Introduction to Commutative Algebra

April 8, 2008
 About This Document

This document was typeset by Jason McCullough and Bart Snapp. It is a

synthesis of class notes taken during a course taught by Professor S.P. Dutta
at the University of Illinois Urbana-Champaign. Special thanks to: Michael
Dewar, Dan Lior, and Patrick Szuta, for making many comments and corrections
concerning these notes.
Please report corrections, suggestions, gripes, complaints, and criticisms to:
snapp@[Link] (Bart Snapp) or jmccullo@[Link] (Jason McCul-
lough)
Contents

0 Background 1
0.1 Operations on Ideals . . . . . . . . . . . . . . . . . . . . . . . . . 1
0.2 Chain Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
0.3 Flat Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
0.4 Localization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1 Primary Decomposition 9
1.1 Primary and Coprimary Modules . . . . . . . . . . . . . . . . . . 9
1.2 The Primary Decomposition Theorem . . . . . . . . . . . . . . . 11
1.2.1 Primary Decomposition and Localization . . . . . . . . . 14
1.2.2 Primary Decomposition and Polynomial Extensions . . . 17
1.2.3 Associated Primes . . . . . . . . . . . . . . . . . . . . . . 18
1.3 Arbitrary Modules . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2 Filtrations and Completions 25

2.1 Limits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.1.1 Direct Limits . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.1.2 Inverse Limits . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.2 Filtrations and Completions . . . . . . . . . . . . . . . . . . . . . 28
2.2.1 Topology and Algebraic Structures . . . . . . . . . . . . . 28
2.2.2 Filtered Rings and Modules . . . . . . . . . . . . . . . . . 29
2.2.3 The Topology Corresponding to a Filtration . . . . . . . . 30
2.2.4 Graded Rings and Modules . . . . . . . . . . . . . . . . . 33
2.3 Adic Completions and Local Rings . . . . . . . . . . . . . . . . . 36
2.4 Faithfully Flat Modules . . . . . . . . . . . . . . . . . . . . . . . 42

3 Dimension Theory 49
3.1 The Graded Case . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.1.1 The Hilbert Polynomial . . . . . . . . . . . . . . . . . . . 51
3.1.2 The Hilbert-Samuel Polynomial . . . . . . . . . . . . . . . 55
3.2 The Topological Approach . . . . . . . . . . . . . . . . . . . . . . 61
3.2.1 Basic Definitions . . . . . . . . . . . . . . . . . . . . . . . 61
3.2.2 The Zariski Topology and the Prime Spectrum . . . . . . 64
3.3 Systems of Parameters and the Dimension Theorem . . . . . . . 68
4 Integral Extensions 75
4.1 Basic Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.2 Normal Domains and DVRs . . . . . . . . . . . . . . . . . . . . . 85
4.3 Dedekind Domains . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.4 The Krull-Akizuki Theorem . . . . . . . . . . . . . . . . . . . . . 91
4.5 Noether’s Normalization Lemma . . . . . . . . . . . . . . . . . . 95

5 Homological Methods 103

5.1 Complexes and Homology . . . . . . . . . . . . . . . . . . . . . . 103
5.1.1 Projective Resolutions . . . . . . . . . . . . . . . . . . . . 108
5.1.2 Injective Resolutions . . . . . . . . . . . . . . . . . . . . . 114
5.2 Tor and Ext . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
5.2.1 Tor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
5.2.2 Ext . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
5.3 Homological Notions of Dimension . . . . . . . . . . . . . . . . . 129
5.3.1 Projective Dimension . . . . . . . . . . . . . . . . . . . . 129
5.3.2 Injective Dimension . . . . . . . . . . . . . . . . . . . . . 131
5.3.3 Global Dimension . . . . . . . . . . . . . . . . . . . . . . 132
5.4 The Local Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
5.4.1 Minimal Free Resolutions . . . . . . . . . . . . . . . . . . 134
5.5 Regular Rings and Global Dimension . . . . . . . . . . . . . . . . 136
5.5.1 Regular Local Rings . . . . . . . . . . . . . . . . . . . . . 136
5.5.2 Regular Rings . . . . . . . . . . . . . . . . . . . . . . . . . 142

A Diagram of Implications 144

B Diagram and Examples of Domains 146

C Table of Invariances 149

Bibliography 151

Index 153
Notation and Terminology

All rings are commutative and contain multiplicative identity, moreover we will
always insist that ring homomorphisms respect the multiplicative identity ele-
ment. Local rings are assumed to be Noetherian. Additionally, all modules are
unitary modules. We have made an attempt to be consistent with our notation:

(1) Rings are often denoted by A and B.

(2) Modules are often denoted by M or N .

(3) Fields are often denoted by k, K, L, or F .

(4) Ideals are denoted by I, J, a, and b, with m usually reserved for maximal
ideals. We will try to reserve p, q, P , and Q for prime ideals.

(5) X is often used to denote indeterminants and in general X := X1 , . . . , Xn

and x := x1 , . . . , xn with the value of n (which is possibly infinite) being
given by the context.

(6) The symbol 1M will denote the identity map 1M : M → M .

(7) The letter η will be often used to denote the canonical or natural map.

e will often stand for the map induced by ϕ.

(8) If ϕ is a map, ϕ

(9) We use the notation ⊆ for set inclusion and use ( for strict inclusion.
(10) The notion ֒→ is used to denote an injective map and ։ denotes a sur-
jective map. If a commutative diagram is drawn, the induced map will be
dashed.

(11) If A is a domain, Frac(A) will stand for the field of fractions of A.

b will often stand the m-adic completion of A.
(12) If (A, m) is a local ring, A
e will often stand for the integral closure of A.
(13) If A is a ring, A

(14) If k is a field, k will often stand for the algebraic closure of k.

Chapter 0

Background

0.1 Operations on Ideals

Definition Given two ideals I, J ⊆ A, the sum of I and J is defined as

I + J = {x + y : x ∈ I and y ∈ J}.

Exercise 0.1 Show that if I and J are ideas in a ring A, then I + J is an ideal.

Definition Given two ideals I, J ⊆ A, the product of I and J is defined as

Xn
I ·J ={ xi · yi : xi ∈ I and yi ∈ J}.
i=1

Exercise 0.2 Show that if I and J are ideas in a ring A, then I · J is an ideal.

Definition Given two ideals I, J ⊆ A, the intersection of I and J is defined

as the set-theoretic intersection of I and J.

Exercise 0.3 Show that if I and J are ideas in a ring A, then I ∩ J is an ideal.

Exercise 0.4 If a, b, c are ideals of A, show that

a(b + c) = ab + ac.

Exercise 0.5 If a and b are ideals of A and M is an A-module, show that

(a + b)M
a(M/bM ) = .
bM
Exercise 0.6 Assuming that a, b, c are ideals of A and that a ⊇ b or a ⊇ c,
prove the modular law:

a ∩ (b + c) = a ∩ b + a ∩ c.

1
0.1. OPERATIONS ON IDEALS

Definition Two ideals I, J ⊆ A are called comaximal if I + J = (1).

Remark Sometimes people use the term coprime for comaximal. We will
refrain from doing this to avoid confusion later on with coprimary ideals.
Exercise 0.7 Show that if I and J are comaximal ideals of A, then IJ = I ∩J.
Definition An element x of a ring is called nilpotent if there exists n ∈ N
such that xn = 0.
Definition The set
√ of nilpotent elements of a ring√A is called the nilradical
of A. We will use 0 to denote this set. Note that 0 is an ideal.
We can generalize the idea of the nilradical as follows:
√
Definition The radical of an ideal I is denoted by I and is defined to be
the set √
I = {x ∈ A : xn ∈ I for some n ∈ N}.
√
Note that I is an ideal.
Proposition 0.8 (Properties of Radicals) If I, J are ideals of A, the following
hold:
√
(1) I ⊆ I.
√ p√
(2) I = I.
√ √ √ √
(3) IJ = I ∩ J = I ∩ J.
√
(4) pn = p.
Proposition 0.9 Given a ring A, the radical of an ideal I is equal to the
intersection of all the prime ideals which contain I.
√
Proof (⊆) Suppose that x ∈ I. Then xn ∈ I and for each prime ideal
containing I, xn ∈ p. Since p is prime, x ∈ p. Thus
√ \
I⊆ p.
p⊇I

(⊇) By the Correspondence Theorem, the prime ideals of A containing I

correspond bijectively to the ideals of A/I, hence we reduce to the case where
I = (0).
Suppose that x is not nilpotent. We’ll show that
\
x∈/ p.
p⊇(0)

Consider the set of ideals of A:

S = {a : xi ∈
/ a for i > 0}

2
 CHAPTER 0. BACKGROUND

Note that (0) ∈ S and that S may be ordered by inclusion. Now let C be any
chain of ideals in S. This chain has an upper bound in S, namely the ideal:
[
a
a∈C

Hence by Zorn’s Lemma, S has a maximal element, call it p. We claim that p is

prime. Suppose that a, b ∈
/ p. Hence (a) + p and (b) + p are ideals not contained
in S. Thus for some m, n ∈ N:

xm ∈ (a) + p and xn ∈ (b) + p

Moreover,
xm+n ∈ (ab) + p
and so we see that ab ∈ / p. Hence p is prime and x ∈
/ p. Thus x is not in the
intersection of the prime ideals of A. ¥

WARNING 0.10 The union of two ideals is not generally an ideal.

Example 0.11 Consider k[X, Y ] where k is a field. Now (X) ∪ (Y ) is not an

ideal as it contains X, Y , but not X + Y .

Despite this fact there are some things we can say about unions of ideals.

Lemma 0.12 (Prime Avoidance) Let A be a ring and I be an ideal of A. If

p1 , . . . , pn are prime ideals such that

I 6⊆ pi for all i,

then
n
[
I 6⊆ pi .
i=1

Remark The above lemma is called prime avoidance as if I 6⊆ pi for all i, then
there is some element of a ∈ I which avoids being contained in any pi .

Definition If a and b are ideals of A, then the colon ideal (b :A a) is defined

as follows:
(b :A a) = {x ∈ A : xa ⊆ b}
Moreover if M is an A-module and N is a submodule of M , then this can be
generalized to modules by defining the colon submodule (N :M a) as follows:

(N :M a) = {x ∈ M : xa ⊆ N }

Remark Sometimes the colon ideal is called the ideal quotient. However, we
will refrain from using that terminology as the word quotient is overused in
mathematics.

3
0.2. CHAIN CONDITIONS

Definition If M is an A-module, the annihilator of M over A, is defined as:

AnnA (M ) := (0 :A M ) = {x ∈ A : xm = 0 for all m ∈ M }

Proposition 0.13 (Properties of the Colon Ideal) If a, b are ideals of A, the

following hold:
(1) b ⊆ (b :A a).
(2) (b :A a)a ⊆ b.
(3) ((c :A b) : a) = (c :A ba) = ((c :A a) :A b).
T T
(4) ( i bi :A a) = i (bi :A a).
P T
(5) (b :A i ai ) = i (b :A ai ).

0.2 Chain Conditions

Definition Given a ring A, an A-module M is Noetherian if it satisfies the
following equivalent conditions:
(1) Every non-empty set of submodules has a maximal element.
(2) M satisfies the ascending chain condition (ACC) on submodules.
(3) Every submodule in M is finitely generated.
Definition A ring A is Noetherian if it is a Noetherian A-module. Note that
the only A-submodules of A are the ideals of the ring A.
Definition Given a ring A, an A-module M is Artinian if it satisfies the
following equivalent conditions:
(1) Every non-empty set of submodules has a minimal element.
(2) M satisfies the descending chain condition (DCC) on submodules.
Definition A ring A is Artinian if it is an Artinian A-module. Note that the
only A-submodules of A are the ideals of the ring A.
Example 0.14 Z is a Noetherian ring which is not an Artinian ring.
Example 0.15 If k is a field k[x1 , . . . , xn , . . . ], is neither Artinian nor Noethe-
rian.
Example 0.16 Any field is an Artinian ring.
Remark As we will soon state, every Artinian ring is also a Noetherian ring.
Example 0.17 A finite Abelian group is a Z-module which is both Noetherian
and Artinian.

4
 CHAPTER 0. BACKGROUND

Proposition 0.18 If A is Noetherian and M is a finitely generated A-module,

then M is Noetherian.
Example 0.19 A = k[x1 , . . . , xn , . . . ] is a finitely generated k[x1 , . . . , xn , . . . ]-
module which is not Noetherian.
Example 0.20 Zp∞ is an Artinian Z-module which is not a Noetherian Z-
module. Recall that Zp∞ is the Z-submodule of Q/Z generated by

{1/pn : p is a prime in Z}.

Definition A chain of A-modules

M = M0 ) M1 ) · · · ) Mn = (0)

is a Jordan-Hölder chain, also known as a composition series, if for each

i, Mi /Mi+1 ≃ A/m for some maximal ideal m in A.
Proposition 0.21 Each composition series for M has the same length.
Definition The length of an A-module, denoted by ℓA (M ), is the length of a
composition series for M . That is, if

M = M0 ) M1 ) · · · ) Mn = (0)

is a composition series, then ℓA (M ) = n.

Proposition 0.22 M has finite length if and only if M is both Artinian and
Noetherian.
Proposition 0.23 Given a short exact sequence of A-modules

0 → M ′ → M → M ′′ → 0

we have that:
(1) M is Noetherian if and only if both M ′ and M ′′ are Noetherian.
(2) M is Artinian if and only if both M ′ and M ′′ are Artinian.
Proposition 0.24 Given a short exact sequence of A-modules

0 → M ′ → M → M ′′ → 0

such that ℓA (M ) is finite, then length is an additive function, that is,

ℓA (M ) = ℓA (M ′ ) + ℓA (M ′′ ).

In particular, ℓA (M ) is finite if and only if ℓA (M ′ ) and ℓA (M ′′ ) are finite.

Theorem 0.25 (Hilbert’s Basis Theorem) If A is a Noetherian ring, then
A[x] is a Noetherian ring.

5
0.3. FLAT MODULES

Corollary 0.26

(1) Z[x1 , . . . , xn ] is Noetherian.

(2) If k is a field, then k[x1 , . . . , xn ] is Noetherian.

Exercise 0.27 Show that if A is Noetherian, then A[[x]] is Noetherian.

Lemma 0.28 If A is a ring with an ideal I which is not prime, then there
exist I1 and I2 each containing I such that I ⊇ I1 I2 .

Lemma 0.29 If A is a Noetherian ring with an ideal I, I must contain a finite

product of prime ideals.

Lemma 0.30 Every Artinian domain is a field.

Theorem 0.31 A ring A is Artinian if and only if A is Noetherian and every

prime ideal is maximal.

Proof See [17]. ¥

Corollary 0.32 If A is an Artinian ring, then A is Noetherian.

0.3 Flat Modules

Definition An A-module F is flat if

M ֒→ M ′

implies that
M ⊗A F ֒→ M ′ ⊗A F.

Remark If F is flat then − ⊗A F and F ⊗A − are exact functors from the

category of A-modules to the category of A-modules.

Proposition 0.33 An A-module F is flat if and only if for all finitely generated
A-modules M and M ′ , M ֒→ M ′ implies that

M ⊗A F ֒→ M ′ ⊗A F.

Proposition 0.34 If A and B are rings, with B an A-module, B is flat over A

if and only if any solution x ∈ B of homogeneous equations
X
ai,j xj = 0 where a∈A

is a linear combination of solutions in A.

Proposition 0.35 Every free module is flat.

Example 0.36 Q is flat over Z but Q is not free over Z.

6
 CHAPTER 0. BACKGROUND

0.4 Localization
Let U be a subset of A which is closed under multiplication and contains 1.
Given an A-module M , we may now write “fractions”
m
u
where m ∈ M and u ∈ U . For m ∈ M and u′ ∈ U , we will say
′

m m′
= ′ when (mu′ − m′ u)z = 0
u u
for some z ∈ U . This defines an equivalence relation and we denote the set
of equivalence classes by U −1 M . We can put the canonical module structure
on U −1 M . If M is an A-algebra, we may put the canonical ring structure on
U −1 M .
WARNING 0.37 The homomorphism A → U −1 A defined via x 7→ x/1 is not
generally injective, consider A = Z/6Z and U = {1, 3}.
Proposition 0.38 (Universal Property of Localization) If ϕ : A → B is a
homomorphism of rings such that ϕ(u) is a unit in B for all u ∈ U , then there
exists a unique homomorphism ϕ e : U −1 A → B making the diagram below
commute.
η
A U −1 A
ϕ
ϕ
e

B
Proposition 0.39 If M is an A-module and U is a multiplicatively closed
subset of A, then there is a canonical isomorphism M ⊗A U −1 A ≃ U −1 M .
Moreover, this isomorphism is functorial. That is, if f : M → N , the following
diagram commutes
ηM
M ⊗A U −1 A U −1 M
f ⊗1U −1 A U −1 f

N ⊗A U −1 A ηN U −1 N

where ηM and ηN represent the canonical isomorphisms.

Proposition 0.40 U −1 A is a flat A-module.
As an immediate corollary we have:
Corollary 0.41 If the following sequence of A-modules is exact
0 → M ′ → M → M ′′ → 0,
then
0 → U −1 M ′ → U −1 M → U −1 M ′′ → 0
is exact.

7
0.4. LOCALIZATION

Definition For a ∈ A, if U = {1, a, a2 , a3 , . . . }, we denote U −1 M by either

Ma or M [ a1 ].
Definition If A is a domain, then the field of fractions is given by:

Frac(A) := (A − {0})−1 A

Definition If U = A − p where p is a prime ideal of A, we denote U −1 M by

Mp . We say this as “M localized at p.”
Definition A ring A is local if it is Noetherian and has a unique maximal
ideal. When dealing with local rings, one often writes

(A, m) or (A, m, k)

to denote the ring and its maximal ideal or the ring, its maximal ideal, and
k = A/m respectively.
Remark Some authors do not insist that local rings are Noetherian. We will
call local rings which are not Noetherian quasilocal.
Proposition 0.42 Ap is a local ring with maximal ideal p.
Definition For a ring A with a prime ideal p we define the residue field of
p, denoted κ(p), by

κ(p) := Ap /pAp = (A/p)p = Frac(A/p).

Now we come to some very important properties of localization:

Proposition 0.43 Given an A-module M , the following are equivalent:
(1) M 6= 0.
(2) Mm 6= 0 for some maximal ideal m of A.
(3) Mp 6= 0 for some prime ideal p of A.
Corollary 0.44 Let M , M ′ , and M ′′ , be A-modules. The following are equiv-
alent:
(1) 0 → M ′ → M → M ′′ → 0 is exact.
(2) 0 → Mp′ → Mp → Mp′′ → 0 is exact for all prime ideals p in A.
′ ′′
(3) 0 → Mm → Mm → Mm → 0 is exact for all maximal ideals m in A.
The above corollary tells us that if we can show that an A-module homo-
morphism is injective (resp. surjective) after localizing at an arbitrary prime or
maximal ideal, then we can conclude that the homomorphism is injective (resp.
surjective). This is why it is sometimes said that the injectivity or surjectivity
of an A-module homomorphism is a local property.

8
Chapter 1

Primary Decomposition

1.1 Primary and Coprimary Modules

In this section we will mostly consider the case when A is Noetherian and A-
modules are finitely generated.

Definition Let A be a Noetherian ring. A nonzero finitely generated A-module

M is coprimary if for all a ∈ A, the map defined via multiplication by a
a
M −→ M

is injective or nilpotent.

Proposition 1.1 If M is coprimary, then the set

a
p = {a ∈ A : M −→ M is nilpotent}

forms a prime ideal in A.

Proof Suppose that a ∈ / p and b ∈

/ p. Then the map defined via multiplication
by a is injective and the map defined via multiplication by b is injective. Hence
the map defined via multiplication by ab is injective and we see that ab ∈/ p. ¥

Definition The coprimary module M which gives the above prime ideal p is
called p-coprimary.

Proposition 1.2 If N is any nonzero submodule of a finitely generated A-

module M and M is p-coprimary, then N is also p-coprimary.

Proof Exercise. ¥

Proposition 1.3 If an A-module M is p-coprimary, then we have an injection

A/p ֒→ M.

9
1.1. PRIMARY AND COPRIMARY MODULES

Proof Consider any m ∈ M such that m 6= 0 and let

I = AnnA (m) = {a ∈ A : am = 0}.
Since A is Noetherian, p is finitely generated, and so we write p = (p1 , . . . , pt ).
Because M is p-coprimary, there exist ni such that for each i,
pni i m = 0.
Thus there exists n such that pn ⊆ I ⊆ p. If p = I, we are done since we have
an injection A/I ֒→ M . If p 6= I, there exists a l such that pl ⊆ I but pl−1 6⊆ I.
Take x ∈ pl−1 − I and consider ϕ : A → M via 1 7→ m. We have that
Ker(ϕ) = I. Hence
A/I ֒→ M.
Since px ⊆ pl ⊆ I, AnnA (x) = p. Hence x has a nonzero image in A/p.
Moreover, if there exists a ∈ A such that ax ∈ I, then since A/I ֒→ M and M
is p-coprimary, we have that ax ∈ p and thus a ∈ p. So
x
A/p ֒→ A/I ֒→ M.
This is the injection we were looking for. ¥
Proposition 1.4 Let M be a finitely generated p-coprimary A-module. If
A/q ֒→ M
for some prime ideal q ⊆ A, then q = p.
Proof Exercise. ¥
Definition Let A be a Noetherian ring. Given finitely generated A-modules
N ֒→ M , N is called primary (resp. p-primary) if M/N is coprimary (resp.
p-coprimary).
Proposition 1.5 If p is a prime ideal, then p is p-primary.
Proof Exercise. ¥
Proposition 1.6 I is a primary ideal of A if and only if whenever xy ∈ I and
y∈/ I, we then have xn ∈ I for some n ∈ N.
Proof (⇒) Assume I is a primary ideal of A. So the map
x
A/I −→ A/I
is either injective or nilpotent. Considering xy ∈ I where y ∈ / I, we see that
xy = 0 and y 6= 0. Thus x must be a nilpotent map, and so xn ∈ I.
(⇐) Assuming whenever xy ∈ I and y ∈ / I, we have xn ∈ I for some n ∈ N
clearly forces the map
x
A/I −→ A/I
to be injective or nilpotent. ¥

10
 CHAPTER 1. PRIMARY DECOMPOSITION

√
Corollary 1.7 If I is p-primary, then I = p.

Proof This follows from Proposition 1.1 and the proof of the forward direction
of Proposition 1.6. ¥

Proposition 1.8 Let A be a Noetherian ring. Suppose I is an ideal and m is

a maximal ideal such that mn ⊆ I ⊆ m. Then I is m-primary.
a
Proof Let a ∈ A. We wish to show that if a ∈ m, the map A/I −→ A/I is
/ m, then the map is injective. If a ∈ m, then an ∈ mn ⊆ I.
nilpotent and if a ∈
n
Thus a (A/I) = (0) and so the map is nilpotent.
Assume a ∈/ m. Since m is maximal, m + aA = A. Thus there are x ∈ m and
y ∈ A such that x + ay = 1. Taking the nth power we get

(x + ay)n = xn + ay ′
= 1,

where y ′ is some element of A. Since xn ∈ mn ⊆ I, a and y ′ are units in A/I.

Thus the map defined by multiplication by a is an isomorphism. In particular,
it is injective. ¥

WARNING 1.9 It is not true in general that an ideal I is p-primary if

pn ⊆ I ⊆ p.

Consider the ring A = k[x, y, z]/(z 2 −xy) where k is a field. Set p = (x, z), where
x = x + (z 2 − xy) and z = z + (z 2 − xy). Then A/p ≃ k[x, y, z]/(x, z) ≃ k[y],
which is a domain. Thus p is a prime ideal.
We claim p2 is not p primary. To see this, note that x ∈ / p2 and y ∈ / p, but
y
xy = z 2 ∈ p2 . Thus the map A/p2 −→ A/p2 is nilpotent. It follows that p2 is
not p-primary.

Exercise 1.10 Suppose A is a UFD and p = (p) where p is a prime element.

Show that pn is p-primary for all n > 0.

Exercise 1.11 Let p = (xz − y 2 , x3 − yz, x2 y − z 2 ) in k[x, y, z] where k is a

field. Show that p is a prime ideal. Is p2 p-primary?

1.2 The Primary Decomposition Theorem

Definition If M is an A-module, a proper submodule N ( M is called irre-
ducible if N 6= N1 ∩ N2 for any submodules N1 , N2 of M that properly contain
N.

Lemma 1.12 Let A be a Noetherian ring and M be a finitely generated A-

module. Any proper submodule N of M can be expressed as a finite intersection
of irreducible submodules of M .

11
1.2. THE PRIMARY DECOMPOSITION THEOREM

Proof Suppose not and let S be the collection of proper submodules of M

that cannot be expressed as a finite intersection of irreducible submodules of
M . By assumption, S = 6 ∅. M is Noetherian, so S has a maximal element
N0 . Then N0 is not irreducible, so there are submodules N1 , N2 of M which
properly contain N0 such that N0 = N1 ∩ N2 . Note that N1 and N2 are proper
submodules of M . Since N0 is maximal in S, N1 and N2 can be expressed
as finite intersections of irreducible submodules. Thus N0 can be expressed as
a finite intersection of submodules, contradicting N0 ∈ S. Therefore, S = ∅
and every proper submodule of M can be expressed as a finite intersection of
irreducible submodules of M . ¥

Lemma 1.13 Let A be a Noetherian ring and M be a finitely generated A-

module. Suppose N is an irreducible submodule of M . Then N is a primary
submodule of M .

Proof To show that N is a primary submodule of M , we must show that

M = M/N is coprimary. Since N is irreducible in M , (0) = N is irreducible in
M . Let a ∈ A and consider the map ϕ : M → M given by ϕ(m) = am.

Ker(ϕ) ⊆ Ker(ϕ2 ) ⊆ · · ·

forms an ascending chain of submodules of M . Since M is Noetherian, M is

Noetherian. Thus the above chain of submodules halts; that is, there is an
integer n such that

Ker(ϕn ) = Ker(ϕn+1 ) = Ker(ϕn+2 ) = · · · .

Set g = ϕn . Then Ker(g) = Ker(g 2 ) from which it follows that

Im(g) ∩ Ker(g) = (0).

Since (0) is irreducible, either Im(g) = (0) or Ker(g) = (0). If Im(g) = (0), then
an M = (0) and ϕ is nilpotent. If Ker(g) = (0), then Ker(ϕ) = (0) and ϕ is
injective. Thus M is coprimary and N is a primary submodule of M . ¥

Lemma 1.14 Let M be a finitely generated A-module. If N1 and N2 are both

p-primary submodules of M , then N1 ∩ N2 is also p-primary.

Proof By definition, M/N1 and M/N2 are p-coprimary. It follows easily from
the definition that M/N1 ⊕ M/N2 is also p-coprimary. Consider the map:

ϕ : M → M/N1 ⊕ M/N2
m 7→ (m + N1 , m + N2 )

Then Ker(ϕ) = N1 ∩ N2 and we have an injection:

M/(N1 ∩ N2 ) ֒→ M/N1 ⊕ M/N2

Thus by Proposition 1.2, M/(N1 ∩ N2 ) is p-coprimary. Therefore N1 ∩ N2 is

p-primary. ¥

12
 CHAPTER 1. PRIMARY DECOMPOSITION

Theorem 1.15 (Primary Decomposition Theorem) Let A be a Noetherian

ring. If N is a proper submodule of a finitely generated A-module M , we can
write
\n
N= Ni
i=1
such that:
(1) Each Ni is pi -primary for some prime ideal pi .
(2) If i 6= j, then pi 6= pj .
Ts
(3) If N = i=1 Ni′ is another such decomposition where Ni′ is p′i -primary for
i = 1, . . . , s, then
{p1 , . . . , pn } = {p′1 , . . . , p′s }
and in particular, n = s.

Proof By Lemmas Tn1.12, 1.13, and 1.14, we can express N as a finite intersec-
tion of submodules i=1 Ni where for each i, Ni is pi -primary, with the prime
ideals {p1 , . . . , pn } distinct.
To finish the proof, we will show:

p ∈ {p1 , . . . , pn } ⇔ A/p ֒→ M/N.

(⇒) Suppose that p ∈ {p1 , . . . , pn }. WLOG assume p = p1 . Write

N2 /N ∩ · · · ∩ Nn /N ≃ (N2 ∩ · · · ∩ Nn )/N
≃ (N1 + (N2 ∩ · · · ∩ Nn ))/N1
⊆ M/N1 ,

with the middle line following from the Second Isomorphism Theorem. Since
we have an injection

N2 /N ∩ · · · ∩ Nn /N ֒→ M/N1 ,

and since M/N1 is p-coprimary, N2 /N ∩ · · · ∩ Nn /N is p-coprimary by Propo-

sition 1.2. Thus by Proposition 1.3, we have injections

A/p ֒→ N2 /N ∩ · · · ∩ Nn /N ֒→ M/N.

(⇐) Now suppose we we have an injection ι : A/p ֒→ M/N . Consider the

map

ϕ : M → M/N1 ⊕ · · · ⊕ M/Nn
m 7→ (m + N1 , . . . , m + Nn )

Clearly Ker(ϕ) = N , and so we see that

n
M
ϕ : M/N ֒→ M/Ni
i=1

13
1.2. THE PRIMARY DECOMPOSITION THEOREM

Ln
is an injection. For i = 1, . . . , n let πi : i=1 M/Ni → M/Ni be the projection
map onto the ith coordinate. Then we have the following commutative diagram:
n
M
ϕ
M/N M/Ni
i=1
ι πi

ιi
A/p M/Ni

We wish to show that ιi is injective for some i = 1, . . . , n. Suppose not, then

Ker(ιi ) 6= (0) for all i. Since A/p is a domain,
(0) 6= Ker(ι1 ) · · · Ker(ιn ) ⊆ Ker(ι1 ) ∩ · · · ∩ Ker(ιn ) = Ker(ϕ ◦ ι).
This contradicts that ϕ ◦ ι is an injection. Thus ιi : A/p ֒→ M/Ni is an injection
for some i. By assumption M/Ni is pi -coprimary, so again by Proposition 1.3,
we see that A/p is pi -coprimary. By Proposition 1.4, p = pi ∈ {p1 , . . . , pn }. ¥

1.2.1 Primary Decomposition and Localization

Proposition 1.16 Let M be a p-coprimary A-module and let U be a multi-
plicatively closed subset of A. The following hold:
(1) If p ∩ U = ∅, then U −1 M is U −1 p-coprimary.
(2) If p ∩ U 6= ∅, then U −1 M = 0.
a
Proof (1) We need to show if ua ∈ U −1 p, then U −1 M −→
u
U −1 M is nilpotent
a −1 1 −1
whenever u ∈ U p. Since u is a unit in U A, we can assume u = 1. If a ∈ p,
a an
then M −→ M is nilpotent. So for some integer n > 0, M −→ M is the zero
an
map. Thus U −1 M −→ U −1 M is the zero map.
a a
/ p, then M −→ M is injective. So U −1 M −→ U −1 M is injective by
If a ∈
the exactness of localization.
(2) If p ∩ U 6= ∅, then there is some u ∈ p ∩ U . Since M is p-coprimary,
u un
M −→ M is nilpotent. So there is an integer n > 0 such that M −→ M is the
un
zero map. Thus U −1 M −→ U −1 M is the zero map. Since un is a unit in U −1 A,
multiplication by un is an isomorphism. Thus U −1 M = (0). ¥
Corollary 1.17 If N is p-primary and p ∩ U = ∅, then U −1 N is p-primary.
Theorem 1.18 Let A be a Noetherian ring and suppose N is a proper sub-
module of a finitely generated A-module M . Let
N = N1 ∩ · · · ∩ Nn
be a primary decomposition for N where Ni is pi -primary. Then
U −1 N = U −1 Ni1 ∩ · · · ∩ U −1 Nim

14
 CHAPTER 1. PRIMARY DECOMPOSITION

is a primary decomposition of U −1 N in U −1 M where m 6 n and ij ∈ {1, . . . , n}

for j = 1, . . . , m.

Proof First note that if N = N1 ∩ N2 , then U −1 N = U −1 N1 ∩ U −1 N2 . To

see this, consider the commutative diagram of exact sequences below.

0 0 0

0 N1 ∩ N2 N1 N1 /(N1 ∩ N2 ) 0

0 N2 M M/N2 0

0 N2 /(N1 ∩ N2 ) M/N1 M/(N1 + N2 ) 0

0 0 0

By tensoring this diagram with U −1 A and by diagram chasing, the claim follows.
Therefore, U −1 N = U −1 N1 ∩ · · · ∩ U −1 Nn . By Proposition 1.16, if {i1 , . . . , im }
are the indices such that pim ∩U = ∅, then U −1 N = U −1 Ni1 ∩· · ·∩U −1 Nim . ¥

Exercise 1.19 Let T be a submodule of U −1 M . Show that there exists a

submodule N of M such that U −1 N = T .

Remark By Theorem 1.18 and the above exercise, we know how to find a
primary decomposition of any submodule of U −1 M .

Lemma 1.20 Suppose that N ( M is p-primary and U is a multiplicatively

closed set. Consider the canonical map:

η : M → U −1 M
m
m 7→
1
The following hold:

(1) If p ∩ U = ∅, then η −1 (U −1 N ) = N .

(2) If p ∩ U 6= ∅, then η −1 (U −1 N ) = M .

Proof (1) Let x ∈ η −1 (U −1 N ), thus x/1 ∈ U −1 N . Hence for some y ∈ N :

x y
=
1 u
Thus there is an element v ∈ U such that

(xu − y)v = 0
xuv = yv.

15
1.2. THE PRIMARY DECOMPOSITION THEOREM

From this we see that xuv ∈ N . Since U is multiplicatively closed, uv ∈ U .

However,
M uv M
−→
N N
is injective as M/N is p-coprimary and uv ∈ / p. Since xuv ∈ N we see that
x ∈ N . Thus η −1 (U −1 N ) ⊆ N . The other containment is clear.
(2) This follows by Proposition 1.16. ¥
Using the techniques of localization and the above lemma, we are able to
say more about different primary decompositions of the same module:
Proposition 1.21 Suppose
N1 ∩ · · · ∩ Nn and N1′ , ∩ · · · ∩ Nn′
are two primary decompositions of N ( M where Ni and Ni′ are pi -primary. If
pi is a minimal prime in {p1 , . . . , pn }, then Ni = Ni′ .
Proof Suppose pi is minimal in {p1 , . . . , pn } and let η : M → U −1 M be the
canonical map. Take U = A − pi . Then
U −1 N = U −1 N1 ∩ · · · ∩ U −1 Nn = U −1 N1′ ∩ · · · ∩ U −1 Nn′ .
Since pi is minimal, pj ∩ U 6= ∅ for all j 6= i. Thus by Proposition 1.16,
U −1 Ni = U −1 Ni′ . Therefore by Lemma 1.20,
Ni = η −1 (U −1 Ni ) = η −1 (U −1 Ni′ ) = Ni′ .
¥
Definition Primes appearing in a primary decomposition that are not minimal
are called the embedded primes.
Proposition 1.22 Let U be a multiplicatively closed subset of A, and consider
the canonical map:
η : M → U −1 M
m
m 7→
1
Suppose N is a submodule of U −1 M such that N is U −1 p-primary for some
prime ideal p. Then η −1 (N ) is p-primary.
Proof Exercise. ¥
Let p be a prime ideal, and η : A → Ap is the canonical map:
η : A → Ap
a
a 7→
1
So pn Ap is a pAp -primary ideal, since pAp is maximal in Ap . By Proposi-
tion 1.22, η −1 (pn Ap ) is p-primary.

16
 CHAPTER 1. PRIMARY DECOMPOSITION

Definition Using the above notation, η −1 (pn Ap ) is called the nth symbolic
power of p and is denoted by p(n) . Note that p(n) ⊇ pn and that p(n) is
p-primary.
Exercise 1.23 Can you find a prime ideal p such that pn 6= p(n) ? If so what
is it? If not why not?

1.2.2 Primary Decomposition and Polynomial Extensions

Definition For any module M we will write denote by M [x] the A[x]-module
M ⊗A A[x].
Proposition 1.24 Let N be a proper submodule of M . Suppose N is p-
coprimary. Then N [x] is p-coprimary.
f (x)
Proof Let f (x) = a0 + a1 x + · · · + at xt . We wish to show that N [x] −→ N [x]
is nilpotent if f (x) ∈ p[x] and injective otherwise. First suppose f (x) ∈ p[x].
an i
Since N is p-coprimary, we can pick n > 0 such that the map N −→ N is the
m
zero map for i = 0, . . . , t. Then for m > tn, f (x) N [x] = 0.
Now suppose f (x) ∈ / p[x]. We proceed with two cases.
a0
Case 1. Suppose a0 ∈ / p. Then N −→ N is injective. So for any nonzero
element of N [x]
n(x) = n0 + n1 x + · · · + ns xs
we may write
n(x) = ni xi + · · · + ns xs
where ni 6= 0 and nj = 0 for j < i. Then

f (x)n(x) = f0 ni xi + (higher degree terms) 6= 0.

f (x)
So N [x] −→ N [x] is injective.
Case 2. Suppose a0 ∈ p. By assumption there is some ai ∈
/ p. So we may
write
f (x) = g(x) + h(x)
where g(x) and h(x) are nonzero and such that all the coefficients of g(x) are in
p and no coefficient of h(x) is in p. Suppose there is some n(x) ∈ N [x] such that
f (x)n(x) = 0. Then (g(x) + h(x))n(x) = 0. Therefore g(x)n(x) = −h(x)n(x).
Similarly:

g(x)2 n(x) = g(x)(g(x)n(x))

= g(x)(−h(x)n(x))
= −h(x)(g(x)n(x))
= h(x)2 n(x)

Inductively, we get that g(x)m n(x) = (−1)m h(x)m n(x). Since g(x) ∈ p[x], there
is some m > 0 such that g(x)m n(x) = 0. Since all coefficients of h(x) lie outside

17
1.2. THE PRIMARY DECOMPOSITION THEOREM

p, (−1)m h(x)m n(x) 6= 0 by Case 1. This is a contradiction. Thus f (x)n(x) 6= 0

f (x)
and again N [x] −→ N [x] is injective. ¥

Corollary 1.25 If N is p-primary, then N [x] is p[x]-primary.

Theorem 1.26 Let A be a Noetherian ring, N be a proper submodule of a

finitely generated A-module M , and
n
\
N= Ni
i=1

be a primary decomposition of N where Ni is pi -primary. Then

n
\
N [x] = Ni [x]
i=1

is a primary decomposition of N [x] where Ni [x] is pi [x]-primary.

Proof Note that if Ni is pi -primary, then Ni [x] is pi [x]-primary by Corol-

lary 1.25. Since A[x] is a free A-module, A[x] is a flat A module. Therefore
N [x] = N1 [x] ∩ · · · ∩ Nn [x]. ¥

1.2.3 Associated Primes

Definition Let ATbe a Noetherian ring and M be a finitely generated A-
n
module. If N = i=1 Ni is a primary decomposition of N ⊆ M such that
Ni is pi -primary, then the prime ideals p1 , p2 , . . . , pn are called the essential
primes of N . If N = 0, then the prime ideals p1 , . . . , pn are called the associ-
ated primes of M and are denoted by AssA (M ), or Ass(M ) when there is no
confusion.

The following are corollaries of the definition and theorems above:

Corollary 1.27 Let A be a Noetherian ring. Given a finitely generated A-

module M , a submodule N is p-primary if and only if AssA (M/N ) = {p}.

Corollary 1.28 Let A be a Noetherian ring and M be a finitely generated

A-module. A prime ideal p is in AssA (M ) if and only if there is an A-module
homomorphism

A/p ֒→ M
1 + p 7→ x,

where x is a nonzero element of M which is killed by p. Note that any nonzero

x ∈ M which is killed by p defines an injection via the above map.

18
 CHAPTER 1. PRIMARY DECOMPOSITION

Exercise 1.29 What is wrong with the following argument: Consider the poly-
nomial ring k[x, y] where k is a field. Since the prime ideal (x) is clearly (x)-
primary, by Corollary 1.27:
Ass(k[x, y]/(x)) = {(x)}
However,
k[x, y] k[x, y]
≃ k ֒→
(x, y) (x)
and so by Corollary 1.28, (x, y) ∈ Ass(k[x, y]/(x)). What!?
Definition A nonzero element a ∈ A is called a zerodivisor on M if there
exists a nonzero element m ∈ M such that am = 0. A nonzero element a ∈ A
is called a nonzerodivisor on M if a is not a zerodivisor.
Exercise 1.30 If (A, m, k) is a local ring and x is a nonzerodivisor on m, then
show there exists a short exact sequence:
0 → k → m → mA → 0
where m = m/xm and mA is the maximal ideal of A = A/xA.
Corollary 1.31 Let A be a Noetherian ring, M be a finitely generated nonzero
A-module, and let
D = {a ∈ A : a is a zerodivisor on M }.
Then [
D ∪ {0} = p.
p∈AssA (M )
S
Proof (⊆) Clearly 0 ∈ p∈AssA (M ) p, since AssA (M ) 6= ∅. Let d ∈ D, then
there exists m ∈ M with m 6= 0 and dm = 0. By the Primary Decomposition
Theorem, we have that
(0) = N1 ∩ · · · ∩ Nn
where each submodule Ni of M is pi -primary. Since m 6= 0, there exists i =
1, . . . , n such that m ∈
/ Ni . Hence the image of m is nonzero in M/Ni . Since
d
M/Ni is pi -primary and since M/Ni −→ M/Ni is not injective,
n
[ [
d ∈ pi ⊆ pi = p.
i=1 p∈AssA (M )

(⊇) Suppose that a is a nonzero element of p ∈ AssA (M ). We have an

injection
A/p ֒→ M
1 + p 7→ x
where x is a nonzero element of M that is killed by p. Since 0 + p = a + p, and
these elements map to 0 and ax respectively, we see that a is a zerodivisor on
M and hence is an element of D. ¥

19
1.2. THE PRIMARY DECOMPOSITION THEOREM

Corollary 1.32 Let A be a Noetherian ring and N be a submodule of a finitely

generated A-module M . Then

AssA (N ) ⊆ AssA (M ) ⊆ AssA (N ) ∪ AssA (M/N ).

Proof If p ∈ AssA (N ), then we have injections A/p ֒→ N ֒→ M . Thus p ∈

AssA (M ). Now suppose p ∈ AssA (M ). Then we have an injection ι : A/p ֒→ M .
Case 1. Suppose that ι(A/p)∩N 6= (0). For any nonzero submodule T ⊆ A/p
t
and any nonzero t ∈ T , the map A/p −→ T is injective since A/p is a domain.
Since T = ι−1 (ι(A/p ∩ N ) is a nonzero submodule of A/p, we have injections
A/p ֒→ T ֒→ N . Thus p ∈ AssA (N ).
Case 2. Suppose that ι(A/p) ∩ N = (0). Then ι induces an injection A/p ֒→
M/N and so p ∈ AssA (M/N ). ¥

Corollary 1.33 Let A be a Noetherian ring and M be a finitely generated

A-module such that M = M1 ⊕ M2 . Then

AssA (M ) = AssA (M1 ) ∪ AssA (M2 ).

Theorem 1.34 (Prime Filtration Theorem) Let A be a Noetherian ring. For

any finitely generated, nonzero A-module M , there exists a filtration of M ,

M = M0 ) M1 ) · · · ) Mn = (0),

such that for all i = 1, . . . , n, Mi−1 /Mi ≃ A/pi where each pi is a prime ideal.
Moreover, given any such filtration, AssA (M ) ⊆ {p1 , . . . , pn }.

Proof Let S be the collection of submodules of M that have a prime filtration

6 ∅, since for any p ∈ AssA (M ), we have an injection
as stated above. S =
ι : A/p ֒→ M so that ι(A/p) ∈ S. Since M is Noetherian, S has a maximal
element, say M0 .
We claim M = M0 . Suppose not, then we have the exact sequence
ϕ
0 −→ M0 −→ M −→ M/M0 −→ 0.

By assumption M/M0 6= (0), so there exists a prime p′ ∈ AssA (M/M0 ). Thus

we have an injection j : A/p′ ֒→ M/M0 . Set T = j(A/p′ ) and Q = ϕ−1 (T ).
Then we have a new exact sequence

0 −→ M0 −→ Q −→ T −→ 0.

Since M0 has a prime filtration, and since Q/M0 ≃ T ≃ A/p′ , Q has a prime
filtration. However, Q ) M0 contradicts that M0 is maximal in S. Thus, we
must have that M = M0 .
The second part of the theorem follows from Corollary 1.32. ¥

Proposition 1.35 Let A be a Noetherian ring, M be a finitely generated A-

module, and I = AnnA (M ). Then any essential prime of I is an associated
prime of M .

20
 CHAPTER 1. PRIMARY DECOMPOSITION

Proof LSince M is finitely generated, we may write M = (α1 , . . . , αn )A. Set

n
ϕ : A → i=1 M defined by

a 7→ (aα1 , . . . , aαn ).
Ln
It’s easy to see that I = Ker(ϕ), and thus A/ Ker(ϕ) ֒→ i=1 M . By Corol-
lary 1.32, it follows that
à n !
M
AssA (A/I) ⊆ AssA M ⊆ AssA (M ).
i=1

By definition, associated primes of A/I are essential primes of I. Thus essential

primes of I are associated primes of M . ¥

Proposition 1.36 Let A be a Noetherian ring and M be a finitely generated

A-module. Define N = {a ∈ A : an M = 0 for some n > 0}. Then
\
N= p.
p∈AssA (M )

Proof Exercise. ¥

Proposition 1.37 Let P be the collection of prime ideals of A that are minimal
in AssA (A). Then
√ \
0= p.
p∈P

Proof Exercise. ¥

Proposition 1.38 Let A be a Noetherian ring, M be a finitely generated A-

module, and p be a prime ideal. The following are equivalent:

(1) p is an essential prime ideal of a submodule N of M .

(2) Mp 6= 0.

(3) p ⊇ AnnA (M ).

(4) p ⊇ q for some prime ideal q ∈ AssA (M ).

Proof Exercise. ¥

Definition The set of prime ideals p satisfying the four equivalent conditions
above are called the support of M , denoted SuppA (M ).

Corollary 1.39 Let A be a Noetherian ring and M be a finitely generated

A-module. The minimal elements of AssA (M ) are the minimal elements of
SuppA (M ).

21
1.2. THE PRIMARY DECOMPOSITION THEOREM

Exercise 1.40 Let A be a Noetherian ring and

0 → M ′ → M → M ′′ → 0

be an exact sequence of finitely generated A-modules. Show that

SuppA (M ) = SuppA (M ′ ) ∪ SuppA (M ′′ ).

Proposition 1.41 Let A be a Noetherian ring and M be a finitely generated

A-module. M has finite length if and only if AssA (M ) consists of maximal ideals
only.

Proof (⇒) Suppose we have a composition series of M ,

M = M0 ( · · · ( Mn = (0).

Then for i = 1, . . . , n, Mi /Mi+1 ≃ A/mi for some maximal ideal mi . Since every
maximal ideal is a prime ideal, this is a prime filtration. Thus by the Prime
Filtration Theorem, Theorem 1.34, AssA (M ) ⊆ {m1 , . . . , mn }.
(⇐) Now assume AssA (M ) consists of maximal ideals only. By the Prime
Filtration Theorem there is a prime filtration of M , say

M = M0 ( · · · ( Mn = (0).

Then for i = 1, . . . , n, Mi /Mi+1 ≃ A/pi for some prime ideal pi . We want to

show pi is maximal for each i. So fix p = pi . Then

(Mi /Mi+1 )p ≃ (A/p)p 6= (0).

Since (Mi )p /(Mi+1 )p ≃ (Mi /Mi+1 )p 6= (0), we have that (Mi )p 6= (0). More-
over, Mi ֒→ M implies that (Mi )p ֒→ Mp by the exactness of localization.
Therefore Mp 6= (0). By Proposition 1.38, p ⊇ q for some prime ideal q ∈
AssA (M ). ¥

Corollary 1.42 (Finite Length Criteria) Let (A, m) be a local ring and M a
finitely generated A-module. Then the following are equivalent:

(1) ℓ(M ) < ∞.

(2) M is Artinian and Noetherian.

(3) Supp(M ) = Ass(M ) = {m}.

(4) There exists t ∈ N such that mt M = 0.

22
 CHAPTER 1. PRIMARY DECOMPOSITION

1.3 Arbitrary Modules

In this section we assume our ring A is still Noetherian but A-modules are no
longer assume to be finitely generated.

Definition Let A be a Noetherian ring. A prime ideal p is an associated

prime of M if there exists an injection A/p ֒→ M . We denote this set of primes
by AssA (M ).
a
Definition If a ∈ A, the map M −→ M is called locally nilpotent if for all
m ∈ M , there is a positive integer n such that an m = 0.

Definition Let A be a Noetherian ring. An A-module M is p-coprimary if

a
the map M −→ M is locally nilpotent for all a ∈ p and is injective for all a ∈
/ p.

Definition Let A be a Noetherian ring. Given A-modules N ֒→ M , N is

p-primary if M/N is p-coprimary.

Note that these definitions agree those in the finitely generated case, and
that the definition of a locally nilpotent map reduces to the definition of a
nilpotent map for finitely generated modules.

Proposition 1.43 Let A be a Noetherian ring. Given an A-module M , there

are submodules N (p) such that
\
(0) = N (p)
p∈AssA (M )

where N (p) is p-primary.

Proof Fix p ∈ AssA (M ). Let

S = {N : N is a submodule of M and p ∈
/ AssA (N )}.

6 ∅ since (0) ∈ S. Since A is Noetherian, S has a maximal element, say

Note S =
N (p). We want to show that N (p) is p-primary. This is equivalent to saying

AssA (M/N (p)) = {p}.

Suppose that this is not the case, that is suppose there exists q ∈ AssA (M/N (p))
and q 6= p. Then
A/q ≃ M ′ /N (p) ⊆ M/N (p).
By Corollary 1.32

AssA (N (p)) ⊆ AssA (M ′ ) ⊆ AssA (M ′ /N (p)) ∪ AssA (N (p)).

/ Ass(N (p)) and as M ′ /N (p) ≃ A/q and q is q-primary,

But by assumption, p ∈
we have by Corollary 1.27 that AssA (M ′ /N (p)) = {q}. Thus p ∈ / AssA (M ′ )

23
1.3. ARBITRARY MODULES

which contradicts the maximality of N (p). Thus N (p) is p-primary. Further-

more, since  
\
AssA  N (p) = ∅
p∈AssA (M )
T
by construction, we have that (0) = p∈AssA (M ) N (p). ¥

Exercise 1.44 Let A be a Noetherian ring and

S = {I : I = AnnA (x) for some nonzero x ∈ M }.

Let J be a maximal element in S. Show that J is a prime ideal. Moreover,

conclude that J ∈ AssA (M ).

Exercise 1.45 Let f : A → B be a homomorphism of Noetherian rings. Let

M be a finitely generated B-module. Show that

AssA (M ) = {f −1 (p) : p ∈ AssB (M )}.

Hence AssA (M ) is finite even if M is not finitely generated over A.

24
Chapter 2

Filtrations and Completions

2.1 Limits

2.1.1 Direct Limits

Definition A nonempty set I is called a directed set if (I, 6) is a partially

ordered set such that for every α, β ∈ I there exists γ ∈ I with α 6 γ and
β 6 γ.

Definition A family of objects (Xα )α∈I is a direct system indexed by a

directed set I if for every α, β ∈ I with α 6 β there exists a morphism
ϕαβ : Xα → Xβ such that:

(1) ϕαα = 1Xα for all α ∈ I.

(2) For any α, β, γ ∈ I where α 6 β 6 γ, the following diagram commutes:

ϕαβ
Xα Xβ

ϕαγ ϕβγ

Xγ

Definition A direct limit, which is an example of a colimit, of a direct

system (Xα )α∈I is an object, denoted by lim(Xα ), with morphisms ϕα : Xα →
−→
lim(Xα ) such that for every α, β ∈ I with α 6 β we have ϕβ ◦ ϕαβ = ϕα .
−→
Further, for every object Y with compatible morphisms ψα : Xα → Y , there

25
2.1. LIMITS

exists a unique morphism ϕ making the diagram below commute for all α 6 β:

ϕ
Y lim(Xα )
−→
ϕβ
ψβ

Xβ
ψα ϕα
ϕαβ

Xα

Example 2.1 If we consider the category of sets, where the morphisms are set
inclusion, then given X0 ⊆ X1 ⊆ · · · ⊆ Xn ⊆ · · · ,
∞
[
lim(Xi ) = Xi .
−→
i=0

Example 2.2 If Xα are Abelian groups, then

L
Xα
lim(Xα ) =
−→ D

where D is the Abelian group generated by x′α − ϕαβ (x ′

α ) where xα ∈ Xα and
L
′ ′
xα and ϕαβ (xα ) are the images of xα and ϕαβ (xi ) in Xα .

Exercise 2.3 Suppose we have direct systems (Aα )α∈I , (Bα )α∈I , and (Cα )α∈I ,
over the directed set I and maps (ϕα ) : (Aα ) → (Bα ) and (ψα ) : (Bα ) → (Cα )
such that for every α ∈ I

ϕα ψα
0 −→ Aα −→ Bα −→ Cα −→ 0

is exact. Then
lim ϕα lim ψα
−→ lim Bα −
0 −→ lim Aα −
→ →
−→ lim Cα −→ 0
−→ −→ −→
is exact. In other words, direct limit is an exact functor from the category of
direct systems of modules over a fixed directed set to the category of modules.

Exercise 2.4 Let U be a multiplicatively closed set and M be an A-module.

Let Mu denote {1, u, u2 , . . .}−1 M . Note that the collection (Mu )u∈U for a direct
system since for any u, u′ ∈ U , we have inclusions Mu ֒→ Muu′ and Mu′ ֒→
Muu′ . Show that U −1 M = lim(Mu ).
−→
Exercise 2.5 Let M be an A-module. Show that M is the direct limit of its
finitely generated submodules.

26
 CHAPTER 2. FILTRATIONS AND COMPLETIONS

2.1.2 Inverse Limits

An inverse limit is the dual notion of a direct limit.
Definition A family of objects (Xα )α∈I is a inverse system indexed by
a directed set I if for every α, β ∈ I with α 6 β there exists a morphism
ϕαβ : Xβ → Xα such that:
(1) ϕαα = 1Xα for all α ∈ I.
(2) For any α, β, γ ∈ I where α 6 β 6 γ, the following diagram commutes:
ϕαβ
Xβ Xα

ϕβγ ϕαγ

Xγ

Definition An inverse limit, which is an example of a limit, of a inverse sys-

tem (Xα )α∈I , is an object, denoted by lim(Xα ), with morphisms ϕα : lim(Xα ) →
←− ←−
Xα such that for all α, β ∈ I with α 6 β we have ϕαβ ◦ ϕβ = ϕα . Further, for
every object Y with compatible morphisms ψα : Y → Xα , there exists a unique
morphism ϕ making the diagram below commute for all α 6 β:
ϕ
Y lim(Xα )
←−
ϕβ
ψβ

Xβ
ψα ϕα
ϕαβ

Xα

Example 2.6 If we consider the category of sets, where the morphisms are set
inclusion, then given X0 ⊇ X1 ⊇ · · · ⊇ Xn ⊇ · · · ,
∞
\
lim(Xi ) = Xi .
←−
i=0

Example 2.7 The inverse limit can be constructed as follows: For a given
inverse system, (Xα )α∈I , write
Y
lim(Xα ) = {(xα )α∈I : if α 6 β, then xα = ϕαβ (xβ )} ⊆ Xα .
←−
α∈I

The reader should check that this construction agrees with the definition of an
inverse limit.
We now will define the ring of formal power series as it will be very useful
in this chapter:

27
2.2. FILTRATIONS AND COMPLETIONS

Definition Given a ring A, we can form the ring of formal power series in
X1 , . . . , Xn over A by considering all infinite sums of the form
∞
X
ai1 ,...,in X1i1 · · · Xnin , where ai1 ,...,in ∈ A.
i=0

Sums such as these form a ring under the canonical rules for summation and
product. We denote the ring of formal power series over A in n variables by
A[[X1 , . . . , Xn ]].

Exercise 2.8 If B = A[X1 , . . . , Xn ] and I = (X1 , . . . , Xn ). Show (A/I t ) form

an inverse system. Moreover, show that

lim(B/I n ) ≃ A[[X1 , . . . , Xn ]].

←−
Exercise 2.9 Suppose we have inverse systems (Aα )α∈I , (Bα )α∈I , and (Cα )α∈I ,
over the directed set I and maps (ϕα ) : (Aα ) → (Bα ) and (ψα ) : (Bα ) → (Cα )
such that for every α ∈ I
ϕα ψα
0 −→ Aα −→ Bα −→ Cα −→ 0

is exact. Then
lim ϕ lim ψ
0 −→ lim Aα ← − α lim B ←
−→ − α
α −→ lim Cα
←− ←− ←−
is exact. In other words, inverse limit is a left exact functor from the category of
inverse systems of modules over a fixed directed set to the category of modules.

2.2 Filtrations and Completions

2.2.1 Topology and Algebraic Structures
Definition A group G is a topological group if there exists some topology
on G such that the maps:

G×G→G G→G
(x, y) 7→ xy x 7→ x−1

are both continuous maps.

Definition A ring A is a topological ring if there exists some topology on

A such that the maps:

A×A→A A×A→A A→A

(a, b) 7→ a + b (a, b) 7→ ab a 7→ a−1

are all continuous maps.

28
 CHAPTER 2. FILTRATIONS AND COMPLETIONS

Definition If A is a topological ring, an A-module M is a topological mod-

ule if there exists some topology M such that the maps:

A×A→A A×M →M
(x, y) 7→ x + y (a, x) 7→ ax

are both continuous maps.

Exercise 2.10 Let G be a topological group with identity element e. If (Nα )

is a system
T of basic neighborhoods of e, show that G is Hausdorff if and only if
{e} = α Nα . Hint: A topological space is Hausdorff if and only if the diagonal
is closed.

2.2.2 Filtered Rings and Modules

Definition If A is a ring, we call a descending chain of additive subgroups

A = A0 ⊇ A1 ⊇ · · · ⊇ An ⊇ · · ·

a filtration of A if
Ai Aj ⊆ Ai+j .
We say that a ring with a filtration is a filtered ring.

Remark Note that from the definition above, the fact that Ai Aj ⊆ Ai+j
necessitates that each Ai is an ideal of A.

Definition If A is a filtered ring with filtration (An ) and M is an A-module,

then M is a filtered module if

M = M0 ⊇ M1 ⊇ · · · ⊇ Mn ⊇ · · ·

is a descending chain of subgroups of M such that

Ai Mj ⊆ Mi+j .

Remark Note that from the definition above, the fact that Ai Mj ⊆ Mi+j
necessitates that each Mj is a submodule of M .

Definition Let M be a filtered A-module with filtration (Mn ) and let N be a

submodule of M . Then setting Nn = N ∩ Mn forms a filtration for N . This is
called the induced filtration.

Definition Let M be a filtered A-module with filtration (Mn ) which surjects

onto another A-module N via a module homomorphism

ϕ : M ։ N.

Setting Nn = ϕ(Mn ), we obtain the image filtration.

29
2.2. FILTRATIONS AND COMPLETIONS

Definition A module homomorphism ϕ : M → N of filtered modules is called

a filtered map if
ϕ(Mn ) ⊆ Nn .

Definition Suppose that ϕ : M → N is a filtered map. Then ϕ is called strict

if:
ϕ(Mn ) = ϕ(M ) ∩ Nn
| {z } | {z }
image filtration induced filtration

2.2.3 The Topology Corresponding to a Filtration

Let M be a filtered A-module—so A and M are both filtered. Treating (Mn )
as a fundamental system of open subsets of (0) we can define a topology on M .
For any x ∈ M , the fundamental system of neighborhoods around x is (x+Mn ).

Exercise 2.11 Show that the topology defined above makes M a topological
module.

Thus by Exercise 2.10, M is Hausdorff if and only if

∞
\
Mn = 0.
n=1

Definition A function of sets d : M × M → [0, ∞) is called a pseudometric

if:

(1) For all x, y ∈ M , d(x, y) = d(y, x).

(2) For all x, y, z ∈ M , d(x, y) + d(y, z) > d(x, z).

If in addition we have that for all x, y ∈ M , d(x, y) = 0 if and only if x = y,

then d is called a metric.

If M has a topology defined by a filtration as above, one may define a

pseudometric on M as follows: Fix any c ∈ (0, 1). For any x, y ∈ M define

d(x, y) := cn

where n is the integer such that (x − y) ∈ Mn − Mn+1 , if no such integer exists,

then set d(x, y) = 0. If M is Hausdorff, then we have defined a metric. We will
c to be the completion of M with respect to the metric defined by the
define M
topology associated to the filtration. We have two different ways of constructing
this completion:

First Construction Recall that a Cauchy sequence is a sequence (xn ) ∈ M

such that for all ε > 0 there exists N such that n, m > N implies

d(xn , xm ) < ε,

30
 CHAPTER 2. FILTRATIONS AND COMPLETIONS

or in other words, for every N0 there exists N such that n, m > N implies that
xn − xm ∈ MN0 . Call a sequences which converges to zero a null sequence.
Thus in our metric, a null sequence is a sequence (xn ) such that for every L
c as
there exists N such that n > N implies that xn ∈ ML . We first construct M
follows:
c := {Cauchy sequences in M }/{null sequences in M }
M

Second Construction Since M is a filtered module, let each Mn in the

filtration be an open neighborhood of 0. Thus each Mn is also closed in M since
[
M − Mn = (x + Mn ),
x∈M
/ n

which is a union of open sets. Then the quotient topology on M/Mn is the
discrete topology since 0, and hence every point, is both open and closed. Thus
M/Mn inherits the discrete topology from the quotient topology and is hence
complete with respect to the metric associated
Q∞ to the given topology. Since the
product of a complete space is complete, n=0 M/Mn is complete under the
product topology. Define
c := lim(M/Mn ) = {xn ∈ M/Mn : if n 6 m, then xm 7→ xn }
M
←−
Y∞
⊆ M/Mn .
n=0

c is then a closed subspace of a complete metric space and hence complete.

M

Exercise 2.12 Check that the two constructions above for the completion of
M are isomorphic as A-modules.

Recall that for X a metric space and Y a subspace of X then Yb = ι(Y )

b
where the bar denotes closure and ι is the inclusion map ι : X → X.

Exercise 2.13 If M is a filtered module with filtration (Mn ) and N is a sub-

module of M , then show
∞
\
N= (N + Mn )
n=0

where N is the closure of N in the filtered topology.

Exercise 2.14 Let M be an A-module. Show that the following are equivalent:

(1) M is Hausdorff.
T∞
(2) n=0 Mn = 0.

(3) M is a metric space and not just a pseudometric space.

31
2.2. FILTRATIONS AND COMPLETIONS

Exercise 2.15 If M is a filtered module with filtration (Mn ), show that

d
M c
n = ι(Mn ) = {(xn ) : xi = 0 for i 6 n and xn+i ∈ Mn for i > 1} ⊆ M .

Proposition 2.16 If M is a filtered A-module with filtration (Mn ), and M c is

its completion, then
c/M
M cn ≃ M/Mn
T∞ d c is Hausdorff even if M is not.
as A-modules. Moreover, n=0 Mn = 0, so M
Q∞
Proof Let πn : n=0 M/Mn → M/Mn denote the projection map. We leave
it to the reader to check that by restricting πn to Mc ⊆ Q∞ M/Mn we have:
n=0
µ ¶
d Mn
Ker(πn ) ≃ M n ≃ lim
←− Mn+t
t

The second statement then follows easily. ¥

Exercise 2.17 If M is a complete filtered module then the series

∞
X
xn
i=0

converges if and only if limn→∞ (xn ) = 0.

Proposition 2.18 If M is a complete filtered module, and N is a closed sub-

module of M , then M/N is complete in the quotient topology.

Proof Let (xn ) be a Cauchy sequence in M/N where x = x + N for x ∈ M .

So there is some increasing integer function f : N → N such that

xn+1 − xn ∈ M f (n) = Mf (n) + N/N.

Thus xn+1 − xn = yn + zn for some yn ∈ N and zn ∈ Mf (n) . Consider the

sequence:

x1 , x1 + z1 , x1 + z1 + z2 , x1 + z1 + z2 + z3 , ...

This is a Cauchy sequence in M . By hypothesis M is complete so the sequence

has a limit, say x. Therefore

x = lim (x1 + z 1 + · · · + z n ) = lim xn+1 .

n→∞ n→∞

Exercise 2.19 Let A be a Noetherian ring. Let m be a maximal ideal in A.

Give A the m-adic topology and give Am the mAm -adic topology. Show that
b≃A
A cm .

32
 CHAPTER 2. FILTRATIONS AND COMPLETIONS

2.2.4 Graded Rings and Modules

Definition A ring A is called a graded ring if it can be written as a direct
sum of subgroups
M∞
An ,
n=0

where Ai Aj ⊆ Ai+j . Further, elements of Ai are called homogeneous elements

of degree i. Note also that a graded ring is a filtered ring with filtration (A′n )
where
X∞
A′n = Ai .
i=n

Definition If A is graded ring:

A = A0 ⊕ A1 ⊕ · · · An ⊕ · · ·
| {z }
A+

Then A+ is called the irrelevant ideal of A.

Exercise 2.20 Show that:

(1) 1A ∈ A0 .

(2) A0 is a ring.

(3) A is Noetherian if and only if A0 is Noetherian and A+ is a finitely gen-

erated ideal of A.

Definition A module M is called a graded module if it can be written as a

direct sum of subgroups
M∞
Mn ,
n=0

where Ai Mj ⊆ Mi+j .

Definition Given a filtered ring A with the filtration (An ), the graded ring
associated to the filtration is defined to be
∞
M
Gr(A) := Ai /Ai+1
i=0

Definition Given a filtered module M with the filtration (Mn ), the graded
module associated to the filtration is defined to be
∞
M
Gr(M ) := Mi /Mi+1
i=0

33
2.2. FILTRATIONS AND COMPLETIONS

Remark c/M
Since M cn ≃ M/Mn ,

b ≃ Gr(A),
Gr(A)
c) ≃ Gr(M ).
Gr(M

Exercise 2.21 Let A be a filtered ring that is Hausdorff under the given fil-
tration. Show that if Gr(A) is an integral domain, then so is A.
Proposition 2.22 Let A be a filtered ring and M, M ′ , M ′′ be filtered A-
modules. If
f g
0 −→ M ′ −→ M −→ M ′′ −→ 0
is exact and f and g are strict, that is,

f (Mn′ ) = f (M ′ ) ∩ Mn and g(Mn ) = g(M ) ∩ Mn′′ ,

then
Gr f Gr g
0 −→ Gr(M ′ ) −→ Gr(M ) −→ Gr(M ′′ ) −→ 0
is exact.

Proof Clearly we have Im(Gr f ) ⊆ Ker(Gr g) since Im(f ) ⊆ Ker(g).

So let x ∈ Ker(Gr g). We can assume that x ∈ Mn /Mn+1 for some n
′
since any x is a finite sum of such homogeneous elements. So g(x) ∈ Mn+1 ∩
g(M ) = g(Mn+1 ) by the strictness of g. So there exists xn+1 ∈ Mn+1 such that
g(x) = g(xn+1 ). So g(x − xn+1 ) = 0. Since
f g
M ′ −→ M −→ M ′′

is exact, there exists y ∈ Mn such that f (y) = x − xn+1 ∈ f (Mn ) by the

strictness of f . Hence (Gr f )(y) = x and so Ker(Gr g) ⊆ Im(Gr f ) also. To see
the injection and surjection on the ends, simply repeat the argument modifying
it as necessary. ¥

Theorem 2.23 Let M be a complete filtered module and let N be a Hausdorff

filtered module. If f is a filtered map and

Gr f : Gr(M ) ։ Gr(N )

is surjective, then f is onto and strict. Moreover, N is complete.

Proof Let y ∈ N . Since N is Hausdorff, there is a n ∈ N such that y ∈

Nn − Nn+1 . By assumption fn : Mn /Mn+1 → Nn /Nn+1 is onto for all n. So
there exists xn ∈ Mn such that f (xn ) = y in Nn /Nn+1 , that is there exists
yn+1 ∈ Nn+1 with yn+1 = y − f (xn ). We may apply the same argument to yn+1
to get xn+1 ∈ Mn+1 such that f (xn+1 ) = yn+1 in Nn+1 /Nn+2 . Continuing in
this way we obtain a sequence (xi )i>0 with xi ∈ Mn+i such that

y − f (xn ) − f (xn+1 ) − · · · − f (xn+i ) ∈ Nn+i+1

34
 CHAPTER 2. FILTRATIONS AND COMPLETIONS

for all i. Since limn→∞ (xn ) = 0, the following sum

∞
X
x= xn+i
i=0

converges. We leave it to the reader to check that x ∈ Mn and f (x) = y. Thus

f is onto, N is complete, and f (Mn ) = Nn so f is strict. ¥

Corollary 2.24 Suppose A is complete, and M is a Hausdorff filtered A-

module. Suppose further that Gr(M ) is a finitely generated Gr(A)-module.
Let x1 , . . . , xd be elements of M such that their images generate Gr(M ). Then
M is generated by x1 , . . . , xd and M is complete.

Proof Because we can find homogeneous generators of a graded module over

a graded ring, let x1 , . . . , xd ∈ M such that xi ∈ Mni − Mni+1 and such that
x1 , . . . , xd generate Gr(M ) over Gr(A). So we have an onto map

f : Ad ։ M
ei 7→ xi

We can define a filtration on Ad by setting

(
Aei if j 6 ni
(Aei )j =
Aj−ni ei if j > ni .

This filtration guarantees that f is a filtered map. Since Gr(f ) is onto by con-
struction, the previous theorem tells us that f is onto and that M is complete.
Moreover, M is generated by x1 , . . . , xd over A. ¥

Corollary 2.25 Let A be complete and M be Hausdorff. If Gr(M ) is Noethe-

rian, then so is M .

Proof Take N ⊆ M to be any submodule. Set Nn = N ∩ Mn . Then the map

Gr(N ) → Gr(M )

is injective by Proposition 2.22. Since Gr(M ) is Noetherian, Gr(N ) is finitely

generated. By the previous corollary, N is finitely generated. ¥

Corollary 2.26 Let A be complete and M be Hausdorff. If Gr(M ) is Noethe-

rian, then every submodule of M is closed in M .

Proof By the previous corollary, every submodule of M is finitely generated

and complete. Since every complete subspace of a Hausdorff space is complete,
we are done by Corollary 2.24. ¥

35
2.3. ADIC COMPLETIONS AND LOCAL RINGS

2.3 Adic Completions and Local Rings

One of the most useful filtrations is the I-adic filtration.
Definition If M is an A-module and I is an ideal of A, then the I-adic
filtration is the filtration:

M ⊇ IM ⊇ I 2 M ⊇ · · · ⊇ I n M ⊇ · · ·

In other words, the filtration (Mn ) is given by Mn = I n M .

Definition If I is an ideal of A, then we denote the associated graded ring by
∞
M
GrI (A) := I i /I i+1 .
i=0

Similarly, given an A-module M , we have the associated graded module

∞
M
GrI (M ) := I i M/I i+1 M.
i=0

Definition If M is complete with respect to the metric defined by the I-adic

filtration, then we say that M is I-adically complete.
Definition If M is an A-module, the I-adic completion is given by
c := lim(M/I n M ).
M
←−
If (A, m) is a local ring and M is an A-module, then by the completion of
M , denoted Mc, we mean the m-adic completion of M .

Exercise 2.27 If B = A[X1 , . . . , Xn ] and I = (X1 , . . . , Xn ), show that the

I-adic completion of B is
b = lim(B/I n ) ≃ A[[X1 , . . . , Xn ]].
B
←−
Exercise 2.28 If p is a prime in Z, show that the p-adic integers are:
b p ≃ lim(Z/pn Z)
Z
←−
Exercise 2.29 To further understand what is going on, show that:

[
C[X] ⊆ C[X](X) ⊆ C{X} ⊆ C[[X]] = C[X]

where C{X} is the set of all convergent power series with respect to the (X)-adic
metric.
As an additional corollary to Theorem 2.23 we have:
Corollary 2.30 If A is a ring with a finitely generated ideal I such that:

36
 CHAPTER 2. FILTRATIONS AND COMPLETIONS

(1) A is I-adically complete and Hausdorff.

(2) A/I is Noetherian.
Then A is Noetherian.

Proof If I is generated by x1 , . . . , xd , then GrI (A) is a quotient of the poly-

nomial ring (A/I)[X1 , . . . , Xd ], and hence Noetherian. Thus A is Noetherian by
Corollary 2.24. ¥

While many of the following theorems will be stated for nonlocal rings, and
I-adic completions, the reader should be aware that the case of a local ring with
the m-adic completion is often most important.
Lemma 2.31 (Artin-Rees) If A is Noetherian with an ideal I, and M is a
finitely generated A-module with submodule N , then there exists m > 0 such
that
N ∩ I m+n M = I n (N ∩ I m M )
for all n > 0.

Proof Note in the above theorem it is always true that

N ∩ I m+n M ⊇ I n (N ∩ I m M ).

For the other containment, set

e = A ⊕ I ⊕ I2 ⊕ · · · ,
A
f = M ⊕ IM ⊕ I 2 M ⊕ · · · ,
M
and
e = N ⊕ N ∩ IM ⊕ N ∩ I 2 M ⊕ · · · .
N
Since A is Noetherian, I is finitely generated and hence A e is Noetherian as we
e
can surject A[X] onto A. Since M is finitely generated over A, M f is finitely gen-
erated over A, e and hence is also Noetherian. Thus N e is finitely generated over
e We may choose generators of N
A. e , η1 , . . . , ηk , such that each is of homogeneous
degree d1 , . . . , dk respectively, that is to say, ηi ∈ N ∩ I di M . Set

m = max{d1 , . . . , dk }.

Suppose x ∈ N ∩ I m+n M for n > 0. Then we may write

x = a1 η1 + a2 η2 + · · · + ak ηk

where deg(ai ) = m + n − di > n. Thus ai ∈ I n . So we can write

x = b(a′1 η1 + · · · + a′k ηk )

where b ∈ I n and ai = ba′i . Therefore N ∩ I m+n M ⊆ I n (N ∩ I m M ). ¥

37
2.3. ADIC COMPLETIONS AND LOCAL RINGS

Definition Let A be a ring and M be an A-module with filtration (Mn ). We

say the filtration (Mn ) is I-good if there exists an integer j such that for all
i > 0:
Mi+j = I i Mj
Remark The Artin-Rees Lemma implies that if N ⊆ M , and M is I-adically
filtered, then the induced filtration on N , (N ∩ I n M ), is I-good. Thus (I n N )
and (N ∩ I n M ) define the same topology on N and hence the completion of N
with respect to the two topologies are identical.
Definition If A is a ring we define the Jacobson radical to be
\
J(A) := m.
m maximal

Exercise 2.32 Show that x ∈ J(A) if and only if 1 − ax is a unit for all a ∈ A.
Exercise 2.33 Show that if A is I-adically complete, then I ⊆ J(A).
Theorem 2.34 (Krull’s Intersection Theorem) Let A be a ring with an ideal
I and M a finitely generated A-module such that IM = M . Then there exists
a ∈ I such that
(1 − a)M = 0.

Proof Let x1 , . . . , xd be a set of generators for M . Since IM = M we have

xi = ai,1 x1 + · · · + ai,d xd

for each i = 1, . . . , d. Therefore

x1 = a1,1 x1 + a1,2 x2 + · · · + a1,d xd

x2 = a2,1 x1 + a2,2 x2 + · · · + a2,d xd
.. .. .. .. ..
. . . . .
xd = ad,1 x1 + ad,2 x2 + · · · + ad,d xd .

We can write this in matrix form as

    
1 − a1,1 −a1,2 ··· −a1,d x1 0
 −a2,1 1 − a · ·· −a2,d  x2  0
 2,2    
 .. .. ..   ..  =  ..  .
 . . .   .  .
−ad,1 −ad,2 ··· 1 − ad,d xd 0

Let B be the above n × n matrix. Then we see that det(B) · xi = 0 for all
i = 1, . . . , d. Observe that det(B) ∈ 1 − I. Thus setting a = 1 − det(B)
completes the proof. ¥

Corollary 2.35 (Nakayama’s Lemma) Let A be a ring and I be an ideal of A

such that I ⊆ J(A). If M is a finitely generated A-module such that IM = M ,
then M = 0.

38
 CHAPTER 2. FILTRATIONS AND COMPLETIONS

Proof By Krull’s Intersection Theorem, Theorem 2.34, there exists a ∈ I

such that (1 − a)M = 0. Since I ⊆ J(A) we have that 1 − a is a unit. Therefore
M = 0. ¥

Corollary 2.36 Let (A, m) be a quasilocal ring and M be a finitely generated

A-module. If M = mM , then M = 0.
Exercise 2.37 Prove Krull’s Intersection Theorem, Theorem 2.34, assuming
you know Nakayama’s Lemma. One can start to see why Nakayama himself
said that the lemma bearing his name is a theorem of Krull and Azumaya.
Exercise 2.38 Let A be a commutative ring and let M be a finitely generated
A-module. Suppose f : M → M is surjective. Then f is an isomorphism. This
a result due to Vasconcelos.
Corollary 2.39 Let (A, m, k) be a local ring. If M is a finitely generated A-
module, then

µ(M ) :={the minimal number of generators of M }

= rankk (M/mM ).

Proof Consider a k-basis {b1 , . . . , bn } of M ⊗A k. We claim that {b1 , . . . , bn }

is a minimal set of generators for M . Write
n
X
M= bi A + mM
i=1
Pn
Setting M = M/ i=1 bi A we then have PM = mM , and so by Corollary 2.35,
n
Nakayama’s Lemma, we see that M = i=1 bi A. To see that this is minimal,
Pn−1
suppose that it is not, then we have i=1 ci A = M . But now {c1 , . . . , cn−1 }
form a basis for M ⊗A k, a contradiction as M ⊗A k is a free module of rank
n. ¥

Compare Corollary 2.39 with the following exercise:

Exercise 2.40 Let (A, m, k) be a complete local ring and let M be a Haus-
dorff A-module. Suppose there exist x1 , . . . , xn ∈ M such that x1 , . . . , xn gen-
erate M/mM over A/m. Then x1 , . . . , xn generate M over A. In particular
if rankk (M/mM ) < ∞, then M is finitely generated over A. Hint: Consider
Corollary 2.24.
To see how the Krull Intersection Theorem gets its namesake read the fol-
lowing corollaries:
Corollary 2.41 Let A be a Noetherian ring and I ⊆ J(A). If M is finitely
generated, then
∞
\
I n M = 0.
n=1

39
2.3. ADIC COMPLETIONS AND LOCAL RINGS

T∞
Proof Let N = i=1 I n M . By the Artin-Rees Lemma, Lemma 2.31, there
exists k > 0 such that for all n we have

N = N ∩ I n+k M = I n (N ∩ I k M ) = I n N.

Thus by Nakayama’s Lemma, N = 0. ¥

Corollary 2.42 Let A be a domain. If I is a proper ideal of A, then

∞
\
I n = 0.
n=1
T
Proof Let J = I n . Then we have J = IJ. By Krull’s Intersection The-
orem, Theorem 2.34, there exists a ∈ I such that (1 − a)J = 0. Since A is a
domain and since I is proper we must have that J = 0. ¥

Exercise 2.43 Let A be a Noetherian ring and let I ( J be two ideals of A.

Suppose that A is J-adically complete. Show that A is also I-adically complete.
Lemma 2.44 If A is Noetherian with an ideal I and

0→N →M →T →0

is an exact sequence of finitely generated A-modules, then

b →M
0→N c → Tb → 0

d denotes I-adic completion.

is exact, where (−)

Proof By the Artin-Rees Lemma, Lemma 2.31, (I n N ) and (N ∩ I n M ) define

the same topology on N . Hence the completions are identical. For all n we have
the following exact sequence
fn gn
0 −→ N/N ∩ I n M −→ M/I n M −→ T /I n T −→ 0.

Since by Exercise 2.9, taking the inverse limit of an inverse system of exact
sequences is left exact, we get the following exact sequence by Proposition 2.22:
f g
0 −→ lim N/N ∩ I n M −→ lim M/I n M −→ lim T /I n T.
←− ←− ←−
This can also be checked directly and in fact, the last map is onto. To see this
take (xn ) ∈ lim T /I n T . We build a preimage of (xn ) by induction. Suppose we
←−
have (yi )16i6n where yi ∈ M/I n M and such that gn (yi ) = xi . By the above
remarks we have the commutative diagram with exact rows
fn+1 gn+1
0 N/N ∩ I n+1 M M/I n+1 M T /I n+1 T 0
ϕ

fn gn
0 N/N ∩ I n M M/I n M T /I n T 0.

40
 CHAPTER 2. FILTRATIONS AND COMPLETIONS

Note that the vertical maps are all surjective. Using this fact, a simple diagram
chasing argument produces yn+1 ∈ M/I n+1 M such that ϕ(yn+1 ) = yn and
gn+1 (yn+1 ) = xn+1 . So by induction we get (yn ) ∈ lim M/I n M that maps to
←−
(xn ) via g. ¥

WARNING 2.45 If A is not Noetherian and I is an ideal in A, then the I-adic

completion is in general neither left nor right exact.
Corollary 2.46 If A is Noetherian with A-modules M and N , and I is an ideal
of A, then
c/N
M \,
b ≃ M/N
d denotes I-adic completion.
where (−)

Proof This follows from Lemma 2.44. ¥

Proposition 2.47 Let A be a Noetherian ring and I is an ideal of A. If M is

b and M
a finitely generated A-module and A c denote the I-adic completion, then

c ≃ M ⊗A A.
M b

Proof Consider the map

ϕ : M ⊗A A b→M c,
x ⊗ (an ) 7→ (an x).

Note that using the properties of tensor product and inverse limits we get that
Ln b L\ n
i=1 A ≃ ( i=1 A). Suppose M is generated by d elements. Then there is an
exact sequence of the form

As → Ad → M → 0.

Then we have the following commutative diagram with exact rows

b
As ⊗A A b
Ar ⊗A A b
M ⊗A A 0
ϕ ψ θ

cs
A cr
A c
M 0.

b is a right exact
The exactness of the first row follows from the fact that − ⊗A A
functor. The exactness of the second row follows from Lemma 2.44. Since ϕ
and ψ are isomorphisms, θ is also an isomorphism by the Five Lemma. ¥
b is the I-adic
Theorem 2.48 Let A be Noetherian and I an ideal of A. If A
completion of A, then
A→A b
b is A-flat.
is flat. That is, A

41
2.4. FAITHFULLY FLAT MODULES

Proof This is clear in light of the previous propositions. ¥

Corollary 2.49 Let M be a finitely generated A-module and let N1 , N2 ⊆ M

be submodules. Then
(1) N\ c c
1 ∩ N2 = N 1 ∩ N 2

(2) N\ c c
1 + N2 = N 1 + N2

Exercise 2.50 b be the I-adic completion of A for I a proper ideal of A.

Let A
Then
(1) For any ideal J of A, Jb ≃ J A.
b

(2) Ib ⊆ J(A).
b

(3) The maximal ideals of A are in bijective correspondence with the maximal
ideals of A that contain I. Further, if m is a maximal ideal with I 6⊆ m.
Then m b
b = A.

2.4 Faithfully Flat Modules

Definition If A is a ring and M is an A-module we call M faithfully flat if
it satisfies any of the equivalent conditions of the following theorem.
Theorem 2.51 If A is a ring and M is an A-module then the following are
equivalent:
(1) The sequence of A-modules

N ′ → N → N ′′

is exact if and only if

M ⊗A N ′ → M ⊗A N → M ⊗A N ′′

is exact.
(2) The sequence of A-modules

0 → N ′ → N → N ′′ → 0

is exact if and only if

0 → M ⊗A N ′ → M ⊗A N → M ⊗A N ′′ → 0

is exact.
(3) M is A-flat and for all A-modules N ,

M ⊗A N = 0 ⇒ N = 0.

42
 CHAPTER 2. FILTRATIONS AND COMPLETIONS

(4) M is A-flat and for all ideals I of A,

M ⊗A A/I = 0 ⇒ A = I.

(5) M is A-flat and for all maximal ideals m of A,

mM ( M.

Proof (1) ⇒ (2) Obvious.

(2) ⇒ (3) Clearly condition (2) implies that M is A-flat. Consider the
sequence of A-modules 0 → N → 0. Then

N = 0 ⇔ 0 → N → 0 is exact,
⇔ 0 → N ⊗A M → 0 is exact,
⇔ M ⊗A N = 0.

(3) ⇒ (4) Take N = A/I.

(4) ⇒ (5) Let m be a maximal ideal. Since m ( A we must have that
M/m ≃ M ⊗A A/m 6= 0. Thus mM ( M .
(5) ⇒ (3) Let N be a nonzero A-module. We must show that M ⊗A N 6= 0.
Let 0 6= x ∈ N and define

ϕ : A → N,
a 7→ a · x.

Let I = Ker(ϕ). If I is a proper ideal we can find a maximal ideal m containing

I. Then
IM ⊆ mM ( M.
Thus M/IM 6= 0 and we have an injection A/I ֒→ N . Since M is flat, M/IM ֒→
M ⊗A N . Therefore M ⊗A N 6= 0.
(3) ⇒ (1) Now suppose that we have a sequence of modules

f g
N ′ −→ N −→ N ′′

and that
M ⊗A f M ⊗A g
M ⊗A N ′ −→ M ⊗A N −→ M ⊗A N ′′
is exact. We have the following commutative diagram:

g◦f
N′ N ′′

π
g◦f
N/ Ker(g ◦ f )

43
2.4. FAITHFULLY FLAT MODULES

Since M is flat and since M ⊗A (g ◦ f ) = 0 we get the following commutative

diagram after applying M ⊗A −:
0
M ⊗A N ′ M ⊗A N ′′

M ⊗A π M ⊗A (g◦f )
M ⊗A (N/ Ker(g ◦ f ))

Therefore M ⊗A (N/ Ker(g ◦ f )) = 0. By assumption then N/ Ker(g ◦ f ) = 0.

So Ker(g ◦ f ) = N and thus g ◦ f = 0.
Now set X = Im(M ⊗A f ) = Ker(M ⊗A g). Since Im(f ) ⊆ Ker(g) ⊆ N and
since M is flat we have the following commutative diagram
γ
M ⊗A Im(f ) M ⊗A Ker(g)

α β

M ⊗A N
In M ⊗A N we have then
X ⊆ Im(α) ⊆ Im(β) ⊆ X.
Thus
0 → M ⊗A Im(f ) → M ⊗A Ker(g) → 0
is an exact sequence. Therefore so is
0 → Im(f ) → Ker(g) → 0.
Therefore Im(f ) = Ker(g), our original sequence is exact. ¥
Example 2.52 If F is a free A-module, then F is faithfully flat. To see this
note M
F = Aα
α∈I
and so if M
0 = F ⊗A N = N
α∈I
we see that N = 0.
WARNING 2.53 Recall that a module M is called faithful if
AnnA (M ) = 0.
It should be easy to see that if an A-module is faithfully flat, then it is both
faithful and flat over A. However, the converse is not true! Consider Q as a
Z-module. We know that Q is Z-flat, and AnnZ (Q) = 0 so Q is faithful, however
Q ⊗Z Z/3Z = 0 but Z/3Z 6= 0.
Hence Q is not faithfully flat over Z.

44
 CHAPTER 2. FILTRATIONS AND COMPLETIONS

Lemma 2.54 If
0→M →N →P →0
is an exact sequence of A-modules and if P is flat, then for any module T

0 → T ⊗A M → T ⊗A N → T ⊗A P → 0

is exact.

Proof Let T be any A-module. Let F be a free module that maps surjectively
onto T so that we have the exact sequence

0 → K → F → T → 0.

Then we have the following commutative diagram

K ⊗A M K ⊗A N K ⊗A P 0

0 F ⊗A M F ⊗A N F ⊗A P 0

0 T ⊗A M T ⊗A N T ⊗A P 0

0 0 0

The first two columns are clearly exact and last column is exact because P is
flat. Free modules are flat, so the middle row is exact. Clearly the first row
is exact also. It remains to check that the bottom row is exact (and not just
right exact). Since K ⊗A P → F ⊗A P is injective, one can check by a diagram
chasing argument that T ⊗A M → T ⊗A N is injective. ¥

Theorem 2.55 If f : A → B is a homomorphism of rings, then the following

are equivalent:
(1) A → B is faithfully flat.
(2) B is A-flat and for every ideal I of A, f −1 (IB) = I.
(3) B is A-flat and for every A-module M , the map

M → M ⊗A B,
x 7→ x ⊗ 1,

is injective.

45
2.4. FAITHFULLY FLAT MODULES

(4) f is injective and B/ Im(f ) is A-flat.

Proof (1) ⇒ (2) By assumption B is A-flat. So let I be an ideal of A. Set

J = f −1 (IB) so that JB = IB. Since I ⊆ J we have the exact sequence

0 → I → J → J/I → 0.

Since B is A-flat, the sequence

0 → I ⊗A B → J ⊗A B → J/I ⊗A B → 0

is also exact. Thus

J/I ⊗A B = JB/IB
= 0.

Since B is faithfully flat, we get by Theorem 2.51 that J/I = 0 and so J = I.

(2) ⇒ (3) Consider the map

fe : M → M ⊗A B,
x 7→ x ⊗A 1.

Let m ∈ Ker(fe) and consider the map

ϕ : A → mA,
1 7→ m.

Now let I = Ker(ϕ) so that A/I ≃ mA. Then since B is A-flat we have the
following commutative diagram:

ϕ
A/I ≃ mA M
0
1A/I ⊗A B fe
ϕ⊗A 1B
A/I ⊗A B M ⊗A B

Since f −1 (IB) = I, we get that the map 1A/I ⊗A B on the left must be injective.
This contradicts that fe ◦ ϕ = 0. Therefore fe must be injective.
(3) ⇒ (4) By (3) the map A → A ⊗A B sending a 7→ a ⊗ 1 is injective. So
let
0 → M ′ → M → M ′′ → 0

be an exact sequence of A-modules. Then we have the following commutative

46
 CHAPTER 2. FILTRATIONS AND COMPLETIONS

diagram

0 0 0

0 M′ M M ′′ 0

0 M ′ ⊗A B M ⊗A B M ′′ ⊗A B 0

0 M ′ ⊗A B/A M ⊗A B/A M ′′ ⊗A B/A 0

0 0 0

where all columns are exact (check it!) and the first two rows are exact by
assumption and by the flatness of B. The last row is also right exact by the
right exactness of − ⊗A B/A. A simple diagram chasing argument shows that
the last row is indeed exact and hence that B/A is A-flat. We leave this to the
reader to check, this diagram chase is sometimes called the Nine Lemma.
(4) ⇒ (1) Let
0 → M ′ → M → M ′′ → 0
be an exact sequence. We then have the same commutative diagram as in the
proof of (3) ⇒ (4). Now we have that the columns are exact by assumption, the
first row is exact by assumption, and the last row is exact since B/A is A-flat.
A diagram chasing argument shows that the middle row is exact and hence B
is A-flat.
We now show that for any A-module M , if M ⊗A B = 0 then M = 0 from
which it follows from Theorem 2.51 that B is faithfully flat over A. Consider
the exact sequence
0 → A → B → B/A → 0.
Let M be an A-module. Since B/A is A-flat, the sequence

0 → M → M ⊗A B → M ⊗A B/A → 0

is exact. Thus if M ⊗A B = 0 then M = 0. ¥

b denotes the
Theorem 2.56 If A is a Noetherian ring with an ideal I, and A
I-adic completion of A, then the following are equivalent:
b is faithfully flat.
(1) A → A

(2) I ⊆ J(A) where J(A) is the Jacobson radical of A.

47
2.4. FAITHFULLY FLAT MODULES

Proof (1) ⇒ (2) Suppose A → A b is faithfully flat. Take any maximal ideal m
b b
of A. Then mA 6= A by the previous theorem. Suppose m 6⊇ I. Then m + I = A,
so we
Pcan write 1 = x + y with x ∈ m and y ∈ I. Since y ∈ I, the element
∞
u = i=0 y i is a convergent, well-defined element of A. b But then

∞
X
xu = (1 − y) y i = 1.
i=0

Thus x is a unit b b = mA b ( A,
b a contradiction. We conclude
T in A. But x ∈ m
m ⊇ I. So I ⊆ m maximal m = J(A).
(2) ⇒ (1) Suppose Tnow that I ⊆ J(A). Then for every finitely generated
A-module M we have I n M = 0. So the I-adic topology on M is Hausdorff.
Therefore M → M c is injective. Since M is finitely generated, M c ≃ M ⊗A A.
b
So M → M ⊗A A b is injective for all finitely generated A-modules M . Thus
M ⊗A A b 6= 0. It remains then to show this for an arbitrary A-module M .
Let M now be any A-module. We can write M as a direct limit of its
finitely generated submodules, M = lim Mi . Since each x ∈ M is contained in
−→
Mi for some i, and since Mi → Mi ⊗A A b is injective for all i, M → M ⊗A A
b is
injective. ¥
b is the m-adic completion then A → A
Corollary 2.57 If A is a local ring and A b
is a faithfully flat extension.
Definition If (A, m) and (B, n) are local rings, a homomorphism ϕ : A → B
is called local if ϕ(m) ⊆ n.

Exercise 2.58 If f : A → B is a local homomorphism of local rings, f is flat

if and only if f is faithfully flat.

Example 2.59 If A = C[X1 , . . . , Xn ] and I = (X1 , . . . , Xn ), then the I-adic

completion of A is C[[X1 , . . . , Xn ]]. Is

C[X] ֒→ C[[X]]

faithfully flat? To answer this we should ask ourselves, is I ⊆ J(C[X])? The

answer to this question is “No!” Thus, C[X] ֒→ C[[X]] is flat but not faithfully
flat. However, the canonical injection (show that this is an injection)

C[X](X) ֒→ C[[X]]

is in fact faithfully flat by the above exercise.

48
Chapter 3

Dimension Theory

In this chapter, we will develop the notion of the dimension of a module three
different ways. The Dimension Theorem will then show that over a local ring,
the three notions are actually equivalent.

3.1 The Graded Case

Definition If f : Z → Z is a function, we say that f is essentially polyno-

mial or polynomial-like if there exists P ∈ Q[x] such that f (n) = P (n) for
sufficiently large n.

Remark Note that it is an easy exercise to see that such a P as in the above
definition is unique. Moreover, we define the degree of f to be the degree of
the polynomial P . If P = 0, then we say f has degree −1.

Definition Given the function f : Z → Z, define

∆f (n) := f (n + 1) − f (n).

Definition We denote by Bk (x) : Z → Z for k > 0 the function defined by

B0 (x) = 1
µ ¶
x x! (x)(x − 1) · · · (x − k + 1)
Bk (x) = := = .
k k!(x − k)! k!

49
3.1. THE GRADED CASE

Note that Bk (x) is essentially polynomial for all k > 0. Also note that
∆Bk (n) = Bk (n + 1) − Bk (n)
(n + 1)! n!
= −
k!(n − k + 1)! k!(n − k)!
(n + 1)n! (n − k + 1)n!
= −
k!(n − k + 1)! k!(n − k + 1)!
(k)n!
=
k!(n − k + 1)!
n!
=
(k − 1)!(n − (k − 1))!
= Bk−1 (n),
for k > 1.
Exercise 3.1 Let f ∈ Q[X]. Then the following are equivalent:
(1) f (n) is a Z-linear combination of the Bk (x).
(2) f (n) ∈ Z for all n > 0.
(3) f (n) ∈ Z for n sufficiently large.
(4) ∆f (n) is a Z-linear combination of the Bk (x).
Lemma 3.2 Let P be the set of essentially polynomial functions and let f :
Z → Z. The following are equivalent:
(1) f ∈ P.
(2) ∆f ∈ P.
(3) ∆r f = 0 for some r > 0.
Proof (1) ⇒ (2) ⇒ (3) Clear from the definitions.
(2) ⇒ (1) First note that if ∆f (n) = 0 for sufficiently large n, we have that
f (n + 1) − f (n) = 0, which implies that f (n + 1) = f (n) for sufficiently large n.
Because f : Z → Z, we see that f must be constant for large n, showing f ∈ P.
Now suppose that ∆f (n) = P (n) for n sufficiently large, where P (x) ∈ Q[x].
By the exercise,
t t
à t !
X X X
P (n) = ai Bk (n) = ai ∆Bk+1 (n) = ∆ ai Bk+1 (n) ,
i=0 i=0 i=0
Pt
where a0 , . . . , at ∈ Z. Let Q(x) = i=0 ai Bk+1 (n). Then if we set g(n) =
f (n) − Q(n) then for n sufficiently large, we have that
∆g(n) = ∆f (n) − ∆Q(n) = P (n) − P (n) = 0.
Hence f (n) = Q(n)+c for n sufficiently large and for some constant c. Therefore
f ∈ P.
(3) ⇒ (1) This follows from (2) ⇒ (1) applied r times. ¥

50
 CHAPTER 3. DIMENSION THEORY

3.1.1 The Hilbert Polynomial

Theorem 3.3 (Hilbert-Serre) Let A0 be an Artinian ring and A be the graded
ring
A = A0 ⊕ A1 ⊕ · · · An ⊕ · · ·
| {z }
A+

where A+ is generated by x1 , . . . , xr ∈ A1 . We see that we may think of A as

the algebra A = A0 [x1 , . . . , xr ]. Let M be a finitely generated graded A-module,

M = M0 ⊕ M1 ⊕ · · · ⊕ Mn ⊕ · · · ,

where each Mn is a finitely generated A0 -module. The following are true:

(1) ℓA0 (Mn ) < ∞ for all n > 0.
(2) Define the Hilbert function χ(M, n) := ℓA0 (Mn ). Then χ(M, n) is
essentially polynomial of degree at most r − 1.
(3) Suppose that M0 generates M over A. Then

∆r−1 χ(M, n) 6 ℓA0 (M0 )

with equality holding if and only if

ψ : M0 ⊗A0 A0 [X1 , . . . , Xr ] → M
m0 ⊗ aX1i1 · · · Xrir 7→ a · xi11 · · · xirr · m0

is an isomorphism where the Xi ’s are indeterminates.

Proof We will show this in several parts:

(1) To show ℓA0 (Mn ) < ∞, it is enough to show that each Mn is finitely
generated over A0 , as A0 is both Artinian and Noetherian. Take any Mn and
suppose that α1 , . . . , αs generate M over A with each αi homogeneous of degree
di . Each element in Mn is then a sum of elements of the form

axi11 · · · xirr αj
Pr
where a ∈ A0 and t=1 it + dj = n. Given n and dj only finitely many it ’s can
be found, so Mn is finitely generated over A0 .

(2) Proceeding by induction on r. If r = 0, then we have A = A0 and

since M is finitely generated over A this implies that Mn = 0 for n sufficiently
large. Thus χ(M, n) = 0 for n sufficiently large and we have that the degree of
χ(M, n) = −1.
Now suppose our statement is true up through r − 1. Consider the exact
sequence
xr
0 −→ N −→ M −→ M −→ L −→ 0

51
3.1. THE GRADED CASE

where N = (0 :M xr ) and L = M/xr M . So both N and L are killed by xr and

hence are both graded modules over

A = A0 [x]/r A0 [x] = A/xr A,

a graded ring with A+ generated by r − 1 elements over A0 .

Now for each n we have an exact sequence
r x
0 −→ Nn −→ Mn −→ Mn+1 −→ Ln+1 −→ 0.

Since length is additive we get

ℓA0 (Ln+1 ) − ℓA0 (Mn+1 ) + ℓA0 (Mn ) − ℓA0 (Nn ) = 0.

And so we see that

ℓA0 (Mn+1 ) − ℓA0 (Mn ) = ℓA0 (Ln+1 ) − ℓA0 (Nn ),

which shows us
∆χ(M, n) = χ(L, n + 1) − χ(N, n).
By the inductive hypothesis, the right-hand side is essentially polynomial of
degree at most r − 2. Thus ∆χ(M, n) is essentially polynomial of degree at most
r − 2. By Lemma 3.2, we have the degree of χ(M, n) is essentially polynomial
of degree at most r − 1. So by induction, we have proved (2).

(3) (⇒) Suppose that M0 generates M over A. That is α1 , . . . , αs generate

M over A and each of the αi ’s have degree 0. Consider the map

ψ : M0 ⊗A0 A0 [X1 , . . . , Xr ] → M
m ⊗ aX1i1 · · · Xrir 7→ m · a · xi11 · · · xirr .

Then ψ is a graded map of degree 0. ψ is onto as M is generated by M0 .

HencePif Fn is the free A0 -module generated by monomials X1i1 · · · Xrir such
that ij = n,
ψn : M0 ⊗A0 Fn → Mn
is onto. So we have

ℓA0 (Mn ) 6 ℓA0 (M0 ⊗A0 Fn )

6 ℓA0 (M0 ) · N,

where N is the number of monomials X1i1 · · · Xrir of degree n. But what is

this value N ? Well whatever it is, it is the same as the number of monomials
X1i1 · · · Xrir of degree n in the expression

(1 + X1 + X12 + · · · )(1 + X2 + X22 + · · · ) · · · (1 + Xr + Xr2 + · · · )

1 1 1
= · ···· · .
1 − X1 1 − X2 1 − Xr

52
 CHAPTER 3. DIMENSION THEORY

But this is nothing more than the coefficient of X n in

1
where X = X1 = X2 = · · · = Xr .
(1 − X)r

So we see that µ ¶
n+r−1
N= .
r−1
Alternatively one can realize that the above formula is the one for choosing with
replacement n items from r types of item. Now by (1) we know that χ(M, n) is
essentially polynomial of degree at most r − 1. So by χ(M, n) we shall instead
now mean the polynomial equal to χ(M, n) for large n. Thus for n sufficiently
large,

(n + r − 1)!
χ(M, n) = ℓA0 (Mn ) 6 ℓA0 (M0 ) · N = ℓA0 (M0 ) ·
(r − 1)!n!

So we see that
(n + r − 1)!
χ(M, n) 6 ℓA0 (M0 ) ·
(r − 1)!n!
and so
(n + r − 1)!
(r − 1)!χ(M, n) 6 ℓA0 (M0 ) · .
n!
Taking the limit as n goes to infinity of both sides, we get

∆r−1 χ(M, n) 6 ℓA0 (M0 ).

Now if ψ as described above is an isomorphism it is clear that

∆r−1 χ(M, n) = ℓA0 (M0 ).

(⇐) Conversely, suppose that ∆r−1 χ(M, n) = ℓA0 (M0 ). We will show that
ψ as described above is an isomorphism. Proceed by induction on ℓA0 (M0 ).
Suppose that ℓA0 (M0 ) = 1 and so M0 = A0 /m0 = k, where m0 is some maximal
ideal of A0 . Consider the exact sequence
ψ
0 −→ L −→ k[X1 , . . . , Xr ] −→ M −→ 0,

where L is an ideal of k[X]. Likewise we have an exact sequence

0 → Ln → k[X1 , . . . , Xr ]n → Mn → 0.

Now we have
ℓA0 (k[X]n ) = ℓA0 (Ln ) + ℓA0 (Mn )
and so
χ(k[X], n) = χ(L, n) + χ(M, n).

53
3.1. THE GRADED CASE

¿From here we see

∆r−1 χ(k[X], n) = ∆r−1 χ(L, n) + ∆r−1 χ(M, n).
However, by (2) we see that ∆r−1 χ(k[X], n) and ∆r−1 χ(M, n) are both of degree
zero, and since ℓA0 (M0 ) = 1, we have that they are both 1. Thus
∆r−1 χ(L, n) = 0
We claim that if L 6= (0), then ∆r−1 χ(L, n) > 0. Suppose there exists
nonzero f ∈ L and set d = deg(f ). Since L is an ideal of k[X]n , we have that
f · k[X]n ⊆ L. Thus
ℓA0 (Ln ) > ℓA0 ((k[X]f )n ).
Since k[X] is a domain, multiplication by f is injective. Therefore, f · k[X] ≃
k[X] as k-modules. Thus
µ ¶
n−d+r−1
ℓA0 ((k[X]f )n ) = ℓA0 (k[X]n−d ) = .
r−1
and so ∆r−1 χ(L, n) > 0.
Now suppose that ℓA0 (M0 ) > 1 and look at
ψ
0 −→ L −→ M0 [X] −→ M −→ 0
where L = Ker(M ). So we need to show that L = (0). Supposing L 6= (0), we
then have
χ(M0 [X], n) = χ(L, n) + χ(M, n),
and so
∆r−1 χ(M0 [X], n) = ∆r−1 χ(L, n) + ∆r−1 χ(M, n).
But ∆r−1 χ(M0 [X], n) = ℓA0 (M0 ) = ∆r−1 χ(M, n). Hence ∆r−1 χ(L, n) = 0.
We claim that this shows that L = (0). Suppose that L 6= (0). Consider a
Jordan-Hölder chain, that is
M0 ) M01 ) · · · ) M0s = (0)
such that M0i /M0i+1 ≃ A0 /mi ≃ ki where ki is a field. Now we have the exact
sequence:
0 → M0i+1 → M0i → ki → 0
Applying − ⊗A0 A0 [X1 , . . . , Xr ] we obtain the exact sequence
0 → M0i+1 [X] → M0i [X] → ki [X] → 0 (⋆)
as A0 [X1 , . . . , Xr ] is a free, hence flat, A0 -module. Since L 6= (0), there exists
i such that L ⊆ M0i [X] but L 6⊆ M0i+1 [X]. By (⋆), the image of L in ki [X] is
nonzero, call it Li . Now we have the surjection L ։ Li and hence we have
χ(Li , n) 6 χ(L, n)
r−1
∆ χ(Li , n) 6 ∆r−1 χ(L, n),

54
 CHAPTER 3. DIMENSION THEORY

but by an argument similar to the one above we see that ∆r−1 χ(Li , n) > 0 and
hence that ∆r−1 χ(L, n) > 0. So we see that L = 0 and hence our map ψ is an
isomorphism. ¥

Definition The polynomial representative of the Hilbert function χ(M, n) =

ℓA0 (Mn ) is called the Hilbert polynomial of M . We will abuse notation and
simply denote this polynomial by χ(M, n).

Remark The previous theorem is saying that if A = k[x1 , . . . , xn ] and if M is

defined as above, then:

(1) Each Mi is a finite dimensional vector space since ℓk (Mn ) = dimk (Mn ) <
∞.

(2) The dimensions of the vector spaces Mi exhibit polynomial growth.

Definition If χ(M, n) is of degree d, then the Hilbert multiplicity of M is

en (M ) := ∆d χ(M, n) = d! ad ,

where χ(M, n) = ad xd + ad−1 xd−1 + · · · + a1 x + a0 .

3.1.2 The Hilbert-Samuel Polynomial

As we have seen, if A is a graded ring and M is a finitely generated graded A-
module, then we can define the Hilbert function χ(M, n) of M . In this section,
we will investigate the Hilbert polynomial of a canonical grading that can be
put on any Noetherian ring. We will call this polynomial the Hilbert-Samuel
polynomial. However, before this can be done, we need more tools.

Lemma 3.4 If A is a quasi-local ring with maximal ideal m, and M , N , are

nonzero finitely generated A-modules, then M ⊗A N 6= 0.

Proof Arguing by contradiction, suppose M 6= 0, N 6= 0, and M ⊗A N = 0.

Then

0 = (M ⊗A N ) ⊗A A/m = M ⊗A (N ⊗A A/m)
= M ⊗A N/mN
= M ⊗A A/m ⊗A/m N/mN
= M/mM ⊗A/m N/mN.

At this point note that M/mM is a finite dimensional vector space over A/m. We
claim that M 6= 0 implies that M/mM 6= 0, for if M/mM = 0, then M = mM ,
and hence by Nakayama’s Lemma, Corollary 2.35, M = 0. Since M/mM 6= 0
as a finite dimensional vector space over A/m and since N/mN 6= 0 as a finite
dimensional vector space over A/m, we see that M/mM ⊗A/m N/mN 6= 0, a
contradiction. ¥

55
3.1. THE GRADED CASE

Lemma 3.5 If A is Noetherian and M, N are finitely generated A-modules,

then
Supp(M ⊗A N ) = Supp(M ) ∩ Supp(N )
Proof (⊆) Suppose that p ∈ Supp(M ⊗A N ). Write
0 6= (M ⊗A N )p ≃ M ⊗A N ⊗A Ap
≃ M ⊗A Np
≃ M ⊗A Ap ⊗Ap Np
≃ Mp ⊗Ap Np .
This shows us that Mp 6= (0) and that Np 6= (0).
(⊇) If p ∈ Supp(M ) ∩ Supp(N ), Mp 6= 0 and Np 6= 0 we see by Lemma 3.4
that Mp ⊗Ap Np 6= 0. This shows us that (M ⊗A N )p 6= 0 and so we see that
p ∈ Supp(M ⊗A N ). ¥
Lemma 3.6 Let A be a Noetherian ring and M a finitely generated A-module
such that for some ideal a of A, ℓ(M/aM ) < ∞. Suppose (Mi ) is a filtration of
M with respect to a, that is an M ⊆ Mn for each n. Then we have the following:
(1) ℓ(M/Mn ) < ∞.
(2) If N ֒→ M , then ℓ(N/aN ) < ∞.
Proof (1) First note that ℓ(M/aM ) < ∞ if and only if Supp(M/aM ) consists
entirely of maximal ideals. Write
Supp(M/aM ) = Supp(M ⊗A A/a) = Supp(M ) ∩ Supp(A/a),
and
Supp(M/an M ) = Supp(M ⊗A A/an ) = Supp(M ) ∩ Supp(A/an ).
But Supp(A/a) = Supp(A/an ). Thus, ℓ(M/an M ) < ∞ for each n if and only
if ℓ(M/aM ) < ∞. By construction we have the following exact sequence
M/an M → M/Mn → 0.
Hence ℓ(M/Mn ) < ∞ for all n > 0.
(2) Now if N ֒→ M ,
Supp(N/aN ) = Supp(N ) ∩ Supp(A/a) ⊆ Supp(M ) ∩ Supp(A/a).
But Supp(M ) ∩ Supp(A/a) = Supp(M/aM ), hence Supp(N/aN ) consists only
of maximal ideals and we see that ℓ(N/aN ) < ∞. ¥
Proposition 3.7 Let A be Noetherian, M finitely generated with a an ideal
of A such that ℓ(M/aM ) < ∞. Let (Mi ) be an a-good filtration of M . Then
P ((Mi ), n) := ℓ(M/Mn )
is essentially polynomial.

56
 CHAPTER 3. DIMENSION THEORY

Proof Write
∆P ((Mi ), n) = P ((Mi ), n + 1) − P ((Mi ), n) = ℓ(M/Mn+1 ) − ℓ(M/Mn ).
Since length is additive, we see that the right-hand side is ℓ(Mn /Mn+1 ) by
looking a the exact sequence
0 → Mn /Mn+1 → M/Mn+1 → M/Mn → 0.
Now
Gr(M ) = M/M1 ⊕ M1 /M2 ⊕ · · · ⊕ Mn /Mn+1 ⊕ · · · ,
Gra (A) = A/a ⊕ a/a2 ⊕ · · · ⊕ an /an+1 ⊕ · · · .
Note that while Gra (A) is Noetherian, and is generated by elements of a/a2 , A/a
is not necessarily Artinian, and hence we cannot use the Hilbert-Serre Theorem,
Theorem 3.3. By assumption we have ℓ(M/aM ) < ∞. Hence
Supp(M/aM ) = Supp(M ) ∩ Supp(A/a)
= Supp(A/ Ann(M )) ∩ Supp(A/a)
= Supp(A/ Ann(M ) ⊗A A/a)
= Supp(A/(Ann(M ) + a)),
and so we see ℓ(A/(Ann(M )+a)) < ∞. But this shows us that A/(Ann(M )+a)
is Artinian. Now take B = A/ Ann(M ) and b = (a + Ann(M ))/ Ann(M ). Each
Mn /Mn+1 is a B/b-module. Now
Grb (B) = B/b ⊕ b/b2 ⊕ · · · ⊕ bn /bn+1 ⊕ · · ·
is a graded ring over B/b which is finitely generated by elements of b/b2 and
B/b is Artinian. Moreover, Gr(M ) is a finitely generated Grb (B)-module
and ℓA (Mn ) = ℓB (Mn ). Hence by the Hilbert-Serre Theorem, Theorem 3.3,
χ(E0 (M ), n) = ℓ(Mn /Mn+1 ) is essentially polynomial and hence ∆P ((Mi ), n)
is essentially polynomial, which shows us that P ((Mi ), n) is essentially polyno-
mial. ¥
Theorem 3.8 (Samuel) If A is Noetherian, M is a finitely generated A-
module, a is an ideal of A such that ℓ(M/aM ) < ∞, and (Mi ) is a filtration of
M such that an M ⊆ Mn , then define
Pa (M, n) := ℓ(M/an M ).
The following are true:
(1) Pa (M, n) > P ((Mi ), n).
(2) Suppose that (Mi ) is a-good. Then
Pa (M, n) = P ((Mi ), n) + R(n)
where the degree of R(n) is strictly less than the degree of Pa ((Mi ), n).
In particular, Pa (M, n) and P ((Mi ), n) have the same degree and same
leading coefficient. Moreover the leading coefficient of R(n) is nonnegative.

57
3.1. THE GRADED CASE

Proof First we will prove (1). Since an M ⊆ Mn we have the exact sequence:
0 → Mn /an M → M/an M → M/Mn → 0
Thus
ℓ(M/an M ) > ℓ(M/Mn ) (⋆)
and so by definition Pa (M, n) > Pa ((Mi ), n).
Now for (2) we will start by defining
R(n) = ℓ(M/an M ) − ℓ(M/Mn ) = ℓ(Mn /an M ).
Since (Mi ) is a-good, we see that there exists m > 0 such that
an+m M ⊆ Mn+m = an Mm ⊆ an M
for some large n. Thus we have the exact sequence
0 → Mn+m /an+m M → M/an+m M → M/Mn+m → 0.
and so ℓ(M/an+m M ) > ℓ(M/Mn+m ). By (⋆) we see that
Pa (M, n + m) > Pa ((Mi ), n + m) > Pa (M, n) > Pa ((Mi ), n)
for large n. Hence Pa (M, n) has the same degree as Pa ((Mi ), n) with the same
leading coefficient. So deg(R(n)) is strictly less than that of Pa (M, n). Since
R(n) > 0 for every n, we see that the leading coefficient of R(n) must be
nonnegative. ¥
Definition If A is Noetherian, M is a finitely generated A-module, a is an
ideal of A such that ℓ(M/aM ) < ∞, then Pa (M, n) as defined above by
Pa (M, n) := ℓ(M/an M ).
is called the Hilbert-Samuel polynomial of M with respect to a.
Definition (First Notion of Dimension) If (A, m) is a local ring and M is a
finitely generated A-module denote
d(M ) := deg(Pm (M, n)).
Corollary 3.9 Properties of the Hilbert-Samuel polynomial Pa (M, n):
(1) deg(Pa (M, n)) 6 m where m is the minimal number of generators of
(a + Ann(M ))/ Ann(M ).
(2) Suppose r is the minimal number of generators of a. Then
∆r Pa (M, n) 6 ℓ(M/aM ).
Moreover, equality holds if and only if the canonical homomorphism
M∞
an M
M/aM [X1 , . . . , Xr ] → n+1
n=0
a M

is an isomorphism.

58
 CHAPTER 3. DIMENSION THEORY

Proof Follows from the Hilbert-Serre Theorem, Theorem 3.3. ¥

Proposition 3.10 If A is Noetherian, and a is an ideal of A let

0→L→M →N →0

be an exact sequence of finitely generated A-modules such that ℓ(M/aM ) < ∞.

Then
Pa (M, n) = Pa (L, n) + Pa (N, n) − R(n)
where deg(R(n)) < deg(Pa (L, n)).

Proof Since ℓ(M/aM ) < ∞, we know that L/aL and N/aN both have finite
length, hence the equation we wish to show is well-defined. We have an exact
sequence:
L M N
0→ → n → n →0
L ∩ an M a M a N
Write Ln = L ∩ an M . By the Artin-Rees Lemma, Lemma 2.31, (Li ) is a-good.
Thus ℓ(M an M ) = ℓ(N/an N ) + ℓ(L/L ∩ an M ) and so

Pa (M, n) = Pa (N, n) + Pa ((Li ), n)

= Pa (N, n) + Pa (L, n) − R(n),

by the previous theorem. ¥

Corollary 3.11 If (A, m) is local, let

0→L→M →N →0

be an exact sequence of finitely generated A-modules such that ℓ(M/mM ) < ∞.

Then
d(M ) = max{d(L), d(N )}.

Proposition 3.12 Let A be Noetherian, M finitely generated, a an ideal of

A, and ℓ(M/aM ) < ∞. Then the degree of Pa (M, n) depends only on M and
Supp(M/aM ).

Proof Let a′ be an ideal such that Supp(M/aM ) = Supp(M/a′ M ). We will

show that deg(Pa (M, n)) = deg(Pa′ (M, n)). Note that

Supp(M/aM ) = Supp(A/(Ann(M ) + a))

and that
Supp(M/a′ M ) = Supp(A/(Ann(M ) + a′ )).
Moreover note that we have that
p \ \ p
Ann(M ) + a = p= p= Ann(M ) + a′ .
p⊇(Ann(M )+a) p⊇(Ann(M )+a′ )

59
3.1. THE GRADED CASE

Since A is Noetherian, and so Ann(M ) + a and Ann(M ) + a′ are both finitely

generated ideals show us that there exists m and m′ greater than zero such that:

(Ann(M ) + a′ )m ⊆ Ann(M ) + a
′
(Ann(M ) + a)m ⊆ Ann(M ) + a′ .

We leave it as an exercise to now check that:

deg(Pa (M, n)) = deg(P(Ann(M )+a) (A, n))

deg(P(Ann(M )+a) (A, n)) = deg(P(Ann(M )+a′ ) (A, n))
deg(P(Ann(M )+a′ ) (A, n)) = deg(Pa′ (M, n))

With the three above equalities proved, the proposition is proved. ¥

Definition We define the Hilbert-Samuel multiplicity of a module M with

respect to a by
ea (M ) := ∆d Pa (M, n),
where d > 0 is the degree of the Hilbert-Samuel polynomial Pa (M, n).
Corollary 3.13 If A is Noetherian, and a is an ideal of A let

0→L→M →N →0

be an exact sequence of finitely generated A-modules such that ℓ(M/aM ) < ∞.

Then
ea (M, n) = ea (L, n) + ea (N, n).

Proof This follows from Proposition 3.10. ¥

Proposition 3.14 If A is Noetherian, M is a finitely generated with ℓ(M/aM ) <

∞ where a is an ideal of A, and m1 , . . . , mr are the maximal ideals containing
Ann(M ) + a in A, then setting ai = aAmi we have
r
X
Pa (M, n) = Pai (Mmi , n).
i=1

In other words, to study the Hilbert-Samuel polynomial, it suffices to work over

local rings.

Proof To start, note that since Pa (M, n) = PAnn(M )+a (M, n), we may assume
that M is an A/ Ann(M )-module. Thus we will assume that AnnA (M ) = 0,
and that ℓ(A/a) < ∞. Hence

AssA (A/q) = {m1 , . . . , mr }

is a set containing only maximal ideals. Now use the Primary Decomposition
Theorem, Theorem 1.15, to write

a = q1 ∩ · · · ∩ qr

60
 CHAPTER 3. DIMENSION THEORY

where each qi is mi -primary. Since A is Noetherian, for each i there exists ti

such that mtii ⊆ qi and so the qi ’s are pairwise comaximal. Thus

a = q1 · · · qr and so an = qn1 · · · qnr .

Thus by the Chinese Remainder Theorem,

A/an ≃ A/qn1 ⊕ · · · ⊕ A/qnr .

Now apply − ⊗A M to obtain

M/an M ≃ M/qn1 ⊕ · · · ⊕ M/qnr (⋆)

and define
ai := aAmi = qi Ami .
Since A/qi is a local ring with maximal ideal mi , we see that

A/qi ≃ Ami /qi Ami = Ami /ai .

Similarly we see that A/qni ≃ Ami /ani . Applying − ⊗A M we see that

M/qni M ≃ Mmi /ani Mmi

and so by (⋆) we obtain

r
M
M/an M ≃ Mmi /ani Mmi .
i=1
Pr
Thus Pa (M, n) = i=1 Pai (Mmi , n). ¥

The previous theorem shows us that the degree of Pa (M, n) is a local prop-
erty and thus when ℓ(M/aM ) < ∞, so is the Hilbert-Samuel multiplicity of M
with respect to a.

3.2 The Topological Approach

3.2.1 Basic Definitions
We will first recall some basic definitions from topology.
Definition If X is a set, a topology on X is a collection of subsets of X such
that:
(1) The union of any number of the sets in the collection is again in the
collection.
(2) The intersection of any two of the sets in the collection is again in the
collection.

61
3.2. THE TOPOLOGICAL APPROACH

(3) X is in the collection.

(4) The empty set is in the collection.

Definition A topological space is a set X equipped with a topology.

Definition A set Y is open in a topological space X if it one of the sets of

the topology of X.

Definition A set Y is closed in a topological space X if it is the complement

of an open set in X.

Definition Let X be a topological space, Y ⊆ X is called irreducible if

Y 6= Y1 ∪ Y2 , where Y1 , Y2 are two proper closed subsets of Y and Y 6= ∅.

Definition The closure in a topological space of a subset Y ⊆ X is the

intersection of all the closed sets containing Y and is denoted by Y .

Proposition 3.15 The following are true:

(1) If Y is irreducible, then Y is irreducible.

(2) Y is irreducible if and only if any two nonempty proper open subsets of
Y must have a nonempty intersection.

(3) If x ∈ X, then {x} is an irreducible closed set.

(4) Y is irreducible if and only if every nonempty open set is dense in Y .

Proof Exercise. ¥

Definition A closed set is called a maximal set if it is not contained in a

larger closed set.

Note that every irreducible closed set is contained in a maximal irreducible

set by Zorn’s Lemma.

Definition A maximal irreducible set in X is called a component of X.

Definition If X is a topological space, X is called Noetherian if any of the

following equivalent conditions hold:

(1) Every nonempty family of open sets has a maximal element.

(2) Any increasing family of open sets terminates.

(3) Every nonempty family of closed sets has a minimal element.

(4) Any decreasing family of closed sets terminates.

Definition A topological space is called compact if every open cover of X

has a finite subcover.

62
 CHAPTER 3. DIMENSION THEORY

Remark Our definition of compact above does not include the assumption that
X is Hausdorff—recall that a Hausdorff space is one where given any two points
you can find open sets containing those points such that the intersection of the
open sets is empty. It used to be the case that all compact topological spaces
were taken to be Hausdorff and topological spaces that would be otherwise be
compact had they been Hausdorff were called quasicompact. However, the
restriction that compact spaces be Hausdorff is becoming less common and so
we will not require compact spaces to be Hausdorff.

Proposition 3.16 Let X be a Noetherian topological space. The following are

true:

(1) Any subset of X is Noetherian.

(2) X is compact.

(3) Any open subset of X is compact.

(4) If in addition X is Hausdorff, then X is a finite set with the discrete

topology.

Proof Exercise. ¥

Proposition 3.17 If X is Noetherian, then X = X1 ∪· · ·∪Xn , where each Xi is

a component of X. Moreover, this decomposition is unique and any irreducible
closed set of X is contained in one of the Xi ’s as above.

Proof If X is Noetherian, then by Proposition 3.16 every subset of X is

Noetherian. Hence let
½ ¾
Y is closed in X and for Y the above
S= Y ⊆X: .
proposition does not hold

Suppose that S = 6 ∅. Then S has a minimal element, call it Z. Z is not

irreducible as Z ∈ S. Thus Z = Z1 ∪ Z2 , where neither Z1 nor Z2 are elements
of S. Hence

Z1 = V1,1 ∪ V1,2 ∪ · · · ∪ V1,r ,

Z2 = V2,1 ∪ V2,2 ∪ · · · ∪ V2,s ,

where each V1,i is a component of Z1 and each V2,i is a component of Z2 . Hence

Y = V1,1 ∪ · · · ∪ V1,r ∪ · · · ∪ V2,1 ∪ · · · ∪ V2,s .

Now throw out V1,i if it is contained in V2,j and vice versa. Hence S has no
minimal elements and thus must be empty.

63
3.2. THE TOPOLOGICAL APPROACH

To see the second part of the proposition, let Y be an irreducible closed set
in X. Then

Y =Y ∩X
= Y ∩ (X1 ∪ · · · ∪ Xn )
[n
= Y ∩ Xi .
i=1

Since Y ∩ Xi , is closed and Y is irreducible, we see that Y ∩ Xi = ∅ for each i

except one. ¥

Definition Let X be a topological space, the dimension of X, denoted by

dim(X) is defined as
½ ¾
there exists a chain X0 ) · · · ) Xd of
dim(X) := sup d : .
length d of irreducible closed subsets of X,

Proposition 3.18 If X is a Noetherian topological space,

dim(X) = sup dim(Xi )

where Xi is a component of X.

3.2.2 The Zariski Topology and the Prime Spectrum

Definition If A is a commutative ring, the prime spectrum of A, denoted
by Spec(A), is defined as

Spec(A) := {p : p is a prime ideal of A}.

Similarly, the set of maximal ideals of a ring A is denoted by

MaxSpec(A) := {m : m is a maximal ideal of A}.

Example 3.19
(1) If k is a field, Spec(k) = {(0)}.
(2) Spec(Z) = {(0), (2), (3), (5), (7), (11), (13), (17), . . . }.
(3) Spec(C[X]) = {(0)} ∪ {(X − a) : a ∈ C}.
(4) If k is an algebraically closed field, Spec(k[X]) = {(0)} ∪ {(X − a) : a ∈ k}.
We should note that in general, the prime spectrum of a ring is not easy to
compute.
Definition Let a be any ideal of A. Define

V (a) := {p ∈ Spec(A) : p ⊇ a}.

64
 CHAPTER 3. DIMENSION THEORY

We define the Zariski topology on Spec(A) as follows: Define V (a) to be the

closed sets of Spec(A). One should check that:

(1) V (a1 ) ∪ V (a2 ) = V (a1 ∩ a2 ) = V (a1 a2 ).

T P
(2) i V (ai ) = V ( i ai ).

Now define the open sets to be the complement of the closed sets, that is, the
open sets of Spec(A) are sets of the form
n o
p does not contain the generators
Spec(A) − V (a) = p ∈ Spec(A) :
of the ideal a
for some a ⊆ A.
On one hand the Zariski topology is very nice. It applies to all rings. How-
ever, there is a price to be paid. The Zariski topology is not Mr. Roger’s
Neighborhood. In general, Spec(A) under the Zariski topology is not Hausdorff.
In particular if A contains a unique minimal prime ideal, then then only closed
set containing it is all of Spec(A). This sort of point is everywhere dense and is
called a generic point.

Definition Let f be any element of A. Define

D(f ) := Spec(A) − V (f ) = {p ∈ Spec(A) : f ∈

/ p}.

Proposition 3.20 {D(f ) : f ∈ A} form a basis of Spec(A).

Proof First we need to check

[
Spec(A) = D(f )
f ∈A

and this is clear.

Next we should check if whenever p ∈ D(f ) and p ∈ D(g), does there exist
h such that
p ∈ D(h) ⊆ D(f ) ∩ D(g)?
This is true as we merely need to set h = f g. ¥

Exercise 3.21 Show that Spec(C[X]) = {(0)}.

Definition Let Z be any subset of Spec(A). Define

\
I(Z) := p.
p∈Z

Proposition 3.22 For any ideal a of A,

\ √
I(V (a)) = p = a.
p⊇a

65
3.2. THE TOPOLOGICAL APPROACH

Proof Exercise. ¥

Proposition 3.23 If Z1 ⊆ Z2 are two closed subsets of Spec(A), then

I(Z2 ) ⊆ I(Z1 ).

Proof Exercise. ¥

Proposition 3.24 For any subset Z of Spec(A),

V (I(Z)) = Z.

Proof Exercise. ¥

Remark From the previous propositions, one sees that there is a bijective in-
clusion reversing correspondence between closed sets of Spec(A) and the radical
ideals of A.

Proposition 3.25 If Z is a closed subset of Spec(A), then Z is irreducible if

and only if I(Z) is a prime ideal.

Proof (⇒) If I(Z) is not prime, then there exists a, b ∈ I(Z) such that
a, b ∈
/ I(Z) such that ab ∈ I(Z). Set

I1 = I(Z) + a,
I2 = I(Z) + b.

Hence V (I1 ) ⊆ Z and V (I2 ) ⊆ Z and more importantly V (I1 ) ∪ V (I2 ) ⊆ Z.

However, I1 · I2 ⊆ I(Z). Hence Z ⊆ V (I1 · I2 ) = V (I1 ) ∪ V (I2 ). Thus
Z = V (I1 ) ∪ V (I2 ) is not irreducible.
(⇐) Now suppose that I(Z) is prime. Since this is a point in the topological
space Spec(A), it is irreducible. ¥

Proposition 3.26 If A is a Noetherian ring, then Spec(A) is a Noetherian

topological space.

Proof Suppose A is Noetherian, then by the Primary Decomposition Theo-

rem, Theorem 1.15, we have

(0) = q1 ∩ · · · ∩ qr

where each qi is pi -primary and so V (qi ) = V (pi ). Eliminate those V (pi ) for
which pi is not minimal in Ass(A). Then reindexing we have

Spec(A) = V (0) = V (p1 ) ∪ · · · ∪ V (ps ).

Since the V (pi ) are the irreducible components of Spec(A) this proves the propo-
sition. ¥

66
 CHAPTER 3. DIMENSION THEORY

WARNING 3.27 The converse of the above proposition is not true. We leave
it as an exercise to show that if k is a field, then

k[X1 , . . . , Xn , . . . ]
A=
(X12 , . . . , Xn2 , . . . )

is not a Noetherian ring but Spec(A) is a Noetherian topological space.

Definition (Second Notion of Dimension) If A is a ring, the Krull dimen-

sion, denoted by dim(A), is the dimension of the topological space Spec(A). To
be explicit:

dim(A) = sup{d : there exists p0 ( p1 ( · · · ( pd such that pi ∈ Spec(A)}.

This notion of dimension is often simply referred to as the dimension of a ring.

If M is an A-module, then

dim(M ) := dim(A/ Ann(M )).

Exercise 3.28 Given a ring A and an A-module M , show that the dimension
of M is the dimension of the topological space SuppA (M ) ⊆ Spec(A).

Example 3.29 While

dim(A) = sup{d : there exists p0 ( p1 ( · · · ( pd such that pi ∈ Spec(A)}.

Nagata gives an example of a ring A such that A is Noetherian but with infinite
dimension in [14].

Example 3.30 In the ring C[X1 , . . . , Xn ] we have the chain of prime ideals

(0) ( (X1 ) ( · · · ( (X1 , . . . , Xn ).

Thus the dimension of C[X1 , . . . , Xn ] is at least n.

Suppose that a ring A has finite dimension. If m1 , . . . , mr are maximal prime

ideals in Spec(A), then dim(A) = d implies that there exists a chain of length d
ending at one of the mi ’s. Thus for some i,

dim(Ami ) = dim(A).

Hence we see that some questions about the dimension of a ring can be reduced
to questions about local rings.

Remark We can characterize dimension in two useful ways:

(1) dim(A) = sup dim(Am ) where m ∈ MaxSpec(A).

(2) dim(A) = sup dim(A/p) where p is a minimal prime ideal of A.

67
3.3. SYSTEMS OF PARAMETERS AND THE DIMENSION THEOREM

3.3 Systems of Parameters and the Dimension

Theorem
Recall that a local ring is a ring that is Noetherian with a unique maximal ideal.

Definition (Third Notion of Dimension) Let A be a local ring and M a

finitely generated A-module. Define
½ ¾
there exists x1 , . . . , xd ∈ m such
s(M ) := inf d : .
that ℓ(M/(x1 , . . . , xd )M ) < ∞

If M is finitely generated as an A-module then, M/mM is finitely generated

as an A/m module. Hence as a vector space over A/m, M/mM has finite
dimension. If A is Noetherian, then we see that M/mM has finite length. This
idea helps motivate our next definition.

Definition Let A be Noetherian and M be a finitely generated A-module.

Then if s(M ) = n, then any sequence x1 , . . . , xn such that

ℓ(M/(x1 , . . . , xn )M ) < ∞

is called a system of parameters of M .

Theorem 3.31 (The Dimension Theorem) Let A be a local ring and M a

finitely generated A-module. Then

dim(M ) = d(M ) = s(M ).

Proof We will show dim(M ) 6 d(M ) 6 s(M ) 6 dim(M ).

dim(M ) 6 d(M ) Proceeding by induction on d(M ). Suppose that d(M ) =

0. Then
d(M ) = Pm (M, n) = ℓ(M/mn M )
is constant for n sufficiently large. Thus for large enough n,

ℓ(M/mn M ) = ℓ(M/mn+1 M )

which shows us that ℓ(mn M/mn+1 M ) = 0. Hence mn M/mn+1 M = 0. So,

mn M = mn+1 M = m(mn M )

and hence by Nakayama’s Lemma, Corollary 2.35, mn M = 0 and hence mn ⊆

AnnA (M ). Since
p
dim(M ) = dim(A/ Ann(M )) = dim(A/ Ann(M )) = dim(A/m),

we see that dim(M ) = 0.

68
 CHAPTER 3. DIMENSION THEORY

Now assume d(M ) = n > 0. Take any maximal chain

p0 ( p1 ( · · · ( pm

in Supp(M ). We need to show m 6 n. We know that p0 is a minimal element of

Supp(M ) and hence is a minimal element in AssA (M ). So we have an injection

A/p0 ֒→ M.

Set N = A/p0 . By Corollary 3.11, d(N ) 6 d(M ). So it suffices to show that

m 6 d(N ). Let x ∈ p1 − p0 . Consider the short exact sequence
x
0 −→ N −→ N −→ N/xN −→ 0.

By Proposition 3.10, we have that

Pm (N, n) = Pm (N, n) + Pm (N/xN, n) − R(n),

where deg(R(n)) < deg(P m(N, n)) = d(N ). But then

Pm (N/xN, n) = R(n).

Thus d(N/xN ) = deg(Pm (M, n)) = deg(R(n)) < n. By the inductive hypothe-
sis, dim(N/xN ) 6 d(N/xN ). Since

p1 ( · · · ( pm

is a strict chain of prime ideals in Supp(N/xN ), we have that

m − 1 6 dim(N/xN ) 6 d(N/xN ) 6 n − 1.

Hence dim(M ) = m 6 n = d(M ).

d(M ) 6 s(M ) If n = s(M ), consider x1 , . . . , xn a system of parameters for

M . In this case we have by definition that

ℓA (M/(x1 , . . . , xn )M ) < ∞.

Consider a = (x1 , . . . , xn ). Now if we consider the image of a in M/ AnnA (M ),

we have that
a = (x1 , . . . , xn )
where xi is the image of xi ∈ M/ AnnA (M ). By Proposition 3.12, d(M ) depends
only on M and
Supp(M/mM ) = Supp(M/aM ),
and thus we see that

d(M ) = deg(Pm (M, n)) = deg(Pa (M, n)) 6 n,

using that a is generated by n elements and applying Corollary 3.9.

69
3.3. SYSTEMS OF PARAMETERS AND THE DIMENSION THEOREM

s(M ) 6 dim(M ) Proceed by induction on dim(M ). Write

dim(M ) = 0 ⇔ AssA (M ) = {m},
⇔ ℓA (M ) < ∞,
and so we see s(M ) = 0.
Now assume that dim(M ) = n. Let pi be the minimal prime ideals in
Supp(M ). By Corollary 1.39, these primes are also minimal in Ass(M ) so
there are only finitely many such primes. By the Prime Avoidance Lemma,
Lemma 0.12, we may pick x ∈ m such that x is not in any of these minimal
primes. Thus
dim(M/xM ) < dim(M )
and thus by induction, s(M/xM ) 6 dim(M/xM ). But s(M ) − 1 6 s(M/xM )
and so
s(m) 6 s(M/xM ) + 1 6 dim(M/xM ) + 1 6 dim(M ).
Putting the above steps together we have shown
dim(M ) 6 d(M ) 6 s(M ) 6 dim(M ),
and hence dim(M ) = d(M ) = s(M ). ¥
Corollary 3.32 If (A, m) is a local ring, then it is has finite dimension.
Proof Since A is Noetherian, m is finitely generated, thus s(A) = dim(A) is
less than or equal to the number of generators of m. ¥
Corollary 3.33 If (A, m) is a local ring and M is a finitely generated A-module,
then
c).
dimA (M ) = dimAb(M
Proof We know from Proposition 2.16
c/m
M/mn M ≃ M c≃M
cn M c/m c,
b nM

b (M , n).
and so Pm (M, n) = Pm c ¥
Corollary 3.34 If (A, m) is a local ring, then
dim(A) = min{i : (a1 , . . . , ai ) = a where a is m-primary}.
Proof Let dim(A) = n, then there exists x1 , . . . , xn a system of parameters
of A such that ℓ(A/x) < ∞. Thus (x1 , . . . , xn ) is m-primary. Since s(A) = n,
we see that we cannot obtain y1 , . . . , yt with t < n such that ℓ(A/y) < ∞. So
for any y1 , . . . , yt with t < n, (y1 , . . . , yt ) is not m-primary. ¥
Corollary 3.35 If A is a Noetherian ring, not necessarily local, consider any
decreasing chain of prime ideals in A
p0 ) p1 ) · · · ) pi ) · · ·
Then there exists n such that pn = pn+1 = · · · .

70
 CHAPTER 3. DIMENSION THEORY

Proof If we localize at p0 , then dim(Ap0 ) < ∞. So for some n

pn Ap0 = pn+1 Ap0 = · · ·

and so pn = pn+1 = · · · . ¥

Definition If A is a Noetherian ring and p is a prime ideal of A, then the

height of p is
½ ¾
there exists a chain of prime ideals
ht(p) = sup n : .
p0 ( p1 ( · · · ( pn−1 ( pn = p

Remark Note that ht(p) = dim(Ap ).

Definition If A is a Noetherian ring and I is any ideal of A, then the height

of I is

ht(I) = inf{ht(p) : I ⊆ p}
= inf{ht(p) : p is minimal in AssA (A/I)}.

Definition If A is a Noetherian ring and I is any ideal, then the coheight,

denoted coht(I) is defined as

coht(I) = dim(A/I).

Exercise 3.36 Check that for any ideal I, ht(I) + dim(A/I) 6 dim(A).

WARNING 3.37 Even if I is a prime ideal, the above inequality may be strict.

Corollary 3.38 (Krull’s Ideal Theorem) Let A be a Noetherian ring and p is

a prime ideal. Then ht(p) 6 n if and only if there exist a1 , . . . , an ∈ p such that
p is a minimal prime containing (a1 , . . . , an ).

Proof Since ht(p) = dim(Ap ), dim(Ap ) 6 n if and only if there exists

x1 xn
,...,
u1 un

in Ap such that
µ ¶
x1 xn
,...,
u1 un
is pAp -primary. This is the case if and only if there exist x1 , . . . , xn ∈ p and p
contains (x1 , . . . , xn ) minimally. ¥

Corollary 3.39 (Krull’s Principal Ideal Theorem) Let A be a Noetherian ring

and x be an element of A which is not a unit or a zerodivisor. Then every prime
which contains (x) minimally has height 1.

71
3.3. SYSTEMS OF PARAMETERS AND THE DIMENSION THEOREM

Proof By Corollary 3.38 the minimal prime containing (x) has height at most
one. The prime p containing (x) cannot have height zero, as then p ∈ Ass(A)
by Corollary 1.39 and hence then every element of p is a zero divisor, which is
not the case. ¥
Remark The above corollary to the Dimension Theorem is sometimes called
Krull’s Hauptidealsatz.
Corollary 3.40 Let (A, m) be a local ring and x be an element of m which is
not a zerodivisor. Then dim(A/xA) = dim(A) − 1.
Proof This follows from the previous corollary and the definition of dimen-
sion. ¥
Exercise 3.41 If (A, m) is a local ring and M is a finitely generated A-module
with x1 , . . . , xi ∈ m, then
dim(M/(x1 , . . . , xi )M ) > dim(M ) − i.
Equality holds if and only if x1 , . . . , xi form part of a system of parameters for
M.
Exercise 3.42 Let A be a Noetherian ring of dimension at least 2. Then A
has infinitely many prime ideals of height 1.
Example 3.43 A ring A is Artinian if and only if dim(A) = 0.
Example 3.44 A PID has dimension one.
Example 3.45 dim(Z) = 1.
Example 3.46 If k is a field, then dim(k[X]) = 1.
Example 3.47 If k is a field, then dim(k[[x]]) = 1.
Lemma 3.48 If A is a ring and P1 ( P2 are two prime ideals of A[X] such
that
P1 ∩ A = P2 ∩ A = p,
then P1 = p[X].
Proof Suppose not. Then
p[X] ( P1 ( P2
and so
(0) ( (P1 /p)[X] ( (P2 /p)[X]
in (A/p)[X]. Set U = A − p. Since P1 ∩ U = P2 ∩ U = ∅, and since localizations
are exact, we have
(0) ( U −1 (P1 /p)[X] ( U −1 (P2 /p)[X]
is a strict chain of prime ideals in (Ap /pAp )[X] = U −1 (A/p)[X]. But this
contradicts that dim(k[X]) = 1 when k is a field as k[X] is a PID, hence all
primes are principal and so are of height one or zero. Thus P1 = p[X]. ¥

72
 CHAPTER 3. DIMENSION THEORY

Lemma 3.49 If A is Noetherian and I is an ideal of A, then

ht(I) = ht(I · A[X]).

Proof By the Primary Decomposition Theorem, Theorem 1.15,

µ ¶
A[X]
Ass = {pi [X] : pi ∈ Ass(A/I)},
IA[X]

as A/I injects into A/I[X]. So it is enough to show that if p is a prime ideal in

A, then
ht(p) = ht(p[X]).
Suppose that ht(p) = n. Then there exist a1 , . . . , an ∈ p such that p ⊇
(a1 , . . . , an ) minimally. By the Primary Decomposition Theorem, Theorem 1.15,
we see that p[X] ⊇ (a1 , . . . , an )[X] minimally. Thus ht(p[X]) 6 n.
On the other hand, if

p0 ( p1 ( · · · ( pn = p

is a chain of prime ideals where ht(p) = n, then

p0 [X] ( p1 [X] ( · · · ( pn [X] = p[X]

is a chain of prime ideals in A[X]. Thus ht(p[X]) > ht(p) and so we see that
ht(p[X]) = ht(p). ¥

Theorem 3.50 If A is a Noetherian ring, dim(A[X]) = dim(A) + 1.

Proof First note that given a chain of primes p0 ( · · · ( pn in A, we can

construct the chain of primes p0 A[X] ( · · · ( pn A[X] ( pn A[X] + xA[X] in
A[X]. It is then clear that dim(A[X]) > dim(A) + 1, so it is enough to show
dim(A[X]) 6 dim(A) + 1. If dim(A) = ∞, then there is nothing to prove. We
will proceed by induction on the dimension of A. If dim(A) = 0, then for Pi a
prime ideal in A[X] write pi = Pi ∩ A. So if

P0 ( P1 ( · · · ( Pn

then since dim(A) = 0, we have p0 = p1 = · · · = pn . Thus by Lemma 3.48, we

have
p0 [x] = P0 = P1 = · · · = Pn−1 ( Pn .
Thus n 6 1 and so we see that dim(A[X]) = 1.
Now suppose that dim(A) = n > 0 and let P0 ( P1 ( · · · ( Pn be a strict
chain of prime ideals in A. Set pi = Pi ∩ A.
Case 1: Suppose that pn−1 ( pn . Then

dim(Apn−1 ) < dim(A).

73
3.3. SYSTEMS OF PARAMETERS AND THE DIMENSION THEOREM

By induction we then have that dim(Apn−1 [X]) = dim(Apn−1 ) + 1 6 dim(A).

In Apn−1 [X] we have a strict chain

P0 Apn−1 ( P1 Apn−1 ( · · · ( Pn−1 Apn−1

and thus
n − 1 6 dim(Apn−1 )[X] 6 dim(A).
Thus n 6 dim(A) + 1 and so dim(A[X]) = dim(A) + 1.
Case 2: Suppose that pn−1 = pn . By Lemma 3.48, we have Pn−1 = pn−1 [X].
By Lemma 3.49, we have ht(pn−1 [X]) = ht(pn−1 ). Thus

dim(A) > ht(pn−1 ) = ht(Pn−1 ) > n − 1.

Thus n 6 dim(A) + 1 and so we see that dim(A[X]) = dim(A) + 1. ¥

Exercise 3.51 If A is Noetherian, show that

dim(A[[X]]) = dim(A) + 1

Hint: Does every maximal ideal in A[[X]] contain X?

Corollary 3.52 We have that if k is a field, then

dim(k[X1 , . . . , Xn ]) = dim(k[[X1 , . . . , Xn ]]) = n,

dim(Z[X1 , . . . , Xn ]) = dim(Z[[X1 , . . . , Xn ]]) = n + 1,
dim(Z(p) [X1 , . . . , Xn ]) = dim(Z(p) [[X1 , . . . , Xn ]]) = n + 1.

74
Chapter 4

Integral Extensions

4.1 Basic Properties

Definition Let A ⊆ B be commutative rings. An element b ∈ B is called
integral over A if b satisfies a monic equation of the form

bn + an−1 bn−1 + · · · + ba1 + a0 = 0

where a0 , . . . , an−1 ∈ A. If every element of B is integral over A, we say that B

is integral over A.

Proposition 4.1 Let A ⊆ B be commutative rings and let b ∈ B. Then the

following are equivalent:

(1) b is integral over A.

(2) A[b] is a finitely generated A-module.

(3) A[b] is contained in a subring of B which is a finitely generated A-module.

(4) There exists an A[b]-submodule of B which is faithful as an A[b]-module

and is finitely generated as an A-module.

Proof (1) ⇒ (2) Suppose x is integral over A. Then

bn + an−1 bn−1 + · · · + a1 b + a0 = 0

for some a0 , . . . , an−1 ∈ A. Thus A[b] is generated by 1, b, b2 , . . . , bn−1 as an

A-module.
(2) ⇒ (3) If A[b] is finitely generated, then taking B = A[b] gives A[b] as a
subring of a ring that is finitely generated as an A-module.
(3) ⇒ (4) If A[b] ⊆ C ⊆ B where C is a subring of B, then C is faithful as
an A[b]-module and C is finitely generated over A.

75
4.1. BASIC PROPERTIES

(4) ⇒ (1) Suppose there exists an A[b]-submodule C of B which is faithful

as an A[b]-module and finitely generated over A. Let c1 , . . . , cn be generators
for C over A. Since C is an A[b]-module, bci ∈ C for all i. So we may write

bc1 = a1,1 c1 + a1,2 c2 + · · · + a1,n cn ,

bc2 = a2,1 c1 + a2,2 c2 + · · · + a2,n cn ,
..
.
bcn = an,1 c1 + an,2 c2 + · · · + an,n cn .

We can write this in matrix form as

    
b − a1,1 −a1,2 · · · −a1,n c1 0
 −a2,1 b − a2,2 · · · −a   c2  0
 2,n    
 .. .. ..   ..  =  ..  .
 . . .   .  .
−an,1 −an,2 · · · b − an,n cn 0

Let X be the n × n matrix shown. Let Yi be the identity n × n matrix with the
ith column replaced by the column in the above equation with entries c1 , . . . , cn .
Then by the above equation, the ith column in the product X · Yi will be the 0
column for all i. Thus det(X · Yi ) = 0. Therefore

det(X)ci = det(X) det(Yi ) = det(X · Yi ) = 0

for all i. Since C is a faithful A[x]-module, det(X) ∈ AnnA[b] (C) = 0. So

det(X) = 0 and this monic equation in b gives an integral dependence for b over
A. ¥

Example 4.2 If X is an indeterminate over Z, it is clear that Z[X] is not an

integral extension of Z. However, Z[X]/(X 2 + 1) ≃ Z[i] is an integral extension
of Z.
Corollary 4.3 Suppose B = A[x1 , . . . , xn ] is a finitely generated A-algebra,
and suppose each xi is integral over A. Then B is a finitely generated A-module.
In this case we say that B is a module finite A-algebra.

Proof Clear from the proposition above. ¥

Definition Let A ⊆ B be commutative rings. Let

B ′ = {b ∈ B : b is integral over A}.

Then B ′ is a subring of B called the integral closure of A in B. If B ′ = A we

say A is integrally closed in B. By the above proposition, B ′ is a subring of B.

If A is a domain, the integral closure of A, without reference to another

ring, means the integral closure of A in Frac(A). We denote the integral closure
of A by A e and when A = A,e we say that A is integrally closed.

76
 CHAPTER 4. INTEGRAL EXTENSIONS

Exercise 4.4 If A is a UFD, then A is integrally closed.

Exercise 4.5 If A is integrally closed in B, and if U is a multiplicatively closed

subset of A, then U −1 A is integrally closed in U −1 B.

Exercise 4.6 Let A ⊆ B ⊆ C be rings where A ⊆ B and B ⊆ C are both

integral extensions. Then A ⊆ C is an integral extension.

Proposition 4.7 Let A be a domain. Suppose that for every maximal ideal m
of A, Am is integrally closed. Then A is integrally closed.

Proof First we show that

\
A= Am
m

where m ∈ MaxSpec(A). Note that for every maximal ideal m, A ⊆ Am ⊆ K

where K = Frac(A). So
\
A⊆ Am .
m

Now take \
x∈ Am
m

and let I = (A :A x). Suppose I 6= A. Then I ⊆ m for some maximal ideal m.

Since x ∈ Am we can write x = a/b with b ∈ / m. But bx = a ∈ A implies that
b ∈ I ⊆ m; a contradiction. Therefore I = A. In particular x ∈ A, so we have
\
A= Am .
m

Now let x ∈ K be integral over A. Then x is integral over Am for all maximal
ideals m. Since Am is integrally closed for every maximal ideal m, we have that
\
x∈ Am = A.
m

So A is integrally closed. ¥

Remark By the previous proposition, we see that the property of being inte-
grally closed is a local property.

Proposition 4.8 Suppose A ⊆ B is an integral extension with B a domain.

Then A is a field if and only if B is a field.

Proof (⇒) Suppose A is a field and consider some nonzero b ∈ B. Then b is

integral over A. So there is a minimal relation of the form

bn + λn−1 bn−1 + · · · + λ1 b + λ0 = 0,

77
4.1. BASIC PROPERTIES

where λ0 , . . . , λn−1 ∈ A and λ0 6= 0. If λ0 = 0, then by factoring out b and

using that B is a domain, we would get an integral relation of smaller degree.
So we can write
1 1
= − (bn−1 + λn−1 bn−2 + · · · + λ1 ).
b λ0
Therefore B is a field.
(⇐) Suppose B is a field and consider some nonzero a ∈ A. Then a−1 ∈ B.
So a−1 is integral over A, meaning we may write
µ ¶n µ ¶n−1 µ ¶
1 1 1
+ λn−1 + · · · + λ1 + λ0 = 0,
a a a

where λ0 , . . . , λn−1 ∈ A. Multiplying by an−1 we have

1
= −λn−1 − λn−2 a − · · · − λ0 an−1 ∈ A.
a
Therefore A is a field. ¥

Definition Let f : A → B be a map of rings and let and let a and b be ideals
in A and B respectively. We say that b lies over a if a = f −1 (b).

Remark This will usually be used in the case where a and b are prime ideals
and A ֒→ B is an integral extension. Then f −1 (b) = a becomes b ∩ A = a.

Proposition 4.9 Suppose A ⊆ B is an integral extension and suppose that P

lies over p. Then P is maximal if and only if p is maximal.

Proof A/p ֒→ B/P is an integral extension. The rest then follows from
Proposition 4.8. ¥

Proposition 4.10 Suppose A ⊆ B is an integral extension and let x be a

nonzerodivisor in B. Then xB ∩ A 6= 0.

Proof Pick an integral equation for x of least degree, say

xn + an−1 xn−1 + · · · + a0 = 0, ai ∈ A.

Then
x(xn−1 + · · · + a1 ) = −a0 .
We must have a0 6= 0, for otherwise, using that x is a nonzerodivisor we have
that xn−1 + · · · + a1 = 0, which is an integral dependence of smaller degree.
Therefore 0 6= a0 ∈ xB ∩ A. ¥

Proposition 4.11 Suppose A ⊆ B is an integral extension. Suppose P is a

prime ideal of B, I is an ideal of B and P ⊆ I. If P ∩ A = I ∩ A then P = I.

78
 CHAPTER 4. INTEGRAL EXTENSIONS

Proof Let p = P ∩ A. Then A/p ֒→ B/P is an integral extension. If I 6= P ,

then I/P 6= 0. So let 0 6= x ∈ I/P . Since B/P is a domain x is a nonzerodivisor
on B/P . So by the previous proposition x(B/P )∩A/p 6= 0. But this contradicts
I ∩ A = p. Therefore I = P . ¥

Proposition 4.12 Suppose A ⊆ B is an integral extension and suppose P lies

over p. Then P contains pB minimally.

Proof Suppose there exists a prime ideal Q such that pB ⊆ Q ⊆ P . Then

pB ∩ A = P ∩ A
=Q∩A
= p.

By the previous proposition Q = P . ¥

Proposition 4.13 Suppose A ⊆ B is an integral extension and suppose

P1 ( P2 ( · · · ( Pn

is a strict chain of prime ideals in B. Then

P1 ∩ A ( P2 ∩ A ( · · · ( Pn ∩ A

is a strict chain of prime ideals in A. In particular dim(B) 6 dim(A).

Proof Follows from the previous proposition. ¥

Definition Let f : A → B be a map of rings. We say f has the Lying-Over

Property if for every prime ideal p of A there exists a prime ideal P of B such
that p = f −1 (P ).

Definition Let f : A → B be a map of rings, a an ideal of A, and b an ideal

of B such that f −1 (b) = a. We say f has the Going-Up Property if given a
chain of ideals a ⊆ p where p is a prime ideal, then there exists a prime ideal
P in B containing b such that f −1 (P ) = p. Pictorially, the situation can be
described by:
f
A B
∪ ∪
p ∃P
∪ ∪
a b
√
Remark Note that if Ker(f ) ⊆ 0, then the Going-Up Property implies the
Lying-Over Property as we may take b = (0) and a = Ker(f ). In particular,
the Going-Up Property implies the Lying-Over Property when f is injective.

79
4.1. BASIC PROPERTIES

For the purposes of integral extensions, f : A → B is an inclusion and

f −1 (P ) becomes P ∩ A. We will see through exercises later in the section that
the generality of the previous definitions will be used when discussing flat and
faithfully flat maps.

Theorem 4.14 (Lying-Over and Going-Up for Integral Extensions) Suppose

A ֒→ B

is an integral extension. Let p be a prime ideal in A. Then there exists a prime

ideal P in B lying over p. Moreover, P may be chosen to contain any ideal I
such that I ∩ A ⊆ p. In other words, integral extensions satisfy the Lying-Over
and Going-Up Properties.

Proof Factoring out I and A ∩ I in B and A respectively, we may assume

that I = 0. Now let U = A − p. Then Ap ֒→ U −1 B is an integral extension.
Pick a maximal ideal m in U −1 B. Then m ∩ Ap is a maximal ideal of Ap by
Proposition 4.9. Thus m ∩ Ap = pAp . Now take P = i−1 B (m) ⊆ B, where iB is
the localization map iB : B → U −1 B. Then P lies over p. ¥

Corollary 4.15 (Integral Extensions Preserve Dimension) Suppose A ֒→ B

is an integral extension, then dim(B) = dim(A).

Proof Let p0 ( p1 ( · · · ( pr be a strict chain of prime ideals of A. By

Theorem 4.14 there is a prime ideal P0 lying over p0 . By Theorem 4.14 there is
a prime ideal P1 lying over p1 containing P0 . Similarly we may pick P2 , . . . , Pn
a necessarily strict ascending chain of prime ideals lying over p2 , . . . , pn . In
particular we have a strict chain of prime ideals P0 ( · · · ( Pn in B. Therefore
dim(B) > dim(A). But by Proposition 4.13, dim(B) 6 dim(A). Therefore
dim(B) = dim(A). ¥

Corollary 4.16 Suppose A ⊆ B is an integral extension. Let I be any ideal in

B lying over an ideal a in A. Then ht(I) 6 ht(a).

Proof Take any minimal prime p ⊇ a so that ht(a) = ht(p). By Theorem 4.14
there is a prime ideal P of B lying over p such that P ⊇ I. By Proposition 4.13
it follows that ht(P ) 6 ht(p). Thus

ht(I) = inf ht(Q) 6 ht(P ) 6 ht(p) = ht(a).

Q⊇I

Exercise 4.17 Suppose f : A → B is a ring map that makes B a faithfully

flat A-module. Then f has the Going-Up Property.

Exercise 4.18 Suppose f : A → B is a ring homomorphism. Then the follow-

ing hold:

80
 CHAPTER 4. INTEGRAL EXTENSIONS

(1) If we define f ∗ as follows

f ∗ : Spec(B) → Spec(A),
P 7→ f −1 (P ),

f ∗ is a continuous map.
(2) Spec(A) is compact.
(3) Suppose f has the Going-Up Property. Then f ∗ is a closed map.
Proposition 4.19 Suppose A is an integral domain, K = Frac(A), L is an
algebraic extension of K, and B is the integral closure of A in L. Then:
(1) L = U −1 B where U = A − {0}; that is, if x ∈ L, then x = b/a where
b ∈ B and 0 6= a ∈ A.
(2) If σ ∈ GalK (L) and if a prime P ⊆ B lies over the prime p ⊆ A, then
σ(P ) also lies over p.

Proof (1) We have the following diagram:

B L

A K

Let x ∈ L. Since L is an algebraic extension of K, there is a relation of the form

xn + λn−1 xn−1 + · · · + λn = 0,

where λ0 , . . . , λn−1 ∈ K. Since K = Frac(A), by taking a common denominator

we can write λi = ai /a for each i where a, a0 , . . . , an−1 ∈ A. Replacing the λi ’s
and multiplying by an we get

(ax)n + an−1 (ax)n−1 + an−2 a(ax)n−2 + · · · + an an−1 = 0.

b
Thus ax is integral over A, and so ax ∈ B. Hence x = a for some b ∈ B.
(2) Let σ ∈ GalK (L). Let x ∈ B. Then

xn + an−1 xn−1 + · · · + a0 = 0.

Applying σ we get

σ(x)n + an−1 σ(x)n−1 + · · · + a0 = 0.

In particular σ(B) is integral over A. Since B is the integral closure of A in L,

σ(B) ⊆ B. Applying σ −1 we see that σ(B) = B. Hence σ is an automorphism
of B. Let P be a prime ideal lying over p. Since P ∩ A = σ(P ) ∩ A = p, σ(P )
also lies over p. ¥

81
4.1. BASIC PROPERTIES

For the next couple theorems we recall some ideas from basic algebra.

Definition Let L be an algebraic extension of K. Then L is said to be normal

over K if L is the splitting field over K of a finite number of polynomials.

Definition Let L be a field extension of K. Then L is said to be separable

over K if L is generated over K by a set of elements each of which is the root
of a separable polynomial in K[x], i.e. an irreducible polynomial with distinct
roots. An extension that is not separable is called inseparable.

Definition Let L be a field extension of K in characteristic p 6= 0. Then L is

n
said to be purely inseparable over K if there exists n > 0 such that αp ∈ K
for all α ∈ L.

Proposition 4.20 Let L be a normal extension of K and let G = GalK (L).

Then L is separable over LG and LG is purely inseparable over K. Note that
LG denotes the subfield of L fixed by G.

Proof See your favorite algebra text, or see [7]. ¥

Theorem 4.21 Let A be an integrally closed domain and K = Frac(A). Let L

be a normal extension of K, let G = GalK (L), and let B be the integral closure
of A in L. Let p ⊆ A be a prime ideal. Then G acts transitively on the set of
prime ideals in B lying over p.

Proof First assume that [L : K] < ∞. Write

G = {σ1 , . . . , σn }.

Let P, Q be primes of B lying over p. It is enough to show that P ⊆ σi (Q) for

some i, for then by Proposition 4.11, P = σi (Q).
Sn Suppose P 6⊆ σi (Q) for any i.
Then by prime avoidance, Lemma 0.12, P 6⊆ i=1 σi (Q). So there exists x ∈ P
with x ∈
/ σi (Q) for all i. Let

y = σ1 (x) · · · σn (x).

Since σi (y) = y for all i, we have y ∈ LG by the previous theorem. Thus y m ∈ K

for some m ∈ N by the previous proposition. But then y m ∈ P ∩ K = p.
Thus y ∈ p ⊆ Q. Thus σi (x) ∈ Q for some i. Hence, x ∈ σi (Q) for some
i, a contradiction. Thus P ⊆ σi (Q) for some i, and by the earlier remarks,
P = σi (Q). S
Now assume that [L : K] = ∞. Then we can write L = i Li where for
each i, Li is a finite normal extension of K. Let P, Q be primes of B lying over
p in A. Let Pi = P ∩ Li and Qi = Q ∩ Qi . Let Gi = GalK (Li ). Let us write
(Li , σi ) 6 (Lj , σj ) if Li ⊆ Lj and σj |Li = σi . This puts a partial ordering on
the set of pairs

{(Li , σi ) : σi ∈ GalK (Li ) and σi (Pi ) = Qi }.

82
 CHAPTER 4. INTEGRAL EXTENSIONS

Further any chain

(Li1 , σi1 ) 6 (Li2 , σi2 ) 6 · · ·
S S
has a least upper bound, namely ( j Lij , j σij ). So we may apply Zorn’s
Lemma to state the existence of a maximal element (L0 , σ0 ). We claim that
L0 = L. So suppose L0 ( L. Pick x ∈ L − L0 . Let L′ be a splitting field for x
over L0 inside L. By assumption
σ0 (P ∩ L0 ) = Q ∩ L0
= Q0 .
Let P ′ = P ∩ L′ and Q′ = Q ∩ L′ and let σ ′ be any extension of σ0 to GalK (L′ ).
Then σ ′ (P ′ ) and Q′ both lie over Q0 . By the finite case there exists σ ′′ ∈
GalL0 (L′ ) such that σ ′′ (σ ′ (P ′ )) = Q′ . Write σ = σ ′′ ◦ σ ′ ∈ GalK (L′ ). Then
σ(P ′ ) = Q′ and (L0 , σ0 ) < (L′ , σ), a contradiction. ¥
Definition Let f : A → B be a map of rings. We say the f has the Going-
Down Property if for all for p, q ∈ Spec(A) and Q ∈ Spec(B) where p ( q
with Q lying over q, then there exists a prime P of B such that P lies over p
and P ( Q. Pictorially the situation can be described by:
f
A B
∪ ∪
q Q
∪ ∪
p ∃P
Theorem 4.22 (Going-Down for Integral Extensions) Suppose A is an inte-
grally closed domain, A ֒→ B is an integral extension of rings, and B is torsion
free over A. Then the Going-Down Property holds.
Proof There are two basic cases for the proof.
Case 1. Suppose B is a domain. Let K = Frac(A) and let L = Frac(B). Let
L to be the normal closure of L and let B to be the integral closure of B in L.
By the Lying-Over Property, we obtain Q in the following diagram:

B B
∪ ∪
Q Q

By the Going-Up Property there is a prime ideal Q′ ⊆ B lying over q and

P ′ ⊆ B lying over p such that P ′ ⊆ Q′ :

A B
∪ ∪
q Q′
∪ ∪
p P′

83
4.1. BASIC PROPERTIES

Since Q and Q′ both lie over q and since L/K is normal by Theorem 4.21,
there exists σ ∈ GalK (L) such that σ(Q′ ) = Q. Thus σ(P ′ ) lies over p and
σ(P ′ ) ⊆ σ(Q′ ) = Q. Now take P = σ(P ′ ) ∩ B. Then P ⊆ Q ∩ L = Q and
P ∩ A = p by construction.
Case 2. Now suppose B is torsion free over A. Suppose p ⊆ q are prime.
We want to apply Case 1, so given

A B
∪ ∪
q Q
∪
p

we first want a prime ideal P ⊆ Q such that P ∩ A = (0). Let U1 = A − {0}

and let U2 = B − Q and set U = U1 U2 . Since B is torsion free over A, 0 ∈
/ U . It
follows that U is a multiplicative set in B and that η : B ֒→ U −1 B. Let P ′ be
any prime ideal of U −1 B. Set P = η −1 (P ′ ) ⊆ B. One should check that P ⊆ Q
and P ∩ A = (0). Setting B = B/P we still have an injection A ֒→ B where B
is a domain. We now have the following diagram

A B
∪ ∪
q Q
∪
p

where Q = Q/P is a prime in B lying over q. By Case 1 there is a prime ideal

P ⊆ B contained in Q lying over p. Finally, if ϕ : B → B, take P ′′ = ϕ−1 (P )
and check that P ′′ ⊆ Q and that P ′′ lies over p as required. ¥

Exercise 4.23 To show that the both the hypotheses are necessary in the
Going-Down Theorem, we give two examples, but leave the details as an exercise.

(1) Let A = k[X, Y ] and B = k[X, Y, Z]/(Z 2 − Z, Y Z) so that A ֒→ B is

an integral extension but B is not torsion free over A. Take Q = (Z −
1, X, Y ) ⊆ B, q = (X, Y ) ⊆ A and p = (x) ⊆ A. Show that Q lies over q
but contains no prime ideal lying over p.

(2) Let A = k[X 2 , XY, Y ] and B = k[X, Y ] so that A ֒→ B is an integral

extension but A is not integrally closed. Take Q = (X − 1, Y ), q =
(XY, Y, X 2 − 1) and p = (X 2 − 1, XY − Y ). Show that Q lies over q but
contains no prime lying over p.

Corollary 4.24 Suppose A is an integrally-closed domain, A ֒→ B is an integral

extension of rings, and B is torsion free over A. If p is a prime ideal in A and
P is a prime ideal in B, then P lies over p if and only if P ⊇ pB minimally.

84
 CHAPTER 4. INTEGRAL EXTENSIONS

Proof (⇒) This follows from Proposition 4.12.

(⇐) Suppose P ⊇ pB minimally. Let q = P ∩ A. Suppose further that q 6= p.
Then by the Going-Down Theorem, there exists a prime Q ⊆ P such that
Q ∩ A = p. Since Q ∩ A = p, Q ⊇ pB. This contradicts that P contains pB
minimally. Thus P ∩ A = p. ¥

Exercise 4.25 Suppose A is an integrally-closed domain, A ֒→ B is an integral

extension of rings, and B is torsion free over A. Let b be an ideal in B and
a = b ∩ A. Then ht b = ht a.

Exercise 4.26 Let A be Noetherian, B be a finitely generated A-algebra, and

f : A ֒→ B be a ring homomorphism. If we suppose the Going-Down Property
holds, then f ∗ : Spec(B) → Spec(A) is an open map.

Exercise 4.27 Let A → B be flat. Then the Going-Down Property holds.

(And hence Spec(B) → Spec(A) is an open map.)

Example 4.28 Let k be a field. Let

P = (x3 − yz, y 2 − xz, z 3 − x2 z) ⊆ k[x, y, z]

and let A = k[x, y, z]/P . What is dim A?

We note that A ≃ k[t3 , t4 , t5 ] and that k[t3 , t4 , t5 ] ֒→ k[t] is an integral
extension. Since dim k[t] = 1, dim A = 1.

4.2 Normal Domains and DVRs

Now we study what happens when we add the Noetherian condition into the
mix.

Definition A Noetherian integrally closed domain is called a normal do-

main.

Example 4.29 k[X1 , X2 , . . . , Xn ] and k[[X1 , . . . , Xn ]] are normal domains.

Definition A local PID is called a discrete valuation ring, often denoted

DVR.

Example 4.30 Z(p) , k[[X]], and k[X](X) , where k is a field, are DVRs.

Proposition 4.31 (Characterization of DVRs) Let (A, m) be a local integral

domain. Then the following are equivalent:

(1) A is a DVR.

(2) A is normal and dim(A) = 1.

(3) A is normal and there exists 0 6= x ∈ A such that m is an essential prime

of Ax. In other words, m ∈ AssA (A/Ax).

85
4.2. NORMAL DOMAINS AND DVRS

(4) m 6= 0, and m is principal.

Proof (1) ⇒ (2) Suppose A is a DVR. Then by assumption A is a PID. Thus
dim(A) = 1. We also know that PID ⇒ UFD ⇒ integrally closed. Since A is a
domain and local, it is thus normal.
(2) ⇒ (3) Suppose A is normal and dim(A) = 1. Take any 0 6= x ∈ m. Since
x ∈ m, x is a system of parameters itself. Since dim(A) = 1, ℓ(A/xA) < ∞. It
follows that AssA (A/xA) = {m}.
(3) ⇒ (4) Suppose A is normal and there exists 0 6= x ∈ A with AssA (A/xA) =
{m}. Then A/m ֒→ A/xA. Let y ∈ A/xA be the image of 1 under this map.
Then my ⊆ xA. Thus myx−1 ⊆ A where myx−1 is an ideal of A.
Suppose myx−1 ⊆ m. Since y ∈ / xA, yx−1 ∈ / A. Write z = yx−1 . The
maximal ideal mP is finitely generated, so we can write m = (x1 , . . . , xn ). So we
can write zxi = Aij xj for each i. So
−ai1 x1 − ai2 x2 + · · · + (z − aii xi ) + · · · − ain xn = 0
for each i = 1, . . . , n. Using the same determinant trick as in Proposition 4.1,
we get that det(A)xi = 0 for all i, where A is the matrix
 
z − a11 −a12 ··· −a1n
 −a21 z − a22 · · · −a2n 
 
 .. .. .. .
 . . . 
−an1 −an2 ··· z − ann
Thus det(A)m = 0. So by Nakayama’s Lemma, Corollary 2.35, det(A) = 0.
Thus z is integral over A. Since A is integrally closed, z ∈ A. This is a
/ A. So we must have that myx−1 = A. But now m = xy −1 A.
contradiction to z ∈
Thus m is principal.
(4) ⇒ (1) Exercise. ¥
Theorem 4.32 (Serre) Let A be Noetherian, then A is normal if and only if
both of the following hold:
(1) Ap is a DVR for all primes p of height 1.
(2) For all 0 6= x ∈ A, if p ∈ Ass(A/Ax) then ht(p) = 1.
Proof (⇒) Suppose A is normal. Pick a prime P with ht(P ) = 1. Then
dim(AP ) = 1 and AP is integrally closed. So AP is a DVR. Now take Q ∈
AssA (A/xA) for some 0 6= x ∈ A. Then QAQ ∈ AssAQ (AQ /xAQ ). Since AQ is
normal and since QAQ is an essential prime of xAQ , we have by the previous
proposition that AQ is a DVR. Thus dim(AQ ) = ht(Q) = 1. T
(⇐) As a scholium to Proposition 4.7 we know that if A = m∈MaxSpec(A) Am
then A is integrally closed. Since A is a domain, for every ideal P of A with
ht(P ) = 1 we have Am ⊆ AP ⊆ Frac(A) for some maximal ideal m. Thus
\ \
A⊆ Am ⊆ AP .
m∈MaxSpec(A) P ∈Spec(A)
ht(P )=1

86
 CHAPTER 4. INTEGRAL EXTENSIONS

T
Thus to show A is normal it is enough to show that A = P : ht(P )=1 AP .
T
Take z ∈ ht(P )=1 AP . We can write z = x/y where x, y ∈ A and y 6= 0.
Thus x ∈ yAP for all P with ht(P ) = 1. Let

yA = q1 ∩ · · · ∩ qr

be a primary decomposition of yA where qi is pi -primary. Then for each i,

pi ∈ AssA (A/yA) and so ht(pi ) = 1. Thus yApi = qi Api for each i. So x ∈ qi Api
for all i. Thus x ∈ yA. So xy ∈ A. Hence A is normal. ¥

Example 4.33 Consider f (x, y) = y 2 − x2 (1 + x) ∈ k[x, y], which is irreducible

by Eisenstein’s Criterion. To see this, note x + 1 is prime in k[x] and divides
all coefficients of y i but the leading term y 2 and (x + 1)2 does not divide the
coefficient of y 0 .
Therefore, A = k[x, y]/(f (x)) is a domain. Is A normal? No, since xy is
integral over A but is not in A. To see this, note xy = (1 + x) 2 is an infinite
1

series. It follows then that A is not normal.

Example 4.34 Consider A = k[x, y, z]/(xz − y 2 , x3 − yz, x2 y − z 2 ). Note
A ≃ k[t3 , t4 , t5 ], and we have the nonsurjective integral extension

k[t3 , t4 , t5 ] ֒→ k[t].

Since k[t] is the integral closure of k[t3 , t4 , t5 ], we see that A is not normal.
Example 4.35 k[x1 , x2 , . . . , xn ] is a UFD, and so it is normal.
Example 4.36 The ring k[x, y, u, v]/(xy − uv) is not normal. This is hard to
show. We will discuss this later.

4.3 Dedekind Domains

Definition An integral domain A is called a Dedekind domain if A is normal
and dim(A) = 1.
Note then that a DVR is just a local Dedekind domain.
Throughout this section A will be an integral domain and K = Frac(A) will
denote the field of fractions.
Definition Let A be an integral domain and K = Frac(A). An A-submodule
M of K is called a fractional ideal of A if all elements of M have a common
denominator; that is,
1
M ⊆A· for some 0 6= d ∈ A.
d
Theorem 4.37 A is a Dedekind domain if and only if the set of fractional
ideals form a group under multiplication.

87
4.3. DEDEKIND DOMAINS

Definition A fractional ideal M of A is called invertible if there is a fractional

ideal M −1 such that M M −1 = A.

Remark If M1 and M2 are fractional ideals of A and K = Frac(A), then the

following are again fractional ideals of A:

(1) M1 + M2

(2) M1 ∩ M2

(3) M1 M2

(4) (M1 :K M2 ) = {x ∈ K : xM2 ⊆ M1 }

Proposition 4.38 Suppose M is an invertible fractional ideal. Then M −1 =

(A :K M ).

Proof We have that M M −1 = A. Thus M −1 ⊆ (A :K M ).

We also have

(A :K M ) = (A :K M )A
= (A :K M )M M −1
⊆ AM −1
= M −1 .

Thus M −1 = (A :K M ). ¥

Proposition 4.39 Suppose M is an invertible fractional ideal. Then M is

finitely generated over A.
P
Proof Since M M −1 = A we can write 1 = i mi ni where mi ∈ M and
ni ∈ M −1 for each i. Now take x ∈ M . Then
X
x= mi ni x
i
X
= xi ai ,
i

where ai = ni x ∈ A. Thus M is finitely generated over A. ¥

Proposition 4.40 If a finite product of fractional ideals is invertible if and

only if each fractional ideal in the product is invertible.

Proof Exercise. ¥

Proposition 4.41 Let I be an ideal of A. Suppose I can be factored into

a product of invertible prime ideals. Then any other factorization of I into a
product of prime ideals is identical.

88
 CHAPTER 4. INTEGRAL EXTENSIONS

Proof Suppose I = P1 · · · Pn where each Pi is an invertible prime ideal and

also I = Q1 · · · Qm where each Qi is a prime ideal. Pick a minimal prime Pi in
{P1 , . . . , Pn }. Then I ⊆ Pi . Since Pi is prime we must have Qj ⊆ Pi for some
j. Similarly we must have Pk ⊆ Qj for some k. Since Pi was minimal, i = k.
Therefore Pi = Qj . Now consider IPi−1 = P1 · · · Pbi · · · Pn and use induction on
the number of factors. ¥

Proposition 4.42 Let A be an integral domain. Suppose every ideal can be

expressed as a product of prime ideals. Then every nonzero prime ideal is
invertible and maximal, in particular A is Noetherian.

Proof First we show that every invertible prime ideal of A is maximal. Let
P be an invertible prime ideal. Choose a ∈
/ P . By assumption we can write

(P + aA)2 = P1 · · · Pn ,
P + a2 A = Q1 · · · Qm .

Let (−) denote the image in A/P . Since (P + aA)2 = P + a2 A, we have that

(a2 ) = P 1 · · · P n = Q1 · · · Qm .

Since the principal ideal (a2 ) is invertible, so are P i and Qj for each i, j by
Proposition 4.40. By Proposition 4.41 we must have that n = m and for each
i there exists j with P i = Qj . Therefore by the Correspondence Theorem,
Pi = Qj in A. Hence P + a2 A = (P + aA)2 . We now have

P ⊆ P + a2 A = (P + aA)2 ⊆ P 2 + aA,

so for any y ∈ P we may write y = z + ax where z ∈ P 2 and x ∈ A. Since

ax = y − z ∈ P and since a ∈
/ P we have that x ∈ P . Hence

P ⊆ P 2 + P a ⊆ P.

Thus

P = P2 + Pa
= P (P + Aa).

Since P is invertible there is a fractional ideal P −1 with P P −1 = A. Hence

A = P −1 P
= P −1 P (P + (a))
= P + (a).

Therefore P + (a) = A and so P is maximal.

Now take any prime ideal P and pick some nonzero a ∈ P . Write Aa =
P1 · · · Pn , a product of invertible prime ideals. Since the principal ideal aA

89
4.3. DEDEKIND DOMAINS

is invertible, so are P1 , . . . , Pn by Proposition 4.40. It is easy to see that P

must contain one of the Pi , otherwise, consider a product of elements xi where
xi ∈ Pi −P and note that x1 . . . xn ∈ P1 . . . Pn −P , a contradiction. By our work
above each Pi is maximal since they are each invertible. Therefore P is itself
maximal. Again by Proposition 4.40 P is invertible since the Pi ’s are invertible.
Finally note thatPsince any nonzero ideal I is invertible, we have II −1 = A.
n −1
Thus we can write Pn i=1 ai bi where ai ∈ I and bi ∈ I for each i. Thus for
any x ∈ I, x = i=1 (xbi )ai . So a1 , . . . , an generate I. Therefore every nonzero
ideal is finitely generated and A is Noetherian. ¥

Theorem 4.43 Let A be a Dedekind domain. Then any fractional ideal M

can be written uniquely as M = pn1 1 · · · pnr r , where pi is a prime ideal and ni ∈ Z
for i = 1, . . . , r.

Proof First we show that every ideal can be written as a product of prime
ideals. Let a be an ideal. We may assume since (0) is prime that (0) 6= a. Let
a = q1 ∩ · · · ∩ qr be a primary decomposition with qi being pi -primary. Since
dim(A) = 1, each pi is maximal. Thus Api is a DVR and so qi Api = pni i Api .
Since pi is maximal, pni i is pi -primary. So pni i Api ∩ A = pni = qi . Therefore

a = pini ∩ · · · ∩ pnr r = pn1 1 · · · pnr r .

Since any fractional ideal M is contained in d1 A for some d ∈ A, by writing

dM and dA as products of primes we can write M is a product of primes with
possibly negative exponents. ¥

Definition Let A be a Dedekind domain. Given a prime p, we define for a

fractional ideal M vp (M ) = n if M = pn pn1 1 · · · pnr r is the factorization of M .
The function vp is called a discrete valuation.
Proposition 4.44 Let vp be a discrete valuation for the Dedekind domain A,
with K = Frac(A). Then for fractional ideals M1 , M2 :
(1) vp (M1 + M2 ) = min{vp (M1 ), vp (M2 )}
(2) vp (M1 · M2 ) = vp (M1 ) + vp (M2 )
(3) vp (M1 ∩ M2 ) = max{vp (M1 ), vp (M2 )}
(4) vp ((M1 :K M2 )) = vp (M1 M2−1 ) = vp (M1 ) − vp (M2 )
(5) M1 ⊆ M2 if and only if vp (M1 ) 6 vp (M2 ) for all primes p.

Proof Exercise. ¥

Corollary 4.45 Let M1 be a fractional ideal in a Dedekind domain A. Then

M1 is an integral ideal, M1 ⊆ A, if and only if vp (M1 ) > 0 for all primes p.
Theorem 4.46 Let A be an integral domain. Then every ideal in A is a
product of prime ideals if and only if the fractional ideals of A form a group
under multiplication.

90
 CHAPTER 4. INTEGRAL EXTENSIONS

Proof (⇒) This is clear from the above.

(⇐) So assume the fractional ideals of A form a multiplicative group. As before
we that all nonzero ideals of A are invertible and hence finitely generated. So
A is Noetherian. Consider the family

F = {ideals a ⊆ A : a is not a product of primes.}.

Suppose F = 6 ∅. Then F has a maximal element, say a. Clearly a cannot be a

maximal ideal so there exists a maximal ideal m with a ( m. So am−1 ( A.
We clearly have a ⊆ am−1 . If a = am−1 then also a = am. But a is
finitely generated so by Nakayama’s Lemma, Corollary 2.35, there is m ∈ m
such that (1 − m)a = (0). This is a contradiction as A is a [Link]
a ( am−1 ( A. Therefore, since a was maximal in F, am−1 ∈ / F. So we can
write am−1 = p1n1 · · · pnr r . So a = mpn−1
1 · · · pnr r , contradicting a ∈ F. Therefore
F = ∅. ¥

Theorem 4.47 Suppose A is an integral domain. Then A is a Dedekind

domain if and only if every ideal can be expressed as a product of prime ideals.

Proof (⇒) This follows from Theorem 4.43.

(⇐) By the previous theorem every nonzero ideal is invertible. By Propo-
sition 4.42 we have that every nonzero prime ideal is maximal. Therefore A is
Noetherian and dim(A) = 1. We must show that A is integrally closed. Set
K = Frac(A). Take x ∈ K − A. Suppose x is integral over A. Then

xn + an−1 xn−1 + · · · + a0 = 0

for some ai ∈ A. Since x ∈ K we can write x = µλ , where λ, µ ∈ A. Since we

can write
xn = −an−1 xn−1 − · · · − a0 ,
and since the right hand side contains denominators µi for i at most n − 1,
we have that mun−1 xn ∈ A. Setting d = µn−1 we have that dxt ∈ A for all
t > 0. Let P be any nonzero prime ideal. Then vP (dxt ) > 0 for all t > 0. By
Proposition 4.44, vP ((d)) + tvP ((x)) > 0 for all t > 0. Thus vP ((x)) > 0 for all
P ∈ Spec(A) − {(0)}. Therefore x ∈ A. So A is integrally closed. ¥

4.4 The Krull-Akizuki Theorem

Definition Given a domain A and an A-module M we define the rank of M ,
denote rankA (M ), to be the following vector space rank

rankA (M ) := {rankK (K ⊗A M )},

where K = Frac(A). If A is not a domain we define the rank of M to be

rankA (M ) := max{rankA/p (M/pM ) : p is a minimal prime of A}

91
4.4. THE KRULL-AKIZUKI THEOREM

Lemma 4.48 Let A be a Noetherian domain with dim(A) 6 1 and M a torsion

free module over A of rank r < ∞. Then for any nonzero a ∈ A,
ℓ(M/aM ) 6 r · ℓ(A/aA).
Proof We first prove the finitely generated case.

Case 1 Assume M is finitely generated over A.

Since M is torsion free over A, the localization map M ֒→ U −1 M is injec-
tive, where U = A − {0}. Let α1 , . . . , αn be a set of generators for M . Then
α1 αn −1
1 , . . . , 1 is a set of generators for U M . By assumption U −1 M is an r
dimensional vector space over K = Frac(A). In particular we may suppose
that α11 , . . . , α1r form a K-basis for U −1 M by throwing away linearly dependent
elements. Thus for i > r we have
X r
αi αj
= λij where λj ∈ K.
1 j=1
1

Finding a common denominator we may write

µij
λij = where µij , s ∈ A
s
for all i, j. So by clearing denominators we get that
r
X
sαi = µij αj for r < i 6 n.
i=1

Let F be the free A-module generated by α1 , . . . , αr . Then we have the exact

sequence
0→F →M →Q→0
where Q is generated by the images of αr+1 , . . . , αn . Since sαj ∈ F for i > r and
since Q ≃ M/F , we have sQ = 0. Thus Q is a finitely generated module over
A/sA. Since dim(A/sA) = 0, ℓ(Q) < ∞. Tensoring the above exact sequence
with A/an A we get the exact sequence
F/an F → M/an A → Q/an Q → 0.
Therefore
ℓ(M/an M ) 6 ℓ(F/an F ) + ℓ(Q/an Q).
Since M/aM ≃ an−1 M/an M and since M is torsion free, we have that ℓ(M/an M ) =
nℓ(M/aM ). Similarly ℓ(F/an F ) = n·ℓ(F/aF ) = nr·ℓ(A/aA). Since ℓ(Q) < ∞,
we must have ℓ(Q/an Q) 6 ℓ(Q). Thus for all n,
1
ℓ(M/aM ) = ℓ(M/an M )
n
1 1
6 ℓ(F/an F ) + ℓ(Q/an Q)
n n
1
= r · ℓ(A/aA) + ℓ(Q).
n

92
 CHAPTER 4. INTEGRAL EXTENSIONS

Therefore
ℓ(M/aM ) 6 r · ℓ(A/aA).

Case 2 We now prove the general case. Suppose that the assertion is not true
for the module M , in other words there exists a ∈ A such that ℓ(M/aM ) >
r · ℓ(A/aA). Then we may choose a finitely generated submodule M ′ ⊆ M such
that ℓ(M ′ /aM ′ ) > rℓ(A/aA). But then

ℓ(M ′ /aM ′ ) > r · ℓ(A/aA)

> rankA (M ′ ) · ℓ(A/aA),

a contradiction of the finitely generated case. ¥

Theorem 4.49 (Krull-Akizuki) Let A be a Noetherian domain, dim(A) 6 1,

K = Frac(A), L/K a finite field extension, A ⊆ B ⊆ L, and B a subring of L.
Then B is Noetherian and dim(B) 6 1.

Proof We prove this theorem in two steps:

Step 1 We first reduce to the case that Frac(A) = K = L.

WLOG we can assume that L = Frac(B). Since [L : K] < ∞, we can find
b1 , . . . , bn ∈ B such that L = K(b1 , . . . , bn ). Set x = bi . Since each bi are
algebraic over K they satisfy relations of the form
an−1 a0
(x)n + (x)n−1 + · · · + = 0,
c c
where a0 , . . . , an−1 , c ∈ A. Therefore

(cx)n + an−1 (cx)n−1 + · · · + cn−1 a0 = 0.

So cx is integral over A. So replacing each bi by an A-multiple which is

integral over A we can assume that b1 , . . . , bn are all integral over A. Set
D = A[b1 , . . . , bn ]. Thus dim(D) = dim(A), Frac(D) = Frac(B) = L and

D ⊆ B ⊆ L.

Since D is finitely generated over A, it is Noetherian. Thus we are reduced to

the case where L = K.

Step 2 We now prove the theorem assuming L = K.

In our situation we have A ⊆ B ⊆ K = Frac(A). Thus rankA (B) = 1. So we
may apply Lemma 4.48 to get that ℓ(B/aB) 6 ℓ(A/aA) < ∞ for any 0 6= a ∈ A.
Let b 6= 0 be an ideal in B. Pick 0 6= b ∈ b. Since b is algebraic over A it satisfies
a relation of the form

am bm + am−1 bm−1 + · · · + a0 = 0 with ai ∈ A.

93
4.4. THE KRULL-AKIZUKI THEOREM

Since B is a domain we can assume a0 6= 0. Thus 0 6= a0 ∈ b ∩ A. So

ℓB (b/a0 B) 6 ℓA (b/a0 B) 6 ℓA (B/a0 B) < ∞. Thus b/a0 B is a finitely generated
B-module; hence b is finitely generated. Thus B is Noetherian. Further, if p
is any nonzero prime ideal of B, then B/pB is Artinian and a domain, hence a
field. So p is maximal. Therefore dim(B) = 1. ¥

Corollary 4.50 Let A be a Noetherian domain with dim(A) 6 1. Then the

integral closure of A is Noetherian.
Remark Many extension of the above theorem that one might want are actu-
ally false. For the following A is a Noetherian domain, K = Frac(A), [L : K] <
∞ and B is the integral closure of A in L.
(1) If A is as above even if dim(A) 6 1, then B is not necessarily a finitely
generated A-module.
(2) If dim(A) > 2 and C is a ring such that A ⊆ C ⊆ L then C is not
necessarily Noetherian.
(3) If dim(A) > 3 then B is not necessarily Noetherian.
See [14] for examples of these.
We include a couple of similar results:
Theorem 4.51 Let A be a Noetherian domain with dim(A) 6 2. Then the
integral closure of A is Noetherian.

Proof See [14], Theorem 33.12. ¥

Theorem 4.52 Let A be a finitely generated k-algebra, K = Frac(A), [L :

K] < ∞ and let B be the integral closure of A in L. Then B is a finitely
generated A-module and is Noetherian.

Proof See [6], Chapter 13. ¥

Definition Given field L which is a finite extension of another field K, the

trace of an element α ∈ L is defined as
X
trL/K (α) := σ(α).
σ∈GalK (L)

Theorem 4.53 Let L/K be a finite field extension. Then L/K is separable if
and only if there exists 0 6= x ∈ L such that trL/K (x) 6= 0.

Proof See [17]. ¥

Theorem 4.54 Let A be a normal domain, K = Frac(A), L/K a finite, sepa-

rable field extension, and let B be the integral closure of A in L. Then B is a
finitely generated A-module

94
 CHAPTER 4. INTEGRAL EXTENSIONS

Proof Let e1 , . . . , en be a basis of L over K such that e1 , . . . , en ∈ B. Let

ee1 , . . . , een be the corresponding dual basis of L, that is trL/K (e
ei ej ) = δij . Fix
b ∈ B. Then thinking of B as a submodule of L∗ = HomK (L, K), we can write
b = λ1 ee1 + · · · + λn een where λ1 , . . . , λn ∈ K. Then λi = trL/K (bei ) ∈ K. But
the trace of an element is the sum of its conjugates, each of which are integral
over A. Thus λi is integral over A for each i. So λi ∈ A for each i. Therefore
B ⊆ Ae en ⊆ L∗ and so B is finitely generated over A.
e1 + · · · + Ae ¥

4.5 Noether’s Normalization Lemma

Theorem 4.55 (Noether’s Normalization Lemma) Let k be a field, A a
finitely generated k-algebra. Let
a1 ( a2 ( · · · ( ar ( A
be an increasing sequence of ideals in A. Then there exist x1 , . . . , xd ∈ A that
are algebraically independent over k such that:
(1) A is integral over C = k[x1 , . . . , xd ].
(2) For all i = 1, . . . , r, we have integers h(i) ∈ [0, d] such that
ai ∩ C = (x1 , . . . , xh(i) ).

Proof We prove this theorem in four steps:

Step 1 Reduce to the case where A is a polynomial ring. Since A is a finitely

generated k-algebra, we may write A = k[y1 , . . . , yn ]. Set B = k[Y1 , . . . , Yn ], a
polynomial ring in n variables. Then we have the surjection
η : B → A,
Yi 7→ yi .
Hence we obtain an increasing sequence of ideals in B:
η −1 (0) ⊆ η −1 (a1 ) ⊆ η −1 (a2 ) ⊆ · · · ⊆ η −1 (ar )
Assuming the theorem is true for the polynomial ring B, we have algebraically
independent elements x1 , . . . , xn ∈ B such that B is integral over D = k[x1 , . . . , xn ]
and such that
I ∩ B = (x1 , . . . , xh(0) )
and
η −1 (ai ) ∩ B = (x1 , . . . , xh(i) ) where for i < j h(0) 6 h(i) 6 h(j).
Setting C = η(D) we see that C = k[zh(0)+1 , . . . , zn ] where zi = η(xi ). Since B
is integral over D, A is integral over C and
ai ∩ C = (zh(0)+1 , . . . , zh(i) ).
It is also clear that the elements zh(0)+1 , . . . , zn remain algebraically independent
over k. Thus we have proved the theorem for A.

95
4.5. NOETHER’S NORMALIZATION LEMMA

Step 2 Assuming A is a polynomial ring we prove the theorem for r = 1 and

a1 = (x1 ) principal with x1 ∈
/ k. (Note that the case where x1 ∈ k is obvious.)
By assumption we can write A = k[Y1 , . . . , Yn ]. Set

xi = Yi − Y1αi for i = 2, . . . , n

where the αi are yet-to-be-determined constants. Then we have the inclusion

C = k[x1 , . . . , xn ] ֒→ A. If Y1 is integral over C then A would be integral over
C and we would be done. So we proceed to show that for some choice of the
αi , Y1 is integral over C.
Since x1 ∈ A we may write
X
x1 = f (Y1 , . . . , Yn ) = ai1 ...in Y1i1 Y2i2 · · · Ynin .

We then have that

f (Y1 , x2 + Y1α2 , x3 + Y1α3 , . . . , xn + Y1αn ) = x1 .

We want to choose the αi so that they the highest degree term in f (Y1 , x2 +
Y1α2 , . . . , xn + Y 1 αn ) is of the form ai1 ...in Y1i1 +α2 i2 +···+αn in . Pick s larger than
all exponents ik appearing in the expansion of f . We leave it to the reader to
check that setting α2 = s, α3 = s2 ,. . .,αn = sn−1 satisfies the above criterion.
With this choice of αi we have that A is integral over C. Thus we have dim(C) =
dim(A) = n.
We must also show that these elements x1 , . . . , xn are algebraically indepen-
dent; in other words, we must show that C = k[x1 , . . . , xn ] is isomorphic to a
polynomial ring. So take an onto map
η
k[X1 , . . . , Xn ] −→ k[x1 , . . . , xn ].

Since dim(k[X1 , . . . , Xn ]) = dim(C) = n, Ker(η) = 0, for otherwise dim(C) =

dim(k[X1 , . . . , Xn ]/ Ker(η)) < n.
Last, we must verify that a1 ∩ C = (x1 ). Clearly a1 ∩ C ⊇ (x1 ). But by the
Going-Down Theorem, ht(a1 ∩ C) = ht(a1 ) = 1. Since (x1 ) is a prime ideal of
height 1 inside a1 ∩ C, we must have a1 ∩ C = (x1 ).

Step 3 We now prove the theorem for r = 1 and a any ideal of the polynomial
ring A = k[Y1 , . . . , Yn ].
Pick nonzero x1 ∈ a and consider the ideal x1 A. Then by Step 2 there are
x2 , . . . , xn ∈ A such that A is integral over C = k[x1 , . . . , xn ] and x1 A∩C = (x1 ).
Now consider a ∩ C. We proceed by induction on the number of variables n.
If n = 1 then the theorem is obvious. So we consider the ideal a∩k[x2 , . . . , xn ]
in the polynomial ring B = k[x2 , . . . , xn ]. By induction there are t2 , . . . , tn ∈ B
such that B is integral over k[t2 , . . . , tn ] and a ∩ k[t2 , . . . , tn ] = (t2 , . . . , td ) for
some d 6 n. Setting D = k[x1 , t2 , . . . , tn ] we have that A is integral over D and

a ∩ D = x1 D + a ∩ k[t2 , . . . , tn ] = (x1 ) + (t2 , . . . , tn ) = (x1 , t2 , . . . , td ).

96
 CHAPTER 4. INTEGRAL EXTENSIONS

Step 4 We now prove the general case of the theorem for the polynomial ring
A = k[Y1 , . . . , Yn ].
We proceed by induction on r. Step 3 finished the base case of r = 1. So by
induction we may assume that there exists algebraically independent elements
x1 , . . . , xn such that A is integral over C = k[x1 , . . . , xn ] and such that

ai ∩ C = (x1 , . . . , xh(i) ) with h(i) 6 h(j) for 1 6 i 6 j 6 r − 1.

Write d = h(r − 1). We may assume that h(r − 1) > 0. Consider a ∩

k[xd+1 , . . . , xn ] in D = k[xd+1 , . . . , xn ]. By Step 3 we can find td+1 , . . . , tn
algebraically independent over k such that D is integral over k[td+1 , . . . , tn ] and

a ∩ k[td+1 , . . . , tn ] = (td , . . . , th(r) ) where h(r) 6 n.

We leave it to the reader to check that A is integral over

B = k[x1 , . . . , xd , td+1 , . . . , tn ]

and that
ai ∩ B = (x1 , . . . , xh(i) ) for i < r
and
ar ∩ B = (x1 , . . . , xd , td+1 , . . . , th(r) ).
This completes the proof. ¥

Definition If k is a field, A is called an affine k-algebra if A is a finitely

generated k-algebra. If in addition A is a domain, then A is called an affine
k-domain.

Definition If K is a field extension of k, then the transcendence degree of

K over k is the cardinality of a maximal algebraically independent set S over
k. Hence, K is an algebraic extension of k(S). We denote the transcendence
degree of K over k by tr degk (K).

Corollary 4.56 Suppose A is an affine k-domain. Then dim(A) = tr degk (K),

where K = Frac(A).

Proof Let dim(A) = n and let

p0 ( · · · ( pn

be a corresponding chain of prime ideals in A. By Noether’s Normalization

Lemma, Theorem 4.55, we have algebraically independent elements x1 , . . . , xd ∈
A such that
k[x1 , . . . , xd ] ֒→ A
is an integral extension and such that

pi ∩ k[x1 , . . . , xd ] = (x1 , . . . , xh(i) )

97
4.5. NOETHER’S NORMALIZATION LEMMA

where i < j implies that h(i) 6 h(j). In particular, since k[x1 , . . . , xd ] ֒→ A is

an integral extension, we have that d = n and pi ∩ k[x1 , . . . , xn ] = (x1 , . . . , xi ).
Taking fraction fields we have
k ֒→ k(x1 , . . . , xn ) ֒→ K
where the second inclusion is an algebraic field extension. Thus tr degk (K) 6 n.
If tr degk (K) < n then we would have an integral extension
k[y1 , . . . , ym ] ֒→ A
with m < n. But this would imply that dim(A) = m < n, a contradiction.
Therefore tr degk (K) = n = dim(A). ¥
Corollary 4.57 Let A be an affine k-algebra and let m be a maximal ideal of
A. Then A/m is a finite-dimensional vector space over k.
Proof By Noether’s Normalization Lemma we can find x1 , . . . , xn ∈ A that
are algebraically independent over k such that
k[x1 , . . . , xn ] ֒→ A
is an integral extension and such that
m ∩ k[x1 , . . . , xn ] = (x1 , . . . , xn ).
Therefore
k = k[x1 , . . . , xn ]/(x1 , . . . , xn ) ֒→ A/m
is an integral extension which is finitely generated as a k-algebra. Thus A/m is
a finitely generated k-module. ¥
Corollary 4.58 Let A = k[X1 , X2 , . . . , Xn ]. Then any maximal ideal m is
generated by n elements. Moreover, m can be written as
m = (f1 (X1 ), f2 (X1 , X2 ), . . . , fn (X1 , . . . , Xn )).
Proof Let m be a maximal ideal of A. Then by the previous corollary
k ֒→ A/m
is a finite field-extension. Write αi for X i ∈ A/m. So we have:
A/m = k[α1 , . . . , αn ]
= k(α1 , . . . , αn ).
Letting Pα
k (X) denote the minimal irreducible polynomial for α over k, consider
the polynomials
α
Pk01 (X1 ) where k0 = k,
α
Pk12 (X2 ) where k1 = k[α1 ],
.. ..
. .
αn
Pkn−1 (Xn ) where kn−1 = k[α1 , . . . , αn−1 ].

98
 CHAPTER 4. INTEGRAL EXTENSIONS

Let fi (X1 , . . . , Xi ) be Pα
ki−1 (Xi ) where α1 , . . . , αi−1 are replaced by X1 , . . . , Xi−1
i

respectively. We now show that

m = (f1 (X1 ), f2 (X1 , X2 ), . . . , fn (X1 , . . . , Xn )).
That
(f1 (X1 ), f2 (X1 , X2 ), . . . , fn (X1 , . . . , Xn )) ⊆ m
is clear. The other containment will follow if we show that
K := k[X1 , . . . , Xn ]/(f1 (X1 ), f2 (X1 , X2 ), . . . , fn (X1 , . . . , Xn )) ≃ A/m.
Note that
k[X1 , . . . , Xn ]
K=
(f1 (X1 ), f2 (X1 , X2 ), . . . , fn (X1 , . . . , Xn ))
k[α1 , X2 , . . . , Xn ]
≃
(f2 (α1 , X2 ), . . . , fn (α1 , X2 , . . . , Xn ))
k[α1 , α2 , . . . , Xn ]
≃
(f3 (α1 , α2 , X3 ), . . . , fn (α1 , α2 , X3 , . . . , Xn ))
..
.
≃ k[α1 , . . . , αn ]
≃ A/m.
So we are done. ¥
Corollary 4.59 Let k be an algebraically closed field and let A = k[X1 , . . . , Xn ].
Then every maximal ideal of A is of the form
m = (X1 − a1 , X2 − a2 , . . . , Xn − an ),
where a1 , a2 , . . . , an ∈ k.
Proof Use the previous corollary. ¥
Exercise 4.60 By the previous Corollary, we have a bijective correspondence
between k n and MaxSpec(k[X1 , . . . , Xn ]) when k is algebraically closed. Show
that k n ≃ MaxSpec(k[X1 , . . . , Xn ]) is dense in Spec(k[X1 , . . . , Xn ]).
Corollary 4.61 Let A be an affine domain, p a prime ideal of A. Then
ht(p) + dim(A/p) = dim(A).
Proof By Noether’s Normalization Lemma we can find x1 , . . . , xn ∈ A which
are algebraically independent over k with A integral over k[x1 , . . . , xn ] and such
that p ∩ k[x1 , . . . , xn ] = (x1 , . . . , xi ). By the Going-Down Theorem ht(p) =
ht((x1 , . . . , xn )) = i. Also since k[x1 , . . . , xn ] ֒→ A is an integral extension, so is
k[xi+1 , . . . , xn ] = k[x1 , . . . , xn ]/(x1 , . . . , xi ) ֒→ A/p.
Thus dim(A/p) = n − i = dim(A) − ht(p). ¥

99
4.5. NOETHER’S NORMALIZATION LEMMA

Remark As a special case of this corollary we get the following result:

If m is a maximal ideal in an affine domain A, then ht(m) = dim(A).

Definition Let A be a ring. If any two saturated chains of primes of A between

two primes p ( q have the same length, then A is called catenary.

Corollary 4.62 Let A be an affine k-algebra, then A is catenary.

Proof Let
p = p0 ( p1 ( · · · ( pr = q

and
p = q0 ( q1 ( · · · ( qs = q

be two saturated chains of prime ideals in A. By considering both chains in

A/p, which is an affine domain, we may assume p = 0. Since A/p is an affine
domain we have by the previous corollary that

dim(A) − r = dim(A/pr )
= dim(A/q)
= dim(A/qs )
= dim(A) − s.

Therefore r = s. ¥

Corollary 4.63 Let A be an affine k-domain, K = Frac(A), L/K a finite field-

extension, and B the integral closure of A in L. Then B is a finitely generated
A-module.

Proof By Noether’s Normalization Lemma we can find x1 , . . . , xn ∈ A such

that A is integral over k[x1 , . . . , xn ] where x1 , . . . , xn are algebraically indepen-
dent over k. So it suffices to show that B is a finitely generated k[x1 , . . . , xn ]-
module. Now let L be the normal closure of L in the algebraic closure of K and
let D be the integral closure of k[x1 , . . . , xn ] in L.
G
Set F = L where G = Galk(x) (L). Then L is separable over F and F is
purely inseparable over k(x1 , . . . , xn ) by Proposition 4.20. Let C be the integral
closure of k[x1 , . . . , xn ] in F . If we can show that C is a finitely generated
k[x1 , . . . , xn ]-module, then C would be Noetherian, and hence normal. It would
then follow from Theorem 4.54 that D was a finitely generated C-module, and
hence that B was a finitely generated A-module.
Therefore we must show that C is a finitely generated k[x1 , . . . , xn ]-module in
the situation where C ⊆ F and F is purely inseparable over E = k(x1 , . . . , xn ).

100
 CHAPTER 4. INTEGRAL EXTENSIONS

We have the following diagram:

D L

G
C F =L

B L

A K

k[x] E = k(x)

Since F/k(x) is purely inseparable, we can write F = E(y1 , . . . , yd ) where E =

i
k(x), such that there exists i > 0 with yjp ∈ E for each j = 1, . . . , d. Thus we
can write
i fj (x1 , . . . , xn )
yjp = ∈ k(x) = E
gj (x1 , . . . , xn )
where fj , gj ∈ k[x1 , . . . , xn ]. ³Let S ´ be the (finite) set of coefficients of the
1
′
polynomials fj , gj . Set k = k S p . Since we are appending a finite number
i

of pi -th roots, [k ′ : k] < ∞. Set

E ′ = k ′ (x1 , . . . , xn ),
1 1
i i
F ′ = k ′ (x1p , . . . , xnp ),
1 1
i i
C ′ = k[x1p , . . . , xnp ].
Then we have the following diagram:

C′ F′

C F

k ′ [x] E′

k[x] E

Clearly k ′ [x1 , . . . , xn ] is a finitely generated k[x1 , . . . , xn ]-module since [k ′ : k] <

∞. Also note that C ′ is a polynomial ring. So C ′ is a UFD and hence normal.

101
4.5. NOETHER’S NORMALIZATION LEMMA

So C ′ is the integral closure of k[x1 , . . . , xn ] in F ′ . Since C is integral over

k[x1 , . . . , xn ], C ′ is the integral closure of C in F ′ . Since C ′ is integral over
k ′ [x1 , . . . , xn ] and is finitely generated as an algebra over k ′ [x1 , . . . , xn ], we have
that C ′ is a finitely generated as a k ′ [x1 , . . . , xn ]-module. It follows that C is a
finitely generated k[x1 , . . . , xn ]-module. ¥

Exercise 4.64 Let f : A → B be a map of finitely generated k-algebras, where

k is a field. Suppose m is a maximal ideal of B. Then f −1 (m) is a maximal
ideal of A.

Exercise 4.65 Let A be a finitely generated k-algebra, where k is a field. For

any ideal I of A we have √ \
I= m,
I⊆m

where m runs through all maximal ideals of A.

Exercise 4.66 Suppose A is a Noetherian domain and B is a finitely generated

A-algebra. Then there exist x1 , . . . , xn ∈ B algebraically independent over A
and 0 6= a ∈ A such that A[ a1 ][x1 , . . . , xn ] ֒→ B[ a1 ] is an integral extension.

Exercise 4.67 Let f1 , . . . , ft ∈ Z[X1 , . . . , Xn ]. Then f1 , . . . , ft have a common

root over C if and only if they have a common root over finite fields of infinitely
many prime characteristics.

Exercise 4.68 Let K be an algebraically closed field and let L ⊇ K be a field-

extension. Then a set of polynomials in K[x1 , . . . , xn ] has a common solution
in L if and only if they have a common solution in K.

102
Chapter 5

Homological Methods

At first, homological methods may seem very abstract. How can something so
abstract be useful? Consider the following:
Definition If (A, m) is local, A is a regular local ring if

dim(A) = µ(m) := {the minimal number of generators of m}.

Example 5.1 Let A = k[X1 , . . . , Xn ]. If

m = (p1 (X1 ), p2 (X1 , X2 ), . . . , pn (X1 , . . . , Xn )),

where pi (X1 , . . . , Xi ) is a polynomial in exactly i variables, then ht(m) = n and

so Am is a regular local ring as

n = ht(m) = dim(Am ) = µ(m).

Now that we have a little background, consider this statement:

Theorem If A is a regular local ring and p is a prime ideal of A, then Ap is a
regular local ring.
Before the advent of homological algebra, many books and papers went
into proving this result, some taking up to 200 pages, even for the case of
A = C[[X1 , . . . , Xn ]]. We will develop the tools of homological algebra and kill
this problem with ease.

5.1 Complexes and Homology

Definition Let A be a ring, by a complex, we mean a sequence of A-modules
and A-module homomorphisms
dn+1n d
· · · −→ Xn+1 −→ Xn −→ Xn−1 −→ · · ·

such that dn ◦ dn−1 = 0 for all n ∈ Z. We denote a complex by X• .

103
5.1. COMPLEXES AND HOMOLOGY

Definition If X• is a complex of A-modules, then the nth homology of X•

is
Ker(dn )
Hn (X• ) := .
Im(dn+1 )

Definition Let A be a ring. By a cocomplex, we mean a sequence of A-

modules and A-module homomorphisms

dn−1 dn
· · · −→ X n−1 −→ X n −→ X n+1 −→ · · ·

such that dn ◦ dn+1 = 0 for all n ∈ Z. We denote a cocomplex by X • .

Definition If X • is a cocomplex of A-modules, then the nth cohomology of

X • is
Ker(dn )
H n (X • ) := .
Im(dn−1 )

Since complexes and cocomplexes are dual notions, we will only discuss the
situation for complexes, and leave the rest as an exercise for the reader.

Definition Let X• and Y• be two complexes over a ring A. A map of com-

plexes
f• : X• → Y•

is a collection of A-module homomorphisms such that the diagram below com-

mutes:

dX
n+1 dX
n
··· Xn+1 Xn Xn−1 ···
fn+1 fn fn−1

··· Yn+1 Yn Yn−1 ···

dY
n+1 dY
n

Exercise 5.2 Show that a map of complexes f• : X• → Y• , defines a collection

of homomorphisms:

fi : Ker(dX Y
i ) → Ker(di ).

fi : Im(dX Y
i+1 ) → Im(di+1 ).
Hn (f• ) : Hn (X• ) → Hn (Y• ).

Definition Two maps of A-complexes

f• : X• → Y• ,
g• : X• → Y• ,

104
 CHAPTER 5. HOMOLOGICAL METHODS

are called homotopic if there exist A-module maps hn : Xn → Yn+1 such that
in the diagram below

dX
n+1 dX
n
··· Xn+1 Xn Xn−1 ···
hn hn−1
fn+1 gn+1 fn gn fn−1 gn−1

··· Yn+1 Yn Yn−1 ···

dY
n+1 dY
n

we have
dYn+1 ◦ hn + hn−1 ◦ dX
n = fn − gn

for all n ∈ Z. We denote this by f• ∼ g• .

Exercise 5.3 Check that f• ∼ g• implies that Hn (f• ) = Hn (g• ).

Definition A sequence of complexes and complex maps

f• g•
0 −→ X•′ −→ X• −→ X•′′ −→ 0

is called an exact sequence if for all n ∈ Z,

fn gn
0 −→ Xn′ −→ Xn −→ Xn′′ −→ 0

is an exact sequence of A-modules and A-module homomorphisms.

Lemma 5.4 Given an exact sequence of complexes,

f• g•
0 −→ X•′ −→ X• −→ X•′′ −→ 0

we obtain a long exact sequence of homologies:

Hn+1 (f• ) Hn+1 (g• )

··· Hn+1 (X•′ ) Hn+1 (X• ) Hn+1 (X•′′ )
∂n+1

Hn (f• ) Hn (g• )
Hn (X•′ ) Hn (X• ) Hn (X•′′ )
∂n

Hn−1 (f• ) Hn−1 (g• )

Hn−1 (X•′ ) Hn−1 (X• ) Hn−1 (X•′′ ) ···

105
5.1. COMPLEXES AND HOMOLOGY

Proof Consider the following commutative diagram with exact rows:

.. .. ..
. . .

′ fn+1 gn+1 ′′
0 Xn+1 Xn+1 Xn+1 0
d′n+1 dn+1 d′′
n+1

fn gn
0 Xn′ Xn Xn′′ 0
d′n dn d′′
n

′ fn−1 gn−1 ′′
0 Xn−1 Xn−1 Xn−1 0

.. .. ..
. . .

First we will show that

Hn (f• ) Hn (g• )
Hn (X•′ ) −→ Hn (X• ) −→ Hn (X•′′ )

is exact. By the construction of Hn (f• ) and Hn (g• ) we know that Im Hn (f• ) ⊆

Ker Hn (g• ). Thus we must show that

Ker Hn (g• ) ⊆ Im Hn (f• ).

′′
Let xn ∈ Xn and suppose that xn ∈ Ker Hn (g• ). Then there exists x′′n+1 ∈ Xn+1
such that
d′′n+1 (x′′n+1 ) = gn (xn ).
By exactness of the rows, there exists xn+1 ∈ Xn+1 such that

gn+1 (xn+1 ) = x′′n+1 .

This xn+1 in turn maps down via dn+1 to some element of Xn , call it yn . By
the commutativity of the diagram

gn (xn − yn ) = 0

and so by the exactness of the rows, there exists x′n ∈ Xn′ such that

fn (x′n ) = xn − yn .

However, since yn = dn+1 (xn+1 ) we have y n = 0 and

xn = xn − y n ∈ Hn (X• ).

106
 CHAPTER 5. HOMOLOGICAL METHODS

Thus, xn ∈ Im Hn (f• ). The method used in the above part of the proof is called
diagram chasing. Often when it is done in practice, the elements found above
are written next to the object they live in on the commutative diagram itself.
Now we need to define the ∂n ’s. Consider x′′n ∈ Hn (X•′′ ) we will define
∂n (x′′n ). Take xn ∈ Xn such that

gn (xn ) = x′′n .

Since x′′n ∈ Ker(d′′n ), d′′n (x′′n ) = 0. Hence if xn−1 = dn (xn ), then

gn−1 (xn−1 ) = 0.

So by the exactness of the rows above, there exists x′n−1 ∈ Xn−1

′
such that

fn−1 (x′n−1 ) = xn−1 .

Hence d′n−1 (x′n−1 ) = 0. Since dn−1 (xn−1 ) = dn−1 ◦ dn (xn ) = 0,

d′n−1 (x′n−1 ) = 0.

Now we define
∂n (x′′n ) := x′n−1 ∈ Hn−1 (X•′ ).
It is left as an exercise for the reader to check that this definition of ∂n is well
defined. Moreover, the reader should check that the sequences
Hn (g• ) ∂
Hn (X• ) −→ Hn (X•′′ ) −→
n
Hn−1 (X•′ ),
∂ Hn−1 (f• )
Hn (X•′′ ) −→
n
Hn−1 (X•′ ) −→ Hn−1 (X• ),

are both exact. ¥

Definition In the above proposition, the ∂n ’s are called connecting homo-

morphisms.
Corollary 5.5 Given a commutative diagram of complexes with short exact
rows:
0 X•′ X• X•′′ 0

0 Y•′ Y• Y•′′ 0
we get a commutative diagram with long exact rows:

··· Hn+1 (X•′′ ) Hn (X•′ ) Hn (X• ) Hn (X•′′ ) Hn−1 (X•′ ) ···

··· Hn+1 (Y•′′ ) Hn (Y•′ ) Hn (Y• ) Hn (Y•′′ ) Hn−1 (Y•′ ) ···

107
5.1. COMPLEXES AND HOMOLOGY

Definition Given a map of complexes f• : X• → Y• , the mapping cone of

f• is the following complex:

· · · −→ Xi ⊕ Yi+1 −→ Xi−1 ⊕ Yi −→ Xi−2 ⊕ Yi−1 −→ · · ·

where the degree i part is Xi−1 ⊕ Yi and the differentials are defined as follows:

di : Xi−1 ⊕ Yi → Xi−2 ⊕ Yi−1

(x, y) 7→ (−dX Y
i−1 (x), di (y) − fi−1 (x))

Exercise 5.6 Show that the mapping cone of a map of complexes is a complex.
Definition Given a complex X• , X• (j) is used to denote a shift, where
Xi (j) := Xi+j .
Exercise 5.7 Given a map of complexes f• : X• → Y• , let C• be the mapping
cone of f• . Then there is a short exact sequence of complexes:

0 Y• C• X• (−1) 0
y (0, y)
(x, y) x

The above short exact sequence of complexes induces a long exact sequence:
Hi (f• )
· · · −→ Hi (X• ) −→ Hi (Y• ) −→ Hi (C• ) −→ · · ·

Remark For more information on the mapping cone see [2, §2.6].

5.1.1 Projective Resolutions

Definition An A-module P is projective if any of the following equivalent
conditions are met:
(1) Given any right exact sequence M → N → 0 of A modules and a homo-
morphism ϕ : P → N , there exits ϕe : P → M such that the diagram
below commutes:
P
ϕ
e
ϕ

M π N 0

(2) HomA (P, −) is an exact functor.

(3) Every short exact sequence 0 → M ′ → M → P → 0 is split exact.
(4) There is a free module F such that F ≃ P ⊕ Q for some A-module Q.
Exercise 5.8 Show that the conditions in the above definition are actually
equivalent.

108
 CHAPTER 5. HOMOLOGICAL METHODS

Exercise 5.9 Show that if A → B is a ring homomorphism and P is a projec-

tive A-module, then P ⊗A B is a projective B-module.

Definition If M is an A-module, a projective resolution of M is a complex

of projective modules P• and a map π : P0 → M such that
2d 1 d π
· · · −→ P2 −→ P1 −→ P0 −→ M −→ 0

is exact.

Given a ring A, every module M has a projective resolution. Firstly, note

that given any module M , we may map a free module F0 onto M . Set

S1 := Ker(F0 → M ).

Now we may map another free module onto S1 . Hence we may inductively
define

Si+1 := Ker(Fi → Si ).

Now we may inductively write the exact sequences:

0 → S1 → F0 → M → 0,
0 → Si+1 → Fi → Si → 0.

Now we put the above short exact sequences together letting each Si connect
the short exact sequences:

0 0

S2
d2 d1 π
··· F1 F0 M 0

S1

0 0

The di ’s above are formed by taking the composition of the relevant maps, while
π is the canonical surjection. Hence we obtain a free resolution of M . Since
every free module is projective, we obtain a projective resolution of M . Note
that if M were a finitely generated module over a Noetherian ring, then we
could insist that each Fi be a finitely generated free module.

Lemma 5.10 Let f : M → N be a homomorphism of A-modules. If P• is

a complex of projective A-modules such that H0 (P• ) = M and Q• is an exact
complex with H0 (Q• ) = N , then there exists a map of complexes f• : P• → Q•
lifting f .

109
5.1. COMPLEXES AND HOMOLOGY

Proof Here is the situation in question:

dM
1 πM
··· P1 P0 M 0
f
dN
1 πN
··· Q1 Q0 N 0

Since P0 is projective, we can obtain f0 by:

P0
f0
f ◦πM

Q0 πN N 0

Now since S1 = Ker(πN ) = Im(dN

1 ), we can obtain f1 by:

P1
f1
f0 ◦dM
1

Q1 S1 0
dN
1

Note that f0 ◦ dM
1 maps into S1 since it is in Ker(πN ). Working inductively, we
repeat a similar procedure to find f• . ¥

Remark Note that the lift f• is not unique.

Lemma 5.11 Let f : M → N be a homomorphism of A-modules, P• be a

complex of projective A-modules such that H0 (P• ) = M , and Q• be an exact
complex with H0 (Q• ) = N . If f• and g• are two lifts of the map f , then f• ∼ g• .

Proof Here is the situation in question:

dM
1 πM
··· P1 P0 M 0
f1 g1 f0 g0 f

··· Q1 Q0 πN N 0
dN
1

Since both f• and g• are chain maps, we have that

f ◦ πM = πN ◦ f0 = πN ◦ g0 ,

and so we see
0 = πN ◦ f0 − πN ◦ g0 = πN ◦ (f0 − g0 ).

110
 CHAPTER 5. HOMOLOGICAL METHODS

Thus Im(f0 − g0 ) ⊆ Ker(πN ) = Im(dN

1 ). By the projectivity of P0 , we obtain
h0
P0
h0
f0 −g0

Q1 Im(dN
1 ) 0
dN
1

such that h0 : P0 → Q1 . Set h−1 to be the zero map and dM

0 := πM . Now

dN M
1 ◦ h0 + h−1 ◦ d0 = f0 − g0 .

Working inductively, suppose that we have constructed homotopy maps for

i < n. We must show

dN M
n+1 ◦ hn + hn−1 ◦ dn = fn − gn .

By the definition of a map of complexes, we have that:

dN M
n ◦ fn = fn−1 ◦ dn ,

dN M
n ◦ gn = gn−1 ◦ dn ,

and so by the inductive hypothesis,

dN M
n ◦ (fn − gn ) = (fn−1 − gn−1 ) ◦ dn

= (dN M M
n ◦ hn−1 + hn−2 ◦ dn−1 ) ◦ dn

= dN M M M
n ◦ hn−1 ◦ dn + hn−2 ◦ dn−1 ◦ dn

= dN M
n ◦ hn−1 ◦ dn .

Thus

dN M N N M
n ◦ (fn − gn − hn−1 ◦ dn ) = dn ◦ (fn − gn ) − dn ◦ hn−1 ◦ dn

= dN M N M
n ◦ hn−1 ◦ dn − dn ◦ hn−1 ◦ dn
= 0.

Therefore
Im(fn − gn − hn−1 ◦ dM N N
n ) ⊆ Ker(dn ) = Im(dn+1 ).

Now we obtain hn as before

Pn
hn
fn −gn −hn−1 ◦dM
n

Qn+1 Im(dN
n+1 ) 0
dN
n+1

such that hn : Pn → Qn+1 and dN M

n+1 ◦ hn + hn−1 ◦ dn = fn − gn . Thus f• is
homotopic to g• . ¥

111
5.1. COMPLEXES AND HOMOLOGY

Lemma 5.12 (Horseshoe Lemma) Let

0 → M ′ → M → M ′′ → 0

is an exact sequence of A-modules, and P•′ → M ′ and P•′′ → M ′′ be projective

resolutions. Then, there exists a projective resolution P• → M such that the
rows in the diagram below are exact:

0 P•′ P• P•′′ 0

0 M′ M M ′′ 0

Proof Take projective resolutions (P•′ , π ′ ) and (P•′′ , π ′′ ) of M ′ and M ′′ re-

spectively and consider the following commutative diagram:

0 0 0

d′1 π′
··· P1′ P0′ M′ 0
f

d1 π
··· P1′ ⊕ P1′′ P0′ ⊕ P0′′ M 0
η0
g

d′′
1 π ′′
··· P1′′ P0′′ M ′′ 0

0 0 0

First we must define π : P0 ։ M such that the diagram commutes. Since P0′′ is
projective, there exist η0 : P0′′ → M such that g ◦ η0 = π ′′ . Define

π(x′ , x′′ ) := f ◦ π ′ (x′ ) + η0 (x′′ ).

This makes the diagram above commute.

Now we will define d1 and then complete the construction by induction.
Since π ′′ ◦ d′′1 = 0, we have that

Im(η0 ◦ d′′1 ) ⊆ Ker(g) = Im(f ).

Thus we have the commutative diagram:

P1′′
η1
η0 ◦d′′
1

f ◦π ′
P0′ Im(f ) 0

112
 CHAPTER 5. HOMOLOGICAL METHODS

And so
f ◦ π ′ ◦ η1 = η0 ◦ d′′1 .
By changing the sign of η1 we see that

f ◦ π ′ ◦ η1 + η0 ◦ d′′1 = 0,

and so we define

d1 (x′ , x′′ ) := (d′1 (x′ ) + η1 (x′′ ), d′′1 (x′′ )).

Plugging everything in we see that π ◦ d1 = 0. Following the diagram around

we see that Im(d1 ) = Ker(π).
Now working inductively, consider the diagram:

0 0 0

d′n d′n−1
′ ′
··· Pn′ Pn−1 Pn−2 ···

dn ′ ′′ dn−1 ′ ′′
··· Pn′ ⊕ Pn′′ Pn−1 ⊕ Pn−1 Pn−2 ⊕ Pn−2 ···

d′′
n
d′′
n−1
′′ ′′
··· Pn′′ Pn−1 Pn−2 ···

0 0 0
′′ ′′
Here we have inductively defined ηn−1 : Pn−1 → Pn−2 similarly to how we
defined η1 and we can write

dn−1 (x′ , x′′ ) := (d′n−1 (x′ ) + ηn−1 (x′′ ), d′′n−1 (x′′ )).

We must now define dn . By construction we have that

d′n−2 ◦ ηn−1 + ηn−2 ◦ d′′n−1 = 0,

and so we see
d′n−2 ◦ ηn−1 ◦ d′′n = −ηn−1 ◦ d′′n−1 ◦ d′′n = 0.
Thus Im(ηn−1 ◦ d′′n ) ∈ Ker(d′n−2 ) = Im(d′n−1 ). Again we are in the following
situation:
Pn′′
ηn
ηn−1 ◦d′′
n

′
Pn−1 Im(d′n−1 ) 0

113
5.1. COMPLEXES AND HOMOLOGY

So we may now define

dn (x′ , x′′ ) := (d′n (x′ ) + ηn (x′′ ), d′′n (x′′ )).

with the sign of ηn chosen so that:

dn ◦ dn−1 (x′ , x′′ ) = (d′n (d′n−1 (x′ ) + ηn−1 (x′′ )) + ηn (d′′n−1 (x′′ )), d′′n (d′′n−1 (x′′ )))
= (d′n (ηn−1 (x′′ )) + ηn (d′′n−1 (x′′ )), 0)
= 0.

Since the direct sum of two projective modules is projective, and since we can
see that Im(dn ) = Ker(dn−1 ), we see that we have constructed the needed exact
sequence. ¥

Compare the above construction to the construction of the mapping cone as

described in the previous section.
Exercise 5.13 Show that given the following commutative diagram of A-
modules with exact rows:

0 M′ M M ′′ 0

0 N′ N N ′′ 0

Then there exist associated projective resolutions that form a commutative di-
agram of A-complexes with exact rows:

0 P•′ P• P•′′ 0

0 Q′• Q• Q′′• 0

Hint: Do you know how to draw cubes?

5.1.2 Injective Resolutions

Definition An A-module E is injective if any of the following equivalent
conditions are met:
(1) Given any left exact sequence 0 → M ′ → M of A-modules and a homo-
morphism ϕ : M ′ → E, there exits ϕ e : M → E such that the diagram
below commutes:
ι
0 M′ M
ϕ
ϕ
e
E

114
 CHAPTER 5. HOMOLOGICAL METHODS

(2) HomA (−, E) is an exact functor.

(3) Every short exact sequence 0 → E → M → M ′′ → 0 is split exact.
Exercise 5.14 Show that the conditions in the above definition are actually
equivalent.
Theorem 5.15 (Baer’s Criterion) Let A be a ring. An A-module E is injective
if and only if given any ideal I of A, a module homomorphism ϕ : I → E can
be extended to a module homomorphism Φ : A → E.
Proof (⇒) If E is injective, and ϕ : I → E, then apply the functor HomA (−, E)
to
0 → I → A.
Since HomA (−, E) is an exact functor, ϕ : I → E can be extended to a module
homomorphism Φ : A → E.
(⇐) Suppose that every A-module homomorphism I → E can be lifted to a
homomorphism A → E. Consider the diagram:

0 L M
ϕ

E
Let L′ be a submodule of M containing L and ϕ′ : L′ → E be a lift of ϕ. In
this case, the ordering: (L′ , ϕ′ ) 6 (L′′ , ϕ′′ ) if L′ ⊆ L′′ and ϕ′′ |L′ = ϕ′ , partially
orders the set S,
S = {(L′ , ϕ′ ) : ϕ′ lifts ϕ to L′ }.
6 ∅ as (L, ϕ) ∈ S. Now considering any chain C in S, it is clear
Note that S =
that [
(L′ , ϕ′ )
(L′ ,ϕ′ )∈C

is an upper bound. Hence by Zorn’s Lemma this set contains a maximal element,
(M ′ , ϕ).
e We will show that M ′ = M . Suppose that m ∈ M − M ′ . Consider the
ideal (M ′ :A m). Note that
(M ′ :A m) → E,
e
a 7→ ϕ(am),
is an A-module homomorphism from (M ′ :A m) to E, thus there exits an A-
module homomorphism Φ : A → E which restricts to the one above. Consider
the submodule
M ′ + Am ⊆ M
and define
f : M ′ + Am → E,
m′ + am 7→ ϕ(m)
e + Φ(a).

115
5.1. COMPLEXES AND HOMOLOGY

To check that f is well defined, consider m1 + a1 m = m2 + a2 m. Then

(a1 − a2 )m = m2 − m1

and so (a1 − a2 ) ∈ (M ′ :A m). Thus

e 1 − a2 )m) = ϕ(m
Φ(a1 − a2 ) = ϕ((a e 2 − m1 ).

So we see that
e 1 ) + Φ(a1 ) = ϕ(m
ϕ(m e 2 ) + Φ(a2 ).
Thus f is well defined and it is a lift extending ϕ to M ′ + Am, which contradicts
the maximality of (M ′ , ϕ),
e and so we must conclude that M ′ = M . ¥

Definition If A is a PID, an A-module M is divisible if given any m ∈ M

and nonzero a ∈ A, there exists q ∈ M such that
m
m=a·q which essentially says = q.
a
From the above definition, we obtain the following corollary to Baer’s Cri-
terion, whose proof we will leave as an exercise to the reader.

Corollary 5.16 Let A be a PID. An A-module is injective if and only if it is

divisible.

Example 5.17 Q is an injective Z-module as it is divisible. Moreover, Q/Z is

an injective Z-module as well.

Theorem 5.18 If A is a ring, then every A-module can be embedded into an

injective A-module.

Proof Step 1. We will show that every A-module M can be embedded into
a divisible Z-module. First note that while M is an A-module, it is also a Z-
module. Hence there exists a free Z-module Z surjecting onto M . Letting K
be the kernel of this surjection, we have that

Z/K ≃ M.

On the other hand, Z canonically embeds into some free Q-module, call it Q.
If we denote this canonical embedding by η : Z → Q, and set D = Q/ Im(η|K )
we may write
M ≃ Z/K ≃ Im(η)/ Im(η|K ) ⊆ D.
Since D is divisible, we have completed Step 1.
Step 2. We will now embed M into HomZ (A, D) where D is defined as in
Step 1. We will denote the embedding of M into D by

ι : M ֒→ D.

116
 CHAPTER 5. HOMOLOGICAL METHODS

Applying HomZ (A, −), we get an injective Z-module homomorphism

ι∗ : HomZ (A, M ) ֒→ HomZ (A, D),

ϕ 7→ ι ◦ ϕ.

Noting that there is a canonical injection of A-modules

HomA (A, M ) ֒→ HomZ (A, M ),

where the A-module structure on HomZ (A, M ) is given by

a · f (x) := f (ax) for a ∈ A and f ∈ HomZ (A, M ),

we see that we have an embedding of A-modules

M ≃ HomA (A, M ) ֒→ HomZ (A, M ) ֒→ HomZ (A, D).

Step 3. We will show that if D is a divisible Z-module, then HomZ (A, D) is

an injective A-module. Note that this step completes the proof of the theorem.
Consider an ideal a of A. By Baer’s Criterion, Theorem 5.15, we need to show
that any A-module homomorphism

ψ : a → HomZ (A, D), extends to

Ψ : A → HomZ (A, D).

Now consider the Z-module homomorphism:

ϕ:a→D
a 7→ ψ(a)(1A )

One should check that this is indeed a Z-module homomorphism. Since D is a

divisible Z-module, by Corollary 5.16, we see that it is an injective Z-module,
and so we write
ι
0 a A
ϕ
ϕ
e
D
and obtain a Z-module homomorphism ϕ
e such that the diagram above com-
mutes. Now define

Ψ : A → HomZ (A, D),

a 7→ f,

e
where f (x) = ϕ(ax). One should check that this defines an A-module homo-
morphism. For a ∈ a and x ∈ A, we have

e
Ψ(a)(x) = f (x) = ϕ(ax) = ϕ(ax) = ψ(ax)(1A ).

117
5.1. COMPLEXES AND HOMOLOGY

Since ψ is an A-module homomorphism, we have

ψ(ax)(1A ) = xψ(a)(1A ) = ψ(a)(x),

where the right-most equality is due to the A-module structure on HomZ (A, D).
Hence we have Ψ : A → HomZ (A, D) and Ψ|a = ψ. Thus we see HomZ (A, D) is
an injective A-module. ¥

Definition If M is an A-module, an injective resolution of M is a complex

of injective modules E• and a map ι : M → E0 such that
ι d0 d1
0 −→ M −→ E 0 −→ E 1 −→ E 2 −→ · · ·

is exact.
If A is a ring and M is an A-module, we can use Theorem 5.18 to construct
an injective resolution as follows. Set

E 0 := {a module which M embeds into},

C 1 := Coker(M ֒→ E 0 ),

and inductively define

E i := {a module which C i embeds into},

C i+1 := Coker(C i ֒→ E i ).

Now we may inductively write the exact sequences:

0 → M → E 0 → C 1 → 0,
0 → C i → E i → C i+1 → 0.

Putting the above exact sequences together we obtain:

0 0

C2
ι d1 d2
0 M E0 E1 ...

C1

0 0

The di ’s above are formed by taking the composition of the relevant maps, while
ι is the canonical injection. Hence we obtain an injective resolution of M .

We now include the corresponding results for injective resolutions that we

had for projective resolutions. The statements and proofs are precisely the duals
of the projective case.

118
 CHAPTER 5. HOMOLOGICAL METHODS

Lemma 5.19 Let f : M → N be a homomorphism of A-modules. If I •

is a cocomplex of injective A-modules such that H 0 (I • ) = M and J • is an
exact cocomplex with H 0 (J • ) = N , then there exists a map of cocomplexes
f • : I • → J • lifting f .
Lemma 5.20 (Horseshoe Lemma) Let

0 → M1 → M2 → M3 → 0

is an exact sequence of A-modules, and M1 → I1• and M3 → I3• be injective

resolutions. Then, there exists an injective resolution M2 → I2• such that the
rows in the diagram below are exact:

0 I1• I2• I3• 0

0 M1 M2 M3 0

Exercise 5.21 Show that given the following commutative diagram of A-

modules with exact rows:

0 M1 M2 M3 0

0 N1 N2 N3 0

Then there exist associated injective resolutions that form a commutative dia-
gram of A-complexes with exact rows:

0 I1• I2• I3• 0

0 J1• J2• J3• 0

Exercise 5.22 Consider the short exact sequence:

0 → Z → Q → Q/Z → 0.

Note that this is an injective resolution of Z. Is Q/Z indecomposable? That is,

is Q/Z a direct sum of Z-modules? If so, what are the summands? If not, why
not?

5.2 Tor and Ext

5.2.1 Tor
To start, let’s recall some of Göthe’s words:

119
5.2. TOR AND EXT

Habe nun, ach! Philosophie,

Juristerei und Medizin,
Und leider auch Theologie!
Durchaus studiert, mit heissem Bemühn.
Da steh ich nun, ich armer Tor!
Und bin so klug als wie zuvor.
– Göthe, Faust act I, scene I

Construction of Tor
Definition Given a ring A and an A-module N , TorA i (−, N ) is the left de-
rived functor of the right exact covariant functor − ⊗A N .
To be more explicit, consider any projective resolution of an A-module M :
d
2 1 d π
· · · −→ P2 −→ P1 −→ P0 −→ M −→ 0

Apply the functor − ⊗A N and chop off the M ⊗A N term to get the complex
P• ⊗A N :
· · · → P2 ⊗A N → P1 ⊗A N → P0 ⊗A N → 0
We now define
Ker(di ⊗ 1)
TorA
i (M, N ) := Hi (P• ⊗A N ) =
Im(di+1 ⊗ 1)
Note that since
P1 → P0 → M → 0
is exact,
P1 ⊗A N → P0 ⊗A N → M ⊗A N → 0
is also exact. Hence
TorA
0 (M, N ) ≃ M ⊗A N.

Proposition 5.23 TorA i (M, N ) does not depend on the choice of projective
resolution used. Hence it is well-defined.

Proof Let P• and Q• be two projective resolutions of M . So we may write:

··· Pi ··· P0 M 0
ϕi ϕ0 1M

··· Qi ··· Q0 M 0
ψi ψ0 1M

··· Pi ··· P0 M 0

Note that the lifts ϕ• and ψ• of 1M are guaranteed to exist by Lemma 5.10.
Thus ψ• ◦ ϕ• : P• → P• is a lift of 1M . But clearly 1P• is another such lift.

120
 CHAPTER 5. HOMOLOGICAL METHODS

Thus ψ• ◦ ϕ• ∼ 1P• . Similarly, ϕ• ◦ ψ• ∼ 1Q• . Applying − ⊗A N to everything

we obtain that
(ψ• ◦ ϕ• ) ⊗ (1N ) ∼ (1P• ) ⊗ (1N ) and (ϕ• ◦ ψ• ) ⊗ (1N ) ∼ (1Q• ) ⊗ (1N )
and so
Hi (P• ⊗A N ) ≃ Hi (Q• ⊗A N ).
Thus TorA
i (M, N ) is well-defined. ¥

Properties of Tor
Exercise 5.24 If N is A-flat or if M is A-flat, show that
TorA
i (M, N ) = 0

for all A-modules M and i > 0. Hint: For the second part, first show that if
0 → M1 → M2 → M3 → 0
is exact and M2 and M3 are flat, so is M1 .
Proposition 5.25 Given an exact sequence of A-modules,
0 → M ′ → M → M ′′ → 0
we obtain a long exact sequence of Tor’s:

··· TorA ′
n+1 (M , N ) TorA
n+1 (M, N ) TorA ′′
n+1 (M , N )
∂n+1

TorA ′
n (M , N ) TorA
n (M, N ) TorA ′′
n (M , N )
∂n

TorA ′
n−1 (M , N ) ············ TorA ′′
1 (M , N )
∂1

M ′ ⊗A N M ⊗A N M ′′ ⊗A N 0
Proof Let P•′ and P•′′ be projective resolutions of M ′ and M ′′ respectively.
By the Horseshoe Lemma there exists a projective resolution P•′ of M such that
0 → P•′ → P• → P•′′ → 0
is an exact sequence of complexes. Since P•′′ is projective, this complex is in
fact split exact, and so we have following short exact sequence of complexes:
0 → P•′ ⊗A N → P• ⊗A N → P•′′ ⊗A N → 0
By Lemma 5.4, we obtain a long exact sequence of homologies, and hence the
result. ¥

121
5.2. TOR AND EXT

Proposition 5.26 Given a ring A and two A-modules M and N , we then have

TorA A
i (M, N ) ≃ Tori (N, M ).

Proof First note that

TorA
0 (M, N ) ≃ M ⊗A N and TorA
0 (N, M ) ≃ N ⊗A M.

Since

M ⊗A N ≃ N ⊗A M
m ⊗ n 7→ n ⊗ m

we have that TorA A

0 (M, N ) ≃ Tor0 (N, M ). Consider

0→S→P →M →0

where P is a free module and S is the kernel of the surjection. Note that since
P is a free module, it is flat, and so

TorA
i (P, N ) = 0 for i > 0 and
TorA
i (N, P ) =0 for i > 0.

Hence by Proposition 5.25 we now have two long exact sequences of Tor’s:

··· TorA
i+1 (S, N ) 0 TorA
i+1 (M, N )
∂i+1

TorA
i (S, N ) 0 TorA
i (M, N )
∂i

TorA
i−1 (S, N ) 0 TorA
i−1 (M, N ) ···

And:
··· TorA
i+1 (N, S) 0 TorA
i+1 (N, M )
∂i+1

TorA
i (N, S) 0 TorA
i (N, M )
∂i

TorA
i−1 (N, S) 0 TorA
i−1 (N, M ) ···
Thus we see that

TorA A
i (M, N ) ≃ Tori−1 (S, N ) if i > 1,
TorA
i (N, M ) ≃ TorA
i−1 (N, S) if i > 1.

122
 CHAPTER 5. HOMOLOGICAL METHODS

We see now that it is enough to show that TorA A

1 (M, N ) ≃ Tor1 (N, M ). Consider
the commutative diagram with exact rows:

0 TorA
1 (M, N ) S ⊗A N P ⊗A N M ⊗A N 0

0 TorA
1 (N, M ) N ⊗A S N ⊗A P N ⊗A M 0

Note that the left-most terms are 0 because M and N are projective. And so
we see that TorA A
1 (M, N ) ≃ Tor1 (N, M ). ¥

Proposition 5.27 Given an exact sequence of A-modules,

0 → N ′ → N → N ′′ → 0

we obtain a long exact sequence of Tor’s:

··· TorA ′
n+1 (M, N ) TorA
n+1 (M, N ) TorA ′′
n+1 (M, N )
∂n+1

TorA ′
n (M, N ) TorA
n (M, N ) TorA ′′
n (M, N )
∂n

TorA ′
n−1 (M, N ) TorA
n−1 (M, N ) TorA ′′
n−1 (M, N ) ···

Proof Note that this follows by Lemma 5.26 and Lemma 5.25. However, we
can also give a direct proof. Let Q• be a projective resolution of N . Since
projective modules are flat, we have the short exact sequence of complexes:

0 → M•′ ⊗A Q• → M• ⊗A Q• → M•′′ ⊗A N → 0

By Lemma 5.4, we obtain a long exact sequence of homologies, and hence the
result. ¥
L L
Exercise 5.28 Tori (M, α Nα ) ≃ α Tori (M, Nα ).

Exercise 5.29 Let B be a flat A-algebra. Then

B ⊗A TorA B
i (M, N ) ≃ Tori (M ⊗A B, N ⊗A B).

In particular, if U is a multiplicatively closed set in A, then

−1
U −1 TorA U
i (M, N ) ≃ Tori (U −1 M, U −1 N ).

123
5.2. TOR AND EXT

5.2.2 Ext
First Construction
Definition Given a ring A and an A-module N , ExtiA (−, N ) is the left de-
rived functor of the left exact contravariant functor HomA (−, N ).
To be more explicit, consider any projective resolution of an A-module M :
2 d 1 d π
· · · −→ P2 −→ P1 −→ P0 −→ M −→ 0

Apply the functor HomA (−, N ) and chop off the HomA (M, N ) term to get the
complex HomA (P• , N ):
d∗
0 1 d∗ 2 d∗
0 −→ HomA (P0 , N ) −→ HomA (P1 , N ) −→ HomA (P2 , N ) −→ · · ·

where d∗0 := 0. We now define:

Ker(d∗i+1 )
ExtiA (M, N ) := H i (HomA (P• , N )) =
Im(d∗i )

The shift in degrees of the differentials in the quotient above, compared to the
definition of cohomology, is due to the fact that d∗i is the (i − 1)th differential
in the cocomplex HomA (P• , N ). Since

P1 → P0 → M → 0

is exact,

1 d∗
0 −→ HomA (M, N ) −→ HomA (P0 , N ) −→ HomA (P1 , N )

is also exact. Hence

Ker(d∗1 )
Ext0A (M, N ) = ≃ HomA (M, N ).
0
Proposition 5.30 ExtiA (M, N ) does not depend on the choice of projective
resolution of M used to compute it. Hence it is well-defined.

Proof Let P• and Q• be two projective resolutions of M . We can lift 1M to

maps of complex ϕ• and ψ• and write:

··· Pi ··· P0 M 0
ϕi ϕ0 1M

··· Qi ··· Q0 M 0
ψi ψ0 1M

··· Pi ··· P0 M 0

124
 CHAPTER 5. HOMOLOGICAL METHODS

Since 1P• and ψ• ◦ ϕ• are both lifts of 1M , we have ψ• ◦ ϕ• ∼ (1P• ). Similarly,

ϕ• ◦ ψ• ∼ (1Q• ). Applying (−)∗ = HomA (−, N ) to everything we obtain that
(ψ• ◦ ϕ• )∗ ∼ (1P• )∗ and (ϕ• ◦ ψ• )∗ ∼ (1Q• )∗
and so
H i (HomA (P• , N )) ≃ H i (HomA (Q• , N )).
Thus ExtA
i (M, N ) is well-defined. ¥

Second Construction
Definition Given a ring A and an A-module N , ExtiA (M, −) is the left
derived functor of the left exact covariant functor HomA (M, −).
To be more explicit, consider any injective resolution of an A-module N :
ι d0 d1
0 −→ N −→ E 0 −→ E 1 −→ E 2 −→ · · ·
Apply the functor HomA (M, −) and chop off the HomA (M, N ) term to get the
complex HomA (M, E • ):
d−1 d0 d1
∗
0 −→ HomA (M, E 0 ) −→
∗
HomA (M, E 1 ) −→
∗
HomA (M, E 2 ) −→ · · ·
where d−1
∗ := 0. We now define:

Ker(di∗ )
ExtiA (M, N ) := H i (HomA (M, E • )) =
Im(di−1
∗ )

Note that since

0 → N → E0 → E1
is exact,
d0
0 −→ HomA (M, N ) −→ HomA (M, E 0 ) −→
∗
HomA (M, E 1 )
is also exact. Hence
Ker(d0∗ )
Ext0A (M, N ) = ≃ HomA (M, N ).
0
Proposition 5.31 ExtiA (M, N ) does not depend on the choice of injective res-
olution of N used to compute it. Hence it is well-defined.
Proof Let I • and J • be two injective resolutions of N . So we may write:
0 N I0 ··· Ii ···
1N ϕ0 ϕi

0 N J0 ··· Ji ···
1N ψ0 ψi

0 N I0 ··· Ii ···

125
5.2. TOR AND EXT

Thus we see that ψ • ◦ ϕ• ∼ (1I • ). Similarly, ϕ• ◦ ψ • ∼ (1J • ). Applying

(−)∗ = HomA (M, −) to everything we obtain that
(ψ • ◦ ϕ• )∗ ∼ (1I • )∗ and (ϕ• ◦ ψ • )∗ ∼ (1J • )∗
and so
H i (HomA (M, I • )) ≃ H i (HomA (M, J • )).
Thus ExtA
i (M, N ) is well-defined. ¥

Properties of Ext
Proposition 5.32 The two constructions of ExtiA (M, N ) given above produce
isomorphic modules and hence are equivalent.
Proof We omit the proof of this result, though it is similar to the proof that
the two definitions of Tor are the same. Readers who are familiar with spectral
sequences can see the result easily by taking a projective resolution P• → M
of M and an injective resolution N → I • and considering the double complex
HomA (P• , I • ). We refer the reader to [16]. ¥
Proposition 5.33 Given an exact sequence of A-modules,
0 → M ′ → M → M ′′ → 0
we obtain a long exact sequence of Ext’s:

0 HomA (M ′′ , N ) HomA (M, N ) HomA (M ′ , N )

∂1

Ext1A (M ′′ , N ) ············ Extn−1 ′

A (M , N )
∂n−1

ExtnA (M ′′ , N ) ExtnA (M, N ) ExtnA (M ′ , N )

∂n

Extn+1 ′′
A (M , N ) Extn+1
A (M, N ) Extn+1 ′
A (M , N )
···

Proof Let P•′ , P• , and P•′′ be projective resolutions of M ′ , M , and M ′′ re-

spectively. Hence we have an exact sequence of complexes:
0 → P•′ → P• → P•′′ → 0
Since P•′′ is projective, our complex is in fact split exact, and so we have following
short exact sequence of complexes:
0 → HomA (P•′′ , N ) → HomA (P• , N ) → HomA (P•′ , N ) → 0
By Lemma 5.4, we obtain a long exact sequence of homologies, and hence the
result. ¥

126
 CHAPTER 5. HOMOLOGICAL METHODS

Proposition 5.34 Given an exact sequence of A-modules,

0 → N ′ → N → N ′′ → 0

we obtain a long exact sequence of Ext’s:

··· Extn−1 ′
A (M, N ) Extn−1
A (M, N ) Extn−1 ′′
A (M, N )
∂n−1

ExtnA (M, N ′ ) ExtnA (M, N ) ExtnA (M, N ′′ )

∂n

Extn+1 ′
A (M, N ) Extn+1
A (M, N ) Extn+1 ′′
A (M, N )
···

Proof Note, one could dualize the above proof or one could take a projective
resolution P• of M and look at the short exact sequence of complexes:

0 → HomA (P• , N ′ ) → HomA (P• , N ) → HomA (P• , N ′′ ) → 0

By Lemma 5.4, we obtain a long exact sequence of homologies, and hence the
result. ¥

Proposition 5.35 If A is a ring, the following are equivalent:

(1) M is projective.
(2) ExtiA (M, N ) = 0 for all A-modules N and for all i > 0.
(3) Ext1A (M, N ) = 0 for all A-modules N .

Proof All that needs to be shown is (3) ⇒ (1). We must show that given
any short exact sequence

0 → N ′ → N → N ′′ → 0

of A modules, we have

HomA (M, N ) ։ HomA (M, N ′′ ).

But by the long exact sequence of Ext and the fact that Ext1A (M, N ) = 0 for
all A-modules N , we have

0 → HomA (M, N ′ ) → HomA (M, N ) → HomA (M, N ′′ ) → 0

which shows M is projective. ¥

Proposition 5.36 If A is a ring, the following are equivalent:

(1) N is injective.

127
5.2. TOR AND EXT

(2) ExtiA (M, N ) = 0 for all A-modules M and for all i > 0.

(3) Ext1A (M, N ) = 0 for all A-modules M .

(4) Ext1A (M, N ) = 0 for all finitely generated A-modules M .

(5) Ext1A (A/I, N ) = 0 for all ideals I ⊆ A.

Proof All that needs to be shown is (5) ⇒ (1). By Baer’s Criterion, Theo-
rem 5.15, we must show that given any ideal I of A, a module homomorphism
ϕ : I → N can be extended to a module homomorphism Φ : A → N . This
amounts to saying that

HomA (A, N ) ։ HomA (I, N ).

Write
0 → I → A → A/I → 0

and apply the functor HomA (−, N ), and note that Ext1A (A/I, N ) = 0 for all
ideals I ⊆ A, to obtain the exact sequence

0 → HomA (A/I, N ) → HomA (A, N ) → HomA (I, N ) → 0

Hence N must be injective. ¥

Exercise 5.37
L Q
(1) ExtiA ( α Mα , N ) ≃ α ExtiA (Mα , N ).
Q Q
(2) ExtiA (M, α Nα ) ≃ α ExtiA (M, Nα ).

In particular, finite direct sums in either variable pass through Ext.

Exercise 5.38 Let B be an A-algebra that is finitely generated and projective

as an A-module or let B be a flat A-algebra where A is Noetherian and M is
finitely generated. Then

B ⊗A ExtiA (M, N ) ≃ ExtiB (M ⊗A B, N ⊗A B).

If U is a multiplicatively closed set in A, A is Noetherian and M is finitely

generated, then

U −1 ExtiA (M, N ) ≃ ExtiU −1 A (U −1 M, U −1 N ).

Hint: For help with this see [3, 6.7].

128
 CHAPTER 5. HOMOLOGICAL METHODS

5.3 Homological Notions of Dimension

5.3.1 Projective Dimension
Definition Given a ring A and an A-module M , the projective dimension
of M is defined to be:

pdA (M ) := inf{n : there exists a projective resolution of M of length n}.

Recall that
0 → Pn → Pn−1 → · · · → P1 → P0 → M → 0
is a resolution of length n if it is an exact complex and each Pi is projective.
Remark Sometimes projective dimension is called homological dimension.
Proposition 5.39 If A is a ring and M is an A-module, then the following are
equivalent:
(1) pdA (M ) 6 n.
(2) ExtiA (M, N ) = 0 for all A-modules N and for all i > n.
(3) Extn+1
A (M, N ) = 0 for all A-modules N .

Proof All that needs to be shown is (3) ⇒ (1). Write

P• : 0 → Sn → Pn−1 → · · · → P1 → P0 → M → 0

where Sn is the kernel of dn−1 . We’ll show that Sn is projective. By Proposi-

tion 5.35, we need only show that Ext1A (Sn , N ) = 0 for all A-module N . Break
up P• into short exact sequences as follows:

0 Sn Pn−1 Sn−1 0
0 Sn−1 Pn−2 Sn−2 0
.. .. ..
. . .
0 S1 P0 M 0

Apply HomA (−, N ) and from the corresponding long exact sequences for Ext
we see

Ext1A (Sn , N ) ≃ Ext2A (Sn−1 , N ) ≃ · · · ≃ ExtnA (S1 , N ) ≃ Extn+1

A (M, N ) = 0.

Hence we see that Sn is projective. ¥

As immediate corollaries to this proposition we have two more characteriza-

tions of projective dimension:
Corollary 5.40 If A is a ring and M is an A-module, then:
½ ¾
Given any projective resolution (P• , d• ) of M ,
pdA (M ) := inf n :
Ker(dn−1 ) is projective

129
5.3. HOMOLOGICAL NOTIONS OF DIMENSION

Corollary 5.41 If A is a ring and M is an A-module, then

pdA (M ) = sup{n : ExtnA (M, N ) 6= 0},

N

where N varies over all A-modules.

Proposition 5.42 Consider an exact sequence of A-modules

0 → M ′ → P → M ′′ → 0

where P is projective. The following are true:

(1) If M ′′ is projective, then so is M ′ .
(2) If pdA (M ′′ ) > 1, then pdA (M ′′ ) = pdA (M ′ ) + 1.

Proof (1) If M ′′ is projective, then the above exact sequence is split, and so
we have P ≃ M ′ ⊕ M ′′ . Since P is projective, it is a summand of a free module,
and so we have
M ′ ⊕ M ′′ ⊕ Q ≃ F
showing that M ′ is also a summand of a free module and hence is also projective.
(2) For some A-module N , apply HomA (−, N ) and look at the long exact
sequence for Ext to see that

ExtiA (M ′ , N ) ≃ Exti+1 ′′
A (M , N ) for i > 1.

The result follows from Proposition 5.39. ¥

Exercise 5.43 If A → B is a ring homomorphism and P is a projective A-

module, then P ⊗A B is a projective B-module.
Proposition 5.44 Given an A-module M , suppose that x ∈ A is a nonzerodi-
visor on both A and M . If pdA (M ) < ∞, then pdA/xA (M/xM ) < ∞.

Proof Consider the following projective resolution of M

P• : 0 → Pn → Pn−1 → · · · → P1 → P0 → M → 0

We would be done if we knew that

P• ⊗A A/xA : 0 → Pn /xPn → · · · → P0 /xP0 → M/xM → 0

was exact, by the above exercise.

Now note that

Hi (P• ⊗A A/xA) = TorA

i (M, A/xA).

Since
x
0 −→ A −→ A −→ A/xA −→ 0 (⋆)

130
 CHAPTER 5. HOMOLOGICAL METHODS

is a free resolution of A/xA, we see that

TorA
i (M, A/xA) = 0 for i > 2.

We must show that TorA1 (M, A/xA) = 0. Applying − ⊗A M to (⋆) above we

obtain
x
0 −→ TorA1 (M, A/xA) −→ M −→ M −→ M/xM −→ 0.

But we see that TorA 1 (M, A/xA) = 0 as the above complex is exact and mul-
tiplication by x is injective. Thus P• ⊗A A/xA is a projective resolution of
M/xM . ¥

Proposition 5.45 Let 0 → M → N → T → 0 be an exact sequence of A-

modules. Then
pdA (N ) 6 max{pdA (M ), pdA (T )}.

Proof This follows easily from the Horseshoe Lemma. ¥

5.3.2 Injective Dimension

Definition Given a ring A and an A-module M , the injective dimension of
M is defined to be:

idA (M ) := inf{n : there exists an injective resolution of M of length n}.

Recall that
0 → M → E 0 → E 1 → · · · → E n−1 → E n → 0
is a resolution of length n if it is an exact cocomplex and each E i is injective.
Proposition 5.46 If A is a ring and N is an A-module, then the following are
equivalent:
(1) idA (N ) 6 n.
(2) ExtiA (M, N ) = 0 for all A-modules M and for all i > n.
(3) Extn+1
A (M, N ) = 0 for all A-modules M .

(4) Extn+1
A (M, N ) = 0 for all finitely generated A-modules M .

(5) Extn+1
A (A/I, N ) = 0 for all ideals I ⊆ A.

Proof This proof is left as an exercise for the reader. Hint: See the proof of
Proposition 5.39. ¥

As immediate corollaries to this proposition we have two more characteriza-

tions of injective dimension:
Corollary 5.47 If A is a ring and N is an A-module, then:
½ ¾
Given any injective resolution (E • , d• ) of M ,
idA (M ) := inf n :
Im(dn−1 ) is injective

131
5.4. THE LOCAL CASE

Corollary 5.48 If A is a ring and N is an A-module, then

idA (N ) = sup{n : ExtnA (M, N ) 6= 0},

M

where M varies over all finitely generated A-modules.

5.3.3 Global Dimension

Definition Given a ring A, the global dimension of A is defined to be:

gd(A) := sup pdA (M )

M

where the M varies over all A-modules.

We now have the following corollary to Proposition 5.39 and Proposition 5.46:
Corollary 5.49 Given a ring A we have that:

gd(A) = sup pdA (M )

M ∈ModA

= sup idA (M )
M ∈ModA

= sup pdA (M ) such that M is finitely generated.

M ∈ModA

5.4 The Local Case

Let (A, m, k) be a local ring. Recall Corollary 2.39 which states that if M is a
finitely generated A-module, then

µ(M ) :={the minimal number of generators of M }

= rankk (M/mM ).

The above fact will be used extensively.

The next theorem is very important as it shows that projective modules and
flat modules are locally free. This means that when you localize a flat module
or a projective module, you get a free module.
Theorem 5.50 If A is a local ring and M is a finitely generated A-module,
then the following are equivalent:
(1) M is free.
(2) M is projective.
(3) M is flat.
(4) TorA
i (M, N ) = 0 for all A-modules N and for all i > 0.

(5) TorA
1 (M, k) = 0.

132
 CHAPTER 5. HOMOLOGICAL METHODS

Proof All that needs to be shown is (5) ⇒ (1). Let F be a free module
mapping onto the minimal generators of M and obtain the short exact sequence

0 → S → F → M → 0.

Applying − ⊗A k we obtain the short exact sequence

0 → S/mS → F/mF → M/mM → 0

since TorA1 (M, k) = 0. However, since we map a basis of F onto a minimal

set of generators of M , we see F/mF ≃ M/mM and so S/mS is 0. Hence by
Corollary 2.35, Nakayama’s Lemma, we see that S = 0 and so M ≃ F . ¥

Proposition 5.51 If (A, m, k) is a local ring and M is a finitely generated

A-module, then the following are equivalent:

(1) pdA (M ) 6 n.

(2) TorA
i (M, N ) = 0 for all A-modules N and for all i > n.

(3) TorA
n+1 (M, k) = 0.

Proof This proof is left as an exercise for the reader. Hint: See the proof of
Proposition 5.39. ¥

Later on we will be able to remove the condition that A is a local ring for
the above proposition.

Corollary 5.52 If (A, m, k) is a local ring and M is a finitely generated A-

module, then
pdA (M ) = sup{n : TorA
i (M, k) 6= 0}.

Proposition 5.53 If (A, m, k) is a local ring, the following are equivalent:

(1) gd(A) 6 n.

(2) TorA
i (M, N ) = 0 for i > n and all A-modules M and N .

(3) TorA
n+1 (k, k) = 0.

Proof All that needs to be shown is (3) ⇒ (1). If TorA n+1 (k, k) = 0, then
pdA (k) 6 n. So by Proposition 5.51, TorA n+1 (M, k) = 0 for all A-modules M . In
particular, TorA
n+1 (M, k) = 0 for all finitely generated A-modules M . Again by
Proposition 5.51 we have that pdA (M ) 6 n for all finitely generated A-modules
M . Thus gd(A) 6 n by Corollary 5.49. ¥

Corollary 5.54 (Main Point) If (A, m, k) is a local ring,

gd(A) = pdA (k).

133
5.4. THE LOCAL CASE

There is an analogous result to Corollary 5.52 for injective dimension over

local rings.

Theorem 5.55 Let (A, m, k) be a local ring and M a finitely generated A-

module. Then
idA (M ) = sup{i : ExtiA (k, M ) 6= 0}.

Proof We refer the reader to [4, Proposition 3.1.14]. ¥

5.4.1 Minimal Free Resolutions

Let (A, m, k) be a local ring and M be a finitely generated A-module. We are
going to discuss the construction of a minimal free resolution of M . Recalling
Corollary 2.39, set

β0 := rankk (M/mM ),
S1 := Ker(Aβ0 → M ),

where the map Aβ0 → M is defined by mapping a basis of Aβ0 onto a minimal
set of generators of M . Inductively define

βi := rankk (Si /mSi ),

Si+1 := Ker(Aβi → Si ),

where at each step, the map Aβi → Si is defined by mapping a basis of Aβi
onto a minimal set of generators of Si . Now we may inductively write the exact
sequences:

0 → S1 → Aβ0 → M → 0,
0 → Si+1 → Aβi → Si → 0.

The integer βi is sometimes called the ith Betti number of M and Si is referred
to as the ith syzygy of M . The rather mysterious word syzygy means yoke.
After putting the above exact sequences together, we can see why syzygy is a
good term to use for the Si ’s, as each Si is connecting two free modules via
A-module homomorphisms:

0 0

S2
d2 d1 π
··· Aβ1 Aβ0 M 0

S1

0 0

134
 CHAPTER 5. HOMOLOGICAL METHODS

The di ’s above are formed by taking the composition Aβi → Si → Aβi−1 , while
π is the canonical surjection. Hence we obtain a free resolution of M , that is
a long exact sequence of free modules ending at M :

d d d π
· · · −→ Aβ3 −→
3
Aβ2 −→
2
Aβ1 −→
1
Aβ0 −→ M −→ 0

A resolution of this form is called a minimal free resolution. Note that the
condition that A is local, and hence Noetherian, is critical for this construction.
By Corollary 5.52, we see that the projective dimension of M is given by

pdA (M ) = sup{n : TorA

i (M, k) 6= 0}.

Since the entries of the matrices defining the di ’s in a minimal free resolution
live in m, they become zero maps when tensored by k. Hence

Ker(di ⊗ 1)
TorA
i (M, k) = = Aβi ⊗A k ≃ k βi .
Im(di+1 ⊗ 1)

Thus we see that if the projective dimension of M is finite, then the degree of
the final nonzero term in the minimal free resolution is equal to pdA (M ). Hence
a minimal free resolution is truly a resolution of minimal length.

Proposition 5.56 Let (A, m, k) be a local ring, M a finitely generated A-

module, and F• be any free resolution of M where each Fi has finite rank and
such that:

(1) π : F0 ։ M and Ker(π) ⊆ mF0 .

(2) Im(di ) = Ker(di−1 ) ⊆ mFi−1 .

Then F• is a minimal free resolution of M .

Proof Consider the exact sequences:

0 → S 1 → F0 → M → 0
0 → Si+1 → Fi → Si → 0

Where Si is the ith syzygy as defined above. Now apply − ⊗A k to obtain

F0 ⊗A k ≃ M ⊗A k,
Fi ⊗A k ≃ Si ⊗A k.

Thus by Corollary 2.39, we see that the rank of F0 is the minimum number of
generators of M . Similarly, we see that Fi is a free module of rank equal to the
minimum number of generators of Si . ¥

135
5.5. REGULAR RINGS AND GLOBAL DIMENSION

5.5 Regular Rings and Global Dimension

5.5.1 Regular Local Rings
Definition If (A, m) is local, A is a regular local ring if

dim(A) = µ(m) = {the minimal number of generators of m}.

Example 5.57 Examples of regular local rings:

(1) Consider A = k[[X1 , . . . , Xn ]] where k is a field. Here dim(A) = n and

m = (X1 , . . . , Xn ). Thus A is a regular local ring.

(2) Any DVR is a regular local ring of dimension 1 since its maximal ideal is
principal. In particular, Z(p) and k[X](X) are regular local rings.

Example 5.58 A nonexample of a regular local ring:

k[[X, Y, U, V ]] k[X, Y, U, V ]m
A= or A=
(XY − U V ) (XY − U V )

In either case m = (X, Y, U, V ) but dim(A) = 3.

Exercise 5.59 Consider

k[X, Y, U, V ]m
A= .
(XY − U V )

Letting p = (X, U ), write down a free resolution of A/p over A. Can you get a
finite free resolution of A/p over A?

Theorem 5.60 Let (A, m, k) be local of dimension n, then the following are
equivalent:

(1) A is regular.

(2) rankk (m/m2 ) = n.

(3)
M∞
mi
k[X1 , . . . , Xn ] ≃ = Grm (A).
i=0
mi+1

Proof (1) ⇔ (2) This follows from the definition of a regular local ring and
Corollary 2.39.
(3) ⇒ (2) If
M∞
mi
k[X1 , . . . , Xn ] ≃ ,
i=0
mi+1

then m/m2 must correspond to degree 1 polynomials hence, m/m2 is the k vector
space generated by basis vectors X1 , . . . , Xn , showing that rankk (m/m2 ) = n.

136
 CHAPTER 5. HOMOLOGICAL METHODS

(1) ⇒ (3) Consider the homomorphism

ϕ : k[X1 , . . . , Xn ] → Grm (A), via

X X
ai1 ,...,in X1i1 · · · Xnin 7→ ai1 ,...,in xi11 · · · xinn ,

where (x1 , . . . , xn ) = m. Note that ϕ is onto. We must show that Ker(ϕ) is

zero.
By Theorem 3.31, the Dimension Theorem, we have that

n = dim(A) = deg(Pm (A, i)) = deg(ℓ(A/mi )).

Since
A ⊇ m ⊇ m2 ⊇ · · · ⊇ mi
we have that

ℓ(A/mi ) = ℓ(A/m) + ℓ(m/m2 ) + · · · + ℓ(mi−1 /mi ).

Thus

∆Pm (A, i) = ℓ(A/mi+1 ) − ℓ(A/mi )

= ℓ(mi /mi+1 )
= χ(Grm (A), i),

and so deg(χ(Grm (A), i)) = n − 1. Thus

0 = ∆n−1 χ(Grm (A), i) = ℓk (k).

Moreover, we have that k generates Grm (A) over Grm (A), and so by Theo-
rem 3.3, the Hilbert-Serre Theorem, we see that the map above is injective. ¥

Corollary 5.61 If A is a regular local ring, then A is an integral domain.

Proof To start, note that A is Hausdorff under the m-adic filtration and we
have that Grm (A) is a domain. It is left as an exercise to show that this implies
that A is a domain. ¥

Definition If (A, m) is a regular local ring, a system of parameters x1 , . . . , xd

is called a regular system of parameters if m = (x1 , . . . , xd ).

Proposition 5.62 Let (A, m) be a regular local ring of dimension n and

x1 , . . . , xj ∈ m.

Then x1 , . . . , xj form part of a regular system of parameters for A if and only

if for all i, 1 6 i 6 j, A/(x1 , . . . , xi ) is a regular local ring of dimension n − i.

137
5.5. REGULAR RINGS AND GLOBAL DIMENSION

Proof Throughout this proof set Ai = A/(x1 , . . . , xi ) and thus

mAi = m/(x1 , . . . , xi ).

(⇒) By the definition of Ai we have that dim(Ai ) > n − i. The maximal

ideal mAi of Ai is generated by n − i elements since you can extend x1 , . . . , xi
to a regular system of parameters. Thus dim(Ai ) 6 n − i and so we see that
dim(Ai ) = n − i.
(⇐) If dim(Ai ) = n − i and Ai is a regular local ring, then mAi can be
generated by n − i elements. Since A is a regular local ring, x1 , . . . , xi must
form part of a regular system of parameters. ¥

Corollary 5.63 If A a regular local ring of dimension n and x1 , . . . , xn is a

regular system of parameters, then (x1 , . . . , xi ) is a prime ideal of height i.

Proof Follows from Corollary 5.61 and Proposition 5.62. ¥

Definition Given a ring A and an A-module M , x1 , . . . , xn ∈ A is called an

M -sequence if the following hold:

(1) (x1 , . . . , xn )M 6= M .

(2) For each i > 0,

M xi M
−→
(x1 , . . . , xi−1 )M (x1 , . . . , xi−1 )M

is an injective map; that is, xi is a nonzerodivisor on M/(x1 , . . . , xi−1 )M

for 1 6 i 6 n.

Example 5.64 If A = k[X1 , . . . , Xn ], then X1 , . . . , Xn form an A-sequence.

Exercise 5.65 If A = k[X, Y, Z], x1 = X, x2 = Y (X − 1), and x3 = Z(X − 1),

then x1 , x2 , x3 form an A-sequence, but x3 , x2 , x1 does not.

The next lemma tells us the relationship between an M -sequence and a

system of parameters:

Lemma 5.66 If A is a local ring, every M -sequence is part of a system of

parameters for M .

Proof This follows by repeatedly applying the exercise after Corollary 3.40.
¥

Corollary 5.67 If A is local and M is finitely generated A-module with x1 , . . . , xi

an M -sequence, then i 6 dim(M ).

It should be pointed out here that unless the ring is nice, the inequality
above is strict.

138
 CHAPTER 5. HOMOLOGICAL METHODS

Proposition 5.68 (A, m) is a regular local ring if and only if m is generated

by an A-sequence.

Proof (⇒) If A is a regular local ring with m = (x1 , . . . , xn ). Then by

Corollary 5.61, we have that A/(x1 , . . . , xi ) is an integral domain, and so the
map defined by xi+1 is an injection. Hence (x1 , . . . , xn ) form an A-sequence.
(⇐) Suppose that m is generated by an A-sequence x1 , . . . , xd . Then we see
that dim(A) 6 d, but from Corollary 5.67 we have that dim(A) > d. Hence we
see that A is a regular local ring. ¥

Lemma 5.69 If (A, m) is local ring and a ∈ m − m2 , then the exact sequence
ι
0 −→ aA/am −→ m/am −→ m/aA −→ 0

splits.

Proof We must define η : m/am → aA/am such that η ◦ ι = 1aA/am . Take η

to be the composite
ϕ ≃
m/am −→ m/m2 −→ A/m −→ aA/am

where ϕ(a) = 1 and ϕ sends the rest of a k-basis for m/m2 to 0. ¥

Lemma 5.70 Suppose that (A, m) is local and gd(A) is finite. Let a ∈ m − m2
such that a is a nonzerodivisor on A. Then gd(A/aA) is finite.

Proof First note that by Corollary 5.54 we have that gd(A) = pdA (A/m).
Thus if gd(A) = 0, we have that A is a field implying that m is zero, and hence
we are done.
Assuming that gd(A) > 0, write

0 → m → A → A/m → 0

and so we see that if pdA (A/m) is finite, then pdA (m) is finite. If a is a nonze-
rodivisor on A, then it is also a nonzerodivisor on m. Hence pdA/aA (m/(am))
is finite by Proposition 5.44. By Lemma 5.69, we have that
M
m/am ≃ A/m m/aA.

By considering the Tor characterization of projective dimension we see that

pdA/aA (A/m) is finite. Hence gd(A/aA) is finite. ¥

Lemma 5.71 If A is local and M is a finitely generated A-module with finite

projective dimension, and a ∈ m is a nonzerodivisor on M , then

pdA (M/aM ) = pdA (M ) + 1.

139
5.5. REGULAR RINGS AND GLOBAL DIMENSION

Proof Write
a
0 −→ M −→ M −→ M/aM −→ 0
apply the functor − ⊗A k and consider the long exact sequence for Tor. Since
a
a ∈ m, the map TorA A
i (M, k) −→ Tori (M, k) is in fact the zero map for all i and
so we have the exact sequence

0 → TorA A A
i (M, k) → Tori (M/aM, k) → Tori−1 (M, k) → 0

for i > 1. Since TorA A

i−1 (M, k) = 0 implies that Tori (M, k) = 0, we must
conclude that

TorA
i (M/aM, k) = 0 whenever TorA
i−1 (M, k) = 0.

The theorem now follows from Theorem 5.51. ¥

Exercise 5.72 If A is a ring and a is an ideal such that

a ⊆ I0 ∪ I1 ∪ · · · ∪ In

where I0 is a prime ideal. Show that there exists a proper subset S of {0, . . . , n}
such that [
a⊆ Ii .
i∈S

Lemma 5.73 Let A be a local ring and suppose every element m − m2 is a

zerodivisor. Then every finitely generated module of finite projective dimension
is free.

Proof To start note that

[
m − m2 ⊆ p
p∈Ass(A)

and so  
[
m⊆ p ∪ m2
p∈Ass(A)

Repeatedly applying the previous exercise we see that either m ∈ Ass(A) or that
m ⊆ m2 . If the latter is the case, then

m = m2

which implies that m = 0 by Corollary 2.35, Nakayama’s Lemma. Hence A

must be a field, and every finitely generated module over a field is free as it is
a vector space.
So now suppose that m ∈ Ass(A). Thus A contains an element x annihilated
by m and we may write, setting k = A/m, the exact sequence:
x
0 −→ k −→ A −→ A/xA −→ 0

140
 CHAPTER 5. HOMOLOGICAL METHODS

Suppose that there exists some finitely generated A-module M of positive pro-
jective dimension. Applying the functor − ⊗A M to the above exact sequence
and considering the long exact sequence for Tor we see that for all i > 1
TorA A
i (k, M ) ≃ Tori+1 (A/xA, M ).

However now we see by Proposition 5.51 that if pdA (M ) = n, then

TorA
n (k, M ) 6= 0 and TorA
n+1 (A/xA, M ) = 0

which is impossible. Thus M must be free. ¥

Theorem 5.74 (A, m, k) is a regular local ring if and only if gd(A) is finite.
In this case
gd(A) = dim(A).
Proof (⇒) Suppose that A is a regular local ring with m = (x1 , . . . , xn ).
Setting xi = x1 , . . . , xi , we see that xn is a regular system of parameters for A
and thus xn forms an A-sequence by Proposition 5.68. Consider the following
short exact sequences:
x1
0 A A A/(x1 ) 0
x2
0 A/(x1 ) A/(x1 ) A/(x2 ) 0
.. .. ..
. . .
xn
0 A/(xn−1 ) A/(xn−1 ) A/(xn ) 0
By Lemma 5.71 we have that
pdA (A) = 0,
pdA (A/(x1 )) = 1,
pdA (A/(x2 )) = 2,
..
.
pdA (A/(xn )) = n.
But A/(xn ) = A/m, and hence by Corollary 5.54, gd(A) = n.
(⇐) Now suppose that gd(A) is finite. Let n = rankk (m/m2 ), and proceed
by induction on n. If n = 0, then m = m2 and by Corollary 2.35, Nakayama’s
Lemma, we see m = 0 and so A must be a field, and hence regular.
Suppose that the statement is true up to n. We must check the case when
rankk (m/m2 ) = n+1. We claim that some element of m−m2 is a nonzerodivisor.
Suppose to the contrary that every element of m − m2 is a zerodivisor. Now
by Lemma 5.73 we see that every module of finite projective dimension is free.
But if gd(A) is finite, then pdA (k) is finite, and hence free, and so we must
conclude that A is a field, a contradiction. Thus there is a ∈ m − m2 which is a
nonzerodivisor on A.
Now by Lemma 5.70, gd(A/aA) is finite and if we set m = m/aA, then we
have n = rankk (m/m2 ). Thus rankk (m/m2 ) = n + 1, and so dim(A) = n + 1. ¥

141
5.5. REGULAR RINGS AND GLOBAL DIMENSION

Now we turn to the question which we started with in this section: If

A = C[[X1 , . . . , Xn ]]

and p is a prime ideal in A, is Ap a regular local ring? We answer this question

in the affirmative with the following corollary:
Corollary 5.75 If A is a regular local ring and p is a prime ideal of A, then
Ap is a regular local ring.

Proof Since A is a regular local ring, gd(A) is finite and hence pdA (A/p) is
finite. Consider a free resolution of A/p:

F• : 0 → Fi → · · · → F1 → F0 → A/p → 0

Apply the functor − ⊗A Ap to get:

(F• )p : 0 → (Fi )p → · · · → (F1 )p → (F0 )p → Ap /pAp → 0

Since Ap is a flat A-module, (F• )p is an exact complex. Thus pdAp (Ap /pAp ) is
finite, and so gd(Ap ) is finite, and hence Ap is a regular local ring. ¥

Theorem 5.76 (Auslander-Buchsbaum) If A is regular local, then A is a

UFD.

Proof We omit the proof of this result and instead refer the reader to [6] or
[12]. ¥

5.5.2 Regular Rings

Definition A Noetherian ring A is regular if gd(A) is finite.
Exercise 5.77 If A is Noetherian and M is a finitely generated A-module,
then show the following:
(1) pdA (M ) = sup pdAm (Mm ).
m∈MaxSpec(A)

(2) Suppose that pdAm (Mm ) is finite for all m ∈ MaxSpec(A). Then pdA (M )
is finite.
Proposition 5.78 If A is a Noetherian ring and M is a finitely generated
A-module, then the following are equivalent:
(1) pdA (M ) 6 n.
(2) TorA
i (M, N ) = 0 for all A-modules N and all i > n.

(3) TorA
n+1 (M, A/m) = 0 for all m ∈ MaxSpec(A).

Proof Follows from the above exercise and Proposition 5.51. ¥

142
 CHAPTER 5. HOMOLOGICAL METHODS

Proposition 5.79 If A is Noetherian, the following are equivalent:

(1) gd(A) 6 n.

(2) For all finitely generated A-modules M and N , TorA

n+1 (M, N ) = 0.

(3) For all finitely generated A-modules M and all maximal ideals m ⊆ A,
TorA
n+1 (M, A/m) = 0.

(4) TorA
n+1 (A/m, A/m) = 0 for all maximal ideals m ⊆ A.

Proof Follows from the above exercise and Proposition 5.53. ¥

Corollary 5.80 If A is a ring then A is regular if and only if pdA (A/m) < ∞
for all maximal ideals m ⊆ A.

Corollary 5.81 If A is a regular ring, then Ap is regular local for all p ∈

Spec(A).

Exercise 5.82 Consider

R[X, Y, Z]
A= .
(X 2 + Y 2 − Z 2 − 1)

Show that A is a regular ring but not a UFD. Conclude that the previous
theorem is false if the local condition is dropped.

Example 5.83 If A is regular, then A[x] is regular. Moreover

gd(A[x]) = gd(A) + 1.

Example 5.84 The following are examples of regular rings.

(1) If k is a field, then k[X1 , . . . , Xn ] is regular.

(2) Z[X1 , . . . , Xn ] is regular.

(3) If A is a Dedekind domain, then A[X1 , . . . , Xn ] is regular.

(4) If k is a field, then k[[X1 , . . . , Xn ]][Y1 , . . . , Ym ] is regular.

143
Appendix A

Diagram of Implications

DVR

PID UFD RLR

DD ND

RD LR

RR

NR

144
 APPENDIX A. DIAGRAM OF IMPLICATIONS

In the diagram on the preceding page, the abbreviations are as follows:

DVR Discrete Valuation Ring

PID Principal Ideal Domain

DD Dedekind Domain

UFD Unique Factorization Domain

RLR Regular Local Ring

ND Normal Domain (Noetherian Integrally Closed Domain)

RD Regular Domain

RR Regular Ring

LR Local Ring

NR Noetherian ring

145
Appendix B

Diagram and Examples of

Domains

All rings are assumed to be domains in the diagram below:

Integrally Closed Noetherian

6
Regular
2
4
10 Dedekind Domain

12
UFD PID
11 13
Local
DVR

14
3 8
Field
15
16 5

Regular Local

146
 APPENDIX B. DIAGRAM AND EXAMPLES OF DOMAINS

Examples:

(1) Not Noetherian, not integrally closed:

• k[X 2 , X 3 , Y1 , Y2 , Y3 , . . .]

(2) Integrally closed, not a UFD, not Noetherian:

• Z[2X, 2X 2 , 2X 3 , . . .]
• k[U, V, Y, Z, X1 , X2 , X3 , . . .]/(U V − Y Z)

(3) A UFD but not Noetherian:

• k[X1 , X2 , X3 , . . .]

(4) Noetherian, not local, not integrally closed:

• k[X 2 , X 3 ]
√
• Z[ 5]

(5) Local, not integrally closed:

• k[X 2 , X 3 ](X 2 ,X 3 )
• k[[X 2 , X 3 ]]

(6) Noetherian, integrally closed, not regular, not a UFD, not local:

• k[W, X, Y, Z]/(W X − Y Z)
• R[W, X, Y, Z]/(X 2 + Y 2 + Z 2 − W 2 )

(7) Local, integrally closed, not regular, not a UFD:

• (k[W, X, Y, Z]/(W X − Y Z))(W,X,Y,Z)

• k[[W, X, Y, Z]]/(W X − Y Z)

(8) Noetherian, a UFD, not regular, not local:

• k[V, W, X, Y, Z]/(V 2 + W 2 + X 2 + Y 2 + Z 2 )
• R[W, X, Y, Z]/(W 2 + X 2 + Y 2 + Z 2 )

(9) Noetherian, a UFD, local, not regular:

• (k[V, W, X, Y, Z]/(V 2 + W 2 + X 2 + Y 2 + Z 2 ))(V,W,X,Y,Z)

(10) Regular, not a Dedekind domain, not a UFD:

• k[X, Y, Z]/(X 2 + Y 2 − 1)

(11) A UFD, regular, not a Dedekind domain, not local:

• k[X, Y ], Z[X]

147
(12) A Dedekind domain, not a UFD and hence not local:
√
• Z[ −5]

(13) A PID but not local:

• Z
• k[X]
• Z[i]

(14) A DVR, not a field:

• Z(p)
• k[X](X)

(15) A Regular local ring, not a Dedekind domain:

• k[[X, Y ]]
• k[X, Y ](X,Y )

(16) A field:

• k
• Q
• R
• C
• Z/pZ

Above k represents any field and m represents any maximal ideal in the given
ring. For further information on examples 2,6,8,9, see [8]. These four examples
are all nontrivial.

148
Appendix C

Table of Invariances

The table below summarizes those basic properties of commutative rings that
are and are not preserved under the basic operations on rings. For example, the
symbol ✓ that appears in the upper left box means that if A is Noetherian, then
A[X] is Noetherian as well. An ✖ in the table merely means “not in general.”

A A[X] A[[X]] A/a A/p U −1 A Ap b

A e
A
Noetherian ✓ ✓ ✓ ✓ ✓ ✓ ✓ ✖
local ✖ ✓ ✓ ✓ ✖ ✓ ✓ ✖
local and complete ✖ ✓ ✓ ✓ ✖ ✖ ✓ ✓
normal domain ✓ ✓ ✖ ✖ ✓ ✓ ✖ ✓
Dedekind domain ✖ ✖ ✖ ✓ ✓ ✓ ✖ ✓
UFD ✓ ✖ ✖ ✖ ✓ ✓ ✖ ✓
PID ✖ ✖ ✖ ✓ ✓ ✓ ✖ ✓
regular local ✖ ✓ ✖ ✖ ✖ ✓ ✓ ✓
DVR or a field ✖ ✖ ✖ ✓ ✓ ✓ ✓ ✓

In the above table, Ab denotes the completion of A with respect to some ideal I
which is taken to be the unique maximal ideal if A is local. Local, as throughout
these notes, is taken to mean Noetherian and local. For the third and fourth
columns a denotes an arbitrary ideal of A while p denotes a prime ideal. Lastly,
Ae denotes the integral closure of A, which is assumed to be a domain in this
column.

149
Bibliography

[1] M. F. Atiyah and I. G. Macdonald, Introduction to commutative algebra,

Addison-Wesley Publishing Co., Reading, Mass.-London-Don Mills, Ont.,
1969.

[2] Nicolas Bourbaki, Éléments de mathématique, Masson, Paris, 1980,

Algèbre. Chapitre 10. Algèbre homologique. [Algebra. Chapter 10. Homo-
logical algebra].

[3] , Commutative algebra. Chapters 1–7, Elements of Mathemat-

ics (Berlin), Springer-Verlag, Berlin, 1998, Translated from the French,
Reprint of the 1989 English translation.

[4] Winfried Bruns and Jürgen Herzog, Cohen-macaulay rings, Cambridge

Studies In Advances Mathematics, vol. 39, Cambridge University Press,
Cambridge, 1993.

[5] Henri Cartan and Samuel Eilenberg, Homological algebra, Princeton Land-
marks in Mathematics, Princeton University Press, Princeton, NJ, 1999,
With an appendix by David A. Buchsbaum, Reprint of the 1956 original.

[6] David Eisenbud, Commutative algebra, Graduate Texts in Mathematics,

vol. 150, Springer-Verlag, New York, 1995, With a view toward algebraic
geometry.

[7] Thomas W. Hungerford, Algebra, Holt, Rinehart and Winston, Inc., New
York, 1974.

[8] Harry C. Hutchins, Examples of commutative rings, Polygonal Publ. House,

Washington, N. J., 1981.

[9] Ernst Kunz, Introduction to commutative algebra and algebraic geometry,

Birkhäuser Boston Inc., Boston, MA, 1985, Translated from the German
by Michael Ackerman, With a preface by David Mumford.

[10] Serge Lang, Algebra, third ed., Graduate Texts in Mathematics, vol. 211,
Springer-Verlag, New York, 2002.

151
[11] Hideyuki Matsumura, Commutative algebra, second ed., Mathematics Lec-
ture Note Series, vol. 56, Benjamin/Cummings Publishing Co., Inc., Read-
ing, Mass., 1980.

[12] , Commutative ring theory, second ed., Cambridge Studies in Ad-

vanced Mathematics, vol. 8, Cambridge University Press, Cambridge, 1989,
Translated from the Japanese by M. Reid.
[13] David Mumford, The red book of varieties and schemes, expanded ed.,
Lecture Notes in Mathematics, vol. 1358, Springer-Verlag, Berlin, 1999,
Includes the Michigan lectures (1974) on curves and their Jacobians, With
contributions by Enrico Arbarello.
[14] Masayoshi Nagata, Local rings, Interscience Tracts in Pure and Applied
Mathematics, No. 13, Interscience Publishers a division of John Wiley &
Sons New York-London, 1962.

[15] Jean-Pierre Serre, Local algebra, Springer Monographs in Mathematics,

Springer-Verlag, Berlin, 2000, Translated from the French by CheeWhye
Chin and revised by the author.

[16] Charles A. Weibel, An introduction to homological algebra, third ed., Cam-

bridge Studies in Advanced Mathematics, vol. 38, Cambridge University
Press, Cambridge, 1997.

[17] Oscar Zariski and Pierre Samuel, Commutative algebra, Volume I, The
University Series in Higher Mathematics, D. Van Nostrand Company, Inc.,
Princeton, New Jersey, 1958, With the cooperation of I. S. Cohen.

[18] , Commutative algebra. Vol. II, The University Series in Higher

Mathematics, D. Van Nostrand Co., Inc., Princeton, N. J.-Toronto-London-
New York, 1960.
Index
b 36
A, (Z :X Y ), 3, 52
adic colon ideal/submodule, 3
filtration, 36 comaximal, 2
affine algebra, 97 compact, 62
affine domain, 97 complete, 36
AnnA (m), 10 completion, 36
annihilator, 4, 10 complex, 103
AnnA (M ), 4 X• , 103
A+ , 51 complex map, 104
Artin-Rees Lemma, 37, 40, 59 component, 62
Artinian composition series, 5
module, 4 cone, 108
ring, 4, 72 connecting homomorphism, 107
AssA (M ), 18, 23 coprimary, 9, 23
associated primes, 18, 23 Correspondence Theorem, 2
Auslander-Buchsbaum
Theorem, 142 d(−), 58
A[[X]], 28 Dedekind domain, 87, 145
degree, 49
Baer’s Criterion, 115 ∆, 49
Betti number, 134 D(f ), 65
Bk (x), 49 diagram chasing, 107
dim(A), 67
catenary, 100 dimension, 67, 141
Cauchy sequence, 30 topological space, 64
chain complex, 103 Dimension Theorem, 68, 137
χ(M, n), 51, 55 dim(M ), 67
closed set, 62 direct limit, 25
closure, 62 direct system, 25
cochain complex, 104 directed set, 25
cocomplex, 104 discrete valuation, 90
X • , 104 discrete valuation ring, 85, 136, 145
coheight, 71 divisible module, 116
cohomology, 104, 124, 125 DVR, 85, 136, 145
H n (X • ), 104
coht(I), 71 ea (M ), 60
colimit, 25 embedded primes, 16

153
en (M ), 55 global dimension, 132, 133, 141
essential primes, 18 Going-Down Property, 83
essentially polynomial, 49 Going-Down Theorem
exact sequence for flat maps, 85
and HomA (−, E), 114 for integral extensions, 83
and HomA (P, −), 108 Going-Up Property, 79, 80
of complexes, 105 Going-Up Theorem
ExtnA (M, N ), 124, 125 for faithfully flat maps, 80
extension for integral extensions, 80
inseparable, 82 I-good, 38
integral, 75 Gr(A), 33
normal, 82 Gr(M ), 33
purely inseparable, 82 GrI (A), 36
separable, 82 GrI (M ), 36
graded
faithfully flat, 42, 44 module, 33
Going-Up Property, 80 associated to a filtration, 33
field of fractions, 8 ring, 33
filtered associated to a filtration, 33
map, 30
strict, 30 Hausdorff, 30, 34, 63
module, 29 height, 71
associated graded ring, 33 Hilbert function, 51, 55
ring, 29 Hilbert multiplicity, 55
associated graded ring, 33 Hilbert polynomial, 55
filtration, 20, 29 Hilbert’s Basis Theorem, 5
adic, 36 Hilbert-Samuel multiplicity, 60
image, 29 Hilbert-Samuel polynomial, 55, 58
induced, 29 Hilbert-Serre Theorem, 51, 137
finite length, 22, 68 homogeneous element, 33
flat, 6, 41, 121 degree, 33
faithfully, 42, 44 homological dimension, 129
Going-Up Property, 85 homology, 104, 120
is locally free, 132 Hn (X• ), 104
formal power series ring, 28 homotopic, 104, 110
Frac(A), 8 f• ∼ g• , 104, 110
fractional ideal, 87 Horseshoe Lemma, 112, 119
invertible, 88 ht(I), 71
free, 44 ht(p), 71
free module
flatness of, 6 id(M ), 131
free resolution, 109, 135 ideal quotient, 3
minimal, 134, 135 image filtration, 29
induced filtration, 29
gd(A), 132 injective, 127
generic point, 65 injective dimension, 131, 132, 134
injective module, 114, 115 locally nilpotent, 23
injective resolution, 118, 125, 131 long exact sequence
inseparable extension, 82 of Ext, 126, 127
integral closure, 76 of homologies, 105
integral extension, 75 of Tor, 121, 123
and dimension, 80 Lying-Over Property, 79, 80
Going-Down Theorem, 83 Lying-Over Theorem
Going-Up Theorem, 80 for integral extensions, 80
Lying-Over Theorem, 80
integrally closed, 76 M -sequence, 138
intersection of ideals, 1 relationship to a system of param-
inverse limit, 27 eters, 138
inverse system, 27 c, 36
M
invertible fractional ideal, 88 Ma , 8
irreducible mapping cone, 108
submodule, 11 maximal set, 62
irreducible set, 62 MaxSpec(A), 64
irrelevant ideal, 33 metric, 30
A+ , 33 minimal free resolution, 134, 135
I(Z), 65 modular law, 1
rank, 91
J(A), 38, 39, 42, 47 module finite, 76
Jacobson radical, 38, 39, 42, 47 Mp , 8
Jordan-Hölder chain, 5, 54 µ, 39, 103, 136

κ(p), 8 Nakayama’s Lemma, 38, 39, 55, 68, 86,

Krull dimension, 67 91, 133, 140, 141
Krull’s Hauptidealsatz, 72 nilpotent, 2
Krull’s Ideal Theorem, 71 nilradical, 2, 21
Krull’s Intersection Theorem, 38 Nine Lemma, 47
Krull’s Principal Ideal Theorem, 71 Noether’s Normalization Lemma, 95
Krull-Akizuki Theorem, 93 Noetherian
module, 4
ℓ, 5, 51, 55 ring, 4, 66, 67, 145
length, 5, 51, 55 topological space, 62, 66, 67
additive property, 5 nonzerodivisor, 19
finite, 22 normal domain, 85, 87, 145
lies over, 78 normal extension, 82
limit, 27 null sequence, 31
local homomorphism, 48
local property, 8 open set, 62
local ring, 8, 68, 145
localization, 7 Pa (M, n), 57, 58
and primary decomposition, 14 Pa ((Mi ), n), 56
universal property, 7 p-coprimary, 9, 23
locally free, 132 pd(M ), 129
PID, 116, 145 strict filtered map, 30
polynomial-like, 49 sum of ideals, 1
power series ring, 28 SuppA (M ), 21, 67
p-primary, 10, 23 support, 21, 22, 67
primary, 10, 23, 70 tensor product, 56
Primary Decomposition Theorem, 13, symbolic power, 17
66 system of parameters, 68
prime avoidance, 3, 70, 82 regular, 137
Prime Filtration Theorem, 20 relationship to an M -sequence, 138
prime spectrum, 64 syzygy, 134
principal ideal domain, 145
product of ideals, 1 topological
projective, 127 group, 28
is locally flat, 132 module, 29, 30
projective dimension, 129, 130, 133, 135 ring, 28
projective module, 108 topological space, 62
projective resolution, 109, 120, 124, 129 topology, 61
pseudometric, 30 TorAn (M, N ), 120, 132, 133
purely inseparable extension, 82 trL/K , 94
trace, 94
quasicompact, 63 transcendence degree, 97
quasilocal, 8 tr deg, 97

radical, 2, 11, 65 UFD, 145

√ unique factorization domain, 145
0, 2
rank, 91 universal property
regular domain, 145 of localization, 7
regular local ring, 103, 136, 141, 145
V (a), 64
regular ring, 142, 145
regular system of parameters, 137 Zariski topology, 64
residue field, 8 zerodivisor, 19
resolution, 109, 118, 135 Zorn’s Lemma, 3, 62, 83, 115
free, 109, 135
minimal, 134, 135
injective, 118, 125, 131
minimal
free, 134, 135
projective, 109, 120, 124, 129
catenary, 100

Samuel’s Theorem, 57
separable extension, 82
sequence, 138
shift of complexes, 108
s(M ), 68
Spec(A), 64