g. Expand
the identity
(1 + eθ1 ) · · · (1 + eθr ) = 0

to get
q + eφ1 + · · · + eφn = 0,

where each φi 6= 0 has the form 1 θ1 + · · · + r θr for some set of

i ∈ {0, 1}, and q = 2r − n ∈ N.
Fix a large prime p, and introduce the polynomial

f(x) := bnp xp−1 (x − φ1 )p · · · (x − φr )p ,

and prove that f ∈ Z[x]. With I(t) defined in Exercise 6, let

J = I(φ1 ) + · · · + I(φr ).

Prove that J is an integer, and in fact J is the sum of two terms J1

and J2 , where p! divides J2 but not J1 (for p large enough), while
(p − 1)! divides both. Conclude that |J| > (p − 1)!.
On the other hand, derive from the previous exercise a bound
of the kind |J| 6 c1 cp2 for suitable constants c1 , c2 , and observe
that this is a contradiction.

420
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 a.9 exercises

9. The problem of squaring the circle starts from a circle (de-

termined by its center and a point on the circumference) and
requires to construct a square having the same area. Use the
previous exercises to show that this is not doable with ruler and
compass.

The theory of cyclic extensions from Propositions A.8.3 and

A.8.7 generalizes to the case where characteristic of the field is
not prime to the degree of the extension, but with a twist. The
results of the following two exercises go under the name of Artin-
Schreier theory.

10. Let K be a field of positive characteristic p, L/K a cyclic ex-

tension of degree p. Prove that L = K(α), where α satisfies a
polynomial xp − x + c ∈ K[x].

11. Let K be a field of positive characteristic p, α ∈ K a root of

the polynomial f(x) = xp − x + c ∈ K[x], and let L := K(α). Prove
that either L = K or L/K is cyclic of degree p. In particular, f is
either irreducible or a product of linear factors.

12. Given a group G, denote G 0 the subgroup of G generated

by the commutators (i.e., elements of the form ghg−1 h−1 for
g, h ∈ G). The group G 0 is called the derived group of G, and the
sequence of groups defined by G(0) = G and Gn+1 = (G(n) ) 0 is
called the derived series of G. Prove that a group G is solvable if
and only if G(n) = {e} for n big enough.

13. Prove that the symmetric group Sn is solvable if and only if

n 6 4 (look at the conjugacy classes in the alternating group A5 ).

14. Let G < Sn be a subgroup of the n-th symmetric group.

Assuming that G contains a transposition and n is prime, prove
that G is the whole Sn .

15. Let f ∈ Q[x] be an irreducible polynomial of prime degree

n. Assume that f has exactly two nonreal roots. Then the Galois
group of f is the whole Sn .

16. Prove that the quintic polynomial x5 + 3x + 3 is not solvable

in radicals over Q. More generally, do this for x5 + px + p, where
p is a positive prime number.

421
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 fields

17. Construct a quartic Galois extension of Q having Galois group

Z/4Z.

18. Construct a quartic Galois extension of Q having Galois group

Z/2Z × Z/2Z.

19. Use Galois theory to find the formula for the solution to a
general equation of degree 3

f(x) = x3 + px + q = 0

over a field K of characteristic different from 2, 3. (You can as-

sume that ζ3 ∈ K. Let a1 , a2 , a3 be the roots of f, L = K(a1 , a2 , a3 ).
Corresponding to the subgroup A3 < S3 there is a field C such
that [L : C] = 3 and [C : K] = 2 – prove that L = C(a1 ). Let
σ be a generator of A3 , acting as σ(a1 ) = a2 , σ(a2 ) = a3 ,
σ(a3 ) = a1 . Follow the proof of Hilbert Theorem 90 and find
that ζ3 = β/σ(β), where

β = a1 + ζ3 a2 + ζ23 a3 .

Conclude that β3 ∈ C; then compute β3 explicitly and express it

in terms of p and q. Use the relations you find to solve for a1 , a2
and a3 in terms of β and β2 .)

20. State and prove a form of Chinese remainder theorem for

rings that are not necessarily commutative, as we used in the
proof of Theorem A.8.6.

21. Give an alternative proof of Artin’s theorem A.8.6 by induc-

tion on the number of characters (given a dependence relation
between n characters, you can get more using the fact that char-
acters are multiplicative).

22. Prove that a topological group G is profinite if and only if it

is the inverse limit (as topological groups) of a family of discrete
finite groups.

23. A topological group G is profinite if and only if it is Haus-

dorff, compact and totally disconnected.

24. Use Corollary A.5.4 to give an alternative proof of the primi-

tive element theorem.

422
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 a.9 exercises

25. Prove the fundamental theorem of algebra, namely show that C is

algebraically closed. (Prove that R does not have finite extensions
of odd degree, and that C does not have quadratic extensions. If
K/C is a finite extension, use Galois theory to show that K = C.)

423
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 BIBLIOGRAPHY

[Aki35] Yasno Akizuki, Einige Bemerkungen über primäre In-

tegritätsbereiche mit Teilerkettensatz, Proc. Phys.-Math.
Soc. Japan. 17 (1935), 327–336.

[Alo99] Noga Alon, Combinatorial nullstellensatz, Combina-

torics, Probability and Computing 8 (1999), no. 1-2,
7–29.

[AM69] Michael Francis Atiyah and Ian Grant MacDonald,

Introduction to commutative algebra, Westview Press,
1969.

[Art71] Michael Artin, On the joins of Hensel rings, Advances

in Math. 7 (1971), 282–296.

[AT51] Emil Artin and John T. Tate, A note on finite ring ex-
tensions, J. Math. Soc Japan 3 (1951), no. 1, 74–77.

[Ax68] James Ax, The elementary theory of finite fields, Annals

of Mathematics 88 (1968), no. 2, 239–271.

[Bau15] Oswald Baumgart, The quadratic reciprocity law, Stud-

ies in logic, no. 73, Birkhäuser Basel, 2015.

[BDRH+ 09] Thomas Bauer, Sandra Di Rocco, Brian Har-

bourne, Michal Kapustka, Andreas Knutsen, Wio-
letta Syzdek, and Tomasz Szemberg, A primer on
Seshadri constants, Contemporary Mathematics 496
(2009), 33–70.

[Bel] J. Bellaïche, Answer on MathOverflow, available at

https://mathoverflow.net/questions/162030/.

[Boo08] Adam Boocher, Gröbner bases and their applications,

Lecture notes, available at http://www.math.utah.
edu/~boocher/writings/kaitlyn.pdf, 2008.

425
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 Bibliography

[Buc65] Bruno Buchberger, Ein Algorithmus zum Auffinden

der Basiselemente des Restklassenringes nach einem
nulldimensionalen Polynomideal, Ph.D. thesis, Univer-
sity of Innsbruck, 1965.
[Car] Dustin Cartwright, Answer on MathOverflow, avail-
able at http://mathoverflow.net/questions/
71699/.
[Car80] A. Caruth, A short proof of the principal ideal theorem,
Quart J. Math Oxford 31 (1980), no. 4, 401.
[CD16] Abel Castillo and Rainer Dietmann, On Hilbert’s irre-
ducibility theorem, 2016, Available at https://arxiv.
org/abs/1602.00314.
[Ces] Kestutis Cesnavicius, Answer on MathOverflow,
available at http://mathoverflow.net/questions/
64399/.
[Che45] Claude Chevalley, Intersection of algebraic and alge-
broid varieties, Trans. Amer. Math. Soc. 57 (1945), 1–
85.
[CK90] Chang C. C. and Howard Jerome Keisler, Model the-
ory, Studies in logic, no. 73, Elsevier, 1990.
[CL] Antoine Chambert-Loir, Answer on MathOver-
flow, available at https://mathoverflow.net/
questions/15611/.
[Cla65] Luther Claborn, Dedekind domains and rings of quo-
tients, Pacific Journal of Mathematics 15 (1965), no. 1,
59–64.
[Cla15] Pete L. Clark, Commutative algebra, available at http:
//math.uga.edu/~pete/integral.pdf, 2015.
[CLO96] David Cox, John Little, and Donal O’Shea, Ideals, va-
rieties, and algorithms, Undergraduate texts in math-
ematics, Springer, 1996.
[Coh46] Irvin S. Cohen, On the structure and ideal theory of com-
plete local rings, Transactions of the American Math-
ematical Society 59 (1946), no. 1, 54–106.

426
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 Bibliography

[Con] Keith Conrad, Ideal factorization, Lecture notes, avail-

able at http://www.math.uconn.edu/~kconrad/
blurbs/gradnumthy/idealfactor.pdf.

[CS46] Irvin S. Cohen and Abraham Seidenberg, Prime ide-

als and integral dependence, Bull. Amer. Math. Soc. 52
(1946), no. 4, 252–261.

[CSS97] Gary Cornell, Joseph H. Silverman, and Glenn

Stevens (eds.), Modular forms and Fermat’s last theo-
rem, Springer-Verlag New York, 1997.

[dJea20] Aise Johan de Jong and et al., The stack project, 2020,
available at https://stacks.math.columbia.edu/.

[Dol03] Igor Dolgachev, Lectures on invariant theory, Lon-

don Mathematical Society Lecture Note Series, Cam-
bridge University Press, 2003, earlier version avail-
able at https://wstein.org/people/dolgachev/
invbook.ps.

[Eak68] Paul M. Eakin, The converse to a well known theorem on

Noetherian rings, Math. Ann. 177 (1968), no. 4, 278–
282.

[Eis50] Ferdinand Gotthold Max Eisenstein, Über die Irre-

ductibilität und einige andere Eigenschaften der Gle-
ichung, von welcher die Theilung der ganzen Lemniscate
abhängt, J. Reine Angew. Math. 39 (1850), no. 39, 160–
179.

[Eis95] David Eisenbud, Commutative algebra with a view to-

ward algebraic geometry, Graduate texts in mathemat-
ics, no. 150, Springer, 1995.

[Ele] George Elencwajg, Answer on MathOverflow, avail-

able at http://mathoverflow.net/questions/
16869/.

[Fil11] Michael Filaseta, The transcendence of e and π, avail-

able at http://people.math.sc.edu/filaseta/
gradcourses/Math785/Math785Notes6.pdf, 2011.

427
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 Bibliography

[Fol94] Gerald B. Folland, A course in abstract harmonic anal-

ysis, Studies in Advanced Mathematics, Taylor &
Francis, 1994.

[For73] Edward Formanek, Faithful Noetherian modules, Proc.

Amer. Math. Soc. 41 (1973), no. 2, 381–383.

[Ful84] William Fulton, Intersection theory, Ergebnisse der

Mathematik und ihrer Grenzgebiete, Springer
Berlin Heidelberg, 1984.

[gne] gnesis, Answer on Math Stack Exchange, available

at https://math.stackexchange.com/questions/
1109732/.

[Gra74] Anne Grams, Atomic rings and the ascending chain con-
dition for principal ideals, Mathematical proceedings
of the Cambridge Philosophical Society 75 (1974),
no. 3, 321–329.

[Gro66] Alexander Grothendieck, Éléments de géométrie al-

gébrique. IV. Étude locale des schémas et des morphismes
de schémas. III., Studies in logic, no. 28, Inst. Hautes
Études Sci. Publ. Math., 1966.

[GT02] Andrew Granville and Thomas J. Tucker, It’s as easy

as abc, Notices of the AMS 49 (2002), no. 10, 1224–
1231.

[Har77] Robin Hartshorne, Algebraic geometry, Graduate

texts in mathematics, no. 52, Springer, 1977.

[Haz09] Michiel Hazewinkel, Witt vectors. Part 1, vol. 6,

pp. 319–472, Elsevier, 2009, Available at https://
arxiv.org/abs/0804.3888.

[Hes05] Lars Hesselholt, Lecture notes on Witt vectors, Lecture

notes, available at https://www.math.nagoya-u.ac.
jp/~larsh/papers/s03/wittsurvey.pdf, 2005.

[HH92] Melvin Hochster and Craig Huneke, Infinite integral

extensions and big Cohen–Macaulay algebras, Annals of
Mathematics 135 (1992), 53–89.

428
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 Bibliography

[HHO18] Jürgen Herzog, Takayuki Hibi, and Hidefumi

Ohsugi, Binomial ideals, Graduate Texts in Mathe-
matics, no. 279, Springer, 2018.

[Hil93] David Hilbert, Über die vollen Invariantensysteme,

Math. Ann. 42 (1893), 313–373.

[Hir97] Morris William Hirsch, Differential topology, Gradu-

ate Texts in Mathematics, Springer New York, 1997.

[Hor76] James Hornell, The multiplicity function of a local ring,

Trans. Amer. Math. Soc. 220 (1976), 321–341.

[Huc76] James A. Huckaba, The integral closure of a Noether-

ian ring, Transactions of the American Mathematical
Society 220 (1976), 159–166.

[Hun68] Thomas W. Hungerford, On the structure of principal

ideal rings, Pacific J. Math. 25 (1968), no. 3, 543–547.

[Jen80] Christian U. Jensen, Peano rings of arbitrary global di-

mension, Journal of the London Mathematical Soci-
ety 2 (1980), 39–44.

[Jot00] P. Jothilingam, Cohen’s theorem and Eakin-Nagata theo-

rem revisited, Comm. Algebra 28 (2000), no. 10, 4861–
4866.

[Kap70] Irving Kaplansky, Commutative rings, University of

Chicago Press, 1970.

[Kata] Makoto Kato, Question on Math Stack Exchange,

available at https://math.stackexchange.com/
questions/131892/.

[Katb] Damien Katz, A guide to Cohen’s structure theo-

rem for complete local rings, Lecture notes, avail-
able at https://pdfs.semanticscholar.org/ae9d/
954d57285883acd94e55b45ed4c349f50ffc.pdf.

[Knaa] Hagen Knaf, Answer on Math Stack Exchange,

available at https://math.stackexchange.com/
questions/24406/.

429
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 Bibliography

[Knab] , Answer on Math Stack Exchange, available

at https://math.stackexchange.com/questions/
363166/.

[KO10] Keith A. Kearnes and Greg Oman, Cardinalities of

residue fields of noetherian integral domains, Communi-
cations in Algebra 38 (2010), no. 10, 3580–3588.

[KR00] Martin Kreuzer and Lorenzo Robbiano, Computa-

tional commutative algebra 1, Springer Berlin Heidel-
berg, 2000.

[Lec57] Christer Lech, On the associativity formula for multi-

plicities, Arkiv för Matematik 3 (1957), no. 4, 301–
314.

[Leq85] Yves Lequain, A local characterization of Noetherian

and Dedekind rings, Proc. Amer. Math. Soc. 94 (1985),
no. 3, 369–370.

[LLL82] Arjen Klaas Lenstra, Hendrik Willem Lenstra, and

László Lovász, Factoring polynomials with rational co-
efficients, Mathematische Annalen 261 (1982), 515—-
534.

[Mar77] Daniel A. Marcus, Number fields, Universitext,

Springer-Verlag New York, 1977.

[Mat70] Hideyuki Matsumura, Commutative algebra, Mathe-

matics lecture notes series, no. 56, W. A. Benjamin,
1970.

[Mat86] , Commutative ring theory, Cambridge Univer-

sity Press, 1986.

[MG02] Daniele Micciancio and Shafi Goldwasser, Complex-

ity of lattice problems, Kluwer Academic Publishers,
2002.

[Mir11] Diego Mirandola, Some basic ideas about Groebner

bases and their application to Euclidean geometry,
Postgraduate students Seminar, available at
http://math.sun.ac.za/wp-content/uploads/

430
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 Bibliography

2011/04/Diego-Mirandola-Groebner-bases.pdf,
2011.

[MM18] Gunter Malle and B. Heinrich Matzat, Inverse Ga-

lois theory, Springer Monographs in Mathematics,
Springer-Verlag, 2018.

[Mur] Daniel Murfet, Cohen’s theorem, Lecture notes,

available at http://therisingsea.org/notes/
CohensTheorem.pdf.

[Nag62] Masayoshi Nagata, Local rings, Wiley, 1962.

[Nag68] , A type of subrings of a Noetherian ring, J. Math.

Kyoto Univ. 8 (1968), 465–467.

[Nag05] A. Naghipour, A simple proof of Cohen’s theorem,

Amer. Math. Monthly 112 (2005), no. 9, 825–826.

[Now97] K. J. Nowak, A proof of the criterion for multiplicity one,

Univ. Iagel. Acta Math. (1997), no. 35, 247–249.

[Pol] Jason Polak, A non-Noetherian subring of a polyno-

mial ring, available at http://blog.jpolak.org/?p=
1427.

[Rab29] J. L. Rabinowitsch, Zum Hilbertschen Nullstellensatz,

Math. Ann. 102 (1929), no. 1, 520.

[Rab14] Joseph Rabinoff, The theory of witt vectors, 2014,

Available at https://arxiv.org/abs/1409.7445.

[Ray70] Michel Raynaud, Anneaux locaux henséliens, Lecture

Notes in Mathematics, no. 169, Springer-Verlag,
1970.

[Ree61] David Rees, a-transforms of local rings and a theorem on

multiplicities of ideals, Mathematical proceedings of
the Cambridge Philosophical Society 57 (1961), no. 1,
8––17.

[Rot79] Joseph J. Rotman, An introduction to homological alge-

bra, Academic Press, 1979.

431
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 Bibliography

[RZ10] Hamid Rahkooy and Zafeirakis Zafeirakopou-

los, On Computing Elimination Ideals Using
Resultants with Applications to Gröbner Bases,
available at https://static1.squarespace.
com/static/559921a3e4b02c1d7480f8f4/t/
585c2df3d1758e618c87fca5/1482436084200/
Rahkooy.pdf, 2010.
[Sam49] Pierre Samuel, La notion de multiplicité en algèbre et en
géométrie algébriques, Ph.D. thesis, Princeton Univer-
sity, 1949.
[Sam53] , Algèbre locale, no. 123, Mémorial Sci. Math.,
1953.
[Sch99] Hans Schoutens, The use of ultraproducts in commuta-
tive algebra, Lecture Notes in Mathematics, Springer
Berlin Heidelberg, 1999.
[Sch03] René Schoof, Number theory, Lecture notes, avail-
able at http://www.mat.uniroma2.it/~schoof/tn.
html, 2003.
[Sei53] Abraham Seidenberg, A note on the dimension theory
of rings, Pacific J. Math 3 (1953), no. 2, 505–512.
[Ser88] Jean-Pierre Serre, Topics in galois theory, Lec-
ture notes by Henri Darmon, available at
http://www.ms.uky.edu/~sohum/ma561/notes/
workspace/books/serre_galois_theory.pdf,
1988.
[Ser00] , Local algebra, Springer Monographs in Math-
ematics, Springer Berlin Heidelberg, 2000.
[Sim10] Denis Simon, Selected applications of LLL in num-
ber theory, pp. 265–282, Springer Berlin Heidelberg,
Berlin, Heidelberg, 2010.
[Sla] Slade, Answer on MathOverflow, available at http://
mathoverflow.net/questions/1267419/.
[Sta69] Harold Mead Stark, On the gap in the theorem of Heeg-
ner, Journal of Number Theory 1 (1969), 16–27.

432
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 Bibliography

[Tha99] Dinesh S. Thakur, Fermat’s last theorem for regu-

lar primes, Lecture notes, available at https://www.
bprim.org/sites/default/files/d3f.pdf, 1999.

[vDdB] Remy van Dobben de Bruyn, Answer on Math-

Overflow, available at https://mathoverflow.net/
questions/242641/.
[Wei95] Charles A. Weibel, An introduction to homological al-
gebra, Cambridge Studies in Advanced Mathematics,
Cambridge University Press, 1995.
[Wit36] Ernst Witt, Zyklische Körper und Algebren der Charac-
teristik p vom Grad pn , J. Reine Angew. Math. 176
(1936), no. 176, 126–140.

[Wof] Eric Wofsey, Answer on Math Stack Exchange,

available at https://math.stackexchange.com/
questions/1515854/.
[Zak82] Abraham Zaks, Atomic rings without a.c.c. on principal
ideals, Journal of Algebra 74 (1982), no. 1, 223–231.

[ZS76a] Oscar Zariski and Pierre Samuel, Commutative al-

gebra I, Graduate Texts in Mathematics, no. 28,
Springer New York, 1976.
[ZS76b] , Commutative algebra II, Graduate Texts in
Mathematics, no. 29, Springer New York, 1976.

433
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 I N D E X O F N OTAT I O N

A∗ invertible elements of the ring A, 7

(a1 , . . . , an ) ideal generated by a1 , . . . , an , 8
Aa localization of A at the powers of a, 37
An (k) affine space of dimension n over the field k, 270
Ann M annihilator ideal of M, 25
|a|p p-adic absolute value of a, 217
 
a
p Legendre symbol of a and p, 208
AP localization of A at the prime ideal P, 37
Ass M set of associated primes of M, 72
|a|st Euclidean absolute value of a, 216
Aut(L) group of automorphisms of the field L, 397
A[[x]] ring of power series over A, 5
A[x] ring of polynomials over A, 3
A{x} ring of convergent power series over A, 6
char A characteristic of the ring A, 12
χIM (n) Hilbert function of M at I, 309
cont(f) content of the polynomial f, 83
DEGLEX graded lexicographic order, 133
DEGREVLEX graded reverse lexicographic order, 133
dim M Krull dimension of M, 304
δM Chevalley dimension of M, 305
dM Poincaré dimension of M, 308
disc(f) discriminant of f, 126
D(Q | P) decomposition group of Q over P, 187
embdim A embedding dimension of A, 330
End(M) module of endomorphisms of M, 3
e(Q, M) multiplicity of M at the primary ideal Q, 336
E(Q | P) inertia group of Q over P, 187
e(Q | P) ramification index of Q over P, 180

435
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 Index of Notation

F(A) total field of fractions of A, 37

f(B/A) conductor ideal of A in B, 169
φn (x) n-th cyclotomic polynomial, 203
Fq finite field with q elements, 382
f(Q | P) degree of inertia of Q over P, 180
G(A) class group of A, 109
Gal(L/K) Galois group of L over K, 397
GrI (A) graded ring associated to the ideal I of A, 44
h(A) class number of A, 109
Hom(M, N) module of homomorphisms from M to N, 21
ht(P) height of the prime ideal M, 315
kIk norm of the ideal I, 180
√
I radical of the ideal I, 9
I(V) ideal of functions vanishing on V, 271
J(A) Jacobson radical of A, 9
k algebraic closure of the field k, 378
per
k perfect closure of the field k, 392
sep
k separable closure of the field k, 388
√ 
K n∆ Kummer extension obtained by adding n-th roots of
∆ to K, 414
(L/K)i maximal inseparable extension of K into L, 392
(L/K)s maximal separable extension of K into L, 390
LC(f) leading coefficient of f, 133
LEX lexicographic order, 132
lim Ai direct limit of the Ai , 234
−→
lim Ai inverse limit of the Ai , 236
←−
`(M) length of the module M, 70
LM(f) leading monomial of f, 133
LT(f) leading term of f, 133
M
c completion of M, 240
hm1 , . . . , mn iA A-module generated by m1 , . . . , mn , 17
mdeg f multidegree of the polynomial f, 133
N(A) nilradical of A, 9
NL/K norm homomorphism L∗ → K∗ , 405
Ok ring of integers of the number field k, 165

436
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 Index of Notation

Pn (k) projective space of dimension n over the field k, 281

Qp field of p-adic numbers, 222
R(f, g) resultant of f and g, 121
rk M rank of the module M, 30
R(V) coordinate ring of the affine variety V, 271
S−1 A localization of the ring A at S, 36
S(f, g) S-polynomial of f and g, 137
Supp M support of M, 99
Syl(f, g) Sylvester matrix of f and g, 121
T(M) torsion submodule of M, 26
trdegK (L) transcendence degree of L over K, 380
TrL/K trace homomorphism L → K, 405
V∨ dual vector space of V, 128
V(I) affine or projective variety defined by the ideal I, 270
vol D volume of the region D, 176
vol L volume of a fundamental domain of the lattice L, 176
vp (a) p-adic valuation of a, 217
hv, wi scalar product of v and w, 213
Wp (A) p-adic Witt vectors over A, 255
WS (A) ring of Witt vectors over A, 262
ζn a primitive n-th root of 1, 204
Zp ring of p-adic integers, 222

437
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 INDEX

Abelian extension, see field ex- associativity formula, 345

tension Auslander-Buchsbaum theo-
absolute value, 216, 224, 227 rem, 335
archimedean, 216 automatic theorem proving, 142
equivalent, 218 Ax-Grothendieck theorem, 276
nonarchimedean, 216
p-adic, 217 Bezout’s identity, 32
ACC, see ascending chain condi- big Witt vectors, see Witt vectors
tion blow-up, 300
additivity formula, 343 Buchberger
affine space, 270 algorithm, 139
affine variety, see algebraic vari- criterion, 138
ety
algebra, 23 Cauchy sequence, 219, 240
finitely generated, 23 Cauchy-Davenport theorem,
algebraic closure, 378 296
algebraic curve, 293 chain, 69
algebraic element, 376 character of a group, 410
algebraic independence, 379 characteristic, 12
algebraic variety class number, 200
affine, 270 coefficient field, 363
projective, 281 coefficient ring, 364
quasiprojective, 298 Cohen’s structure theorem, 363,
altitude formula, see Nagata’s 366
height formula Cohen’s theorem, 74
annihilator, 25 Cohen-Macaulay rings, viii
Artin’s theorem on indepen- cokernel, 20
dence of characters, combinatorial Nullstellensatz,
410 296
Artin-Rees lemma, 243 commutators, 421
Artin-Schreier theory, 421 complete intersection, 300
Artin-Tate lemma, 76 complete metric space, 219
Artinian ring, see ring completion, 220, 240, 247, 279
ascending chain condition, 57 completion of a group, 240
associated prime, see prime complex, 20

439
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 Index

composite of fields, 376 DVR, see discrete valuation ring

conductor ideal, 169
content of a polynomial, 83 Eakin-Nagata theorem, 77
convex body, 197 Eisenstein’s criterion, 84
coordinate ring of an affine va- element
riety, 271 homogeneous, 44
cusp, 292 nilpotent, 7, 101
cyclic extension, see field exten- elimination theory, 122, 140
sion embedding dimension, 330
cyclotomic Euclidean algorithm, 32
field, 204 exact sequence, 20
polynomial, 203 short, 20
split, 25
DCC, see descending chain con- extension of an ideal, 23
dition
decomposition group, 187 Fermat’s last theorem, 117, 212
Dedekind domain, see Dedek- field
ing ring algebraically closed, 378
Dedekind field, 209 finite, 381
Dedekind ring, see ring of fractions, see fractions
degree, 339 residue, 41
degree of inertia, 180 field extension
derived group, 421 Abelian, 408, 416
descending chain condition, 57 algebraic, 377
determinant trick, 152 cyclic, 408
determinantal variety, 282 degree, 377
Dickson’s lemma, 145 finite, 377
dimension Galois, 397
Chevalley, 287, 305, 306 Kummer, 413
Krull, 286, 304, 305 normal, 394
Poincaré, 288, 308 solvable, 408
transcendence, 286 transcendental, 377
Dimension theorem, 310 filtration, 46
direct limit, see limit stable, 242
direct product, see module flatness, viii
direct sum, see module Formanek’s theorem, 77
discrete valuation ring, 105, 232 fractions
discriminant, 126–128, 130 field of, 37
of a number ring, 175 total ring of, 37
divisibility chain, 80 Frobenius element, 211, 267,
divisor-stable set, 262 403
dual basis, 128 Frobenius endomorphism, 383
dual of a group, 415 fundamental domain, 176

440
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 Index

fundamental theorem of alge- Hilbert-Samuel multiplicity, see

bra, 423 multiplicity
homomorphism
Galois correspondence, 398 graded, 46
Galois extension, see field exten- of modules, 18
sion of rings, 11
Galois group, 397 Hungerford’s theorem, 372
absolute, 402 hypersurface, 270, 282
Gauss’ lemma, 83, 168
generic freeness, 73, 170 ideal, 7
ghost component, 256 coprime, 15
going down theorem, 161 definition, 309
going up theorem, 159 fractional, 107
Gorenstein rings, viii homogeneous, 45
Gröbner basis, 136 irreducible, 90
reduced, 144 irrelevant, 44, 295
graded lexicographic order, see maximal, 7
order monomial, 49, 135
graded reverse lexicographic or- of the localization, 38
der, see order primary, 89
Grassmann varieties, 298 prime, 7
greatest common divisor, 82 principal, 8
group ideal class group, 109, 199
complete, 240 ideal membership problem, 131
topological, 240 inertia group, 187
inertia-ramification formula,
Hauptidealsatz, 315 184, 185, 372
Heegner-Stark theorem, 209 inseparable degree, 389
height, 315 integral closure, 150, 153, 160,
height formula, see Nagata’s 231, 372
height formula integral domain, 6
heights, 265 integral element, 149, 160
Hensel ring, 249 intersection theory, 341
Hensel’s lemma, 224, 248, 264 inverse Galois problem, 404
Hermite’s theorem, 210 inverse limit, see limit
Heron’s formula, 146 invertible, 7
Hilbert function, 309 irreducible
Hilbert polynomial, 309 element, 80
Hilbert’s basis theorem, 63 polynomial, 84
Hilbert’s irreducibility theorem, isolated subgroup, 267
404 isomorphism
Hilbert’s Theorem 90, 409 of modules, 19
Hilbert-Poincaré series, 306 of rings, 11

441
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 Index

Jacobson radical, see radical Möbius

Jordan-Hölder theorem, 71 function, 48
Jothilingam’s theorem, 76 inversion, 48
module, 16
Kaplansky’s theorem, 86 Artinian, 57
Koszul complex, viii direct product, 24
Kronecker’s theorem, 266 direct sum, 24
Krull dimension, see dimension faithful, 150
Krull intersection theorem, 247, finitely generated, 22
264 free, 24
Krull’s principal ideal theorem, generated by a set, 17
315 graded, 45
Krull-Akizuki theorem, 166, irreducible, 101
171 Noetherian, 57
Kummer extension, see field ex- of homomorphisms, 21
tension primary, 101
Kummer’s theorem, 194 projective, 50, 111
Kummer’s theorem on Abelian support of, 99
extensions, 416 torsion-free, 26
monoid, 43
lattice, 176
monomial order, see order
leading
morphism
coefficient, 63, 133
affine, 277
monomial, 133
projective, 283
term, 133
multidegree, 133
Legendre symbol, 208
multiplicative set, 36
length, 70
multiplicity, 336, 339
lexicographic order, see order
function, 352
limit
multiplicity 1 criterion, 338
direct, 234
inverse, 236
LLL algorithm, 213 Nagata’s height formula, 328
local property, 43 Nagata’s lemma, 115
local ring, see ring Nakayama’s lemma, 22, 152
localization, 36, 279 Netwon polytope, 146
of modules, 41 nilradical, 9
universal property, 37 node, 292
Noether normalization lemma,
main theorem of Galois theory, 162
398, 402 Noetherian ring, see ring
Mason’s theorem, 117 Noetherian topology, 271
maximum common divisor, 32 norm
minimal polynomial, 377 Euclidean, 30
Minkowski’s theorem, 197 of a field extension, 405

442
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 Index

of an ideal, 180 primitive element theorem, 385

normal point, 335 primitive polynomial, 83
Northcott’s theorem, 266 principal ideal domain, 54, 107,
Nullstellensatz, 272, 295 157
number field, 165 principal ideal ring, 372
number ring, 165 product formula, 265
profinite group, 400
order projection, 13, 19, 298
elimination, 145 projective closure, 282
graded lexicographic, 133 projective module, see module
graded reverse lexico- projective space, 281
graphic, 133 projective variety, see algebraic
lexicographic, 132 variety
monomial, 131 purely inseparable extension,
weight, 145 389
order function, 351
ordered group, 227 quadratic reciprocity, 208
Ostrowski’s theorem, 224, 227 quasiprojective variety, see alge-
braic variety
p-adic numbers, 222 quotient
Pell’s equation, 211 of modules, 19
perfect closure, 370, 392 of rings, 13
perfect field, 259, 370, 391
Plücker map, 299 Rabinowitsch trick, 273
place, 218 radical
polynomial ideal, 9
solvable by radicals, 412 Jacobson, 9
symmetric, 147 of an ideal, 9
Pontryagin duality, 415 ramification index, 180
power series, see ring ramified prime, 180, 191
primary decomposition, 90, 91, rank, 30, 112
101 rational map, 298
minimal, 91, 92, 97, 102, rational normal curve, 284
104 reduction of an ideal, 372
prime regular sequence, viii
associated, 72, 92 regular variety, 291
element, 8, 80 resultant, 121, 123, 125
embedded, 95 ring, 1
ideal, see ideal Artinian, 57
in integral extensions, 159, atomic, 116
161 Boolean, 75
minimal, 64, 103 commutative, 3
regular, 212 complete, 366
prime field, 381 Dedekind, 105, 156, 164

443
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22
 Index

equicharacteristic, 364 system of parameters, 305, 330

Euclidean, 30
generated by a set, 6 tangent cone, 339
graded, 44 Teichmüller representative, 251,
integrally closed, 150, 154 257
local, 22, 38 tensor product, 26
Noetherian, 57, 63 torsion, 26
normal, 154 torsion submodule, 26
of power series, 5 total ring of fractions, see frac-
regular, 330 tions
semilocal, 304 trace, 405
unmixed, 338, 372 transcendence basis, 379
valuation, 228 transcendence degree, 380
with unit, 3 transcendental element, 376
ring of integers, 165 triangular inequality, 216
rng, 3
ruler and compass construc- uniformizer, 293
tions, 418 unique factorization
domain, 80
for ideals, 106
S-polynomial, 137
theorem, 80, 86
Samuel’s formula, 347
unmixed ring, see ring
segment of a group, 267
Segre map, 297 valuation, 228, 351
semilocal ring, see ring discrete, 232
separable closure, 388 p-adic, 217
separable degree, 386 valuation ring, see ring
separable extension, 384 variety
series of composition, 69 irreducible, 285
signature, 177 Veronese map, 284, 297
singular variety, 291 Verschiebung, 257
Smith normal form, 34 volume of a lattice, 176
snake lemma, 49
solvable extension, see field ex- Witt polynomials, 253, 261
tension Witt vectors, 250, 255, 262
solvable group, 408 big, 262
splitting field, 394
superficial element, 355, 357 Zariski tangent space, 290, 291
support, see module Zariski topology, 271, 281
Sylvester matrix, 121 Zariski’s lemma, 274
symbolic power, 98, 118 zero divisor, 7, 101

444
AMS Open Math Notes: Works in Progress; Reference # OMN:202003.110823; Last Revised: 2020-07-24 13:55:22