A Primer of Commutative Algebra

James S. Milne
August 15, 2014, v4.01
Abstract
These notes collect the basic results in commutative algebra used in the rest of my
notes and books. Although most of the material is standard, the notes include a few
results, for example, the affine version of Zariskis main theorem, that are difficult to
find in books.

Contents
1 Rings and algebras . . . . . . . . . . . .
2 Ideals . . . . . . . . . . . . . . . . .
3 Noetherian rings . . . . . . . . . . . . .
4 Unique factorization . . . . . . . . . . . .
5 Rings of fractions . . . . . . . . . . . . .
6 Integral dependence . . . . . . . . . . . .
7 The going-up and going-down theorems . . . .
8 Noethers normalization theorem . . . . . . .
9 Direct and inverse limits . . . . . . . . . .
10 Tensor Products . . . . . . . . . . . . .
11 Flatness . . . . . . . . . . . . . . . .
12 Finitely generated projective modules . . . . .
13 Zariskis lemma and the Hilbert Nullstellensatz . .
14 The spectrum of a ring . . . . . . . . . . .
15 Jacobson rings and max spectra . . . . . . . .
16 Artinian rings . . . . . . . . . . . . . .
17 Quasi-finite algebras and Zariskis main theorem. .
18 Dimension theory for finitely generated k-algebras .
19 Primary decompositions . . . . . . . . . .
20 Dedekind domains . . . . . . . . . . . .
21 Dimension theory for noetherian rings . . . . .
22 Regular local rings . . . . . . . . . . . .
23 Flatness and fibres . . . . . . . . . . . .
24 Completions. . . . . . . . . . . . . . .
A Solutions to the exercises. . . . . . . . . . .
References . . . . . . . . . . . . . . . . .
Index . . . . . . . . . . . . . . . . . . .

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

3
3
9
14
18
24
30
33
35
38
43
49
57
60
68
73
74
81
86
91
96
100
102
105
107
108
109

c
2009,
2010, 2011, 2012, 2013, 2014 J.S. Milne. Single paper copies for noncommercial personal use
may be made without explicit permission from the copyright holder. Available at www.jmilne.org/math/.

CONTENTS

Notations and conventions

Our convention is that rings have identity elements,1 and homomorphisms of rings respect
the identity elements. A unit of a ring is an element admitting an inverse. The units of a ring
A form a group, which we denote by2 A . Throughout ring means commutative ring.
Following Bourbaki, we let N D f0; 1; 2; : : :g. Throughout, k is a field and k al is an algebraic
closure of k.
X
X
X
X

Y
def
DY
Y
'Y

X is a subset of Y (not necessarily proper).

X is defined to be Y , or equals Y by definition.
X is isomorphic to Y .
X and Y are canonically isomorphic
(or there is a given or unique isomorphism).

Prerequisites
A knowledge of the algebra usually taught in advanced undergraduate or first-year graduate
courses.

References
A reference to monnnn is to question nnnn on mathoverflow.net.

Historical Notes
Sometime Ill add these. For the moment, I refer the reader to Bourbaki AC, Historical Note;
Matsumura 1986, Introduction; Nagata 1962, Appendix A2.

Acknowledgements
I thank the following for providing corrections and comments for earlier versions of these
notes: Florian Herzig, Chun Yin Hui, Keenan Kidwell, Leon Lampret, Andrew McLennan,
Shu Otsuka, Dmitri Panov, Bhupendra Nath Tiwari, Wei Xu.

1 An element e of a ring A is an identity element if ea D a D ae for all elements a of the ring. It is usually
denoted 1A or just 1. Some authors call this a unit element, but then an element can be a unit without being a
unit element. Worse, a unit need not be the unit.
2 This notation differs from that of Bourbaki, who writes A for the multiplicative monoid A X f0g and A
for the group of units. We shall rarely need the former, and  is overused.

RINGS AND ALGEBRAS

Rings and algebras

A ring is an integral domain if it is not the zero ring and if ab D 0 in the ring implies that
a D 0 or b D 0.
Let A be a ring. A subring of A is a subset that contains 1A and is closed under addition,
multiplication, and the formation of negatives. An A-algebra is a ring B together with a
homomorphism iB W A ! B. A homomorphism of A-algebras B ! C is a homomorphism
of rings 'W B ! C such that '.iB .a// D iC .a/ for all a 2 A.
Elements x1 ; : : : ; xn of an A-algebra B are said to generate it if every element of B
can be expressed as a polynomial in the xi with coefficients in iB .A/. This means that the
homomorphism of A-algebras AX1 ; : : : ; Xn ! B acting as iB on A and sending Xi to xi
is surjective.
When A  B and x1 ; : : : ; xn 2 B, we let Ax1 ; : : : ; xn denote the A-subalgebra of B
generated by the xi .
A ring homomorphism A ! B is of finite type, and B is a finitely generated A-algebra,
if B is generated by a finite set of elements as an A-algebra. This means that B is a quotient
of a polynomial ring AX1 ; : : : ; Xn . An A-algebra B is finitely presented if it is the quotient
of a polynomial ring AX1 ; : : : ; Xn by a finitely generated ideal.
A ring homomorphism A ! B is finite, and B is a finite3 A-algebra, if B is finitely
generated as an A-module. If A ! B and B ! C are finite ring homomorphisms, then so
also is their composite A ! C .
Let k be a field, and let A be a k-algebra. If 1A 0, then the map k ! A is injective,
and we can identify k with its image, i.e., we can regard k as a subring of A. If 1A D 0, then
the ring A is the zero ring f0g.
Let AX be the ring of polynomials in the symbol X with coefficients in A. If A is an
integral domain, then deg.fg/ D deg.f / C deg.g/, and so AX is also an integral domain;
moreover, AX  D A .
Let A be both an integral domain and an algebra over a field k. If A is finite over k, then
it is a field. To see this, let a be a nonzero element of A. Because A is an integral domain,
the k-linear map x 7! axW A ! A is injective, and hence is surjective if A is finite, which
shows that a has an inverse. More generally, if every element a of A is algebraic over k,
then ka is finite over k, and hence contains an inverse of a; again A is a field.
An A-module M is faithful if aM D 0, a 2 A, implies a D 0.

Exercises
E XERCISE 1.1. Let k be an infinite field, and let f be a nonzero polynomial in kX1 ; : : : ; Xn .
Show that there exist a1 ; : : : ; an 2 k such that f .a1 ; : : : ; an / 0.

Ideals

Let A be a ring. An ideal a in A is a subset such that

a is a subgroup of A regarded as a group under addition;
a 2 a, r 2 A H) ra 2 a:
3 This is Bourbakis terminology (AC V 1, 1). Finite homomorphisms of rings correspond to finite maps of
varieties and schemes. Some other authors say module-finite.

IDEALS

The ideal generated by a subset S of A is the intersection of all ideals a containing S it

is
Peasy to verify that this is in fact an ideal, and that it consists of all finite sums of the form
ri si with ri 2 A, si 2 S. The ideal generated by the empty set is the zero ideal f0g. When
S D fa; b; : : :g, we write .a; b; : : :/ for the ideal it generates.
An ideal is principal if it is generated by a single element. Such an ideal .a/ is proper if
and only if a is not a unit. Thus a ring A is a field if and only if 1A 0 and the only proper
ideal in A is .0/.
Let a and b be ideals in A. The set fa C b j a 2 a; b 2 bg is an ideal, denoted a C b. The
ideal generated
by fab j a 2 a; b 2 bg is denoted by ab. Clearly ab consists of all finite
P
sums ai bi with ai 2 a and bi 2 b, and if a D .a1 ; : : : ; am / and b D .b1 ; : : : ; bn /, then
ab D .a1 b1 ; : : : ; ai bj ; : : : ; am bn /. Note that ab  aA D a and ab  Ab D b, and so
ab  a \ b:

(1)

The kernel of a homomorphism A ! B is an ideal in A. Conversely, for every ideal a

in a ring A, the set of cosets of a in A (regarded as an additive group) forms a ring A=a,
and a 7! a C a is a homomorphism 'W A ! A=a whose kernel is a. There is a one-to-one
correspondence
fideals of A containing ag

b7!'.b/
'

1 .b/

! fideals of A=ag:

(2)

[b

For an ideal b of A, ' 1 '.b/ D a C b.

The ideals of A  B are all of the form a  b with a and b ideals in A and B. To see
this, note that if c is an ideal in A  B and .a; b/ 2 c, then .a; 0/ D .1; 0/.a; b/ 2 c and
.0; b/ D .0; 1/.a; b/ 2 c. Therefore, c D a  b with
a D fa j .a; 0/ 2 cg;

b D fb j .0; b/ 2 cg:

An ideal p in A is prime if p A and ab 2 p ) a 2 p or b 2 p. Thus p is prime if and

only if the quotient ring A=p is nonzero and has the property that
ab D 0 H) a D 0 or b D 0;
i.e., A=p is an integral domain. In particular, the zero ideal is prime if and only if the ring is
an integral domain. When p is prime, we write .p/ for the field of fractions of A=p.
L EMMA 2.1. Let p be a prime ideal in A.
(a) If p contains a product of elements of A, then it contains one of the elements.
(b) If p contains a finite product of ideals, then it contains one of the ideals.
P ROOF. (a) In the integral domain A=p, a finite product of elements is 0 only if one of its
terms is zero.
(b) Suppose that p  a1


an . If p contains none of the ai , then there exist ai 2 p X ai ,
i D 1; : : : ; an . But then a1


an 2 p, which is a contradiction.
An ideal m in A is maximal if it is a maximal element of the set of proper ideals in A.
Therefore an ideal m is maximal if and only if the quotient ring A=m is nonzero and has no
nonzero proper ideals (by (2)), and so is a field. Note that
m maximal H) m prime.

IDEALS

A multiplicative subset of a ring A is a subset S with the property:

1 2 S;

a; b 2 S H) ab 2 S:

For example, the following are multiplicative subsets:

the multiplicative subset f1; f; : : : ; f r ; : : :g generated by an element f of A;
the complement of a prime ideal (or of a union of prime ideals);
def
1 C a D f1 C a j a 2 ag for any ideal a of A.
P ROPOSITION 2.2. Let S be a subset of a ring A, and let a be an ideal disjoint from S.
The set of ideals in A containing a and disjoint from S contains maximal elements (i.e., an
element not properly contained in any other ideal in the set). If S is multiplicative, then
every such maximal element is prime.
P ROOF. The set of ideals containing a and disjoint from S is nonempty (it contains a).
If A is noetherian (see 3 below), automatically contains maximal elements. Otherwise,
S
we apply Zorns lemma. Let b1  b2 


be a chain of ideals in , and let b D bi .
Then b 2 , because otherwise some element of S lies in b, and hence in some bi , which
contradicts the definition of . Therefore b is an upper bound for the chain. As every chain
in has an upper bound, Zorns lemma implies that has a maximal element.
Now assume that S is a multiplicative subset of A, and let c be maximal in . Let
bb 0 2 c. If b is not in c, then c C .b/ properly contains c, and so it is not in . Therefore
there S contains an element in c C .b/, say,
f D c C ab;

c 2 c;

a 2 A:

Similarly, if b 0 is not in c, then S contains an element

f 0 D c 0 C a0 b;

c 0 2 c;

a0 2 A:

Now
ff 0 D cc 0 C abc 0 C a0 b 0 c C aa0 bb 0 2 c;
which contradicts
ff 0 2 S:
Therefore, at least one of b or b 0 is in c, which is therefore prime.
C OROLLARY 2.3. Every proper ideal in a ring is contained in a maximal ideal.
P ROOF. Apply the proposition with S D f1g.
An element f of a ring is nilpotent if f r D 0 for some r  1. A ring is reduced if it has
no nonzero nilpotents. The radical rad.a/ of an ideal a in a ring A is
ff 2 A j f r 2 a some r  1g:
An ideal a is said to be radical if it equals its radical. Thus an ideal a is radical if and only
if A=a is reduced. Since integral domains are reduced, prime ideals (a fortiori, maximal
ideals) are radical. The radical of .0/ consists of the nilpotent elements of A it is called
the nilradical of A.
If b $ b0 under the one-to-one correspondence (2) between ideals of A and ideals of
A=a, then A=b ' .A=a/=b0 , and so b is prime (resp. maximal, radical) if and only if b0 is
prime (resp. maximal, radical).

IDEALS

P ROPOSITION 2.4. Let a be an ideal in a ring A.

(a) The radical of a is an ideal.
(b) rad.rad.a// D rad.a/.
P ROOF. (a) If f 2 rad.a/, then clearly af 2 rad.a/ for all a 2 A. Suppose that a; b 2 rad.a/,
with say ar 2 a and b s 2 a. When we expand .a C b/rCs using the binomial theorem, we
find that every term has a factor ar or b s , and so lies in a.
(b) If ar 2 rad.a/, then ars D .ar /s 2 a for some s > 0, and so a 2 rad.a/.
Note that (b) of the proposition shows that rad.a/ is radical. In fact, it is the smallest
radical ideal containing a.
If a and b are radical, then a \ b is radical, but a C b need not be: consider, for example,
a D .X 2 Y / and b D .X 2 C Y /; they are both prime ideals in kX; Y (by 4.11 below), but
a C b D .X 2 ; Y /, which contains X 2 but not X.
P ROPOSITION 2.5. The radical of an ideal is equal to the intersection of the prime ideals
containing it. In particular, the nilradical of a ring A is equal to the intersection of the prime
ideals of A.
P ROOF. If a D A, then the set of prime ideals containing it is empty, and soT
the intersection
is A. Thus we may suppose that a is a proper ideal of A. Then rad.a/  pa p because
prime ideals are radical and rad.a/ is the smallest radical ideal containing a.
For the reverse inclusion, let f rad.a/. According to Proposition 2.2, there exists a
prime
Tideal containing a and disjoint from the multiplicative subset f1; f; : : :g. Therefore
f pa p.
D EFINITION 2.6. The Jacobson radical J of a ring is the intersection of the maximal ideals
of the ring:
\
J.A/ D fm j m maximal in Ag:
A ring A is local if it has exactly one maximal ideal m. For such a ring, the Jacobson
radical is m.
P ROPOSITION 2.7. An element c of A is in the Jacobson radical of A if and only if 1
is a unit for all a 2 A.

ac

P ROOF. We prove the contrapositive: there exists a maximal ideal m such that c m if and
only if there exists an a 2 A such that 1 ac is not a unit.
): Suppose that c is not in the maximal ideal m. Then m C .c/ D A, and so 1 D m C ac
for some m 2 m and a 2 A. Now 1 ac 2 m, and so it is not a unit.
(: If 1 ac is not a unit, then it lies in some maximal ideal m of A (by 2.3). Now
c m, because otherwise 1 D .1 ac/ C ac 2 m.
P ROPOSITION 2.8 ( PRIME AVOIDANCE ). Let p1 ; : : : ; pr , r  1, be ideals in A with p2 ; : : : ; pr
prime. If an ideal a is not contained in any of the pi , then it is not contained in their union.
P ROOF. S
When r D 1, there is nothing to prove, and so we may assume that r > 1.S
Suppose
that a  1j r pj and that no pj can be deleted from the union. In particular, a 6 j i pj ,

IDEALS

and so there exists an ai 2 aX

Consider

j i pj .

Then ai 2 pi , because otherwise ai 2 aX

a D a1


ar

1 C ar

1j r pj .

2 a:

I claim that a belongs to no pi , which is a contradiction. Because none of the elements

a1 ; : : : ; ar 1 lies in pr and pr is prime, their product does not lie in pr (2.1); but ar 2 pr ,
and so a pr . Next consider an ideal pi with i  r 1. In this case a1


ar 1 2 pi because
the product involves ai , but ar pi , and so again a pi .
A SIDE 2.9. 4 In general, the condition in (2.8) that the ideals p2 ; : : : ; pr be prime is necessary: the
ideal .x; y/ in F2 x; y is the union of three proper nonprime ideals. However, when A contains an
infinite field k, the condition can be dropped. In the above proof, let
S V be the (finite-dimensional)
k-vector space generated by the ai , and let Vi D pi \ V . Then V  1ir Vi , but the Vi are proper
subspaces of V , and so this is impossible as k is infinite.

Extension and contraction of ideals

Let 'W A ! B be a homomorphism of rings.
N OTATION 2.10. For an ideal b of B, ' 1 .b/ is an ideal in A, called the contraction of b
to A, which is often denoted bc . For an ideal a of A, the ideal in B generated by '.a/ is
called the extension of a to B, and is often denoted ae . When ' is surjective, '.a/ is already
an ideal, and when A is a subring of B, bc D b \ A.
2.11. There are the following equalities (a; a0 ideals in A; b; b0 ideals in B):
.a C a0 /e D ae C a0e ;

.aa0 /e D ae a0e ;

.b \ b0 /c D bc \ b0c ;

rad.b/c D rad.bc /:

2.12. Let a be an ideal of A and b an ideal of B. Obviously (i) a  aec and (ii) bce  b. On
applying e to (i), we find that ae  aece , and (ii) with b replaced by ae shows that aece  ae ;
therefore ae D aece . Similarly, bcec D bc : It follows that extension and contraction define
inverse bijections between the set of contracted ideals in A and the set of extended ideals in
B:
a7!ae
fbc  A j b an ideal in Bg c ! fae  B j a an ideal in Ag
b

Note that, for every ideal b in B, the map

radical) if b is prime (resp. radical).

A=bc

[b

! B=b is injective, and so bc is prime (resp.

The Chinese remainder theorem

Recall the classical form5 of the theorem: let d1 ; :::; dn be integers, relatively prime in pairs;
then for any integers x1 ; :::; xn , the congruences
x  xi mod di
have a simultaneous solution x 2 Z; moreover, if x is oneQ
solution, then the other solutions
are the integers of the form x C md with m 2 Z and d D di :
4 Asides

can be ignored.
credited to Qin Jiushao (1208-1261), one of the greatest mathematicians of his era (Notices AMS,
May 2013, p.596).
5 Often

IDEALS

We want to translate this into a statement about ideals. Integers m and n are relatively
prime if and only if .m; n/ D Z, i.e., if and only if .m/ C .n/ D Z. This suggests defining
ideals a and b in a ring A to be relatively
prime (or coprime) if a C b D A.
T
If m1 ; :::; m
are
integers,
then
.m
/
i D .m/
Tk
Q
Q where m is the least common multiple of
the mi . Thus
Q .mi /  . mi /, which
T equals
Q .mi /. If the mi are relatively prime in pairs,
then m D mi , and so we have .mi / D .mi /. Note that in general,
a1
a2


an  a1 \ a2 \ ::: \ an ;
but the two ideals need not be equal.
These remarks suggest the following statement.
T HEOREM 2.13 (C HINESE R EMAINDER T HEOREM ). Let a1 ; : : : ; an be ideals in a ring A.
If ai is relatively prime to aj whenever i j , then the map
a 7! .: : : ; a C ai ; : : :/W A ! A=a1 


 A=an
Q
Q
T
is surjective with kernel ai (so ai D ai ).

(3)

P ROOF. Suppose first that n D 2. As a1 C a2 D A, there exist ai 2 ai such that a1 C a2 D 1.

Then a1 x2 C a2 x1 maps to .x1 mod a1 ; x2 mod a2 /, which shows that (3) is surjective.
Moreover, for c 2 a1 \ a2 , we have
c D a1 c C a2 c 2 a1
a2
which proves that a1 \ a2 D a1 a2 . Thus
A=a1 a2 ' A=a1  A=a2 :
We now use induction to prove the theorem for n > 2. For i  2, there exist elements
ai 2 a1 and bi 2 ai such that
ai C bi D 1:
Q
The product i 2 .ai C bi / lies in a1 C a2


an and equals 1, and so
a1 C a2


an D A:
Therefore,
A=a1


an D A=a1
.a2


an /
' A=a1  A=a2


an
' A=a1  A=a2 


 A=an

by the n D 2 case
by induction.

Exercises
E XERCISE 2.14. Let M be an A-module. Define the product of two elements of A M by
.a; m/.a0 ; m0 / D .aa0 ; am0 C a0 m/:
Show that this makes A M into a ring. Show that the ideals of A M contained in M are
exactly the A-submodules of A M .6
6 This

construction shows that modules over A and their submodules can be realized as ideals in the ring
A M , which is useful for deducing results about modules from results about ideals. Nagata calls this the
principle of idealization (Nagata 1962, p.2).

NOETHERIAN RINGS

Noetherian rings

P ROPOSITION 3.1. The following three conditions on a ring A are equivalent:

(a) every ideal in A is finitely generated;
(b) every ascending chain of ideals a1  a2 


eventually becomes constant, i.e.,
am D amC1 D


for some m.
(c) every nonempty set of ideals in A has a maximal element.
S
P ROOF. (a) ) (b): If a1  a2 


is an ascending chain, then a D ai is an ideal, and
hence has a finite set fa1 ; : : : ; an g of generators. For some m, all the ai belong am , and then
am D amC1 D


D a:
(b) ) (c): Let be a nonempty set of ideals in A. If has no maximal element, then
the axiom of dependent choice7 shows that there exists a strictly ascending sequence of
ideals in , which contradicts (b).
(c) ) (a): Let a be an ideal in A, and let be the set of finitely generated ideals
contained in a. Then is nonempty because it contains the zero ideal, and so it contains
a maximal element c D .a1 ; : : : ; ar /. If c a, then there exists an element a 2 a X c, and
.a1 ; : : : ; ar ; a/ will be a finitely generated ideal in a properly containing c. This contradicts
the definition of c, and so c D a.
A ring A is noetherian if it satisfies the equivalent conditions of the proposition. For
example, fields and principal ideal domains are noetherian. On applying (c) to the set of all
proper ideals containing a fixed proper ideal, we see that every proper ideal in a noetherian
ring is contained in a maximal ideal. We saw in (3.6) that this is, in fact, true for every ring,
but the proof for non-noetherian rings requires Zorns lemma.
A quotient A=a of a noetherian ring A is noetherian, because the ideals in A=a are all of
the form b=a with b an ideal in A, and every set of generators for b generates b=a.
P ROPOSITION 3.2. Let A be a ring. The following conditions on an A-module M are
equivalent:
(a) every submodule of M is finitely generated (in particular, M is finitely generated);
(b) every ascending chain of submodules M1  M2 


eventually becomes constant.
(c) every nonempty set of submodules of M has a maximal element.
P ROOF. Essentially the same as that of (3.1).
An A-module M is noetherian if it satisfies the equivalent conditions of the proposition.
Let A A denote A regarded as a left A-module. Then the submodules of A A are exactly the
ideals in A, and so A A is noetherian (as an A-module) if and only if A is noetherian (as a
ring).
P ROPOSITION 3.3. Let A be a ring, and let

0 ! M 0 ! M ! M 00 ! 0
be an exact sequence of A-modules.
7 This

says: Let R be a binary relation on a nonempty set X, and suppose that, for each a in X, there exists
a b such that aRb; then there exists a sequence .an /n2N of elements of X such that an RanC1 for all n. It is
strictly stronger than the axiom of countable choice but weaker than the axiom of choice. See the Wikipedia
(axiom of dependent choice).

NOETHERIAN RINGS

10

(a) If N  P are submodules of M such that .M 0 / \ N D .M 0 / \ P and .N / D

.P /, then N D P .
(b) If M 0 and M 00 are finitely generated, so also is M .
(c) M is noetherian if and only if M 0 and M 00 are both noetherian.
P ROOF. (a) Let p 2 P . The second condition implies that there exists an n 2 N such that
.n/ D .p/. Then .p n/ D 0, and so p n lies in M 0 , and hence in M 0 \ P D
M 0 \ N  N . Thus p D .p n/ C n 2 N .
(b) Let S 0 be a finite set of generators for M , and let S 00 be a finite subset of M such
that S 00 generates M 00 . The submodule N of M generated by S 0 [ S 00 is such that
M 0 \ N D M 0 and N D M 00 . Therefore (a) shows that N D M .
(c) ): An ascending chain of submodules of M 0 or of M 00 gives rise to an ascending
chain in M , and therefore becomes constant.
(: Consider an ascending chain of submodules of M . As M 00 is Noetherian, the image
of the chain in M 00 becomes constant, and as M 0 is Noetherian, the intersection of the chain
with M 0 becomes constant. Now the (a) shows that the chain itself becomes constant.
For example, a direct sum
M D M1 M2
of A-modules is noetherian if and only if M1 and M2 are both noetherian.
P ROPOSITION 3.4. Every finitely generated module over a noetherian ring is noetherian.
P ROOF. Let M be a module over a noetherian ring A. If M is generated by a single element,
then M  A=a for some ideal a in A, and the statement is obvious. We argue by induction on
the minimum number n of generators of M . Clearly M contains a submodule N generated
by n 1 elements such that the quotient M=N is generated by a single element, and so the
statement follows from (3.3c).
Hence, every submodule of a finitely generated module over a noetherian ring is finitely
generated. This statement is false for nonnoetherian rings, as any non finitely generated ideal
in the ring demonstrates.
P ROPOSITION 3.5. Every finitely generated module M over a noetherian ring A contains a
finite chain of submodules M  Mr 


 M1  0 such that each quotient Mi =Mi 1 is
isomorphic to A=pi for some prime ideal pi .
P ROOF. The annihilator of an element x of M is
def

ann.x/ D fa 2 A j ax D 0g:
It is an ideal in A, which is proper if x 0.
Let a D ann.x/ be maximal among the annihilators of nonzero elements of M . I claim
that a is prime. Let ab 2 a, so that abx D 0. Then a  .a/ C a  ann.bx/. If b a, then
bx 0, and so a D ann.bx/ by maximality, which implies that a 2 a.
We now prove the proposition. Note that, for every x 2 M , the submodule Ax of
M is isomorphic to A=ann.x/. If M is nonzero, then there exists a nonzero x such that
ann.x/ is maximal among the annihilators of nonzero elements of M , and so M contains
a submodule M1 D Ax isomorphic to A=p1 with p1 prime. Similarly, M=M1 contains
a submodule M2 =M1 isomorphic A=p2 for some prime ideal p2 , and so on. The chain
0  M1  M2 


terminates because M is noetherian (by 3.4).

NOETHERIAN RINGS

11

A SIDE 3.6. The proofs of (2.2) and (3.5) are two of many in commutative algebra in which an ideal,
maximal with respect to some property, is shown to be prime. For a general examination of this
phenomenon, see Lam and Reyes, J. Algebra 319 (2008), no. 7, 30063027.

T HEOREM 3.7 (H ILBERT BASIS T HEOREM ). Every finitely generated algebra over a noetherian ring is noetherian.
P ROOF. Let A be a noetherian ring, and let B be a finitely generated A-algebra. We
argue by induction on the minimum number of generators for B. As Ax1 ; : : : ; xn D
Ax1 ; : : : ; xn 1 xn , it suffices to prove the theorem for n D 1. But then B is a quotient of
AX , and so it suffices to prove that AX is noetherian.
Recall that for a polynomial
f .X / D c0 X r C c1 X r

C


C cr ;

ci 2 A;

c0 0;

c0 is called the leading coefficient of f .

Let a be an ideal in AX , and let a.i / be the set of elements of A that occur as the leading
coefficient of a polynomial in a of degree i (we also include 0). Then a.i / is obviously an
ideal in A, and a.i 1/  a.i / because, if cX i 1 C


2 a, then X.cX i 1 C


/ 2 a.
Let b be an ideal of AX contained in a. Then b.i /  a.i /, and if equality holds for all
i , then b D a. To see this, let f be a polynomial in a of degree i . Because b.i / D a.i /, there
exists a g 2 b such that deg.f g/ < deg.f /. In other words, f D g C f1 with g 2 b and
deg.f1 / < deg.f /. Now f1 D g1 C f2 with g1 2 b and deg.f2 )< deg.f1 /. Continuing in
this fashion, we find that f D g C g1 C g2 C


2 b.
As A is noetherian, the sequence of ideals
a.1/  a.2/ 


 a.i / 



becomes constant, say, a.d / D a.d C 1/ D : : : (and a.d / contains the leading coefficient of
every polynomial in a). For each i  d , there exists a finite generating set fci1 ; ci 2 ; : : :g for
a.i/, and for each .i; j /, there exists a polynomial fij 2 a of degree i with leading coefficient
cij . The ideal b of AX generated by the fij is contained in a and has the property that
b.i/ D a.i / for all i. Therefore b D a, and a is finitely generated.
C OROLLARY 3.8. When A is noetherian, every finitely generated A-algebra is finitely
presented.
P ROOF. Every finitely generated A-algebra B is of the form AX1 ; : : : ; Xn =a for some n
and ideal a in AX1 ; : : : ; Xn . Because AX1 ; : : : ; Xn is noetherian, the ideal a is finitely
generated, and so B is finitely presented.
In particular, the polynomial ring kX1 ; : : : ; Xn over a field k is noetherian. This is the
original theorem of Hilbert.
NAKAYAMA S L EMMA 3.9. Let A be a ring, let a be an ideal in A, and let M be an Amodule. Assume that a is contained in all maximal ideals of A and that M is finitely
generated.
(a) If M D aM , then M D 0:
(b) If N is a submodule of M such that M D N C aM , then M D N .

NOETHERIAN RINGS

12

P ROOF. (a) Suppose that M 0. Choose a minimal set of generators fe1 ; : : : ; en g for M ,
n  1, and write
e1 D a1 e1 C


C an en , ai 2 a:
Then
.1

a1 /e1 D a2 e2 C


C an en

and, as 1 a1 lies in no maximal ideal, it is a unit. Therefore e2 ; : : : ; en generate M , which

contradicts the minimality of the original set.
(b) The hypothesis implies that M=N D a.M=N /, and so M=N D 0.
Recall (2.6) that the Jacobson radical J of A is the intersection of the maximal ideals of
A, and so the condition on a is that a  J. In particular, the lemma holds with a D J; for
example, when A is a local ring, it holds with a the maximal ideal in A.
def

C OROLLARY 3.10. Let A be a local ring with maximal ideal m and residue field k D A=m,
and let M be a finitely generated module over A. The action of A on M=mM factors through
k, and elements a1 ; : : : ; an of M generate it as an A-module if and only if the elements
a1 C mM; : : : ; an C mM
generate M=mM as k-vector space.
P ROOF. If a1 ; : : : ; an generate M , then it is obvious that their images generate the vector
space M=mM . Conversely, suppose that a1 C mM; : : : ; an C mM span M=mM , and let N
be the submodule of M generated by a1 ; : : : ; an . The composite N ! M ! M=mM is
surjective, and so M D N C mM . Now Nakayamas lemma shows that M D N .
C OROLLARY 3.11. Let A be a noetherian local ring with maximal ideal m. Elements
a1 ; : : : ; an of m generate m as an ideal if and only if a1 C m2 ; : : : ; an C m2 generate m=m2 as
a vector space over A=m. In particular, the minimum number of generators for the maximal
ideal is equal to the dimension of the vector space m=m2 .
P ROOF. Because A is noetherian, m is finitely generated, and we can apply the preceding
corollary with M D m.
E XAMPLE 3.12. Nakayamas lemma may fail if M is not finitely generated. For example,
let Z.p/ D f m
n j p does not divide ng and consider the Z.p/ -module Q. Then Z.p/ is a local
ring with maximal ideal .p/ (see 5 below) and Q D pQ but Q 0.
D EFINITION 3.13. Let A be a noetherian ring.
(a) The height ht.p/ of a prime ideal p in A is the greatest length d of a chain of distinct
prime ideals
p D pd  pd 1 


 p0 :
(4)
(b) The (Krull) dimension of A is supfht.p/ j p  A;

p primeg.

Thus, the Krull dimension of a ring A is the supremum of the lengths of chains of prime
ideals in A (the length of a chain is the number of gaps, so the length of (4) is d ). It is
sometimes convenient to define the Krull dimension of the zero ring to be 1.

NOETHERIAN RINGS

13

Let A be an integral domain. Then

dim.A/ D 0 .0/ is maximal A is a field.
The height of a nonzero prime ideal in a principal ideal domain is 1, and so such a ring has
Krull dimension 1 (unless it is a field).
We shall see in 21 that the height of every prime ideal in a noetherian ring is finite.
However, the Krull dimension of the ring may be infinite, because it may contain a sequence
of prime ideals whose heights tend to infinity (Krull 1938).8
L EMMA 3.14. In a noetherian ring, every set of generators for an ideal contains a finite
generating set.
P ROOF. Let S be a set of generators for a, and let a0 be maximal among the ideals generated
by finite subsets of S. Then a0 contains every element of S (otherwise it wouldnt be
maximal), and so equals a.
T HEOREM 3.15 (K RULL I NTERSECTION T HEOREM ). Let
T a ben an ideal in a noetherian
ring A. If a is contained in all maximal ideals of A, then n1 a D f0g:
P ROOF. We shall show that, for every ideal a in a noetherian ring,
\
\
an D a

an :
n1

(5)

n1

When
a is contained in all maximal ideals of A, Nakayamas lemma then shows that
T
n D 0.
a
n1
Let a1 ; : : : ; ar generate a. Then an consists of finite sums
X
ci1


ir a1i1


arir ; ci1


ir 2 A:
i1 C


Cir Dn

In other words, an consists of the elements of A of the form g.a1 ; : : : ; ar / for some homogeneous polynomial g 2 AX1 ; : : : ; Xr of degree n.
Let Sm denote the
m such
T set ofn homogeneous polynomials f .X1 ; : : : ; Xr / of degree S
that f .a1 ; : : : ; ar / 2 n1 a , and let c be the ideal in AX1 ; : : : ; Xr generated by m Sm .
Because AX1 ; : : : ; Xr is noetherian,
c is finitely generated, and so c is generated by a finite
S
set ff1 ; : : : ; fT
s g of elements of
m Sm (3.14). Let di D deg fi , and let d D max di .
Let b 2 n1 an ; then b 2 ad C1 , and so b D f .a1 ; : : : ; ar / for some homogeneous
polynomial f of degree d C 1. By definition, f 2 Sd C1  c, and so there exist gi 2
AX1 ; : : : ; Xr such that
f D g1 f1 C


C gs fs

in AX1 ; : : : ; Xr :

As f and the fi are homogeneous, we can omit from each gi all terms not of degree
deg f deg fi , since these terms cancel out. In other words, we can choose the gi to be
homogeneous of degree deg f deg fi D d C 1 di > 0. In particular, the constant term of
gi is zero, and so gi .a1 ; : : : ; ar / 2 a. Now
X
\
b D f .a1 ; : : : ; ar / D
gi .a1 ; : : : ; ar /
fi .a1 ; : : : ; ar / 2 a

an ;
i

which completes the proof of (5).

8 In

Nagata 1962, p.203, there is the following example. Let N D I0 t I1 t : : : be a partition of N into finite
sets with strictly increasing cardinality. Let A D kX0 ; X1 ; : : : be the polynomial ring in a set of symbols
indexedS
by N, and let pi be the prime ideal in A generated by the Xj with j in Ii . Let S be the multiplicative
set A X pi . Then S 1 A is noetherian and regular, and the prime ideal S 1 pi has height jIi j.

UNIQUE FACTORIZATION

14

The equality (5) can also be proved using primary decompositions see (19.14).
P ROPOSITION 3.16. In a noetherian ring, every ideal contains a power of its radical; in
particular, some power of the nilradical of the ring is zero.
P ROOF. Let a1 ; : : : ; an generate rad.a/. For each i , some power of ai , say airi , lies in a.
Then every term of the expansion of
.c1 a1 C


C cn an /r1 C


Crn ;

ci 2 A;

has a factor of the form airi for some i , and so lies in a. Thus rad.a/r1 C


Crn  a.
A SIDE 3.17. In a noetherian ring, every ideal is finitely generated, but there is little that one can say
in general about the number of generators required. For example, in kX every ideal is generated by
a single element, but in kX; Y the ideal .X; Y /n requires at least n C 1 generators.
A SIDE 3.18. The following example shows that the Krull intersection theorem fails for nonnoetherian rings. Let A be the ring of germs9 of C 1 functions at 0 on
T the real line. Then A is a local ring
with maximal ideal m equal to the set of germs zero at 0, and n1 mn consists of the germs whose
derivatives at zero are all zero. In particular, it contains the nonzero function e

1=x 2

Exercises
E XERCISE 3.19. Consider the subalgebra
A D k C kX; Y X D kX; X Y; X Y 2 ; : : :
of kX; Y . Show that A is not noetherian (hence subrings of noetherian rings need not be
noetherian, and subalgebras of finitely generated algebras need not be finitely generated).

Unique factorization

Let A be an integral domain. An element a of A is said to be irreducible if it is neither zero

nor a unit and admits only trivial factorizations, i.e.,
a D bc H) b or c is a unit.
The element a is said to be prime if it is neither zero nor a unit and .a/ is a prime ideal, i.e.,
ajbc H) ajb or ajc:
An integral domain A is called a unique factorization domain (or a factorial domain)
if every nonzero nonunit a in A can be written as a finite product of irreducible elements in
exactly one way up to units and the order of the factors. The uniqueness condition means
that if
Y
Y
aD
ai D
bj
i 2I

j 2J

with each ai and bj irreducible, then there exists a bijection i 7! j.i /W I ! J such that
bj.i / D ai  unit for each i . Every principal ideal domain is a unique factorization domain
(proved in most algebra courses).
9A

germ of a function at 0 is represented by a function f on an open neighbourhood U of 0; two pairs .f; U /

and .f 0 ; U 0 / represent the same germ if and only if f and f 0 agree on some neighbourhood of 0 in U \ U 0 .

UNIQUE FACTORIZATION

15

P ROPOSITION 4.1. Let A be an integral domain, and let a be an element of A that is neither
zero nor a unit. If a is prime, then a is irreducible, and the converse holds when A is a
unique factorization domain.
Thus, .a/ is a prime ideal if a is irreducible and A is a unique factorization domain.
P ROOF. Assume that a is prime. If a D bc, then a divides bc and so a divides b or c.
Suppose the first, and write b D aq. Now a D bc D aqc, which implies that qc D 1 because
A is an integral domain, and so c is a unit. We have shown that a is irreducible.
For the converse, assume that a is irreducible and that A is a unique factorization domain.
If ajbc, then bc D aq for some q 2 A. On writing each of b, c, and q as a product of
irreducible elements, and using the uniqueness of factorizations, we see that a differs from
one of the irreducible factors of b or c by a unit. Therefore a divides b or c.
C OROLLARY 4.2. Let A be an integral domain. If A is a unique factorization domain, then
every prime ideal of height 1 is principal.
P ROOF. Let p be a prime ideal of height 1. Then p contains a nonzero element, and hence
an irreducible element a. We have p  .a/  .0/. As .a/ is prime and p has height 1, we
must have p D .a/.
The converse is true for noetherian integral domains (21.4).
P ROPOSITION 4.3. Let A be an integral domain in which every nonzero nonunit element is
a finite product of irreducible elements. If every irreducible element of A is prime, then A is
a unique factorization domain.
P ROOF. We have to prove the uniqueness of factorizations. Suppose that
a1


am D b1


bn

(6)

with the ai and bi irreducible elements in A. As a1 is prime, it divides one of the bi , which
we may suppose to be b1 , say b1 D a1 u. As b1 is irreducible, u is a unit. On cancelling a1
from both sides of (6), we obtain the equality
a2


am D .ub2 /b3


bn :
Continuing in this fashion, we find that the two factorizations are the same up to units and
the order of the factors.
P ROPOSITION 4.4. Let A be an integral domain in which every ascending chain of principal
ideals becomes constant (e.g., a noetherian integral domain). Then every nonzero nonunit
element in A is a finite product of irreducible elements.
P ROOF. The hypothesis implies that every nonempty set of principal ideals has a maximal
element (cf. the proof of 3.1). Assume that A has nonfactorable elements, and let .a/ be
maximal among the ideals generated by such elements. Then a is not itself irreducible, and
so a D bc with neither b nor c units. Now .b/ and .c/ both properly contain .a/, and so b
and c are both factorable, which contradicts the nonfactorability of a.

UNIQUE FACTORIZATION

16

P ROPOSITION 4.5. Let A be a unique factorization domain with field of fractions F . If an

element f of AX factors into the product of two nonconstant polynomials in F X , then it
factors into the product of two nonconstant polynomials in AX .
In other words, if f is not the product of two nonconstant polynomials in AX , then it
is irreducible in F X .
P ROOF. Let f D gh in F X . For suitable c; d 2 A, the polynomials g1 D cg and h1 D dh
have coefficients in A, and so we have a factorization
cdf D g1 h1 in AX .
If an irreducible element p of A divides cd , then, looking modulo .p/, we see that
0 D g1
h1 in .A=.p// X .
According to Proposition 4.1, the ideal .p/ is prime, and so .A=.p// X is an integral
domain. Therefore, p divides all the coefficients of at least one of the polynomials g1 ; h1 ,
say g1 , so that g1 D pg2 for some g2 2 AX . Thus, we have a factorization
.cd=p/f D g2 h1 in AX .
Continuing in this fashion, we can remove all the irreducible factors of cd , and so obtain a
factorization of f in AX .
The proof shows that every factorization f D gh in F X of an element f of AX
arises from a factorization f D .cg/.c 1 h/ in AX with c 2 F .
Let A be a unique factorization domain. A nonzero polynomial
f D a0 C a1 X C


C am X m
in AX is said to be primitive if the coefficients ai have no common factor other than units.
Every polynomial f in F X can be written f D c.f /
f1 with c.f / 2 F and f1 primitive.
The element c.f /, which is well-defined up to multiplication by a unit, is called the content
of f . Note that f 2 AX if and only if c.f / 2 A.
P ROPOSITION 4.6. Let A be a unique factorization domain. The product of two primitive
polynomials in AX is primitive.
P ROOF. Let
f D a0 C a1 X C


C am X m
g D b0 C b1 X C


C bn X n ;
be primitive polynomials, and let p be a prime element of A. Let ai0 be the first coefficient
of f not P
divisible by p and bj0 the first coefficient of g not divisible by p. Then all the
terms in i Cj Di0 Cj0 ai bj are divisible by p, except ai0 bj0 , which is not divisible by p.
Therefore, p doesnt divide the .i0 C j0 /th-coefficient of fg. We have shown that no prime
element of A divides all the coefficients of fg, which must therefore be primitive.
Each of the last two propositions is referred to as Gausss lemma (Gauss proved them
with A D Z).

UNIQUE FACTORIZATION

17

P ROPOSITION 4.7. Let A be a unique factorization domain with field of fractions F , and
let f; g 2 F X . Then
c.fg/ D c.f /
c.g/.
Hence every factor in AX of a primitive polynomial is primitive.
P ROOF. Let f D c.f /f1 and g D c.g/g1 with f1 and g1 primitive. Then
fg D c.f /c.g/f1 g1
with f1 g1 primitive, and so c.fg/ D c.f /c.g/.
C OROLLARY 4.8. The irreducible elements in AX are the irreducible elements c of A
and the nonconstant primitive polynomials f such that f is irreducible in F X .
P ROOF. Obvious from (4.5) and (4.7).
T HEOREM 4.9. If A is a unique factorization domain, then so also is AX .
P ROOF. Let f 2 AX , and write f D c.f /f1 . Then c.f / is a product of irreducible
elements in A. If f1 is not irreducible, then it can be written as a product of two polynomials
of lower degree, which are necessarily primitive (4.7). Continuing in this fashion, we find
that f1 is a product of irreducible primitive polynomials, and hence that f is a product of
irreducible elements in AX .
It remains to show that each irreducible element of AX is prime (see 4.3). There are
two cases (4.8).
Let c be an irreducible element of A. If a divides the product gh of g; h 2 AX , then it
divides c.gh/ D c.g/c.h/. As a is prime, it divides c.g/ or c.h/, and hence also g or h.
Let f be a nonconstant primitive polynomial in AX such that f is irreducible in F X .
If f divides the product gh of g; h 2 AX , then it divides g or h in F X . Suppose the
first, and write f q D g with q 2 F X . Because f is primitive, c.q/ D c.f /c.q/, and
c.f /c.q/ D c.f q/ D c.g/ 2 A, and so q 2 AX . Therefore f divides g in AX .
Let k be a field. A monomial in X1 ; : : : ; Xn is an expression of the form
X1a1


Xnan ; aj 2 N:
P
The total degree of the monomial is
ai . The degree, deg.f /, of a nonzero polynomial f .X1 ; : : : ; Xn / is the largest total degree of a monomial occurring in f with nonzero
coefficient. Since
deg.fg/ D deg.f / C deg.g/;
kX1 ; : : : ; Xn is an integral domain and kX1 ; : : : ; Xn  D k  . Therefore, an element f of
kX1 ; : : : ; Xn is irreducible if it is nonconstant and f D gh H) g or h is constant.
T HEOREM 4.10. The ring kX1 ; : : : ; Xn is a unique factorization domain.
P ROOF. Note that
AX1 ; : : : ; Xn D AX1 ; : : : ; Xn

1 Xn :

(7)

This simply says that every polynomial f in n symbols X1 ; : : : ; Xn can be expressed uniquely
as a polynomial in Xn with coefficients in kX1 ; : : : ; Xn 1 ,
f .X1 ; : : : ; Xn / D a0 .X1 ; : : : ; Xn

r
1 /Xn

C


C ar .X1 ; : : : ; Xn

1 /:

The theorem is trivially true when n D 0, and (7) allows us to deduce it from (4.9) for all n.

RINGS OF FRACTIONS

18

C OROLLARY 4.11. A nonzero proper principal ideal .f / in kX1 ; : : : ; Xn is prime if and

only f is irreducible.
P ROOF. Special case of (4.1).

Rings of fractions

Recall that a multiplicative subset of a ring is a nonempty subset closed under the formation
of finite products. In particular, it contains 1 (the empty product).
Let S be a multiplicative subset of a ring A. Define an equivalence relation on A  S by
.a; s/  .b; t / u.at

bs/ D 0 for some u 2 S:

Write as for the equivalence class containing .a; s/, and define addition and multiplication of
equivalence classes according to the rules:
a
s

C bt D

atCbs
st ;

ab
s t

ab
st :

It is easily checked these do not depend on the choices of representatives for the equivalence
classes, and that we obtain in this way a ring
S

A D f as j a 2 A; s 2 S g
iS

and a ring homomorphism a 7! a1 W A ! S

1A

whose kernel is

fa 2 A j sa D 0 for some s 2 S g:
If S contains no zero-divisors, for example, if A is an integral domain and 0 S, then iS is
injective. At the opposite extreme, if 0 2 S, then S 1 A is the zero ring.
iS

1A

A homomorphism A ! B factors through A ! S

B consists of units. More formally:
P ROPOSITION 5.1. The pair .S

1 A; i

S/

if and only if the image of S in

has the following universal property:

A

every element of S maps to a unit in S 1 A, and

any other ring homomorphism W A ! B with this
property factors uniquely through iS

a
1

1A
9

B:

P ROOF. Let W A ! B be such a homomorphism, and let W S

such that iS D . Then
sa
1s

iS

1A ! B

be a homomorphism

H) . 1s /. as / D . a1 / H) .s/. as / D .a/

and so
. as / D .a/.s/

(8)

This shows that there can be at most one such that iS D . We define by the formula
(8). Then
a
s

b
t

H) u.at

bs/ D 0 some u 2 S

H) .a/.t /

.b/.s/ D 0

because .u/ 2 B  ;

which shows that is well-defined, and it is easy to check that it is a homomorphism.

RINGS OF FRACTIONS

19

As usual, this universal property determines the pair .S 1 A; iS / uniquely up to a unique

isomorphism.10
When A is an integral domain and S D A X f0g, the ring S 1 A is the field of fractions
F of A. In this case, for any other multiplicative subset T of A not containing 0, the ring
T 1 A can be identified with the subring of F consisting of the fractions at with a 2 A and
t 2T.
E XAMPLE 5.2. Let h 2 A. Then Sh D f1; h; h2 ; : : :g is a multiplicative subset of A, and we
let Ah D Sh 1 A. Thus every element of Ah can be written in the form a= hm , a 2 A, and
a
hm

b
hn

hN .ahn

bhm / D 0;

some N:

If h is nilpotent, then Ah D 0, and if A is an integral domain with field of fractions F and

h 0, then Ah is the subring of F of elements that can be written in the form a= hm , a 2 A,
m 2 N:
P ROPOSITION 5.3. For every ring A and h 2 A, the map
isomorphism
AX =.1 hX / ! Ah :

ai X i 7!

P ai

hi

defines an

P ROOF. If h D 0, both rings are zero, and so we may assume h 0. In the ring
def

Ax D AX =.1

hX /;

1 D hx, and so h is a unit. Let W A ! B be a homomorphism of rings such that .h/ is a

unit in B. The homomorphism
P
P
i
i
i ai X 7!
i .ai /.h/ W AX ! B
factors through Ax because 1 hX 7! 1 .h/.h/ 1 D 0, and this is the unique extension
of to Ax. Therefore Ax has the same universal property as Ah , and so the two are
(uniquely) isomorphic by an A-algebra isomorphism that makes h 1 correspond to x.
Let S be a multiplicative subset of a ring A, and let S 1 A be the corresponding ring of
fractions. For every ideal a in A, the ideal generated by the image of a in S 1 A is
S

a D f as j a 2 a;

s 2 Sg:

If a contains an element of S , then S 1 a contains 1, and so is the whole ring. Thus some of
the ideal structure of A is lost in the passage to S 1 A, but, as the next proposition shows,
some is retained.
10 Recall the proof: let .A ; i / and .A ; i / have the universal property in the proposition; because every
1 1
2 2
element of S maps to a unit in A2 , there exists a unique homomorphism W A1 ! A2 such that i1 D i2
(universal property of A1 ; i1 /; similarly, there exists a unique homomorphism 0 W A2 ! A1 such that 0 i2 D i1 ;
now
0 i1 D 0 i2 D i1 D idA1 i1 ;

and so 0 D idA1 (universal property of A1 ; i1 ); similarly, 0 D idA2 , and so and 0 are inverse
isomorphisms (and they are uniquely determined by the conditions i1 D i2 and 0 i2 D i1 ).

RINGS OF FRACTIONS

20

P ROPOSITION 5.4. Let S be a multiplicative subset of the ring A, and consider extension
a 7! ae D S 1 a and contraction a 7! ac D fa 2 A j a1 2 ag of ideals with respect to the
homomorphism iS W A ! S 1 A. Then
ace D a
ec

for all ideals of S

Da

if a is a prime ideal of A disjoint from S:

Moreover, the map p 7! pe is a bijection from the set of prime ideals of A disjoint from S
onto the set of all prime ideals of S 1 A; the inverse map is p 7! pc .
P ROOF. Let a be an ideal in S 1 A. Certainly ace  a. For the reverse inclusion, let b 2 a.
We can write b D as with a 2 A, s 2 S. Then a1 D s. as / 2 a, and so a 2 ac . Thus b D as 2 ace ,
and so a  ace .
Let p be a prime ideal of A disjoint from S . Clearly pec  p. For the reverse inclusion,
0
let a 2 pec so that a1 D as for some a0 2 p, s 2 S . Then t .as a0 / D 0 for some t 2 S, and
so ast 2 p. Because st p and p is prime, this implies that a 2 p, and so pec  p.
Let p be a prime ideal of A disjoint from S , and let Sx be the image of S in A=p.
Then .S 1 A/=pe ' Sx 1 .A=p/ because S 1 A=pe has the correct universal property, and
Sx 1 .A=p/ is an integral domain because A=p is an integral domain and Sx doesnt contain
0. Therefore pe is prime. From (2.12) we know that pc is prime if p is, and so p 7! pe and
p 7! pc are inverse bijections on the two sets.
C OROLLARY 5.5. If A is noetherian, then so also is S

1A

for any multiplicative set S:

P ROOF. As bc is finitely generated, so also is .bc /e D b.

Let spec.A/ denote the set of prime ideals in A. Then (5.4) says that
spec.S

A/ ' fp 2 spec.A/ j p \ S D ;g.

P ROPOSITION 5.6. Let 'W A ! B be a ring homomorphism. A prime ideal p of A is the

contraction of a prime ideal in B if and only if p D pec .
2.12

P ROOF. Suppose p D qc with q prime. Then pec D qcec D qc D p. Conversely, suppose

that p D pec , and let S D A X p. Let s 2 S ; if '.s/ 2 pe , then s 2 pec D p, contradicting the
definition of S . Therefore '.S / is disjoint from pe . It is a multiplicative subset of B, and
so there exists a prime ideal q in B containing pe and disjoint from '.S / (apply 2.2). Now
' 1 .q/ contains p and is disjoint from S , and so it equals p.
def

E XAMPLE 5.7. Let p be a prime ideal in A. Then Sp D A X p is a multiplicative subset of

A, and we let Ap D Sp 1 A. Thus each element of Ap can be written in the form ac , c p, and
a
c

b
d

s.ad

bc/ D 0, some s p:

A prime ideal of is disjoint from Sp if and only if it is contained p, and so

spec.Ap / ' fq 2 spec.A/ j q  pg.
Therefore, Ap is a local ring with maximal ideal m D pe D f as j a 2 p; s pg.

RINGS OF FRACTIONS

21

P ROPOSITION 5.8. Let m be a maximal ideal of a ring A, and let n D mAm be the maximal
ideal of Am : For all n, the map
a C mn 7! a C nn W A=mn ! Am =nn
is an isomorphism. Moreover, it induces isomorphisms
mr =mn ! nr =nn
for all pairs .r; n/ with r  n.
P ROOF. The second statement follows from the first, because of the exact commutative
diagram .r < n/:
0

mr =mn

A=mn

A=mr

'

nr =nn

'

Am =nn

Am =nr

0:

We consider extension and contraction with respect to a 7! a1 W A ! Am . Note that

nn D .mn /e , and so the kernel of A=mn ! Am =nn is .mn /ec =mn . Let a 2 .mn /ec . Then
a
b
n
n
1 D s with b 2 m and s 2 S , and so t sa 2 m for some t 2 S . Therefore t sa D 0 in
n
A=m . Every maximal ideal of A containing m contains rad.mn / D m, and so equals m.
Therefore the only maximal ideal in A=mn is m=mn . But t s is not in m=mn , and so it must
be a unit in A=mn . Therefore a D 0 in A=mn , which means that a 2 mn . We deduce that
A=mn ! Am =nn is injective.
It remains to prove that A ! Am =nn is surjective. Let as 2 Am , a 2 A, s 2 A X m. As
we just showed, the only maximal ideal of A containing mn is m, and so no maximal ideal
contains both s and mn . Therefore .s/ C mn D A, and so sb C q D 1 for some b 2 A and
q 2 mn . Hence
s.ba/ D a.1 q/:
(9)
On passing to Am and multiplying by s

1,

ba
1

As

aq
s

2 nn , this shows that

a
s

we find that

a
s

aq
s :

mod nn is in the image of A ! Am =nn .

P ROPOSITION 5.9. In a noetherian ring A, only 0 lies in all powers of all maximal ideals:
\
fmn j m maximal, n 2 Ng D f0g:
P ROOF. Let a be an element of a noetherian ring A. If a 0, then its annihilator fb j ba D 0g
is a proper ideal in A, and so it is contained in some maximal ideal m. Then a1 is nonzero in
Am , and so a1 .mAm /n for some n (by the Krull intersection theorem 3.15), which implies
that a mn (by 5.8).

RINGS OF FRACTIONS

22

Modules of fractions
Let S be a multiplicative subset of the ring A, and let M be an A-module. Define an
equivalence relation on M  S by
.m; s/  .n; t / u.t m

sn/ D 0 for some u 2 S:

Write m
s for the equivalence class containing .m; s/, and define addition and scalar multiplication by the rules:
m
s

C nt D

mtCns
st ;

am
s t

am
st ;

m; n 2 M;

s; t 2 S;

a 2 A:

It is easily checked these do not depend on the choices of representatives for the equivalence
classes, and that we obtain in this way an S 1 A-module
S
and a homomorphism m 7!

m
1 WM

M D fm
s j m 2 M; s 2 S g
iS

1M

!S

of A-modules whose kernel is

fa 2 M j sa D 0 for some s 2 S g:
A homomorphism M ! N of A-modules factors through M ! S
every element of S acts invertibly on N . More formally:
P ROPOSITION 5.10. The pair .S

1 M; i

S/

1M

if and only if

has the following universal property:

the elements of S act invertibly on S 1 M , and

every homomorphism M ! N from M to an Amodule N on which the elements of S act invertibly factors uniquely through iS

iS

1M

N:

P ROOF. Similar to that of Proposition 5.1.

In particular, for any homomorphism W M ! N of A-modules, there is a unique
homomorphism S 1 W S 1 M ! S 1 N such that S 1 iS D iS :
iS

1M

S
iS

N
In other words, S
such that

1 N:
1 A-modules

is the unique homomorphism of S

.S

In this way, M

1M

1 /. m /
s

.m/
s ;

1M

!S

1N

m 2 M;

becomes a functor from A-modules to S

1 A-modules.

P ROPOSITION 5.11. The functor M

S 1 M is exact. In other words, if the sequence of
A-modules

M 0 ! M ! M 00
is exact, then so also is the sequence of S
S

M0

!S

1 A-modules
1

!S

M 00 :

RINGS OF FRACTIONS

23

P ROOF. Because D 0, we have 0 D S 1 . / D S 1 S 1 . Therefore Im.S 1 / 

1 / where m 2 M and s 2 S. Then
Ker.S 1 /. For the reverse inclusion, let m
s 2 Ker.S
.m/
0
s D 0 and so, for some t 2 S, we have t ..m// D 0. Then .t m/ D 0, and so t m D .m /
0
0
for some m 2 M . Now
.m0 /
1
m
tm
/:
s D t s D t s 2 Im.S
E XAMPLE 5.12. Let M be an A-module. For h 2 A, let Mh D Sh 1 M where Sh D
f1; h; h2 ; : : :g. Then every element of Mh can be written in the form hmr , m 2 M , r 2 N, and
0
0
m
D hmr 0 if and only if hN .hr m hr m0 / D 0 for some N 2 N.
hr
P ROPOSITION 5.13. Let M be a finitely generated A-module. If S
exists an h 2 S such that Mh D 0.

1M

D 0, then there

P ROOF. To say that S 1 M D 0 means that, for each x 2 M , there exists an sx 2 S such
def
that sx x D 0. Let x1 ; : : : ; xn generate M . Then h D sx1


sxn lies in S and has the property
that hM D 0. Therefore Mh D 0.
P ROPOSITION 5.14. Let M be an A-module. The canonical map
Y
M ! fMm j m a maximal ideal in Ag
is injective.
P ROOF. Let m 2 M map to zero in all Mm . The annilator a D fa 2 A j am D 0g of m is
an ideal in A. Because m maps to zero Mm , there exists an s 2 A X m such that sm D 0.
Therefore a is not contained in m. Since this is true for all maximal ideals m, a D A (by 2.3),
and so it contains 1. Now m D 1m D 0.
C OROLLARY 5.15. An A-module M D 0 if Mm D 0 for all maximal ideals m.
P ROOF. Immediate consequence of the lemma.
P ROPOSITION 5.16. A sequence

M 0 ! M ! M 00

(10)

is exact if and only if

m

Mm0 ! Mm ! Mm00

(11)

is exact for all maximal ideals m.

P ROOF. The necessity is a special case of (5.11). For the sufficiency, let N D Ker./= Im./.
Because the functor M
Mm is exact,
Nm D Ker.m /= Im.m /:
If (11) is exact for all m, then Nm D 0 for all m, and so N D 0 (by 5.15). But this means
that (10) is exact.
C OROLLARY 5.17. A homomorphism M ! N of A-modules is injective (resp. surjective)
if and only if Mm ! Nm is injective (resp. surjective) for all maximal ideals m:

INTEGRAL DEPENDENCE

24

P ROOF. Apply the proposition to 0 ! M ! N (resp. M ! N ! 0).

P ROPOSITION 5.18. Let N be the nilradical of A. For any multiplicative subset S of A,
S 1 N is the nilradical of S 1 A.
P ROOF. Let a 2 A and s 2 S . If . as /n D 0, then t an D 0 for some t 2 S , and so
S 1 N. Conversely, if a 2 N, then clearly as is in the nilradical of S 1 A.

a
s

ta
ts

C OROLLARY 5.19. A ring A is reduced if and only if Am is reduced for all maximal ideals
m in A.
P ROOF. Combine (5.18) with (5.15).

Exercises
E XERCISE 5.20. (Bourbaki AC, II, 2, Exercise 1.) A multiplicative subset S of a ring A is
said to be saturated if
ab 2 S H) a and b 2 S:
(a) Show that the saturated multiplicative subsets of A are exactly the subsets S such that
A X S is a union of prime ideals.
(b) Let S be a multiplicative subset of A, and let Sz be the set of a 2 A such that ab 2 S
for some b 2 A. Show that Sz is a saturated multiplicative subset of A (hence it is the
smallest such subset containing S ), and that A X Sz is the union of the prime ideals
of A not meeting S. Show that for any A-module M , the canonical homomorphism
S 1 M ! Sz 1 M is bijective. In particular, S 1 A ' Sz 1 A.
E XERCISE 5.21. Let A ! B be a homomorphism of rings, and let p be a prime ideal of A.
Show that the prime ideals of B lying over p are in natural one-to-one correspondence with
the prime ideals of B A .p/.

Integral dependence

Let A be a subring of a ring B. An element of B is said to be integral over A if it is a root

of a monic11 polynomial with coefficients in A, i.e., if it satisfies an equation
n C a1 n

C


C an D 0;

ai 2 A:

More generally, an element of an A-algebra B is integral over A if it is integral over the

image of A in B. If every element of B is integral over A, then B is said to be integral over
A.
In the next proof, we shall need to apply a variant of Cramers rule. We define the
determinant of an m  m matrix C D .cij / with coefficients cij in a ring A by the usual
formula
X
det.C / D
sign. /c1 .1/


cm .m/ :
 2Sm
11 A

polynomial is monic if its leading coefficient is 1, i.e., f .X/ D X n C terms of degree less than n.

INTEGRAL DEPENDENCE

25

Clearly, det.C / is linear in each column, and det.C / D 0 if two columns are equal because
then each term occurs twice but with opposite signs. If x1 ; : : : ; xm is a solution to the system
of linear equations
m
X
cij xj D 0; i D 1; : : : ; m;
j D1

with coefficients in a ring A, then

det.C /
xj D 0;

j D 1; : : : ; m;

(12)

where C is the matrix of coefficients. To prove this, expand out the left hand side of
0
1
P
c11 : : : c1 j 1
c1 j C1 : : : c1m
i c1i xi
B
::
::
::
:: C D 0
det @ :::
:
:
:
: A
P
cm1 : : : cm j 1
i cmi xi cm j C1 : : : cmm
using the properties of determinants mentioned above.
P ROPOSITION 6.1. Let A be a subring of a ring B. An element of B is integral over A
if and only if there exists a faithful A-submodule of B that is finitely generated as an
A-module.
P ROOF. )W Suppose that
n C a1 n

C


C an D 0;

ai 2 A:

Then the A-submodule M of B generated by 1, , ..., n 1 has the property that M  M ,

and it is faithful because it contains 1.
(W Let M be an A-module in B with a finite set fe1 ; : : : ; en g of generators such that
M  M and M is faithful as an A-module. Then, for each i ,
P
ei D aij ej , some aij 2 A:
We can rewrite this system of equations as
.

a11 /e1

a21 e1 C .

a12 e2

a13 e3

D 0

a22 /e2

a23 e3

D 0



D 0:

Let C be the matrix of coefficients on the left-hand side. Then Cramers rule (12) tells
us that det.C /
ei D 0 for all i . As M is faithful and the ei generate M , this implies that
det.C / D 0. On expanding out det.C /, we obtain an equation
n C c1 n

C c2 n

C


C cn D 0;

ci 2 A:

P ROPOSITION 6.2. An A-algebra B is finite if it is generated as an A-algebra by a finite

number of elements, each of which is integral over A.
P ROOF. We may replace A with its image in B. Suppose that B D A1 ; : : : ; m and that
ini C ai1 ini

C


C ai ni D 0;

aij 2 A;

i D 1; : : : ; m.

Any monomial in the i divisible by some ini is equal (in B) to a linear combination of
monomials of lower degree. Therefore, B is generated as an A-module by the monomials
rm
1r1


m
, 1  ri < ni .

INTEGRAL DEPENDENCE

26

C OROLLARY 6.3. An A-algebra B is finite if and only if it is finitely generated and integral
over A.
P ROOF. (: Immediate consequence of (6.2).
): We may replace A with its image in B. Then B is a faithful A-module for all
2 B (because 1B 2 B), and so (6.1) shows that every element of B is integral over A. As
B is finitely generated as an A-module, it is certainly finitely generated as an A-algebra.
P ROPOSITION 6.4. Consider rings A  B  C . If B is integral over A and C is integral
over B, then C is integral over A.
P ROOF. Let 2 C . Then

n C b1 n

C


C bn D 0

for some bi 2 B. Now Ab1 ; : : : ; bn is finite over A (see 6.2), and Ab1 ; : : : ; bn is finite
over Ab1 ; : : : ; bn , and so it is finite over A. Therefore is integral over A by (6.1).
T HEOREM 6.5. Let A be a subring of a ring B. The elements of B integral over A form an
A-subalgebra of B.
P ROOF. Let and be two elements of B integral over A. As just noted, A; is finitely
generated as an A-module. It is stable under multiplication by and and it is faithful
as an A -module and as an A-module (because it contains 1A ). Therefore (6.1)
shows that and are integral over A.
D EFINITION 6.6. Let A be a subring of the ring B. The integral closure of A in B is the
subring of B consisting of the elements integral over A. When A is an integral domain, the
integral closure of A in its field of fractions is called the integral closure of A (tout court).
P ROPOSITION 6.7. Let A be an integral domain with field of fractions F , and let E be a
field containing F . If 2 E is algebraic over F , then there exists a nonzero d 2 A such that
d is integral over A.
P ROOF. By assumption, satisfies an equation
m C a1 m

C


C am D 0;

ai 2 F:

Let d be a common denominator for the ai , so that dai 2 A for all i , and multiply through
the equation by d m :
d m m C a1 d m m

C


C am d m D 0:

We can rewrite this as

.d/m C a1 d.d/m

C


C am d m D 0:

As a1 d; : : : ; am d m 2 A, this shows that d is integral over A.

C OROLLARY 6.8. Let A be an integral domain and let E be an algebraic extension of the
field of fractions of A. Then E is the field of fractions of the integral closure of A in E.

INTEGRAL DEPENDENCE

27

P ROOF. In fact, the proposition shows that every element of E is a quotient =d with
integral over A and d 2 A.
D EFINITION 6.9. An integral domain A is is said to be integrally closed or normal if it is
equal to its integral closure in its field of fractions F , i.e., if
2 F;

integral over A H) 2 A:

P ROPOSITION 6.10. Every unique factorization domain is integrally closed.

P ROOF. Let A be a unique factorization domain. An element of the field of fractions of
A not in A can be written a=b with a; b 2 A and b divisible by some prime element p not
dividing a. If a=b is integral over A, then it satisfies an equation
.a=b/n C a1 .a=b/n

C


C an D 0;

ai 2 A:

On multiplying through by b n , we obtain the equation

a n C a1 a n

b C


C an b n D 0:

The element p then divides every term on the left except an , and hence must divide an .
Since it doesnt divide a, this is a contradiction (as A is a unique factorization domain).
Let F  E be fields, and let 2 E be algebraic over F . The minimum polynomial of
over F is the monic polynomial in F X of smallest degree having as a root. Then f is
the (unique) monic generator of the kernel of the homomorphism X 7! W F X ! F ,
and so this map defines an isomorphism F X =.f / ! F , i.e.,
F x ' F ;

x $ :

A conjugate of is an element 0 in some field containing F such that f . 0 / D 0. Then f

is the minimum polynomial of 0 over F , and so there is an isomorphism
F ' F 0 ;

$ 0.

P ROPOSITION 6.11. Let A be a normal integral domain, and let E be a finite extension of
the field of fractions F of A. An element of E is integral over A if and only if its minimum
polynomial over F has coefficients in A.
P ROOF. Let be integral over A, so that
m C a1 m

C


C am D 0;

some ai 2 A;

m > 0.

Let f be the minimum polynomial of over F , and let L a field containing F and splitting
f . For any conjugate 0 of in L, there is an isomorphism  W F ! F 0 sending to
0 . On applying  to the above equation we obtain an equation
0m C a1 0m

C


C am D 0

demonstrating that 0 is integral over A. As the coefficients of f are polynomials in the

conjugates of in L, it follows from (6.5) that the coefficients of f are integral over A.
They lie in F , and A is integrally closed in F , and so they lie in A. This proves the only if
part of the statement, and the if part is obvious.

INTEGRAL DEPENDENCE

28

C OROLLARY 6.12. Let A be a normal integral domain with field of fractions F , and let f
be a monic polynomial in AX . Then every monic factor of f in F X has coefficients in
A.
P ROOF. It suffices to prove this for an irreducible monic factor g of f in F X . Let be a
root of g in some extension field of F . Then g is the minimum polynomial of over F . As
is a root of f , it is integral over A, and so g has coefficients in A.
We shall need a more general form of (6.12).
L EMMA 6.13. Let A be a ring, and let B be an A-algebra. Let f; g 2 BT be monic
polynomials such that g divides f . If the coefficients of f are integral over A, then so also
are those of g.
P ROOF. There exists a ring B 0 containing B such that f splits in B 0 T . This can be
constructed in the same as way as the splitting field of a polynomial over a field.12 The roots
of f in B 0 are integral over the A-subalgebra of B generated by the coefficients of f , and
hence over A (see 6.4). As the roots of g are also roots of f , they are integral over A. The
coefficients of g are polynomials in its roots, and hence are integral over A (see 6.5).
P ROPOSITION 6.14. Let A  B be rings, and let A0 be the integral closure of A in B. For
any multiplicative subset S of A, S 1 A0 is the integral closure of S 1 A in S 1 B.
P ROOF. Let b=s 2 S

1 A0

with b 2 A0 and s 2 S . Then

b n C a1 b n

C


C an D 0

for some ai 2 A, and so

 n
 n
b
a1 b
C
s
s s

C


C

an
D 0:
sn

Therefore b=s is integral over S 1 A. This shows that S 1 A0 is contained in the integral
closure of S 1 A.
For the converse, let b=s (b 2 B, s 2 S ) be integral over S 1 A. Then
 n
 n
b
a1 b
C
s
s1 s

C


C

an
D 0:
sn

for some ai 2 A and si 2 S . On multiplying this equation by s n s1


sn , we find that

s1


sn b 2 A0 , and therefore that b=s D s1


sn b=ss1


sn 2 S 1 A0 .
C OROLLARY 6.15. Let A  B be rings, and let S be a multiplicative subset of A. If A is
integrally closed in B, then S 1 A is integrally closed in S 1 B.
P ROOF. Special case of the proposition in which A0 D A.
deg.f /  1, we take B 0 D B. Otherwise, let B1 D BT =.f /. As f is monic, deg.fg/ D deg.f / C
deg.g/ for any polynomial g 2 BT , and so .f / \ B D f0g. Therefore the map B ! B1 is injective. On the
other hand, f has a root b D T mod.f / in B1 , and so f D .T b/f1 in B1 T with f1 monic. If deg.f1 / > 1,
apply the same argument to it and B1 . Continuing in this fashion, we eventually arrive at the required ring B 0 .
12 If

INTEGRAL DEPENDENCE

29

P ROPOSITION 6.16. The following conditions on an integral domain A are equivalent:

(a) A is integrally closed;
(b) Ap is integrally closed for all prime ideals p;
(c) Am is integrally closed for all maximal ideals m.
P ROOF. The implication (a))(b) follows from (6.15), and (b))(c) is obvious. For (c))(a),
let A0 be the integral closure of A in its field of fractions F . Then .A0 /m is the integral
closure of Am in F (by 6.14). If (c) holds, then Am ! .A0 /m is surjective for all maximal
ideals m in A, which implies that A ! A0 is surjective (by 5.17), and so A is integrally
closed.
We shall need to use the next statement in the proof of Zariskis main theorem (Chapter
17).
P ROPOSITION 6.17. Every polynomial ring over a normal integral domain is a normal
integral domain.
P ROOF. It suffices to prove that if A is a normal integral domain, then AT is a normal
integral domain. Let F be the field of fractions of A. If an element of the field of fractions
F .T / of AT is integral over AT , then it is integral over F T , and so lies in F T (see
6.10). We can now apply the next proposition with B D F .
P ROPOSITION 6.18. Let B be an A-algebra. If a polynomial in BT is integral over AT ,
then each of its coefficients is integral over A.
P ROOF. We may replace A with its image in B. Suppose that P 2 BT is a root of the
polynomial
q.X / D X n C f1 X n 1 C


C fn ; fi 2 AT :
Let r be greater than the degrees of the polynomials P; f1 ; : : : ; fn . Let P1 .T / D P .T / T r ,
and let
def
q1 .X / D q.X C T r / D X n C g1 X n 1 C


C gn ; gi 2 AT :
Then P1 is a root of q1 .X /,
P1n C g1 P1n

C


C gn D 0;

and so
gn D P1
.P1n

C g1 P1n

C


C gn

1 /:

The choice of r implies that both P1 and gn are monic (as polynomials in T ). As gn has
coefficients in A, Lemma 6.13 shows that the coefficients of P1 are integral over A. This
implies that the coefficients of P are integral over A.

Exercises
E XERCISE 6.19. A ring A is said to be normal if Ap is a normal integral domain for all
prime ideals p in A. Show that a noetherian ring is normal if and only if it is a finite product
of normal integral domains.
E XERCISE 6.20. Prove the converse of (6.18).
E XERCISE 6.21. Let A be an integral domain, and let A0 be its integral closure. Show that
the integral closure of AT is A0 T .

THE GOING-UP AND GOING-DOWN THEOREMS

30

The going-up and going-down theorems

The going-up theorem

P ROPOSITION 7.1. Let A  B be integral domains, with B integral over A. Then B is a
field if and only if A is a field.
P ROOF. Suppose that A is a field, and let b be a nonzero element of B. Then
b n C a1 b n

C


C an D 0

for some ai 2 A, and we may suppose that n is the minimum degree of such a relation. As
B is an integral domain, an 0, and the equation
b
.b n

C a1 b n

C


C an

1 /an

D 1

shows that b has an inverse in B.

Conversely, suppose that B is a field, and let a be a nonzero element of A. Then a has
an inverse a 1 in B, and
a

C a1 a

.n 1/

C


C an D 0

for some ai 2 A. On multiplying through by an

a

1,

we find that

C a1 C a2 a


C an an

D 0,

and so
a

D .a1 C a2 a


C an an

/ 2 A:

R EMARK 7.2. The second part of the proof shows that A \ B  D A .

C OROLLARY 7.3. Let A  B be rings with B integral over A. Let q be a prime ideal of B,
and let p D q \ A. Then q is maximal if and only if p is maximal.
P ROOF. Apply the proposition to A=p  B=q.
C OROLLARY 7.4 ( INCOMPARABILITY ). Let A  B be rings with B integral over A, and
let q  q0 be prime ideals of B. If q \ A D q0 \ A, then q D q0 .
In other words, if B  A is integral over A, then there is no containment relation between
the prime ideals of B lying over a given prime ideal of A.
P ROOF. Let p D q \ A D q0 \ A. Then Ap  Bp , and Bp is integral over Ap . The ideals
qBp  q0 Bp are both prime ideals of Bp lying over pAp , which is maximal, and so qBp D q0 Bp
(by 7.3). Now

c 5.4
5.4
q D .qBp /c D q0 Bp D q0 :
P ROPOSITION 7.5. Let A  B be rings with B integral over A, and let p be a prime ideal
of A. Then there exists a prime ideal q of B such that p D q \ A.

THE GOING-UP AND GOING-DOWN THEOREMS

31

P ROOF. We have Ap  Bp , and Bp is integral over Ap . Let n be a maximal ideal in Bp

(which exists by 2.3), and let q be the inverse image of n in B. We claim that q \ A D p.
The ideal n \ Ap is maximal (7.3), but pAp is the unique maximal ideal of Ap , and so
n \ Ap D pAp . From the commutative diagram
B

Bp

Ap ;

we see that q \ A is the inverse image of pAp in A. But the inverse image of pAp in A is p
(as pec D p; see 5.4).
T HEOREM 7.6. Let A  B be rings with B integral over A. Let p  p0 be prime ideals of
A, and let q be a prime ideal of B such that q \ A D p. Then there exists a prime ideal q0 of
B containing q and such that q0 \ A D p0 :
q0

 p0 :

P ROOF. We have A=p  B=q, and B=q is integral over A=p. According to the (7.5), there
exists a prime ideal q00 in B=q such that q00 \ .A=p/ D p0 =p. The inverse image q0 of q00 in B
has the required properties.
C OROLLARY 7.7. Let A  B be rings with B integral over A, and let p1 


 pn be
prime ideals in A. Let
q1 


 qm .m < n/
(13)
be prime ideals in B such that qi \ A D pi for all i  m. Then (13) can be extended to a
chain of prime ideals
q1 


 qn
such that qi \ A D pi for all i  n:
q1

qm

qn

p1

pm

pn

P ROOF. Immediate consequence of Corollary 7.6.

Theorem 7.6 and its corollary 7.7 are known as the going-up theorem (of Cohen and
Seidenberg).
A SIDE 7.8. The going-up theorem (7.6) fails for the rings Z  ZX : consider the prime ideals
.0/  .2/ of Z, and the prime ideal q D .1 C 2X / of ZX ; then q \ Z D .0/, but a prime ideal q0 of
ZX containing q and such that q0 \ Z D .2/ would have to contain .2; 1 C 2X / D ZX (mo159544).

THE GOING-UP AND GOING-DOWN THEOREMS

32

The going-down theorem

Before proving the going-down theorem, we need to extend some of the definitions and
results from earlier in this section.
Let A  B be rings, and let a be an ideal of A. An element b of B is said to be integral
over a if it satisfies an equation
b n C a1 b n

C


C an D 0

(14)

with the ai 2 a. The set of elements of B integral over a is called the integral closure of a in
B. The proof of Proposition 6.1 shows that b 2 B is integral over a if there exists a faithful
Ab-submodule M of B, finitely generated as an A-module, such that bM  aM .
Note that if b m is integral over a, so also is b (the equation (14) for b m can be read as a
similar equation for b).
L EMMA 7.9. Let A0 be the integral closure of A in B. Then the integral closure of a in B is
the radical of aA0 .
P ROOF. Let b 2 B be integral over a. From (14) we see that b 2 A0 and that b n 2 aA0 , and
so b is in the radical of aA0 .
Conversely, let b be in the radical of aA0 , so that
X
ai xi ; some m > 0, ai 2 a; xi 2 A0 :
bm D
i

def

As each xi is integral over A, M D Ax1 ; : : : ; xn is a finite A-algebra (see 6.2). As b n M 

aM , we see that b n is integral over a, which implies that b is integral over a.
In particular, the integral closure of a in B is an ideal in A0 , and so it is closed under the
formation of sums and (nonempty) products.
P ROPOSITION 7.10. Let A be a normal integral domain, and let E extension of the field
of fractions F of A. If an element of E is integral over an ideal a in A, then its minimum
polynomial over F has coefficients in the radical of a.
P ROOF. Let be integral over a, so that
n C a1 n

C


C an D 0

for some n > 0 and ai 2 a. As in the proof of (6.11), the conjugates of satisfy the same
equation as , and so are also integral over a. The coefficients of the minimum polynomial
of over F are polynomials without constant term in its conjugates, and so they are also
integral over a. As these coefficients lie in F , they lie in the integral closure of a in F , which
is the radical of a (by 7.9).
T HEOREM 7.11. Let A  B be integral domains with A normal and B integral over A. Let
p  p0 be prime ideals in A, and let q be a prime ideal in B such that q \ A D p. Then q
contains a prime ideal q0 in B such that q0 \ A D p0 :
q0

 p0 :

NOETHERS NORMALIZATION THEOREM

33

P ROOF. The prime ideals of B contained in q are the contractions of prime ideals in Bq (see
5.4), and so we have show to that p0 is the contraction of a prime ideal of Bq , or, equivalently
(see 5.6), that


A \ p0 Bq D p0 .
Let b 2 p0 Bq . Then b D y=s with y 2 p0 B and s 2 B X q. By (7.9), y is integral over p0 ,
and so (by 7.10) the minimum equation
y m C a1 y m

C


C am D 0

(15)

of y over the field of fractions F of A has coefficients ai 2 p0 .

Suppose that b 2 A \ p0 Bq . Then b 1 2 F , and so, on replacing y with bs in (15) and
dividing through by b m , we obtain the minimum equation for s over F :
s m C .a1 =b/s m

C


C .am =b m / D 0:

(16)

But s is integral over A, and so (by 6.11), each coefficient ai =b i 2 A. Suppose that
b p0 . The coefficients ai =b i 2 p0 , and so (16) shows that s m 2 p0 B  pB  q, and so s 2 q,
which contradicts its definition. Hence b 2 p0 , and so A \ p0 Bq D p0 as required.
C OROLLARY 7.12. Let A  B be integral domains with A normal and B integral over A.
Let p1 


 pn be prime ideals in B, and let
q1 


 qm

(m < n)

(17)

be prime ideals in B such that qi \ A D pi for all i . Then (17) can be extended to a chain of
prime ideals
q1 


 qn
such that qi \ A D pi for all i :
q1

qm

qn

p1

pm

pn

P ROOF. Immediate consequence of the theorem.

Theorem 7.11 and its corollary 7.12 are known as the going-down theorem (of Cohen
and Seidenberg). The going-down theorem also holds for flat A-algebras see (11.15).

Noethers normalization theorem

T HEOREM 8.1 (N OETHER NORMALIZATION THEOREM ). Every finitely generated algebra A over a field k contains a polynomial algebra R such that A is a finite R-algebra.
In other words, there exist elements y1 ; : : : ; yr of A that are algebraically independent over
k and such that A is finite over ky1 ; : : : ; yr .
Let A D kx1 ; : : : ; xn . If the xi are algebraically independent, then there is nothing
to prove. Otherwise, the next lemma shows that A is finite over a subring kx10 ; : : : ; xn0 1 .
Continuing in this fashion, we arrive at a proof.

NOETHERS NORMALIZATION THEOREM

34

L EMMA 8.2. Let A D kx1 ; : : : ; xn be a finitely generated k-algebra, and let fx1 ; : : : ; xd g
be a maximal algebraically independent subset of fx1 ; : : : ; xn g. If n > d , then there exist an
d
m 2 N such that A is finite over its subalgebra kx1 xnm ; : : : ; xd xnm ; xd C1 ; : : : ; xn 1 .
P ROOF. By assumption, the set fx1 ; : : : ; xd ; xn g is algebraically dependent, and so there
exists a nonzero f 2 kX1 ; : : : ; Xd ; T such that
f .x1 ; : : : ; xd ; xn / D 0:

(18)

Because the set fx1 ; : : : ; xd g is algebraically independent, T occurs in f , and so we can

write
f .X1 ; : : : ; Xd ; T / D a0 T r C a1 T r 1 C


C ar
with ai 2 kX1 ; : : : ; Xd , a0 0, and r > 0.
If a0 2 k, then (18) shows that xn is integral over kx1 ; : : : ; xd . Hence x1 ; : : : ; xn are
integral over kx1 ; : : : ; xn 1 , and so A is finite over kx1 ; : : : ; xn 1 (see 6.3). Thus the
lemma holds with with m D 0.
If a0 k, then we make a change of variables so that it becomes constant. Specifically,
for a suitable m 2 N, the polynomial
2

g.X1 ; : : : ; Xd ; T / D f .X1 C T m ; X2 C T m ; : : : ; Xd C T m ; T /
def

takes the form

g.X1 ; : : : ; Xd ; T / D c0 T r C c1 T r

C


C cr

with c0 2 k  (see the next lemma). As

g.x1

xnm ; : : : ; xd

xnm ; xn / D 0,

(19)

this shows that xn is integral over kx1 xnm ; : : : ; xd xnm . The elements xi , i  d , are too,
d
because xi D .xi xnm /Cxnm , and so A is finite over kx1 xnm ; : : : ; xd xnm ; xd C1 ; : : : ; xn 1 .
L EMMA 8.3. Let f 2 kX1 ; : : : ; Xd ; T . For a suitable m 2 N,
2

f .X1 C T m ; X2 C T m ; : : : ; Xd C T m ; T /
takes the form c0 T r C c1 T r

1 C


Cc
r

with c0 2 k  :

P ROOF. Let
f .X1 ; : : : ; Xd ; T / D

cj1


jn X1 1


Xdd Tnjn :

(20)

Let S be the set of .d C 1/-tuples .j1 ; : : : ; jd ; jn / such that cj1 :::jn 0, and choose m so
that m > maxi ji for all .j1 ; : : : ; jd / 2 S . Note that
d

.X1 C T m /j1


.Xd C T m /jd Tnjn

2j

D T mj1 Cm

d
2 C


Cm jd Cjn

C terms of lower degree in T:

When .j1 ; : : : ; jn / runs over the elements of S, the exponents

mj1 C m2 j2 C


C md jd C jn

(21)

are distinct, because they are distinct base-m expansions of natural numbers. Now
g.X1 ; : : : ; Xd ; T / D c0 T1N C c1 T1N
with c 2 k  and N equal to the largest value of (21).

C

DIRECT AND INVERSE LIMITS

35

R EMARK 8.4. When k is infinite, it is possible to prove a somewhat stronger result: let
A D kx1 ; : : : ; xn ; then there exist algebraically independent elements f1 ; : : : ; fr that are
linear combinations of the xi such that A is finite over kf1 ; : : : ; fr . See my Algebraic
Geometry notes.
A SIDE 8.5. The map ky1 ; : : : ; yr ! A in (8.1) is flat if and only if A is Cohen-Macaulay (for
example, regular). See (23.10).
Let X be the variety obtained by removing the origin from C2 and identifying the points .1; 1/
and . 1; 1/. Then G D Z=2 acts on by .x; y/ 7! . x; y/ and the quotient is smooth, but X is not
Cohen-Macaulay (two planes intersecting in a point is not Cohen-Macaulay). Therefore the quotient
map. See mo173538.

Direct and inverse limits

Direct limits
D EFINITION 9.1. A partial ordering  on a set I is said to be directed, and the pair .I; /
is called a directed set, if for all i; j 2 I there exists a k 2 I such that i; j  k.
D EFINITION 9.2. Let .I; / be a directed set, and let A be a ring.
A direct system of A-modules indexed by .I; /
is a family .Mi /i2I of A-modules together with a
family .ji W Mi ! Mj /i j of A-linear maps such
j

that ii D idMi and k ji D ki all i  j  k.

An A-module M together with a family . i W Mi !
M /i 2I of A-linear maps satisfying i D j ji
all i  j is said to be a direct limit of the system ..Mi /; .ji // if it has the following universal
property: for any other A-module N and family . i W Mi ! N / of A-linear maps such that
i D j ji all i  j , there exists a unique morphism W M ! N such that i D i for all
i.

Mk

ki
j

Mi

ji

Mj

i
j

Mi

ji

Mj

As usual, the universal property determines the direct limit (if it exists) uniquely up to a
j
unique isomorphism. We denote it lim.Mi ; i /, or just lim Mi .
!
!
C RITERION
An A-module M together with A-linear maps i W Mi ! M such that i D j ji for all
j

i  j is the S
direct limit of a system .Mi ; i / if and only if
(a) M D i 2I i .Mi /, and
(b) if mi 2 Mi maps to zero in M , then it maps to zero in Mj for some j  i .
C ONSTRUCTION
L
L
Consider the direct sum i 2I Mi of the modules Mi . Thus, the elements of i 2I Mi are
the families .mi /i 2I with mi D 0 for all but finitely many i . We can identify Mi0 with the

DIRECT AND INVERSE LIMITS

36

L
submodule
of
of elements .mi /i with mi D 0 for i i0 . Then every
i 2I Mi P
L
L element of
i 2I Mi is a finite sum
i 2I mi with mi 2 Mi . Let M be the quotient of
i 2I Mi by the
A-submodule M 0 generated by the elements
mi

ji .mi /, mi 2 Mi ;

i < j.

Let i .mi / D mi C M 0 . Then certainly i D j ji for all i  j . For every A-module N

and A-linear maps j W Mj ! N , there is a unique map
M
Mi ! N;
i 2I

namely, mi 7! i .mi /, sending mi to i .mi /, and this map factors through M and is
the unique A-linear map with the required properties.
Direct limits of A-algebras, etc., are defined similarly.
P

A N EXAMPLE
P ROPOSITION 9.3. For every multiplicative subset S of a ring A, S
h runs over the elements of S (partially ordered by division).

1A

' lim Ah , where

!

P ROOF. An element h of a ring that divides a unit is itself a unit (if u D hq, then 1 D
h.qu 1 /). Therefore, if hjh0 in A, say, h0 D hq, then h becomes a unit in Ah0 , and so
(see 5.1) there is a unique homomorphism Ah ! Ah0 respecting the maps A ! Ah and
A ! Ah0 , namely, ha 7! ag
. In this way, the rings Ah form a direct system indexed by the
h0
set S . When h 2 S, the homomorphism A ! S 1 A extends uniquely to a homomorphism
a
7! ha W Ah ! S 1 A, and these homomorphisms are compatible with the maps in the direct
h
system (apply 5.1 again). The criterion p. 35 shows that S 1 A is the direct limit of the Ah .
E XACTNESS
P ROPOSITION 9.4. The direct limit of a system of exact sequences of modules is exact.
This means the following: suppose that .Mi ; ji /, .Ni ; ji /, and .Pi ; ji / are direct
systems with repect to the directed set I , and let
.ai /

.bi /

.Mi ; ji / ! .Ni ; ji / ! .Pi ; ji /

be a sequence of maps of direct systems; if the sequences
ai

bi

Mi ! Ni ! Pi
are exact for all i , then the direct limit sequence
lim ai

lim bi

lim Mi !! lim Ni !! lim Pi

!
!
!
is exact.
P ROOF. Let .ni / 2 lim Ni . If .bi .ni // D 0, then there exists an i0 such that bi .ni / D 0 for
!
all i  i0 . Let mi D 0 unless i  i0 , in which case we let mi be the unique element of Mi
such that ai .mi / D ni . Then .mi / maps to .ni /. This proves the exactness.

DIRECT AND INVERSE LIMITS

37

Inverse limits
Inverse limits are the same as direct limits except that the directions of the arrows is reversed.
Thus, formally, the theory of inverse limits is the same as that of inverse limits. However, in
concrete categories, they behave very differently. For example, the inverse limit of a system
of exact sequences of modules need not be exact.
We shall consider inverse limits only in the case that the indexing set if N with its usual
ordering. In this case, an inverse system of A-modules is nothing more than a sequence of
modules and A-homomorphisms
M0

M1

Mn

:

A homomorphism .Mn ; n / ! .Nn ; n / of inverse systems is a sequence of A-homomorphisms

n W Mn ! Nn such that n nC1 D n n for all n 2 N.
Given an inverse system .Mn ; n / of A-modules, we define lim Mn and lim1 Mn to be
the kernel and cokernel of the A-module homomorphism
Y
Y
.: : : ; mn ; : : :/ 7! .: : : ; mn n .mnC1 /; : : :/W Mn !
Mn :
P ROPOSITION 9.5. For any inverse system .Mn ; n / and A-module N ,
Hom.lim Mn ; N / ' lim Hom.Mn ; N /:
P ROOF. This is easy to check directly.
P ROPOSITION 9.6. Every inverse system of exact sequences
0 ! .Mn ; n / ! .Nn ; n / ! .Pn ; n / ! 0;
gives rise to an exact sequence
0 ! lim Mn ! lim Nn ! lim Pn ! lim1 Mn ! lim1 Nn ! lim1 Pn ! 0:
P ROOF. The sequence
0!

Mn !

Nn !

Pn ! 0

is exact, and so this follows from the snake lemma.

C OROLLARY 9.7. If the maps n W MnC1 ! Mn are all surjective, then lim1 Mn D 0.
Q
P ROOF. Let .mi / 2 i 2N Mi . We have show that there exists an infinite sequence .xi /i 2N ,
xi 2 Mi , such that
xi i .xi C1 / D mi
(22)
for all i 2 N. We consider finite sequences fx0 ; : : : ; xn g, xi 2 Mi , satisfying (22) for i < n.
For example, f0g is such a sequence. Such a sequence fx0 ; : : : ; xn g can always be extended:
use the surjectivity of nC1 to find an xnC1 2 MnC1 such that
n .xnC1 / D xn

mn :

Now the axiom of dependent choice shows that there exists a sequence .xi /n2N , xi 2 Mi ,
satisfying (22) for all n.
A SIDE 9.8. Direct (resp. inverse) limits are also called inductive (resp. projective) limits or colimits
(resp. limits).

10

10

TENSOR PRODUCTS

38

Tensor Products

Tensor products of modules

Let A be a ring, and let M , N , and P be A-modules. A map W M  N ! P of A-modules
is said to be A-bilinear if
.x C x 0 ; y/ D .x; y/ C .x 0 ; y/;
0

x; x 0 2 M;

y2N

x 2 M;

y; y 0 2 N

.ax; y/ D a.x; y/;

a 2 A;

x 2 M;

y2N

.x; ay/ D a.x; y/;

a 2 A;

x 2 M;

y 2 N;

.x; y C y / D .x; y/ C .x; y /;

i.e., if  is A-linear in each variable.

An A-module T together with an A-bilinear map
M N

W M  N ! T

0

is called the tensor product of M and N over A if it has the

following universal property: every A-bilinear map
0

 WM N ! T

T
9 linear

T 0:

factors uniquely through .

As usual, the universal property determines the tensor product uniquely up to a unique
isomorphism. We write it M A N . Note that
HomA-bilinear .M  N; T / ' HomA-linear .M A N; T /:
C ONSTRUCTION
Let M and N be A-modules, and let A.M N / be the free A-module with basis M  N . Thus
each element A.M N / can be expressed uniquely as a finite sum
X
ai .xi ; yi /; ai 2 A; xi 2 M; yi 2 N:
Let P be the submodule of A.M N / generated by the following elements
.x C x 0 ; y/
0

.x; y C y /

.x; y/

.x 0 ; y/;
0

x; x 0 2 M;

y2N

.x; y/

.x; y /;

x 2 M;

y; y 0 2 N

.ax; y/

a.x; y/;

a 2 A;

x 2 M;

y2N

.x; ay/

a.x; y/;

a 2 A;

x 2 M;

y 2 N;

and define
M A N D A.M N / =P:
Write x y for the class of .x; y/ in M A N . Then
.x; y/ 7! x yW M  N ! M A N

10

TENSOR PRODUCTS

39

is A-bilinear we have imposed the fewest relations necessary to ensure this. Every element
of M A N can be written as a finite sum13
X
ai .xi yi /; ai 2 A; xi 2 M; yi 2 N;
and all relations among these symbols are generated by the following relations
.x C x 0 / y D x y C x 0 y
x .y C y 0 / D x y C x y 0
a.x y/ D .ax/ y D x ay:
The pair .M A N; .x; y/ 7! x y/ has the correct universal property because any bilinear
map  0 W M  N ! T 0 defines an A-linear map A.M N / ! T 0 , which factors through
A.M N / =K, and gives a commutative triangle.
S YMMETRIC MONOIDAL STRUCTURE
P ROPOSITION 10.1. Let M; N; P be modules over a ring A.
(a) (Existence of an identity object) There is a unique isomorphism
W A M ! M
such that .a m/ D am for all a 2 A, m 2 M .
(b) (Associativity) There is a unique isomorphism
W M .N P / ! .M N / P
such that .m .n p// D ..m n/ p/ for all m 2 M , n 2 N , p 2 P .
(c) (Symmetry) There is a unique isomorphism
W M N ! N M
such that .m n/ D n m for all m 2 M , n 2 N .
P ROOF. We prove (b). The uniqueness is obvious because the elements m.np/ generate
M .N P / as an A-module. The map
.m; n; p/ 7! m .n p/W M  N  P ! M .N P /

(23)

is A-trilinear. Let W M  N  P ! Q be a second A-trilinear map. For a fixed m 2 M ,

the map .n; p/ ! .m; n; p/W N  P ! Q is A-bilinear, and so it extends uniquely to an
A-linear map m W N P ! Q. Now the map .m; n p/ 7! m .n p/W M  .N P / ! Q
is A-bilinear, and so it extends uniquely to an A-linear map M .N P / ! Q. This shows
that (23) is universal among A-trilinear maps from M  N  P to an A-module. Similarly,
the A-trilinear map
.m; n; p/ 7! .m n/ pW M  N  P ! .M N / P
is universal, from which the statement follows (see the footnote p.19).
The proofs of (a) and (c) are similar, but easier.
13 An

element of the tensor product of two vector spaces is not necessarily a tensor product of two vectors,
but sometimes a sum of such. This might be considered a mathematical shenanigan but if you start with the
state vectors of two quantum systems it exactly corresponds to the notorious notion of entanglement which so
displeased Einstein. Georges Elencwajg on mathoverflow.net.

10

TENSOR PRODUCTS

40

E XTENSION OF SCALARS
Let A be a commutative ring and let B be an A-algebra (not necessarily commutative) such
that the image of A ! B lies in the centre of B. Then M
B A M is a functor from
left A-modules to left B-modules. Let M be an A-module and N a B-module; an A-linear
map W M ! N defines a B-linear map W B A M ! N such that b m 7! b
.m/, and
$ is an isomorphism:
HomA-linear .M; N / ' HomB-linear .B A M; N /.

(24)

If .e /2I is a family of generators (resp. basis) for M as an A-module, then .1 e /2I is

a family of generators (resp. basis) for B A M as a B-module.
def
The functor M
MB D B A M commutes with taking tensor products:
.M A N /B ' MB B NB :

(25)

To see this, note that

MB B NB D .B A M / B .B A N / (definition)
' ..B A M / B B/ A N
' .B A M / A N

(associativity)

(obvious)

' B A .M A N / (associativity)
D .M A N /B

(definition).

B EHAVIOUR WITH RESPECT TO DIRECT LIMITS

P ROPOSITION 10.2. Direct limits commute with tensor products:
lim Mi A lim Nj '
!
!
i 2I

j 2J

lim
!

Mi A Nj :

.i;j /2I J

P ROOF. Using the universal properties of direct limits and tensor products, one sees easily
that lim.Mi A Nj / has the universal property to be the tensor product of lim Mi and lim Nj .
!
!
!

Tensor products of algebras

Let k be a ring, and let A and B be k-algebras. A k-algebra C together with homomorphisms
i W A ! C and j W B ! C is called the tensor product of A and B if it has the following
universal property:
for every pair of homomorphisms (of k-algebras)
f W A ! R and gW B ! R, there exists a unique
homomorphism .f; g/W C ! R such that .f; g/
i D and .f; g/ j D ,

A
f

C
9 .f; g/

B
g

If it exists, the tensor product, is uniquely determined up to a unique isomorphism by this

property. We write it A k B. Note that the universal property says that
Hom.A k B; R/ ' Hom.A; R/  Hom.B; R/
(k-algebra homomorphisms).

(26)

10

TENSOR PRODUCTS

41

C ONSTRUCTION
Regard A and B as k-modules, and form the tensor product Ak B. There is a multiplication
map A k B  A k B ! A k B for which
.a b/.a0 b 0 / D aa0 bb 0 ;

all a; a0 2 A;

b; b 0 2 B:

This makes A k B into a ring, and the homomorphism

c 7! c.1 1/ D c 1 D 1 c
makes it into a k-algebra. The maps
a 7! a 1W A ! A k B and b 7! 1 bW B ! A k B
are homomorphisms, and they make A k B into the tensor product of A and B in the above
sense.
E XAMPLE 10.3. The algebra A, together with the maps
k !A

idA

A,

is k k A (because it has the correct universal property). In terms of the constructive

definition of tensor products, the map c a 7! caW k k A ! A is an isomorphism.
E XAMPLE 10.4. The ring kX1 ; : : : ; Xm ; XmC1 ; : : : ; XmCn , together with the obvious inclusions
kX1 ; : : : ; Xm ,! kX1 ; : : : ; XmCn

- kXmC1 ; : : : ; XmCn

is the tensor product of the k-algebras kX1 ; : : : ; Xm and kXmC1 ; : : : ; XmCn . To verify
this we only have to check that, for every k-algebra R, the map
Hom.kX1 ; : : : ; XmCn ; R/ ! Hom.kX1 ; : : :; R/  Hom.kXmC1 ; : : :; R/
induced by the inclusions is a bijection. But this map can be identified with the bijection
RmCn ! Rm  Rn :
In terms of the constructive definition of tensor products, the map
kX1 ; : : : ; Xm k kXmC1 ; : : : ; XmCn ! kX1 ; : : : ; XmCn
sending f g to fg is an isomorphism.
R EMARK 10.5. (a) Let k ,! k 0 be a homomorphism of rings. Then
k 0 k kX1 ; : : : ; Xn ' k 0 1 X1 ; : : : ; 1 Xn ' k 0 X1 ; : : : ; Xn :
If A D kX1 ; : : : ; Xn =.g1 ; : : : ; gm /, then
k 0 k A ' k 0 X1 ; : : : ; Xn =.g1 ; : : : ; gm /:
(b) If A and B are algebras of k-valued functions on sets S and T respectively, then the
definition
.f g/.x; y/ D f .x/g.y/; f 2 A, g 2 B, x 2 S, y 2 T;
realizes A k B as an algebra of k-valued functions on S  T .

10

TENSOR PRODUCTS

42

The tensor algebra of a module

Let M be a module over a ring A. For each A  0, set
T r M D M A


A M
so that T 0 M D A and T 1 M D M , and define
M
TM D

r0

(r factors),

T r M:

This can be made into a noncommutative A-algebra, called the tensor algebra of M , by
requiring that the multiplication map
T r M  T s M ! T rCs M
send .m1


mr ; mrC1


mrCs / to m1


mrCs .
The pair .TM; M ! TM / has the following universal property: every A-linear map from M to an A-algebra R (not necessarily commutative) extends uniquely to an A-algebra homomorphism TM ! R.
If M is a free A-module with basis x1 ; : : : ; xn , then TM is
the (noncommutative) polynomial ring over A in the noncommuting symbols xi (because this A-algebra has the same universal
property as TM ).

M
A-linear

TM
9A-algebra

The symmetric algebra of a module

The symmetric algebra Sym.M / of an A-module M is the quotient of TM by the ideal
generated by all elements of T 2 M of the form
m n n m; m; n 2 M:
L
It is a graded algebra Sym.M / D r0 Symr .M / with Symr .M / equal to the quotient of
M r by the A-submodule generated by all elements of the form
m1


mr

m .1/


m .r/ ;

mi 2 M;

 2 Br (symmetric group).

The pair .Sym.M /; M ! Sym.M // has the following

universal property: every A-linear map M ! R from M
M
Sym.M /
to a commutative A-algebra R extends uniquely to an
A-algebra homomorphism Sym.M / ! R (because it ex9A-algebra
A-linear
tends to an A-algebra homomorphism TM ! R, which
factors through Sym.M / because R is commutative).
R
If M is a free A-module with basis x1 ; : : : ; xn , then
Sym.M / is the polynomial ring over A in the (commuting) symbols xi (because this A-algebra has the same universal property as TM ).

Exercises
E XERCISE 10.6. Look up symmetric monoidal category in the Wikipedia and show that
the category of A-modules equipped with the bifunctor and the maps , , and in (10.1)
is such a category.

11

11

FLATNESS

43

Flatness

P ROPOSITION 11.1. Let M be an A-module, and let 0 ! N 0 ! N ! N 00 ! 0 be an

exact sequence of A-modules. Then the sequence
1

M A N 0 ! M A N ! M A N 00 ! 0
is exact.


P ROOF. The surjectivity of 1 is obvious. Let M A N ! Q be the cokernel of 1 .

Because
.1 / .1 / D 1 . / D 0;
there is a unique A-linear map f W Q ! M A N 00 such that f  D 1 . We shall
construct an inverse g to f .
Let m 2 M , and let n 2 N . If .n/ D 0, then n D .n0 / for some n0 2 N 0 ; hence m n D
m .n0 /, and so .m n/ D 0. It follows by linearity that .m n1 / D .m n2 / if
.n1 / D .n2 /, and so the A-bilinear map
M  N ! Q;

.m; n/ 7! .m n/

factors through M  N 00 . It therefore defines an A-linear map gW M A N 00 ! Q. To

show that f and g are inverse, it suffices to check that g f D idQ on elements of the
form .m n/ and that f g D idM A N 00 on elements of the form m .n/ both are
obvious.
The map M A N 0 ! M A N in (11.1) need not be injective. For example, when we
tensor the exact sequence of Z-modules
m

0 ! Z ! Z ! Z=mZ ! 0
with Z=mZ, only the sequence
Z=mZ

x7!mxD0

! Z=mZ

x7!x

! Z=mZ ! 0

is exact.
Moreover, M A N may be zero even when neither M nor N is nonzero. For example,
Z=2Z Z Z=3Z D 0
because it is killed by both 2 and 3.14 In fact, M A M may be zero without M being zero.
For example,
Q=Z Z Q=Z D 0.
To see this, let x; y 2 Q=Z; then nx D 0 for some n 2 Z, and y D ny 0 for some y 0 2 Q=Z;
now
x y D x ny 0 D nx y 0 D 0 y 0 D 0:
14 It

was once customary in certain circles to require a ring to have an identity element 1 0 (see, for example,
Northcott 1953, p.3). However, without the zero ring, tensor products dont always exist. Bourbakis first
example of a ring is the zero ring.

11

FLATNESS

44

D EFINITION 11.2. An A-module M is flat if

N 0 ! N injective H) M A N 0 ! M A N injective.
It is faithfully flat if, in addition,
M A N D 0 H) N D 0:
A homomorphism of rings A ! B is said to be flat (resp. faithfully flat) when B is flat
(resp. faithfully flat) as an A-module.
Thus, an A-module M is flat if and only if the functor M A takes short exact
sequences to short exact sequences. In other words, M is flat if and only if M A is an
exact functor, i.e.,
N 0 ! N ! N 00 exact H) M A N 0 ! M A N ! M A N 00 exact.
An A-algebra B is said to be flat if B is flat as an A-module.
E XAMPLE 11.3. The functor M A takes direct sums to direct sums, and direct sums of
exact sequences are exact; in particular, direct sums of injective maps are injective. Therefore
direct sums of flat modules are flat, and direct summands of flat modules are flat. All free
A-modules are flat. Therefore all vector spaces over a field are flat, and nonzero vector
spaces are faithfully flat.
E XAMPLE 11.4. Quotient maps A ! A=a are rarely flat. If A is a product, A D A1  A2 ,
then the projection map A ! A1 is obviously flat. When A is noetherian, all flat quotient
maps are of this form (Exercise 23.12).
P ROPOSITION 11.5. Let A ! B be a faithfully flat homomorphism of rings. A sequence
of A-modules
0 ! N 0 ! N ! N 00 ! 0
(27)
is exact if
0 ! B A N 0 ! B A N ! B A N 00 ! 0

(28)

is exact.
P ROOF. Let N0 be the kernel of N 0 ! N . Because A ! B is flat, B A N0 is the kernel of
B A N 0 ! B A N , which is zero by assumption; because A ! B is faithfully flat, this
implies that N0 D 0. We have proved the exactness at N 0 , and the proof of the exactness
elsewhere is similar.
R EMARK 11.6. There is a converse to the proposition: suppose that
(27) is exact , (28) is exact;
then A ! B is faithfully flat. The implication ) shows that A ! B is flat. Now let N be
an A-module, and consider the sequence
0 ! 0 ! N ! 0 ! 0.
If B A N D 0, then this sequence becomes exact when tensored with B, and so is itself
exact, which implies that N D 0. This shows that A ! B is faithfully flat.

11

FLATNESS

45

C OROLLARY 11.7. Let A ! B be faithfully flat. An A-module M is flat (resp. faithfully

flat) if B A M is flat (resp. faithfully flat) as a B-module.
def

P ROOF. Assume that MB D B A N is flat, and let N 0 ! N be an injective map of

A-modules. The functor B A from A-modules to B-modules commutes with tensor
products, and so
B A .M A N 0 ! M A N / ' MB B .NB0 ! NB /.
The map at right is injective because A ! B is flat and MB is flat. Now (11.5) shows that
M A N 0 ! M A N is injective. Thus M is flat.
Assume that MB is faithfully flat, and let N be an A-module. If M A N D 0, then
MB B NB is zero because it is isomorphic to .M A N /B . Now NB D 0 because MB is
faithfully flat, and so N D 0 because A ! B is faithfully flat.
P ROPOSITION 11.8. Let i W A ! B be a faithfully flat homomorphism. For every A-module
M , the sequence
d0

d1

0 ! M ! B A M ! B A B A M
with

d0 .m/ D 1 m;
d1 .b m/ D 1 b m

(29)

b 1m

is exact.
P ROOF. Assume first that there exists an A-linear section to A ! B, i.e., an A-linear map
f W B ! A such that f i D idA , and define
k0 W B A M ! M;

k0 .b m/ D f .b/m

k1 W B A B A M ! B A M;

k1 .b b 0 m/ D f .b/b 0 m:

Then k0 d0 D idM , which shows that d0 is injective. Moreover,

k1 d1 C d0 k0 D idBA M
which shows that, if d1 .x/ D 0, then x D d0 .k0 .x//, as required.
We now consider the general case. Because A ! B is faithfully flat, it suffices to prove
that the sequence (29) becomes exact after tensoring in B. But the sequence obtained from
(29) by tensoring with B is isomorphic to the sequence (29) for the homomorphism of rings
b 7! 1 bW B ! B A B and the B-module B A M , because, for example,
B A .B A M / ' .B A B/ B .B A M /:
Now B ! B A B has an B-linear section, namely, f .b b 0 / D bb 0 , and so we can apply
the first part.
C OROLLARY 11.9. If A ! B is faithfully flat, then it is injective with image the set of
elements on which the maps

b 7! 1 b
W B ! B A B
b 7! b 1
agree.

11

FLATNESS

46

P ROOF. This is the special case M D A of the Proposition.

P ROPOSITION 11.10. Let A ! A0 be a homomorphism of rings. If A ! B is flat (or
faithfully flat), then so also is A0 ! B A A0 .
P ROOF. For any A0 -module M ,
.B A A0 / A0 M ' B A .A0 A0 M / ' B A M;
from which the statement follows.
P ROPOSITION 11.11. For every multiplicative subset S of a ring A and A-module M ,
S
The homomorphism a 7! a1 W A ! S

A A M ' S

1A

M:

is flat.

P ROOF. To give an S 1 A-module is the same as giving an A-module on which the elements
of S act invertibly. Therefore S 1 A A M and S 1 M satisfy the same universal property
(see 10, especially (24)), which proves the first statement. As M
S 1 M is exact (5.11),
1
so also is M
S A A M , which proves the second statement.
P ROPOSITION 11.12. A homomorphism of rings 'W A ! B is flat if A'
for all maximal ideals n in B.

1 .n/

! Bn is flat

P ROOF. Let N 0 ! N be an injective homomorphism of A-modules, and let n be a maximal

ideal of B. Then p D ' 1 .n/ is a prime ideal in A, and Ap A .N 0 ! N / is injective (11.11).
Therefore, the map
Bn A .N 0 ! N / ' Bn Ap .Ap A .N 0 ! N //
is injective, and so the kernel M of B A .N 0 ! N / has the property that Mn D 0. Let
x 2 M , and let a D fb 2 B j bx D 0g. For each maximal ideal n of B, x maps to zero in Mn ,
and so a contains an element not in n. Hence a D B, and so x D 0.
P ROPOSITION 11.13. The following conditions on a flat homomorphism 'W A ! B are
equivalent:
(a) ' is faithfully flat;
(b) for every maximal ideal m of A, the ideal '.m/B B;
(c) every maximal ideal m of A is of the form ' 1 .n/ for some maximal ideal n of B.
P ROOF. (a) ) (b): Let m be a maximal ideal of A, and let M D A=m; then
B A M ' B='.m/B:
As B A M 0, we see that '.m/B B.
(b) ) (c): If '.m/B B, then '.m/ is contained in a maximal ideal n of B. Now
' 1 .n/ is a proper ideal in A containing m, and hence equals m.
(c) ) (a): Let M be a nonzero A-module. Let x be a nonzero element of M , and let
def
def
a D ann.x/ D fa 2 A j ax D 0g. Then a is an ideal in A, and M 0 D Ax ' A=a. Moreover,
B A M 0 ' B='.a/
B and, because A ! B is flat, B A M 0 is a submodule of B A M .
Because a is proper, it is contained in a maximal ideal m of A, and therefore
'.a/  '.m/  n
for some maximal ideal n of A. Hence '.a/
B  n B, and so B A M  B A M 0 0.

11

FLATNESS

47

Let spm.A/ denote the set of maximal ideals in A (see Chapter 15). In more geometric
terms, the proposition says that a flat homomorphism 'W A ! B is faithfully flat if and only
if the map spm.'/W spm .B/ ! spm .A/ is surjective. In fact, as we now prove, spec.B/ !
spec .A/ is also surjective.
P ROPOSITION 11.14. Let 'W A ! B be a faithfully flat homomorphism. Every prime ideal
p of A is of the form ' 1 .q/ for some prime ideal q of B.
P ROOF. The prime ideals of B lying over p are in natural one-to-one correspondence with
the prime ideals of B A .p/ (5.21). But the ring B A .p/ is nonzero because A ! B is
faithfully flat, and so it has a prime (even maximal) ideal.
P ROPOSITION 11.15. Let A ! B be a flat homomorphism. Let p0  p be prime ideals in
A, and let q be a prime ideal in B such that qc D p. Then there exists a prime ideal q0  q in
B such that q0c D p0 .
P ROOF. The homomorphism Ap ! Bp is faithfully flat, and p0 Ap is prime (5.4), and so
there exists a prime ideal P of Bp lying over p0 Ap (by 11.14). The contraction of P to B is
contained in q and lies over p0 .
The proposition says that the going-down theorem (7.11), hence also its corollary (7.12),
holds for flat homomorphisms. The going-up theorem fails for flat homomorphisms (7.8).
T HEOREM 11.16 (G ENERIC FLATNESS ). Let A a noetherian integral domain, and let B
be a finitely generated A-algebra. Then for some nonzero elements a of A and b of B, the
homomorphism Aa ! Bb is faithfully flat.
P ROOF. Let F be the field of fractions of A. We first assume that B  F A B.
As F A B is a finitely generated F -algebra, the Noether normalization theorem (8.1)
shows that there exist elements x1 ; : : : ; xm of F A B such that F x1 ; : : : ; xm is a polynomial ring over F and F A B is a finite F x1 ; : : : ; xm -algebra. After multiplying each
xi by an element of A, we may suppose that it lies in B. Let b1 ; : : : ; bn generate B as an
A-algebra. Each bi satisfies a monic polynomial equation with coefficients in F x1 ; : : : ; xm .
Let a 2 A be a common denominator for the coefficients of these polynomials. Then each bi
is integral over Aa . As the bi generate Ba as an Aa -algebra, this shows that Ba is a finite
Aa x1 ; : : : ; xm -algebra (by 6.2). Therefore, after replacing A with Aa and B with Ba , we
may suppose that B is a finite Ax1 ; : : : ; xm -algebra.
B

injective

finite

F A B

E Ax1 ;:::;xm B
finite

finite

Ax1 ; : : : ; xm

F x1 ; : : : ; xm

F:

def

E D F .x1 ; : : : ; xn /

Let E D F .x1 ; : : : ; xm / be the field of fractions of Ax1 ; : : : ; xm , and let b1 ; : : : ; br be

elements of B that form a basis for E Ax1 ;:::;xm B as an E-vector space. Each element
of B can be expressed as a linear combination of the bi with coefficients in E. Let q be

11

FLATNESS

48

a common denominator for the coefficients arising from a set of generators for B as an
Ax1 ; : : : ; xm -module. Then b1 ; : : : ; br generate Bq as an Ax1 ; : : : ; xm q -module. In other
words, the map
P
.c1 ; : : : ; cr / 7! ci bi W Ax1 ; : : : ; xm rq ! Bq
(30)
is surjective. This map becomes an isomorphism when tensored with E over Ax1 ; : : : ; xm q ,
which implies that each element of its kernel is killed by a nonzero element of Ax1 ; : : : ; xm q
and so is zero (because Ax1 ; : : : ; xn q is an integral domain). Hence the map (30) is an
isomorphism, and so Bq is free of finite rank over Ax1 ; : : : ; xm q . Let a be some nonzero
coefficient of the polynomial q, and consider the maps
Aa ! Aa x1 ; : : : ; xm ! Aa x1 ; : : : ; xm q ! Baq :
The first and third arrows realize their targets as nonzero free modules over their sources,
and so are faithfully flat. The middle arrow is flat by (11.11). Let m be a maximal ideal in
Aa . Then mAa x1 ; : : : ; xm does not contain the polynomial q because the coefficient a of q
is invertible in Aa . Hence mAa x1 ; : : : ; xm q is a proper ideal of Aa x1 ; : : : ; xm q , and so the
map Aa ! Aa x1 ; : : : ; xm q is faithfully flat (apply 11.13). This completes the proof when
B  F A B.
We now prove the general case. Note that F A B is the ring of fractions of B with
respect to the multiplicative subset A X f0g (see 11.11), and so the kernel of B ! F A B is
the ideal
n D fb 2 B j ab D 0 for some nonzero a 2 Ag:
This is finitely generated (Hilbert basis theorem 3.7), and so there exists a nonzero c 2 A
such that cb D 0 for all b 2 n. I claim that the homomorphism Bc ! F Ac Bc is injective.
If cbr lies in its kernel, then cas cbr D 0 in Bc for some nonzero cas 2 Ac , and so c N ab D 0
in B for some N ; therefore b 2 n, and so cb D 0, which implies that cbr D 0 already in Bc .
Therefore, after replacing A, B, and M with Ac , Bc , and Mc , we may suppose that the map
B ! F A B is injective. On identifying B with its image, we arrive at the situation of the
theorem.

Exercises
E XERCISE 11.17.
Q Let f1 ; : : : ; fm be elements of a ring A. Show that the canonical homomorphism A ! i Afi is faithfully flat if and only if .f1 ; : : : ; fm / D A. Let f1 ; : : : ; fm
satisfy this condition, and let M be an A-module. Deduce from (11.8) that the sequence
Y
Y
Mfi !
Mfi fj
0!M !
1i m

1i;j m

is exact (the first map sends m to .ni / with ni equal to the image of m in Mfi , and the
second map sends .mi / to .nij / with
nij D .image of mi in Mfi fj /

.image of mj in Mfi fj )).

E XERCISE 11.18. Let .Ai ; ji / be a direct system of rings, and let .Mi ; ji / be a direct
system of abelian groups with the same indexing set. Suppose that each Mi has the structure

12

FINITELY GENERATED PROJECTIVE MODULES

49

of an Ai -module, and that the diagrams

Ai  Mi

Mi

ji ji

ji

Aj  Mj

Mj

commute for all i  j . Let A D lim Ai and M D lim Mi .

!
!
(a) Show that M has a unique structure of an A-module for which the diagrams
Ai  Mi

Mi

i  i

AM

commute for all i .

(b) Show that M is flat as an A-module if each Mi is flat as an Ai -module.
(Bourbaki AC, I, 2, Prop. 9.)

12

Finitely generated projective modules

In many situations, the correct generalization of finite-dimensional vector space is not

finitely generated module but finitely generated projective module. From a different perspective, finitely generated projective modules are the algebraists analogue of the differential
geometers vector bundles (see 12.8).

Projective modules
D EFINITION 12.1. An A-module P is projective if, for each surjective A-linear map
f W M ! N and A-linear map gW P ! N , there exists an A-linear map hW P ! M (not
necessarily unique) such that f h D g:
P
g

9h

0:

In other words, P is projective if every map from P onto a quotient of a module M lifts to a
map to M . Equivalently, P is projective if the functor
M

Hom.P; M /

(A-linear maps)

is exact.
As
L
L
Hom. i Pi ; M / ' i Hom.Pi ; M /

12

FINITELY GENERATED PROJECTIVE MODULES

50

we see that a direct sum of A-modules is projective if and only if each summand is projective.
As A itself is projective, this shows that free A-modules are projective and every direct
summand of a free module is projective. Conversely, let P be a projective module, and write
it as a quotient of a free module,
f

F ! P ! 0I
because P is projective, there exists an A-linear map hW P ! F such that f h D idP ; then
F ' Im.h/ Ker.f / ' P Ker.f /;
and so P is a direct summand of F . We conclude: the projective A-modules are exactly the
direct summands of free A-modules.

Finitely presented modules

D EFINITION 12.2. An A-module M is finitely presented if there exists an exact sequence
Am ! An ! M ! 0, some m; n 2 N.
A finite
P family .ei /i 2I of generators for an A-module M defines a homomorphism
.ai / 7! i 2I ai ei W AI ! M . The elements of the kernel of this homomorphism are called
the relations between the generators. Thus, M is finitely presented if it admits a finite family
of generators whose module of relations is finitely generated. Obviously
finitely presented ) finitely generated,
and the converse is true when A is noetherian (by 3.4).
P ROPOSITION 12.3. If M is finitely presented, then the kernel of every surjective homomorphism Am ! M , m 2 N, is finitely generated.
In other words, if M is finitely presented, then the module of relations for every finite
generating set is finitely generated.
P ROOF. Let An ! M be a surjective homomorphism with finitely generated kernel N . We
have to show that the kernel N 0 of every other surjective homomorphism Am ! M is finitely
generated. Consider the diagram:
0

An

N
f

N0

Am

idM

The map g exists because An is projective, and it induces the map f . From the diagram, we
get an exact sequence
f

N ! N 0 ! Am =gAn ! 0,
either from the snake lemma or by a direct diagram chase. The image of N in N 0 is finitely
generated, and so N 0 is an extension of finitely generated modules. Therefore it is finitely
generated (3.3(b)).

12

FINITELY GENERATED PROJECTIVE MODULES

51

P ROPOSITION 12.4. A finitely generated projective module is finitely presented.

P ROOF. Let M be finitely generated and projective. There exists a surjective homomorphism
An ! M (because M is finitely generated), whose kernel N is a direct summand of An
(because M is projective). As N is a quotient of An , it is finitely generated.

Finitely generated projective modules

According to the above discussion, the finitely generated projective modules are exactly the
direct summands of free A-modules of finite rank.
T HEOREM 12.5. The following conditions on an A-module are equivalent:
(a) M is finitely generated and projective;
(b) M is finitely presented and Mm is a free Am -module for all maximal ideals m of A;
(c) there exists a finite family .fi /i 2I of elements of A generating the ideal A and such
that, for all i 2 I , the Afi -module Mfi is free of finite rank;
(d) M is finitely presented and flat.
Moreover, when A is an integral domain and M is finitely presented, they are equivalent to:
(e) dimk.p/ .M A .p// is the same for all prime ideals p of A (here .p/ denotes the
field of fractions of A=p).
P ROOF. (a))(d). Free modules are flat, and direct summands of flat modules are flat (11.3).
Therefore, projective modules are flat, and we know (12.4) that finitely generated projective
modules are finitely presented.
(b))(c). Let m be a maximal ideal of A, and let x1 ; : : : ; xr be elements of M
whose images in Mm form a basis for Mm over Am . The kernel N 0 and cokernel N of the
homomorphism
X
W Ar ! M; g.a1 ; : : : ; ar / D
ai xi ;
are both finitely generated, and Nm0 D 0 D Nm . Therefore, there exists an f 2 A X m such
that Nf0 D 0 D Nf (5.13). Now becomes an isomorphism when tensored with Af .
The set T of elements f arising in
Pthis way is contained in no maximal ideal, and so
generates the ideal A. Therefore,
1 D i2I ai fi for certain ai 2 A and fi 2 T .
Q
(c))(d).QLet B D i 2I Afi . Then B is faithfully flat over A (Exercise 11.17), and
B A M D Mfi , which is clearly a flat B-module. It follows that M is a flat A-module
(apply 11.7).
(c))(e). This is obvious.
(e))(c): Fix a prime ideal p of A. For some f p, there exist elements x1 ; : : : ; xr of
Mf whose images in M A k.p/ form a basis. Then the map
P
W Arf ! Mf ; .a1 ; : : : ; ar / D ai xi ;
defines a surjection Arp ! Mp (Nakayamas lemma; note that .p/ ' Ap =pAp ). The cokernel
N of is finitely generated, and so gN D 0 for some g 2 A X p. The map becomes
surjective once f has been replaced fg. For any prime ideal q of Af , the map k.q/r !
M A k.q/ defined by is surjective, and hence is an isomorphism because dim.M A
k.q// D r. Thus Ker./  qArf for every q, which implies that it is zero as Af is reduced.
Therefore Mf is free. As in the proof of (b), a finite set of such f s will generate A.

12

FINITELY GENERATED PROJECTIVE MODULES

52

To prove the remaining implications, (d))(a) and (b) we shall need the following
lemma.
L EMMA 12.6. Let
0!N !F !M !0

(31)

be an exact sequence of A-modules with N a submodule of F .

(a) lf M and F are flat over A, then N \ aF D aN (inside F ) for all ideals a of A.
(b) Assume
that F is free with basis .yi /i 2I and that M is flat. If
P
Pthe element n D
i 2I ai yi of F lies in N , then there exist ni 2 N such that n D
i 2I ai ni :
(c) Assume that M is flat and F is free. For every finite set fn1 ; : : : ; nr g of elements of
N , there exists an A-linear map f W F ! N with f .nj / D nj ; j D 1; : : : , r.
P ROOF. (a) Consider
aN

aF

aM

'

N \ aF

'

aF

aM

The first row is obtained from (31) by tensoring with a, and the second row is a subsequence
of (31). Both rows are exact. On tensoring a ! A with F we get a map a F ! F , which is
injective because F is flat. Therefore a F ! aF is an isomorphism. Similarly, a M !
aM is an isomorphism. From the diagram we get a surjective map a N ! N \ aF , and
so the image of a N in aF is N \ aF . But this image is aN .
(b) Let a be the ideal
P generated by the ai . Then n 2 N \ aF D aN , and so there are
ni 2 N such that n D ai ni :
(c) We use induction on r. Assume first that r D 1, and write
P
n1 D i 2I0 ai yi
where .yi /i 2I is a basis for F and I0 is a finite subset of I . Then
P
n1 D i 2I0 ai n0i
for some n0i 2 N (by (b)), and f may be taken to be the map such that f .yi / D n0i for i 2 I0
and f .yi / D 0 otherwise. Now suppose that r > 1, and that there are maps f1 ; f2 : F ! N
such that f1 .n1 / D n1 and
f2 .ni

f1 .ni // D ni

f1 .ni /;

i D 2; : : : r:

Then
f W F ! N;

f D f1 C f2

f2 f1

has the required property.

We now complete the proof of the theorem.
(d))(a). Because M is finitely presented, there is an exact sequence
0!N !F !M !0

12

FINITELY GENERATED PROJECTIVE MODULES

53

in which F is free and N and F are both finitely generated. Because M is flat, (c) of the
lemma shows that this sequence splits, and so M is projective.
(d))(b): We may suppose that A itself is local, with maximal ideal m. Let x1 ; : : : ; xr 2
M be such that their images in M=mM form a basis for this over the field A=m. Then the
xi generate M (by Nakayamas lemma 3.9), and so there exists an exact
g

0!N !F !M !0
in which F is free with basis fy1 ; : : : ; yr g and g.yi / D xi . According to (a) of the lemma,
mN D N \ .mF /, which equals N because N  mF . Therefore N is zero by Nakayamas
lemma 3.9.
E XAMPLE 12.7. (a) When regarded as a Z-module, Q is flat but not projective (it is not
finitely generated, much less finitely presented, and so this doesnt contradict the theorem).
(b) Let R be a product of copies of F2 indexed by N, and let a be the ideal in R consisting
of the elements .an /n2N such that an is nonzero for only finitely many values of n (so a is a
direct sum of copies of F2 indexed by N). The R-module R=a is finitely generated and flat,
but not projective (it is not finitely presented).
A SIDE 12.8. The equivalence of (a) and (c) in the theorem has the following geometric interpretation:
for an affine scheme X, the functor M
M.X / is an equivalence from the category of locally free
OX -modules of finite rank to the category of finitely generated OX .X /-modules. (See Section 50 of
J.-P. Serre, Ann. of Math. (2) 61 (1955), 197278. This is also where Serre asked whether a finitely
generated projective kX1 ; : : : ; Xn -module is necessarily free. That it is was proved (independently)
by Quillen and Suslin. For a beautiful exposition of Quillens proof, see A. Suslin, Quillens solution
of Serres problem. J. K-Theory 11 (2013), 549552.)
A SIDE 12.9. Nonfree projective finitely generated modules are common: for example, the ideals
in a Dedekind domain are projective and finitely generated, but they are free only if principal. The
situation with modules that are not finitely generated is quite different: if A is a noetherian ring with
no nontrivial idempotents, then every nonfinitely generated projective A-module is free (Bass, Hyman.
Big projective modules are free. Illinois J. Math. 7 1963, 2431, Corollary 4.5). The condition on the
idempotents is needed because, for a ring A  B, the module A.I /  B .J / is not free when the sets I
and J have different cardinalities.

Duals
The dual HomA-linear .M; A/ of an A-module M is denoted M _ .
P ROPOSITION 12.10. For any A-modules M , S , T with M finitely generated and projective, the canonical maps
HomA-linear .S; T A M / ! HomA-linear .S A M _ ; T /
_

T A M ! HomA-linear .M ; T /
_

M T

! .M T /

M !M
are isomorphisms.

__

(32)
(33)
(34)
(35)

12

FINITELY GENERATED PROJECTIVE MODULES

54

P ROOF. The canonical map (32) sends f W S ! T A M to the map f 0 W S A M _ ! T

such that f 0 .s g/ D .T g/.f .s//. It becomes the canonical isomorphism
HomA-linear .S; T n / ! HomA-linear .S n ; T /
when M D An . It follows that (32) is an isomorphism whenever M is a direct summand of
a finitely generated free module, i.e., whenever M is finitely generated and projective.
The canonical map (33) sends t m to the map f 7! f .m/t. It is the special case of
(32) in which S D A.
The canonical map (34) sends f g 2 M _ T _ to the map mt 7! f .m/g.t /W M
T ! A, and the canonical map (35) sends m to the map f 7! f .m/W M _ ! A. Again, it is
obviously an isomorphism if one of M or T is free of finite rank, and hence also if one is a
direct summand of such a module.
We let evW M _ A M ! A denote the evaluation map f m 7! f .m/.
L EMMA 12.11. Let M and N be modules over commutative ring A, and let eW N A M !
A be an A-linear map. There exists at most one A-linear map W A ! M A N such that the
composites
M
N

M e

eN

! M N M

N M N

! M

(36)

are the identity maps on M and N respectively. When such a map exists,
T A N ' HomA-linear .M; T /

(37)

for all A-modules T . In particular,

.N; e/ ' .M _ ; ev/.

(38)

P ROOF. From e we get an A-linear map

T eW T A N A M ! T;
which allows us to define an A-linear map
x 7! fx W T A N ! HomA-lin .M; T /

(39)

by setting
fx .m/ D .T e/.x m/;

x 2 T A N , m 2 M .

An A-linear map f W M ! T defines a map f N W M A N ! T A N , and so a map

W A ! M A N defines an A-linear map
f 7! .f N /..1//W HomA-lin .M; T / ! T A N:

(40)

When the first (resp. the second) composite in (36) is the identity, then (40) is a right

12

FINITELY GENERATED PROJECTIVE MODULES

55

(resp. a left) inverse to (39).15 Therefore, when a map exists with the required properties,
the map (39) defined by e is an isomorphism. In particular, e defines an isomorphism
x 7! fx W M A N ! HomA-lin .M; M /;
which sends .a/ to the endomorphism x 7! ax of M . This proves that is unique.
To get (38), take T D M in (37).
Let A ! B be a ring homomorphism. Let eW N A M ! A be an A-linear map. Because
def
the functor M
MB D B A M commutes with tensor products (see p.40), W A !
M A N satisfies the conditions of (12.11) relative to e, then B W B ! MB B NB satisfies
the conditions of (12.11) relative to eB .
P ROPOSITION 12.12. An A-module M is finitely generated and projective if and only if
there exists an A-linear map W A ! M M _ such that

.M ev/ . M / D idM
(41)
.M _ / .ev M _ / D idM _ :
_
P ROOF. (H: On taking
A M ' End.M / (A-linear
P T D M in (37), we see that M P
endomorphisms). If i 2I fi mi corresponds to idM , so that i 2I fi .m/mi D m for all
m 2 M , then
P

m7!.fi .m//

! AI

.ai /7!

ai mi

!M

is a factorization of idM . Therefore M is a direct summand of a free module of finite rank.

H) : Suppose first that M is free with finite basis .ei /i 2I , and let .ei0 /i 2I be the dual
basis of M _ . The linear map W A ! M M _ , 1 7! ei ei0 , satisfies the conditions (41).
Moreover, it is the unique map satisfying (41) see (12.11). In particular, it is independent
of the choice of ei .
For the general case, we choose a family .fi /1i m as in (12.5c). In particular, Mfi is
a free Afi -module, and so is defined for each module Mfi ; the uniqueness assertion in
Lemma 12.11 then implies that the for the different Mfi patch together to give a for M .
15 Assume

satisfies the condition in the statement of the lemma.

Let x 2 T A N ; by definition, .fx N /..1// D .T e N /.x .1//. On tensoring the second sequence
in (36) with T , we obtain maps
T A N ' T A N A A

T N

! T A N A M A N

T eN

! T A N

whose composite is the identity map on T A N . As x D x 1 maps to x .1/ under T N , this shows
that .fx N /..1// D x.
Let f 2 HomA-lin .M; T /, and consider the commutative diagram
T A N A M

T e

f N M

M A N A M

M e

M:

For m 2 M , the two images of .1/ m in T are f .m/ and f.f N /..1// .m/, and so f D f.f N /..1// .

12

FINITELY GENERATED PROJECTIVE MODULES

56

In more detail, consider the diagram

Y
A
Afi

1i m

Afi fj

1i;j m
.i /

M M_

.M M _ /fi

1i m

.M M _ /fi fj :

1i;j m

In the top row, the first arrow sends a to .ai / with ai equal to the image of a in Afi , and the
upper arrow (resp. lower arrow) sends .ai / to .ai;j / with ai;j equal to the image of ai in
Afi fj (resp. the image of aj in Afi fj ). The bottom row is obtained from the top row by
tensoring with M M _ . The vertical map .i / is the product of the (unique) maps satisfying
(41). The vertical
Q map atQright can be described as the extension of scalars of .i / via the
upper arrow i Afi ! i;j Afi fj or the extension of scalars
of .i / via the lower arrow
Q
they are the same because they both equal the uniqueQ i;j Afi fj -linear
map satisfying
Q
the condition (41). As A and M are the submodules of i Afi and i Mfi on which the
pairs of arrows agree (Exercise 11.17), the map .i / induces an A-linear A ! M M _ ,
which satisfies (41). [This argument becomes more transparent when expressed in terms of
sheaves.]
A SIDE 12.13. A module M over a ring A is said to be reflexive if the canonical map M ! M __ is
an isomorphism. We have seen that for finitely generated modules projective implies reflexive,
but the converse is false. In fact, for a finite generated module M over an integrally closed noetherian
integral domain A, the following are equivalent (Bourbaki AC, VII, 4, 2):
(a) M is reflexive;
(b) M is torsion-free and equals the intersection of its localizations at the prime ideals of A of
height 1;
(c) M is the dual of a finitely generated module.
For noetherian rings of global dimension  2, for example, for regular local rings of Krull
dimension  2, every finitely generated reflexive module is projective: for every finitely generated
module M over a noetherian ring A, there exists an exact sequence
Am ! An ! M ! 0
with m; n 2 N; on taking duals and forming the cokernel, we get an exact sequence
0 ! M _ ! An ! Am ! N ! 0I
if A has global dimension  2, then M _ is projective, and if M is reflexive, then M ' .M _ /_ .
A SIDE 12.14. For a finitely generated torsion-free module M over an integrally closed noetherian
integral domain A, there exists a free submodule L of M such that M=L is isomorphic to an ideal a in
A (Bourbaki AC, VII, 4, Thm 6). When A is Dedekind, every ideal is projective, and so M ' L a.
In particular, M is projective. Therefore, the finitely generated projective modules over a Dedekind
domain are exactly the finitely generated torsion-free modules.

S UMMARY 12.15. Here is a summary of the assumptions under which the canonical morphisms of A-modules below are isomorphisms. If P is finitely generated projective:
'

P ! P __

13

ZARISKIS LEMMA AND THE HILBERT NULLSTELLENSATZ

57

A module P is finitely generated projective if and only if the following canonical map is an
isomorphism
'

P _ P ! End.P /:
If P or P 0 is finitely generated projective:
'

P _ P 0 ! Hom.P; P 0 /:
If both P and P 0 or both P and M or both P 0 and M 0 are finitely generated projective
'

Hom.P; M / Hom.P 0 ; M 0 / ! Hom.P P 0 ; M M 0 /:

In particular, for P or P 0 finitely generated projective
'

P _ P 0_ ! .P P 0 /_ :
(Georges Elencwajg on mathoverflow.net).

13

Zariskis lemma and the Hilbert Nullstellensatz

Zariskis lemma
In proving Zariskis lemma, we shall need to use that the ring kX contains infinitely many
distinct monic irreducible polynomials. When k is infinite, this is obvious, because the
polynomials X a, a 2 k, are distinct and irreducible. When k is finite, we can adapt Euclids
argument: if p1 ; : : : ; pr are monic irreducible polynomials in kX , then p1


pr C 1 is
divisible by a monic irreducible polynomial distinct from p1 ; : : : ; pr .
T HEOREM 13.1 (Z ARISKI S L EMMA ). Let k  K be fields. If K is finitely generated as a
k-algebra, then it is algebraic over k (hence finite over k, and it equals k if k is algebraically
closed).
P ROOF. We shall prove this by induction on r, the smallest number of elements required to
generate K as a k-algebra. The case r D 0 being trivial, we may suppose that
K D kx1 ; : : : ; xr with r  1:
If K is not algebraic over k, then at least one xi , say x1 , is not algebraic over k. Then, kx1
is a polynomial ring in one symbol over k, and its field of fractions k.x1 / is a subfield of
K. Clearly K is generated as a k.x1 /-algebra by x2 ; : : : ; xr , and so the induction hypothesis
implies that x2 ; : : : ; xr are algebraic over k.x1 /. According to Proposition 6.7, there exists a
c 2 kx1 such that cx2 ; : : : ; cxr are integral over kx1 .
Let f 2 k.x1 /. Then f 2 K D kx1 ; : : : ; xr and so, for a sufficiently large N , c N f 2
kx1 ; cx2 ; : : : ; cxr . Therefore c N f is integral over kx1 by (6.5), which implies that
c N f 2 kx1 because kx1 is integrally closed in k.x1 / (6.10). But this contradicts the
fact that kx1 (' kX ) has infinitely many distinct monic irreducible polynomials that can
occur as denominators of elements of k.x1 /. Hence K is algebraic over k.
C OROLLARY 13.2. Let A be a finitely generated k-algebra. Every maximal ideal in A is
the kernel of a homomorphism from A into a finite field extension of k.

13

ZARISKIS LEMMA AND THE HILBERT NULLSTELLENSATZ

58

P ROOF. Indeed, A=m itself is a finite field extension of k.

C OROLLARY 13.3. Let k  K  A be k-algebras with K a field and A finitely generated
over k. Then K is algebraic over k.
P ROOF. Let m be a maximal ideal in A. Then m \ K D .0/, and so k  K  A=m. The
theorem shows that the field A=m is algebraic over k, and hence K is also.
A SIDE 13.4. There is a very short proof of Zariskis lemma when k is uncountable. Let k  K be
fields. If K is finitely generated as a k-algebra, then its dimension as a k-vector space is countable.
On the other hand, if x 2 K is transcendental over k, then the elements x 1 c , c 2 k, are linearly
independent (assume a linear relation, and clear denominators). When k is uncountable, this gives a
contradiction.

Alternative proof of Zariskis lemma

The following is a simplification of Swans simplication16 of a proof of Munshi.
L EMMA 13.5. For an integral domain A, there does not exist an f 2 AX such that AX f
is a field.
P ROOF. Suppose, on the contrary, that AX f is a field. Then f A, and we can write
.f 1/ 1 D g=f n with g 2 AX and n  1. Then
.f
with h 2 AX , and so .f
(it has degree  1).

1/g D f n D .1 C .f
1/.g

1//n D 1 C .f

h/ D 1. Hence f

1/h

1 is a unit in AX , which is absurd

P ROPOSITION 13.6. Let A be an integral domain, and suppose that there exists a maximal
ideal m in AX1 ; : : : ; Xn such that A \ m D .0/. Then there exists a nonzero a 2 A such
that Aa is a field and AX1 ; : : : ; Xn =m is a finite extension of Aa .
Note that the condition A \ m D .0/ implies that A (hence also Aa ) is a subring of the field
def
K D AX1 ; : : : ; Xn =m, and so the statement makes sense.
P ROOF. We argue by induction on n. When n D 0, the hypothesis is that .0/ is a maximal
ideal in A; hence A is a field, and the statement is trivial. Therefore, suppose that n  1, and
regard AX1 ; : : : ; Xn as a polynomial ring in n 1 symbols over AXi for some fixed i . If
m \ AXi D .0/, then, by induction, there exists an f 2 AXi such that AXi f is a field,
contradicting Lemma 13.5. We conclude that, for each i , there exists a nonzero element
ai Xini C



in m \ AXi . The image xi of Xi in K satisfies the equation
ai xin C


D 0;
and so K is integral over its subring Aa1


an . This implies that Aa1


an is a field (see 7.1),
and K is finite over it because it is integral and finitely generated (6.3).
We now prove Zariskis lemma. Let m be a maximal ideal in kX1 ; : : : ; Xn . Then
k \ m D .0/ because k is a field. According to the proposition, there exists a nonzero a 2 k
such that kX1 ; : : : ; Xn =m is a finite extension of ka , but, because k is a field, ka D k.
16 For

a leisurely exposition of Munshis proof, see May, J. Peter, Munshis proof of the Nullstellensatz. Amer.
Math. Monthly 110 (2003), no. 2, 133140.

13

ZARISKIS LEMMA AND THE HILBERT NULLSTELLENSATZ

59

The Nullstellensatz
Recall that k al denotes an algebraic closure of the field k.
T HEOREM 13.7 (N ULLSTELLENSATZ ). Every proper ideal a in kX1 ; : : : ; Xn has a zero
in .k al /n , i.e., there exists a point .a1 ; : : : ; an / 2 .k al /n such that f .a1 ; : : : ; an / D 0 for all
f 2 a.
P ROOF. We have to show that there exists a k-algebra homomorphism kX1 ; : : : ; Xn ! k al
containing a in its kernel. Let m be a maximal ideal containing a. Then kX1 ; : : : ; Xn =m is
a field, which is finitely generated as a k-algebra. Therefore it is finite over k by Zariskis
lemma, and so there exists a k-algebra homomorphism kX1 ; : : : ; Xn =m ! k al . The composite of this with the quotient map kX1 ; : : : ; Xn ! kX1 ; : : : ; Xn =m contains a in its
kernel.
C OROLLARY 13.8. When k is algebraically closed, the maximal ideals in kX1 ; : : : ; Xn
are exactly the ideals .X1 a1 ; : : : ; Xn an /, .a1 ; : : : ; an / 2 k n .
P ROOF. Clearly, kX1 ; : : : ; Xn =.X1 a1 ; : : : ; Xn an / ' k, and so .X1 a1 ; : : : ; Xn
an / is maximal. Conversely, because k is algebraically closed, a maximal ideal m of
kX1 ; : : : ; Xn has a zero .a1 ; : : : ; an / in k n . Let f 2 kX1 ; : : : ; Xn ; when we write f as a
polynomial in X1 a1 ; : : : ; Xn an , its constant term is f .a1 ; : : : ; an /. Therefore
f 2 m H) f 2 .X1
and so m D .X1

a1 ; : : : ; Xn

a1 ; : : : ; Xn

an /,

an /.

T HEOREM 13.9 (S TRONG N ULLSTELLENSATZ ). For an ideal a in kX1 ; : : : ; Xn , let Z.a/

be the set of zeros of a in .k al /n . If a polynomial h 2 kX1 ; : : : ; Xn is zero on Z.a/, then
some power of h lies in a.
P ROOF. 17 We may assume h 0. Let g1 ; : : : ; gm generate a, and consider the system of
m C 1 equations in n C 1 variables, X1 ; : : : ; Xn ; Y;

gi .X1 ; : : : ; Xn / D 0; i D 1; : : : ; m
1 Y h.X1 ; : : : ; Xn / D 0:
If .a1 ; : : : ; an ; b/ satisfies the first m equations, then .a1 ; : : : ; an / 2 Z.a/; consequently,
h.a1 ; : : : ; an / D 0, and .a1 ; : : : ; an ; b/ doesnt satisfy the last equation. Therefore, the equations are inconsistent, and so, according to the Nullstellensatz (13.7), the ideal
.g1 ; : : : ; gm ; 1
17 This

Y h/ D kX1 ; : : : ; Xn ; Y :

argument is known as Rabinowitschs trick (J. L. Rabinowitsch, Zum Hilbertschen Nullstellensatz,

Math. Ann. 102 (1930), p.520). Rabinowitsch simplified his name to Rainich. He was a faculty member
at the University of Michigan from 19251956, where the following story is folklore: Rainich was giving a
lecture in which he made use of a clever trick which he had discovered. Someone in the audience indignantly
interrupted him pointing out that this was the famous Rabinowitsch trick and berating Rainich for claiming to
have discovered it. Without a word Rainich turned to the blackboard and wrote RABINOWITSCH. He then
began erasing letters. When he was done what remained was RA IN I CH. He then went on with his lecture. See
also mo45185.

14

THE SPECTRUM OF A RING

60

This means that there exist fi 2 kX1 ; : : : ; Xn ; Y such that

1D

m
X

fi
gi C fmC1
.1

Y h/.

(42)

i D1

On applying the homomorphism


Xi 7! Xi
W kX1 ; : : : ; Xn ; Y ! k.X1 ; : : : ; Xn /
Y 7! h 1
to (42), we obtain the identity
X
1D
fi .X1 ; : : : ; Xn ; h
i

/
gi .X1 ; : : : ; Xn /

(43)

in k.X1 ; : : : ; Xn /. Clearly
fi .X1 ; : : : ; Xn ; h

/D

polynomial in X1 ; : : : ; Xn
hNi

for some Ni . Let N be the largest of the Ni . On multiplying (43) by hN we obtain an

identity
X
hN D
(polynomial in X1 ; : : : ; Xn /
gi .X1 ; : : : ; Xn /;
i

which shows that

hN

2 a.

P ROPOSITION 13.10. The radical of an ideal a in a finitely generated

k-algebra A is equal
T
to the intersection
of
the
maximal
ideals
containing
it:
rad.a/
D
m.
In particular, if A
ma
T
is reduced, then m maximal m D 0.
P ROOF. Because of the correspondence between the ideals in a ring and in a quotient of the
ring ((2), p. 4), it suffices to T
prove this for A D kX1 ; : : : ; Xn .
The inclusion rad.a/  ma m holds in any ring (because maximal ideals are radical
and rad.a/ is the smallest radical ideal containing a).
For the reverse inclusion, let h lie in all maximal ideals containing a, and let .a1 ; : : : ; an / 2
Z.a/. The image of the evaluation map
f 7! f .a1 ; : : : ; an /W kX1 ; : : : ; Xn ! k al
is a subring of k al which is algebraic over k, and hence is a field (see 1). Therefore,
the kernel of the map is a maximal ideal, which contains a, and therefore also contains h.
This shows that h.a1 ; : : : ; an / D 0, and we conclude from the strong Nullstellensatz that
h 2 rad.a/.

14

The spectrum of a ring

Definition
Let A be a ring, and let V be the set of prime ideals in A. For an ideal a in A, let
V .a/ D fp 2 V j p  ag:

14

THE SPECTRUM OF A RING

61

P ROPOSITION 14.1. There are the following relations:

(a) a  b H) V .a/  V .b/I
(b) V .0/ D V ; V .A/ D ;I
(c) V .ab/
b/ D V .a/ [ V .b/I
P D V .a \T
(d) V . i 2I ai / D i 2I V .ai / for every family of ideals .ai /i 2I .
P ROOF. The first two statements are obvious. For (c), note that
ab  a \ b  a; b H) V .ab/  V .a \ b/  V .a/ [ V .b/:
For the reverse inclusions, observe that if p V .a/ [ V .b/, then there exist an f 2 a X pPand
a g 2 b X p; but then fg 2 ab X p, andP
so p V .ab/. For (d) recall that, by definition, ai
consists of all finite sums of the form fi , fi 2 ai . Thus (d) is obvious.
Statements (b), (c), and (d) show that the sets V .a/ satisfy the axioms to be the closed
subsets for a topology on V : both the whole space and the empty set are closed; a finite
union of closed sets is closed; an arbitrary intersection of closed sets is closed. This topology
is called the Zariski topology on V . We let spec.A/ denote the set of prime ideals in A
endowed with its Zariski topology.
For h 2 A, let
D.h/ D fp 2 V j h pg.
Then D.h/ is open in V , being the complement of V ..h//. If S is a set of generators for an
ideal a, then
[
V X V .a/ D
D.h/;
h2S

and so the sets D.h/ form a base for the topology on V . Note that
D.h1


hn / D D.h1 / \


\ D.hn /:
For every element h of A, spec.Ah / ' D.h/ (see 5.4), and for every ideal a in A, spec.A/=a '
V .a/ (isomorphisms of topological spaces).

Idempotents and decompositions of spec.A/

An element e of a ring A is idempotent if e 2 D e. For example, 0 and 1 are both idempotents
they are called the trivial idempotents. Idempotents e1 ; : : : ; en are orthogonal if ei ej D 0
for i j . Every sum of orthogonal idempotents is again idempotent. A set fe1 ; : : : ; en g
of orthogonal idempotents is complete if e1 C


C en D 1. Every finite set of orthogonal
idempotents fe1 ; : : : ; en g can be made into a complete set of orthogonal idempotents by
adding the idempotent e D 1 .e1 C


C en /.
L EMMA 14.2. The topological space spec .A/ is disconnected if and only if A contains a
nontrivial idempotent e, in which case
spec.A/ D D.e/ t D.1

e/:

P ROOF. Let e be a nontrivial idempotent, and let f D 1 e. For a prime ideal p, the map
A ! A=p must send exactly one of e or f to a nonzero element. This shows that spec A is a
disjoint union of the sets D.e/ and D.f /, each of which is open. If D.e/ D spec A, then

14

THE SPECTRUM OF A RING

62

e would be a unit (2.3), and hence can be cancelled from ee D e to give e D 1. Therefore
D.e/ spec A, and similarly, D.f / spec A.
Conversely, suppose that spec A is disconnected, say, the disjoint union of two nonempty
closed subsets V .a/ and V .b/. Because the union is disjoint, no prime ideal contains both
a and b, and so a C b D A. Thus a C b D 1 for some a 2 a and b 2 b. As ab 2 a \ b, all
prime ideals contain ab, which is therefore nilpotent (2.5), say .ab/m D 0. Any prime ideal
containing am contains a; similarly, any prime ideal containing b m contains b; thus no prime
ideal contains both am and b m , which shows that .am ; b m / D A. Therefore, 1 D ram C sb m
for some r; s 2 A. Now
.ram /.sb m / D rs.ab/m D 0;
.ram /2 D .ram /.1

sb m / D ram ,

.sb m /2 D sb m
ram C sb m D 1;
and so fram ; sb m g is a complete set of orthogonal idempotents. Clearly V .a/  V .ram /
and V .b/  V .sb m /. As V .ram / \ V .sb m / D ;, we see that V .a/ D V .ram / and V .b/ D
V .sb m /, and so each of ram and sb m is a nontrivial idempotent.
Let U be an open and closed subset of spec.A/. The proof of the lemma shows that
U D D.e/ for some idempotent e 2 A. Let U 0 D spec.A/ X U . The image of e in O.U 0 /
lies in all prime ideals of O.U 0 /; hence is nilpotent; hence is 0. The image ex of e in O.U /
lies in no prime ideals of O.U /; hence 1 ex D 0; hence ex D 1. As spec.A/ D U [ U 0 , this
shows that e is uniquely determined by U .
P ROPOSITION 14.3. Let X D spec.A/. There are natural one-to-one correspondences
between the following objects.
(a) Decompositions
X D X1 t : : : t Xn
of X into a finite disjoint union of open subsets.
(b) Decompositions
A D A1 


 An
of A into a finite product of rings (Ai  A).
(c) Decompositions
1 D e1 C


C en
of 1 into the sum of a complete sets fe1 ; : : : ; en g of orthogonal idempotents in A.
The sets Xi in (a) are connected no ring Ai in (b) has a nontrivial idempotent no
idempotent ei in (c) can be written as a sum of two nontrivial idempotents.
P ROOF. (b)$(c). If A D A1 


 An (direct product of rings), then the elements
i

ei D .0; : : : ; 1; : : : ; 0/;

1  i  n;

form a complete set of orthogonal idempotents in A. Conversely, if fe1 ; : : : ; en g is a complete

set of orthogonal idempotents in A, then Aei becomes a ring18 with the addition and
multiplication induced by that of A, and A ' Ae1 


 Aen .
18 But

Aei is not a subring of A if n 1 because its identity element is ei 1A : However, the map a 7!
aei W A ! Aei realizes Aei as a quotient of A.

14

THE SPECTRUM OF A RING

63

(c)$(a). Let fe1 ; : : : ; en g be a complete set of orthogonal idempotents, and let p be a

prime ideal in A. Because A=p is an integral domain, exactly one of the ei s maps to 1 in
A=p and the remainder map to zero. This proves that spec .A/ is the disjoint union of the
sets D.ei /.
Now consider a decomposition
spec .A/ D X1 t : : : t Xn
with each Xi open. We use induction on n to show that it arises from a unique complete set
of orthogonal idempotents. When n D 1, there is nothing to prove, and when n  2, we write
spec A D X1 t .X2 t : : : t Xn /.
From Lemma 14.2 et seq. we know that there exist unique orthogonal idempotents e1 , e10 2 A
such that e1 C e10 D 1 and
X1 D D.e1 /
X2 t : : : t Xn D D.e10 / D spec Ae10 :
By induction, there exist unique orthogonal idempotents e2 ; : : : ; en in the ring Ae10 such that
e2 C


C en D e10 and Xi D D.ei / for i D 2; : : : ; n. Now fe1 ; : : : ; en g is a complete set of
orthogonal idempotents in A such that Xi D D.ei / for all i .
(b)$(a). The ideals in a finite product of rings
 An are all of the form
Q A DQA1 


Q
a1 


 an with ai an ideal in Ai (cf. p.8). As i Ai = i ai ' i A=ai , we see that the
prime ideals are those of the form
A1 


 Ai

1  ai

 AiC1 


 An

F
with ai prime. It follows that spec.A/ D i spec.Ai / (disjoint union of open subsets).
Let spec .A/ D X1 t : : : t Xn , and let 1 D e1 C


C enQ
be the corresponding decomposition of 1. Then OX .Xi / ' OX .X /ei , and so OX .X / ' i OX .Xi /.

Properties of spec.A/
We study more closely the Zariski topology on spec.A/. For each subset S of A, let V .S /
denote the set of prime ideals containing S , and for each subset W of spec.A/, let I.W /
denote the intersection of the prime ideals in W :
S  A;
W  spec.A/;

V .S / D fp 2 spec.A/ j S  pg;
\
I.W / D
p:
p2W

Thus V .S / is a closed subset of spec.A/ and I.W / is a radical ideal in A. If V .a/  W , then
a  I.W /, and so V .a/  V I.W /. Therefore V I.W / is the closure of W (smallest closed
subset of spec.A/ containing W ); in particular, V I.W / D W if W is closed.
P ROPOSITION 14.4. Let V be a closed subset of spec.A/.
(a) There is an order-inverting one-to-one correspondence W $ I.W / between the closed
subsets of spec.A/ and the radical ideals in A.
(b) The closed points of V are exactly the maximal ideals in V .
(c) Every open covering of V has a finite subcovering.

14

THE SPECTRUM OF A RING

64

(d) If A is noetherien, then every ascending chain of open subsets U1  U2 


of V

eventually becomes constant; equivalently, every descending chain of closed subsets
of V eventually becomes constant.
P ROOF. (a) and (b) are obvious.
(c) Let .Ui /i 2I be an open covering of spec.A/. On covering each Ui with basic
open subsets,
S we get a covering .D.hj //j 2J of spec.A/ by basic open subsets. Because
spec.A/ D j D.hj /, the ideal generated by
S the hj is A, and so 1 D a1 hj1 C


C am hjm
for some a1 ; : : : ; am 2 A. Now spec.A/ D 1lm D.hjl /, and it follows that spec.A/ is
covered by finitely many of the sets Ui .
(d) We prove the second statement. A sequence V1  V2 


of closed subsets of V
gives rise to a sequence of ideals I.V1 /  I.V2 /  : : :, which eventually becomes constant.
If I.Vm / D I.VmC1 /, then V I.Vm / D V I.VmC1 /, i.e., Vm D VmC1 .
A topological space V having property (c) is said to be quasi-compact (by Bourbaki at
least; others call it compact, but Bourbaki requires a compact space to be Hausdorff). A
topological space V having the property in (d) is said to be noetherian. This condition is
equivalent to the following: every nonempty set of closed subsets of V has a minimal element.
Clearly, noetherian spaces are quasi-compact. Since an open subspace of a noetherian space
is again noetherian, it will also be quasi-compact.
D EFINITION 14.5. A nonempty topological space is said to be irreducible if it is not the
union of two proper closed subsets.
Equivalent conditions: any two nonempty open subsets have a nonempty intersection;
every nonempty open subset is dense.
If an irreducible space W is a finite union of closed subsets, W D W1 [ : : : [ Wr , then
W D W1 or W2 [ : : : [ Wr ; if the latter, then W D W2 or W3 [ : : : [ Wr , etc.. Continuing in
this fashion, we find that W D Wi for some i.
The notion of irreducibility is not useful for Hausdorff topological spaces, because the
only irreducible Hausdorff spaces are those consisting of a single point two points would
have disjoint open neighbourhoods.
P ROPOSITION 14.6. A closed subset W of spec.A/ is irreducible if and only if I.W / is
prime. In particular, the spectrum of a ring A is irreducible if and only if the nilradical of A
is prime.
P ROOF. ): Let W be an irreducible closed subset of spec.A/, and suppose that fg 2 I.W /.
Then fg lies in each p in W , and so either f 2 p or g 2 p; hence W  V .f / [ V .g/, and so
W D .W \ V .f // [ .W \ V .g//:
As W is irreducible, one of these sets, say W \ V .f /, must equal W . But then f 2 I.W /.
We have shown that I.W / is prime.
(: Assume I.W / is prime, and suppose that W D V .a/ [ V .b/ with a and b radical
ideals we have to show that W equals V .a/ or V .b/. Recall that V .a/ [ V .b/ D V .a \ b/
(see 14.1c) and that a \ b is radical; hence I.W / D a \ b (by 15.3). If W V .a/, then there
exists an f 2 a X I.W /. For all g 2 b,
fg 2 a \ b D I.W /:
Because I.W / is prime, this implies that b  I.W /; therefore W  V .b/.

14

THE SPECTRUM OF A RING

65

Thus, in the spectrum of a ring, there are one-to-one correspondences

radical ideals $ closed subsets
prime ideals $ irreducible closed subsets
maximal ideals $ one-point sets:
E XAMPLE 14.7. Let f 2 kX1 ; : : : ; Xn . According to Theorem 4.10, kX1 ; : : : ; Xn is a
unique factorization domain, and so .f / is a prime ideal if and only if f is irreducible (4.1).
Thus
V .f / is irreducible f is irreducible.
On the other hand, suppose that f factors as
Y m
f D
fi i ; fi distinct irreducible polynomials.
Then
\ m
.fi i /; .fimi / distinct ideals,
\
rad..f // D .fi /; .fi / distinct prime ideals,
[
V .f / D
V .fi /; V .fi / distinct irreducible algebraic sets.
.f / D

P ROPOSITION 14.8. Let V be a noetherian topological space. Then V is a finite union of

irreducible closed subsets, V D V1 [ : : : [ Vm . If the decomposition is irredundant in the
sense that there are no inclusions among the Vi , then the Vi are uniquely determined up to
order. The Vi are exactly the maximal irreducible subsets of V .
P ROOF. Suppose that V can not be written as a finite union of irreducible closed subsets.
Then, because V is noetherian, there will be a closed subset W of V that is minimal
among those that cannot be written in this way. But W itself cannot be irreducible, and so
W D W1 [ W2 , with each Wi a proper closed subset of W . Because W is minimal, both W1
and W2 can be expressed as finite unions of irreducible closed subsets, but then so can W .
We have arrived at a contradiction.
Suppose that
V D V1 [ : : : [ Vm D W1 [ : : : [ Wn
S
are two irredundant decompositions. Then Vi D j .Vi \ Wj /, and so, because Vi is
irreducible, Vi D Vi \ Wj for some j . Consequently, there exists a function f W f1; : : : ; mg !
f1; : : : ; ng such that Vi  Wf .i / for each i . Similarly, there is a function gW f1; : : : ; ng !
f1; : : : ; mg such that Wj  Vg.j / for each j . Since Vi  Wf .i /  Vgf .i / , we must have
gf .i / D i and Vi D Wf .i / ; similarly fg D id. Thus f and g are bijections, and the
decompositions differ only in the numbering of the sets.
Let W be a maximal irreducible subset of V . Then
W D .V1 \ W / [ : : : [ .Vm \ W /:
Each set Vi \ W is closed in W , and so W D Vi \ W for some i , i.e., W  Vi for some i .
Because W is maximal, it equals Vi .
The Vi given uniquely by the proposition are called the irreducible components of V .
In Example 14.7, the V .fi / are the irreducible components of V .f /.

14

THE SPECTRUM OF A RING

66

C OROLLARY 14.9. Every radical ideal a in a noetherian ring A is a finite intersection of

prime ideals, a D p1 \ : : : \ pn ; if there are no inclusions among the pi , then the pi are
uniquely determined up to order. Every prime ideal of A containing a contains some pi .
P ROOF. In view of the correspondence between radical (resp. prime) ideals in A and closed
(resp. irreducible closed) subsets in Spec.A/, this is a restatement of the proposition.
In particular, a noetherian ring has only finitely many minimal prime ideals, and their
intersection is the radical of the ring.
C OROLLARY 14.10. A noetherian topological space has only finitely many connected
components (each of which is open).
P ROOF. Each connected component is closed, hence noetherian, and so is a finite union of
its irreducible components. Each of these is an irreducible component of the whole space,
and so there can be only finitely many.
R EMARK 14.11. (a) An irreducible topological space is connected, but a connected topological space need not be irreducible. For example, Z.X1 X2 / is the union of the coordinate axes
in k 2 , which is connected but not irreducible. A closed subset V of spec.A/ is not connected
if and only if there exist proper ideals a and b such that a \ b D I.V / and a C b D A.
(b) A Hausdorff space is noetherian if and only if it is finite, in which case its irreducible
components are the one-point sets.
(c)
T In a noetherian ring, every proper ideal a has a decomposition into primary ideals:
a D qi (see 19). For radical ideals, this becomes a simpler decomposition
prime
Q minto
i
ideals, as in the corollary.
For
an
ideal
.f
/
in
kX
;
:
:
:
;
X

with
f
D
f
,
it
is the
1
n
i
T
decomposition .f / D .fimi / noted in Example 14.7.

Maps of spectra
Let 'W A ! B be a homomorphism of rings, and let p be a prime ideal of B. Then B=pB
is an integral domain and the map A=' 1 .p/ ! B=p is injective, and so ' 1 .p/ is a prime
ideal in A. Therefore ' defines a map
' a W Spec.B/ ! Spec.A/:
This map is continuous because .' a / 1 .D.f // D D.'.f //. In this way, Spec becomes a
contravariant functor from the category of commutative rings to the category of topological
spaces.
D EFINITION 14.12. A subset C of a noetherian topological space X is constructible if it
is a finite union of subsets of the form U \ Z with U open and Z closed.
The constructible subsets of An are those that can be defined by a finite number of
statements of the form
f .X1 ; : : : ; Xn / D 0
combined using only and, or, and not. This explains the name.
P ROPOSITION 14.13. Let C be a constructible set whose closure Cx is irreducible. Then C
contains a nonempty open subset of Cx .

14

THE SPECTRUM OF A RING

67

S
P ROOF. We are given that C D .Ui \ Zi / with each Ui open and each Zi closed.
S We
x
may assume that each set Ui \ Zi in this decomposition is nonempty. Clearly C  Zi ,
and as Cx is irreducible, it must be contained in one of the Zi . For this i
C  Ui \ Zi  Ui \ Cx  Ui \ C  Ui \ .Ui \ Zi / D Ui \ Zi :
Thus Ui \ Zi D Ui \ Cx is a nonempty open subset of Cx contained in C .
T HEOREM 14.14. Let A be a noetherian ring, and let 'W A ! B be a finitely generated
A-algebra. The map ' a W Spec.B/ ! Spec.A/ sends constructible sets to constructible sets.
In particular, if U is a nonempty open subset of Spec.B/, then ' a .U / contains a nonempty
open subset of its closure in Spec.A/.
P ROOF. The in particular statement of the theorem is proved for finitely generated kalgebras in (15.8) below and for noetherian rings in (21.11) below.
We now explain how to deduce the main statement of the theorem from the in particular
statement. Let X D Spec.A/ and Y D Spec.B/, and let C be a constructible subset of Y .
Let Yi be the irreducible components of Y . Then C \ Yi is constructible in Yi , and ' a .Y / is
the union of the ' a .C \ Yi /; it is therefore constructible if the ' a .C \ Yi / are. Hence we
may assume that Y is irreducible. Moreover, C is a finite union of its irreducible components,
and these are closed in C ; they are therefore constructible. We may therefore assume that C
also is irreducible; Cx is then an irreducible closed subvariety of Y .
We shall prove the theorem by induction on the dimension of Y . If dim.Y / D 0, then the
statement is obvious because Y is a point. If Cx Y , then dim.Cx / < dim.Y /, and because
C is constructible in Cx , we see that ' a .C / is constructible (by induction). We may therefore
assume that Cx D Y . But then Cx contains a nonempty open subset of Y , and so we know
that ' a .C / contains an nonempty open subset U of its closure. Replace X with the closure
of ' a .C /, and write
' a .C / D U [ ' a .C \ .' a / 1 .X X U //:
Then .' a / 1 .X X U / is a proper closed subset of Y (the complement of X U is dense in
X and ' a is dominant). As C \ .' a / 1 .X X U / is constructible in .' a / 1 .X X U /, the set
' a .C \ .' a / 1 .X X U // is constructible in X by induction, which completes the proof.
Let p and p0 be prime ideals in a ring A. If p  p0 (i.e., p0 2 V .p/), then we say that p0 is
a specialization of p and that p is a generalization of p0 .
P ROPOSITION 14.15. Let A be a noetherian ring, and let X D Spec.A/. A constructible
subset Z of X is closed if it is closed under specialization.
x and let p D I.W /; then W D V .p/, i.e.,
P ROOF. Let W be an irreducible component of Z,
W consists of the specializations of p. Then W \ Z is constructible and it is dense in W ,
and so it contains a nonempty open subset U of W (14.13). Hence p 2 U and, because Z
x it
is closed under specialization, W  Z. As Z contains all irreducible components of Z,
contains Z.
P ROPOSITION 14.16. Let A be a noetherian ring, and let 'W A ! B be a finitely generated
A-algebra. If ' satisfies the going-down theorem, then the map ' a W Spec.B/ ! Spec.A/ is
open (i.e., sends open subsets to open subsets).
P ROOF. Let U be an open subset of Spec.B/, Then ' a .U / is constructible (14.14), and the
going-down theorem says that it is closed under generalization. Therefore Spec.A/ X ' a .U /
is constructible and closed under specialization, and hence closed.

15

JACOBSON RINGS AND MAX SPECTRA

15

68

Jacobson rings and max spectra

D EFINITION 15.1. A ring A is Jacobson if every prime ideal in A is an intersection of

maximal ideals.
A field is Jacobson.
T The ring Z is Jacobson because every nonzero prime ideal is
maximal and .0/ D p prime .p/. A principal ideal domain (more generally, a Dedekind
domain) is Jacobson if it has infinitely many maximal ideals.19 A local ring is Jacobson if
and only if its maximal ideal is its only prime ideal.
P ROPOSITION 15.2. Every finitely generated algebra over a field is Jacobson.
P ROOF. Apply (13.10).
P ROPOSITION 15.3. In a Jacobson ring, the radical of an ideal is equal to the intersection
of the maximal ideals containing it. In particular, an element is nilpotent if it is contained in
all maximal ideals.
P ROOF. Proposition 2.5 says that the radical of an ideal is an intersection of prime ideals,
and so this follows from the definition of a Jacobson ring.
A SIDE 15.4. Every ring of finite type over a Jacobson ring is a Jacobson ring (EGA IV, 10.4.6).

Max spectra
Let A be ring. The set spm.A/ of maximal ideals in A acquires a topology in exactly the
same way as spec.A/. Namely, the closed sets for the topology are the subsets
V .a/ D fm j m  ag
of spm.A/ with a an ideal in A.
Everything in 14 holds, with essentially the same proofs, for the max spectra of Jacobson
rings. For example, in the proof of (14.2), we used that an element of A is nilpotent if it
is contained in all prime ideals. The is true with maximal for prime provided A is
Jacobson.
In particular, for a Jacobson ring A, there are natural one-to-one correspondences between
the decompositions of spm.A/ into a finite disjoint union of open subspaces,
the decompositions of A into a finite direct products of rings, and
the complete sets of orthogonal idempotents in A.
A SIDE 15.5. By definition, spm.A/ is the subspace of spec.A/ consisting of the closed points.
When A is Jacobson, the map U 7! U \ spm.A/ is a bijection from the set of open subsets of spec.A/
onto the set of open subsets of spm.A/; therefore spm.A/ and spec.A/ have the same topologies
only the underlying sets differ.
r

a principal ideal domain, a nonzero element a factors as a D up11


psrs with u a unit and the pi prime.
The only prime divisors of a are p1 ; : : : ; ps , and so a is contained in only finitely many prime ideals. Similarly,
r
in a Dedekind domain, a nonzero ideal a factors as a D p11


prss with the pi prime ideals (cf. 20.7 below), and
p1 ; : : : ; pr are the only prime ideals containing a. On taking a D .a/, we see that again a is contained in only
finitely many prime ideals.
19 In

15

JACOBSON RINGS AND MAX SPECTRA

69

A SIDE 15.6. Let k D R or C. Let X be a set and let A be a k-algebra of k-valued functions on X.
In analysis, X is called the spectrum of A if, for each k-algebra homomorphism 'W A ! k, there
exists a unique x 2 X such that '.f / D f .x/ for all f 2 A, and every x arises from a '.
Let A be a finitely generated algebra over an arbitrary algebraically closed field k, and let
X D spm.A/. An element f of A defines a k-valued function
m 7! f

mod m

on X. When A is reduced, Proposition 13.10 shows that this realizes A as a ring of k-valued functions
on X. Moreover, because (45) is an isomorphism in this case, for each k-algebra homomorphism
'W A ! k, there exists a unique x 2 X such that '.f / D f .x/ for all f 2 A. In particular, when
k D C and A is reduced, spm.A/ is the spectrum of A in the sense of analysis.

The max spectrum of a finitely generated k-algebra

Let k be a field, and let A be a finitely generated k-algebra. For any maximal ideal m of A,
def
the field .m/ D A=m is a finitely generated k-algebra, and so .m/ is finite over k (Zariskis
lemma, 13.1). In particular, it equals .m/ D k when k is algebraically closed.
Now fix an algebraic closure k al . The image of any k-algebra homomorphism A ! k al
is a subring of k al which is an integral domain algebraic over k and therefore a field (see
1). Hence the kernel of the homomorphism is a maximal ideal in A. In this way, we get a
surjective map
Homk-alg .A; k al / ! spm.A/:
(44)
Two homomorphisms A ! k al with the same kernel m factor as
A ! k.m/ ! k al ;
and so differ by an automorphism20 of k al . Therefore, the fibres of (44) are exactly the orbits
of Gal.k al =k/. When k is perfect, each extension k.m/=k is separable, and so each orbit
has k.m/W k elements, and when k is algebraically closed, the map (44) is a bijection.
Set A D kX1 ; : : : ; Xn =a. Then to give a homomorphism A ! k al is the same as giving
an n-tuple .a1 ; : : : ; an / of elements of k al (the images of the Xi ) such that f .a1 ; : : : ; an / D 0
for all f 2 a, i.e., an element of the zero-set V .a/ of a. The homomorphism corresponding
to .a1 ; : : : ; an / maps k.m/ isomorphically onto the subfield of k al generated by the ai s.
Therefore, we have a canonical surjection
V .a/ ! spm.A/

(45)

whose fibres are the orbits of Gal.k al =k/. When the field k is perfect, each orbit has
ka1 ; : : : ; an W k-elements, and when k is algebraically closed, V .a/ ' spm.A/.

Maps of max spectra

Let 'W A ! B be a homomorphism of rings, and let p be a prime ideal of B. Then B=p is an
integral domain and A=' 1 .p/ ! B=p is injective, and so ' 1 .p/ is a prime ideal in A. In
this way, spec becomes a functor from rings to topological spaces. Unfortunately, when p is
f and g be two k-homomorphisms from a finite field extension k 0 of k into k al . We consider the set
of pairs .K; / in which is a k-homomorphism from a subfield K of k al containing f .k 0 / into k al such that
f D g. The set is nonempty, and Zorns lemma can be applied to show that it has a maximal element .K 0 ; 0 /.
For such an element K 0 will be algebraically closed, and hence equal to k al .
20 Let

15

JACOBSON RINGS AND MAX SPECTRA

70

maximal, ' 1 .p/ need not be maximal consider for example the inclusion map Z ! Q
and the ideal .0/ in Q. Therefore, spm is not a functor on the category of all rings, but it is a
functor on the category of finitely generated over a fixed field.
L EMMA 15.7. Let 'W A ! B be a homomorphism of k-algebras, and let m be a maximal
ideal in B. If B is finitely generated over k, then the ideal ' 1 .m/ is maximal in A.
P ROOF. Because B is finitely generated over k, its quotient B=m by any maximal ideal m
is a finite field extension of k (Zariskis lemma, 13.1). Therefore the image of A in B=m is
an integral domain finite over k, and hence is a field (see 1). As this image is isomorphic to
A=' 1 .m/, this shows that the ideal ' 1 .m/ is maximal in A.
Therefore ' defines a map
'  W spm.B/ ! spm.A/;

m 7! '

.m/;

which is continuous because .'  / 1 .D.f // D D.'.f //. In this way, spm becomes a
functor from finitely generated k-algebras to topological spaces.
T HEOREM 15.8. Let 'W A ! B be a homomorphism of finitely generated k-algebras. Let
U be a nonempty open subset of spm.B/, and let '  .U / be the closure of its image in
spm.A/. Then '  .U / contains a nonempty open subset of each irreducible component of
'  .U / .
P ROOF. Let W D spm.B/ and V D spm.A/, so that '  is a continuous map W ! V .
We first prove the theorem in the case that ' is an injective homomorphism of integral
domains. For some b 0, D.b/  U . According to Proposition 15.9 below, there exists
a nonzero element a 2 A such that every homomorphim W A ! k al such that .a/ 0
extends to a homomorphism W B ! k al such that .b/ 0. Let m 2 D.a/, and choose to
be a homomorphism A ! k al with kernel m. The kernel of is a maximal ideal n 2 D.b/
such that ' 1 .n/ D m, and so D.a/  '  .D.b//.
We now prove the general case. If W1 ; : : : ; Wr are the irreducible components of W ,
then '  .W / is a union of the sets '  .Wi / , and any irreducible component C of '  .U /
is contained in one of '  .Wi / , say '  .W1 / . Let q D I.W1 / and let p D ' 1 .q/. Because
W1 is irreducible, they are both prime ideals. The homomorphism 'W A ! B induces an
injective homomorphism 'W
x A=p ! B=q, and 'x can be identified with the restriction of ' 
to W1 . From the first case, we know that 'x .U \ W1 / contains a nonempty open subset of
C , which implies that '  .U / does also.
In the next two statements, A and B are arbitrary commutative rings they need not be
k-algebras.
P ROPOSITION 15.9. Let A  B be integral domains with B finitely generated as an algebra
over A, and let b be a nonzero element of B. Then there exists an element a 0 in A
with the following property: every homomorphism W A ! from A into an algebraically
closed field such that .a/ 0 can be extended to a homomorphism W B ! such that
.b/ 0.
We first need a lemma.

15

JACOBSON RINGS AND MAX SPECTRA

71

L EMMA 15.10. Let B  A be integral domains, and assume B D At D AT =a. Let c  A

be the ideal of leading coefficients of the polynomials in a. Then every homomorphism
W A ! from A into an algebraically closed field such that .c/ 0 can be extended
to a homomorphism of B into .
P ROOF. If a D 0, then c D 0, and every extends. Thus we may assume a 0. Let be a
homomorphism A ! such that .c/ 0. Then there exist polynomials am T m C


C a0
in a such that .am / 0, and we choose one, denoted f , of minimum degree. Because
B 0, the polynomial f is nonconstant.
Extend to a homomorphism AT ! T , again denoted , by sending T to T , and
consider the subset .a/ of T .
F IRST CASE : .a/ DOES NOT CONTAIN A NONZERO CONSTANT. If the -subspace
of T spanned by .a/ contained 1, then so also would .a/,21 contrary to hypothesis.
Because
P
P
T
ci .gi / D ci .gi T /; ci 2 ; gi 2 a;
this -subspace an ideal, which we have shown to be proper, and so it has a zero c in .
The composite of the homomorphisms

AT ! T ! ;

T 7! T 7! c;

factors through AT =a D B and extends .

S ECOND CASE : .a/ CONTAINS A NONZERO CONSTANT. This means that a contains a
polynomial
g.T / D bn T n C


C b0

such that .b0 / 0;

.b1 / D .b2 / D


D 0:

On dividing f .T / into g.T / we obtain an equation

d
am
g.T / D q.T /f .T / C r.T /;

d 2 N;

q; r 2 AT ;

deg r < m:

When we apply , this becomes

.am /d .b0 / D .q/.f / C .r/:
Because .f / has degree m > 0, we must have .q/ D 0, and so .r/ is a nonzero constant.
After replacing g.T / with r.T /, we may suppose that n < m. If m D 1, such a g.T / cant
exist, and so we may suppose that m > 1 and (by induction) that the lemma holds for smaller
values of m.
For h.T / D cr T r C cr 1 T r 1 C


C c0 , let h0 .T / D cr C


C c0 T r . Then the Amodule generated by the polynomials T s h0 .T /, s  0, h 2 a, is an ideal a0 in AT . Moreover,
a0 contains a nonzero constant if and only if a contains a nonzero polynomial cT r , which
implies t D 0 and A D B (since B is an integral domain).
When a0 does not contain a nonzero constant, we set B 0 D AT =a0 D At 0 . Then a0
contains the polynomial g 0 D bn C


C b0 T n , and .b0 / 0. Because deg g 0 < m, the
induction hypothesis implies that extends to a homomorphism B 0 ! . Therefore, there
exists a c 2 such that, for all h.T / D cr T r C cr 1 T r 1 C


C c0 2 a,
h0 .c/ D .cr / C .cr

1 /c C


C c0 c

D 0:

On taking h D g, we see that c D 0, and on taking h D f , we obtain the contradiction

.am / D 0.
21 Use

that, if a system of linear equation with coefficients in a field k has a solution in some larger field, then
it has a solution in k.

15

JACOBSON RINGS AND MAX SPECTRA

72

P ROOF ( OF 15.9). Suppose that we know the proposition in the case that B is generated by
a single element, and write B D At1 ; : : : ; tn . Then there exists an element bn 1 such that any
homomorphism W At1 ; : : : ; tn 1 ! such that .bn 1 / 0 extends to a homomorphism
W B ! such that .b/ 0. Continuing in this fashion (with bn 1 for b), we eventually
obtain an element a 2 A with the required property.
Thus we may assume B D At . Let a be the kernel of the homomorphism T 7! t ,
AT ! At .
Case (i). The ideal a D .0/. Write
b D f .t / D a0 t n C a1 t n

C


C an ;

ai 2 A;

and take a D a0 . If W A ! is such that .a0 / 0, P

then there P
exists a c 2 such that
f .c/ 0, and we can take to be the homomorphism di t i 7! .di /c i .
Case (ii). The ideal a .0/. Let f .T / D am T m C


C a0 , am 0, be an element of
a of minimum degree. Let h.T / 2 AT represent b. Since b 0, h a. Because f is
irreducible over the field of fractions of A, it and h are coprime over that field. In other
words, there exist u; v 2 AT and a nonzero c 2 A such that
uh C vf D c:
It follows now that cam satisfies our requirements, for if .cam / 0, then can be extended
to W B ! by the lemma, and .u.t /
b/ D .c/ 0, and so .b/ 0.
R EMARK 15.11. In case (ii) of the last proof, both b and b
there exist equations
a0 b m C


C am D 0;
a00 b

C


C an0 D 0;

are algebraic over A, and so

ai 2 A;

a0 0I

ai0 2 A;

a00 0:

One can show that a D a0 a00 has the property required by the proposition (cf. Atiyah and
Macdonald 1969 5.23, p.66).
A SIDE 15.12. Let A be a noetherian ring, and let 'W A ! B be a finitely generated A-algebra. Then
the statement of (15.8) holds for '  W spm.B/ ! spm.A/ with much the same proof.
A SIDE 15.13. Let A be a ring and 'W A ! B a finitely generated A-algebra. If A is Jacobson, so
also is B, and ' induces a map spm.B/ ! spm.A/.
A SIDE 15.14. In general, the map A ! AX does not induce a map spm.AX / ! spm.A/. Consider for example a discrete valuation ring A with maximal ideal ./ (e.g., Z.p/ with maximal ideal
.p/). The ideal .X 1/ is maximal, because AX =.X 1/ is the field of fractions of A (by 5.3),
but .X 1/ \ A D .0/, which is not maximal.

Exercises
E XERCISE 15.15. Let A denote the polynomial ring QX1 ; X2 ; : : : in countably many
symbols.
(a) Show that A is not a Jacobson ring (consider the kernel of a surjective homomorphism
from A to a countable
T local domain, e.g., QX .X / ).
(b) Show that .0/ D fm j m a maximal ideal in Ag.
See mo151011.

16

16

ARTINIAN RINGS

73

Artinian rings

A ring A is artinian if every descending chain of ideals a1  a2 


in A eventually

becomes constant; equivalently, if every nonempty set of ideals has a minimal element.
Similarly, a module M over a ring A is artinian if every descending chain of submodules
N1  N2 


in M eventually becomes constant.
P ROPOSITION 16.1. An artinian ring has Krull dimension zero; in other words, every prime
ideal is maximal.
P ROOF. Let p be a prime ideal of an artinian ring A, and let A0 D A=p. Then A0 is an
artinian integral domain. Let a be a nonzero element of A0 . The chain .a/  .a2 / 



eventually becomes constant, and so an D anC1 b for some b 2 A0 and n  1. We can cancel
an to obtain 1 D ab. Thus a is a unit, and so A0 is a field, and p is maximal:
C OROLLARY 16.2. In an artinian ring, the nilradical and the Jacobson radical coincide.
P ROOF. The first is the intersection of the prime ideals (2.5), and the second is the intersection of the maximal ideals (2.6).
P ROPOSITION 16.3. An artinian ring has only finitely many maximal ideals.
P ROOF. Let m1 \ : : : \ mn be minimal among finite intersections of maximal ideals in an
artinian ring, and let m be another maximal ideal in the ring. If m is not equal to one of the
mi , then, for each i , there exists an ai 2 mi X m. Now a1


an lies in m1 \ : : : \ mn but not
in m (because m is prime), contradicting the minimality of m1 \ : : : \ mn .
P ROPOSITION 16.4. In an artinian ring, some power of the nilradical is zero.
P ROOF. Let N be the nilradical of the artinian ring A. The chain N  N2 


eventually
becomes constant, and so Nn D NnC1 D


for some n  1. Suppose that Nn 0. Then
there exist ideals a such that a
Nn 0, for example N, and we choose an a that is minimal
among such ideals. There exists an a 2 a such that a
Nn 0, and so a D .a/ (by minimality).
Now .aNn /Nn D aN2n D aNn 0 and aNn  .a/, and so aNn D .a/ (by minimality
again). Hence a D ax for some x 2 Nn . Now a D ax D ax 2 D


D a0 D 0 because x 2 N.
This contradicts the definition of a, and so Nn D 0.
L EMMA 16.5. Let A be a ring in which some finite product of maximal ideals is zero. Then
A is artinian if and only if it is noetherian.
P ROOF. Suppose that m1


mn D 0 with the mi maximal ideals (not necessarily distinct),
and consider
A  m1 


 m1


mr
def

 m1


mr 


 m1


mn D 0:

The action of A on the quotient Mr D m1


mr 1 =m1


mr factors through the field A=mr ,
and the subspaces of the vector space Mr are in one-to-one correspondence with the ideals
of A contained between m1


mr 1 and m1


mr . If A is either artinian or noetherian, then
Mr satisfies a chain condition on subspaces and so it is finite-dimensional as a vector space
and both artinian and noetherian as an A-module. Now repeated applications of Proposition
3.3 (resp. its analogue for artinian modules) show that if A is artinian (resp. noetherian),
then it is noetherian (resp. artinian) as an A-module, and hence as a ring.

17

QUASI-FINITE ALGEBRAS AND ZARISKIS MAIN THEOREM.

74

T HEOREM 16.6. A ring is artinian if and only if it is noetherian of dimension zero.

P ROOF. ): Let A be an artinian ring. After (16.1), it remains to show that A is noetherian,
but according to (16.2), (16.3), and (16.4), some finite product of maximal ideals is zero,
and so this follows from Lemma 16.5.
(: Let A be a noetherian ring of dimension zero. Because A is noetherian, its radical N
is a finite intersection of prime ideals (14.9), each of which is maximal because dim A D 0.
Hence N is a finite intersection of maximal ideals (2.5), and since some power of N is zero
(3.16), we again have that some finite product of maximal ideals is zero, and so can apply
Lemma 16.5.
T HEOREM 16.7. Every artinian ring is (uniquely) a product of local artinian rings.
P ROOF. Let A be artinian, and let m1 ; : : : ; mr be the distinct maximal ideals in A. We saw
n
in the proof of (16.6) that some product mn1 1


mnr r D 0. For i j , the ideal mni i C mj j is
not contained in any maximal ideal, and so equals A. Now the Chinese remainder theorem
2.13 shows that
A ' A=mn1 1 


 A=mnr r ,
and each ring A=mni i is obviously local.
P ROPOSITION 16.8. Let A be a local artinian ring with maximal ideal m. If m is principal,
so also is every ideal in A; in fact, if m D .t /, then every ideal is of the form .t r / for some
r  0.
P ROOF. Because m is the Jacobson radical of A, some power of m is zero (by 16.4); in
particular, .0/ D .t r / for some r. Let a be a nonzero ideal in A. There exists an integer r  0
such that a  mr but a 6 mrC1 . Therefore there exists an element a of a such that a D ct r
for some c 2 A but a .t rC1 /. The second condition implies that c m, and so it is a unit;
therefore a D .a/.
E XAMPLE 16.9. The ring A D kX1 ; X2 ; X3 ; : : :=.X1 ; X22 ; X33 ; : : :/ has only a single prime
ideal, namely, .x1 ; x2 ; x3 ; : : :/, and so has dimension zero. However, it is not noetherian
(hence not artinian).
A SIDE 16.10. Every finitely generated module over a principal Artin ring is a direct sum of cyclic
modules (see mo22722).

17

Quasi-finite algebras and Zariskis main theorem.

In this section we prove a fundamental theorem of Zariski.22 Throughout, k is a field and A

is a commutative ring.
22 Our

2010.

exposition of the proof follows those in Raynaud 1970 and in Hochsters course notes from Winter,

17

QUASI-FINITE ALGEBRAS AND ZARISKIS MAIN THEOREM.

75

Quasi-finite algebras
P ROPOSITION 17.1. Let B be a finitely generated k-algebra. A prime ideal q of B is an
isolated point of spec.B/ if and only if Bq is finite over k.
P ROOF. To say that q is an isolated point of spec.B/ means that there exists an f 2 B X q
such that spec.Bf / D fqg. Now Bf is noetherian with only one prime ideal, namely,
def
m D qBf , and so it is artinian (16.6). The quotient Bf =m is a field which is finitely
generated as a k-algebra, and hence is finite over k (Zariskis lemma 13.1). Because Bf is
artinian,
Bf  m  m2 



can be refined to a finite filtration whose quotients are one-dimensional vector spaces over
Bf =m. Therefore Bf is a finite k-algebra. As f q, we have Bq D .Bf /q , which equals
Bf because Bf is local. Therefore Bq is also a finite k-algebra.
For the converse, suppose that Bq is finite over k, and consider the exact seqence
0 ! M ! B ! Bq ! N ! 0

(46)

of B-modules. When we apply the functor Sq 1 to (46), it remains exact (5.11), but the
middle arrow becomes an isomorphism, and so Mq D 0 D Nq . Because B is noetherian,
the B-module M is finitely generated, with generators e1 ; : : : ; em say. As Mq D 0, there
def
exists, for each i , an fi 2 B X q such that fi ei D 0. Now f 0 D f1 : : : fm has the property
that f 0 M D 0, and so Mf 0 D 0.
Because Bq is a finite k-algebra, N is finitely generated as a k-module, and therefore
also as a B-module. As for M , there exists an f 00 2 B X q such that Mf 00 D 0. Now
def
f D f 0 f 00 2 B X q has the property that Mf D 0 D Nf . When we apply the functor Sf 1
to (46), we obtain an isomorphism Bf ' Bq , and so spec.Bf / D spec.Bq / D fqg, which
shows that q is an isolated point.
P ROPOSITION 17.2. Let B be a finitely generated k-algebra. The space spec.B/ is discrete
if and only if B is a finite k-algebra.
P ROOF. If B is finite over k, then it is artinian and so (16.7)
Y
B D fBm j m maximalg (finite product),
and
spec.B/ D

G
m

spec.Bm / D

G
m

fmg (disjoint union of open subsets).

Therefore each point is isolated in spec.B/.

Conversely, if spec.B/ is discrete then it is a finite disjoint union,
G
spec.B/ D
spec.Bfi /; fi 2 B;
1i n

Q
with spec.Bfi / D fqi g. Hence B D 1i n Bfi (by 14.3) with Bfi D Bqi . According to
Proposition (17.1), each k-algebra Bqi is finite over k, and so B is finite over k.
D EFINITION 17.3. Let B be a finitely generated A-algebra.

17

QUASI-FINITE ALGEBRAS AND ZARISKIS MAIN THEOREM.

76

(a) Let q be a prime ideal of B, and let p D qc . The ring B is said to be quasi-finite over
A at q if Bq =pBq is a finite .p/-algebra.
(b) The ring B is said to be quasi-finite over A if it is quasi-finite over A at all the prime
ideals of B.
P ROPOSITION 17.4. Let B be a finitely generated A-algebra. Let q be a prime ideal of B,
and let p D qc . Then B is quasi-finite over A at q if and only if q is an isolated point of
spec.B A .p//.
P ROOF. As
Bq =pBq ' .B=pB/q=p ' .B A .p//q=p ;
this is an immediate consequence of (17.1) applied to the .p/-algebra B A .p/.
The prime ideals of B=pB correspond to the prime ideals of B whose contraction to A
contains p, and the prime ideals of B A .p/ correspond to the prime ideals of B whose
contraction to A is p. To say that B is quasi-finite over A at q means that q is both maximal
and minimal among the prime ideals lying over p (i.e., that each point of spec.B A .p//
is closed).
P ROPOSITION 17.5. A finitely generated A-algebra B is quasi-finite over A if and only if,
for all prime ideals p of A, B A .p/ is finite over .p/.
P ROOF. Immediate consequence of (17.2).
E XAMPLE 17.6. Let C be a finitely generated A-algebra. If C is finite over A, then
C A .p/ is finite over .p/ for all prime ideals p of A, and so C is quasi-finite over A.
In particular, spec.C A .p// is discrete for all primes p of A, and so if B is a finitely
generated C -algebra such that the map spec.B/ ! spec.C / is an open immersion, then B
is also quasi-finite over A. Zariskis main theorem says that all quasi-finite A-algebras arise
in this way.
The next two lemmas will be used in the proof of Zariskis main theorem.
L EMMA 17.7. Let A ! C ! B be homomorphisms of rings such that the composite
A ! B is of finite type, and let q be a prime ideal of B. If B is quasi-finite over A at q, then
it is quasi-finite over C at q.
P ROOF. Let pA and pC be the inverse images of q in A and C respectively. Then spec.B C
.pC // is subspace of spec.B A .pA //, and so if q is an isolated point in the second space,
then it is an isolated point in the first space.
L EMMA 17.8. Let A  C  B be rings. Let q be a prime ideal of B, and let r D q \ C and
p D q \ A.
(a) If q is minimal among the primes lying over p and there exists a u 2 C X q such that
Cu D Bu , then r is minimal among the primes lying over p.
(b) If B is integral over a finitely generated A-subalgebra B0 and q is maximal among the
prime ideals lying over p, then r is maximal among the prime ideals lying over p.
(c) Assume that B is integral over a finitely generated A-subalgebra B0 , and that there
exists a u 2 C X q such that Cu D Bu . If B is quasi-finite over A at q, then C is
quasi-finite over A at r.

17

QUASI-FINITE ALGEBRAS AND ZARISKIS MAIN THEOREM.

77

P ROOF. (a) If r0 is a prime ideal of C lying over p and strictly contained in r, then by
extending r0 to Cu D Bu and then contracting the result to B, we obtain a prime ideal q0 of
B lying over p and strictly contained in q.
(b) We may replace A, C , and B with their localizations at p, and so assume that A is
local with maximal ideal p. Then
A=p  C =r  B=q
and we also have
A=p  B0 =r0  B=r
where r0 D q \ B0 : As q is maximal among the prime ideals lying over p, B=q is a field. As
B=q is integral over B0 =r0 , the latter is also a field (see 7.1), and it is finitely generated as an
A=p-algebra. Zariskis lemma (13.1) now shows that B0 =r0 is a finite algebraic extension of
A=p, and so B=q is an algebraic extension of A=p. It follows that C =r is a field, and so r is
maximal among the prime ideals in C over p.
(c) Combine (a) and (b) (with the remark following (17.3)).
A SIDE 17.9. Geometrically, to say that A ! B is quasi-finite means that the map Spec B ! Spec A
has finite fibres. The condition that A ! B be finite is much stronger: it not only requires that
Spec B ! Spec A have finite fibres but also that it be universally closed. See, for example, my notes
on algebraic geometry.

Statement of Zariskis main theorem

T HEOREM 17.10. Let B be a finitely generated A-algebra, and let A0 be the integral closure
of A in B. Then B is quasi-finite over A at a prime ideal q if and only if A0f ' Bf for some
f 2 A0 X q.
The sufficiency is obvious; the proof of the necessity will occupy the rest of this section.
First, we list some consequences.
C OROLLARY 17.11. Let B be a finitely generated A-algebra. The set of prime ideals of B
at which B is quasi-finite over A is open in spec.B/.
P ROOF. Let q be a prime ideal of B such that B is quasi-finite over A at q. The theorem
shows that there exists an f 2 A0 X q such that A0f ' Bf . Write A0 as the union of the
finitely generated A-subalgebras Ai of A0 containing f :
[
A0 D
Ai :
i

Because A0 is integral over A, each Ai is finite over A (see 6.3). We have

[
Bf ' A0f D
Aif :
i

Because Bf is a finitely generated A-algebra, Bf D Aif for all sufficiently large Ai . As the
Ai are finite over A, Bf is quasi-finite over A, and spec.Bf / is an open neighbourhood of q
consisting of quasi-finite points.
C OROLLARY 17.12. Let B be a finitely generated A-algebra, quasi-finite over A, and let
A0 be the integral closure of A in B. Then

17

QUASI-FINITE ALGEBRAS AND ZARISKIS MAIN THEOREM.

78

(a) the map Spec B ! Spec A0 is an open immersion, and

(b) there exists an A-subalgebra A00 of A0 , finite over A, such that Spec B ! Spec A00 is
an open immersion.
P ROOF. (a) Because B is quasi-finite over A at every point of spec.B/, the theorem implies
that there exist fi 2 A0 such that the open sets spec.Bfi / cover spec.B/ and A0f ' Bfi for
i
all i . As spec.B/ quasi-compact, finitely many sets spec.Bfi / suffice to cover spec.B/, and
it follows that spec.B/ ! spec.A0 / isSan open immersion.
(b) We have seen that spec.B/ D 1i n spec.Bfi / for certain fi 2 A0 such that A0f '
i
Bfi . The argument in the proof of (17.11) shows that there exists an A-subalgebra A00 of A0 ,
finite over A, which contains f1 ; : : : ; fn and is such that Bfi ' A00f for all i . Now the map
i
spec.B/ ! spec.A00 / is an open immersion.
Theorem 17.10, its corollary 17.12, and various global versions of these statements are
referred to as Zariskis main theorem.

A variant of Zariskis main theorem

P ROPOSITION 17.13. Let A  C  B be rings such that A integrally closed in B, C is
finitely generated over A, and B is finite over C . If B is quasi-finite over A at a prime ideal
q, then Bp D Ap with p D q \ A.
P ROOF THAT 17.13 IMPLIES 17.10
Let A, A0 , and B be as in the Theorem 17.10. We apply the proposition to A0  B D B
Lemma 17.7 shows that the ring B is quasi-finite over A0 at q. The proposition shows that
Bp0 D A0p0 with p0 D q \ A0 . Let b1 ; : : : ; bn generate B as an A0 -algebra, and let bi0 denote
the image of bi in Bp0 D A0p0 . Then bi0 D ai =f for some ai 2 A0 and f 2 A0 X p0 . The bi0 are
in the image of the map A0f ! Bf , which is therefore surjective. But A0f ! Bf is injective
because A  B, and so the map is an isomorphism. This completes the proof of the theorem.

Proof of Proposition 17.10

We proceed by proving four special cases of Proposition 17.10.
L EMMA 17.14. Let A  Ax D B be rings such that A is integrally closed in B. If B is
quasi-finite over A at a prime ideal q, then Bp D Ap with p D q \ A.
P ROOF. The hypotheses remain true when we invert the elements of S X p to obtain Ap 
Ap x D Bp . Thus, we may suppose that A is local with maximal ideal p, and we have to
prove that B D A. As A is integrally closed in B and B D Ax, it suffices to show that x is
integral over A.
Let k D A=p and consider the k-algebra
def

kx
x D Ax A k D B A .p/:
By assumption, q is an isolated point in spec.kx
x /. Consequently, x is algebraic over k,
because otherwise kx
x would be a polynomial ring over k, and its spectrum would have no
isolated points. Therefore there exists a polynomial F 2 AX with nonconstant image in

17

QUASI-FINITE ALGEBRAS AND ZARISKIS MAIN THEOREM.

79

kX such that F .x/ 2 pAx. Now F F .x/ is a polynomial in AX that vanishes on x

and has at least one coefficient not in p. Choose such a polynomial H of minimum degree
m, and write it
H.X / D am X m C


C a0 :
The equation am

1 H.x/

D 0 can be written

.am x/m C am

1 .am x/

m 1

m
C


C a0 am

D 0:

It shows that am x is integral over A, and so lies in A. Now the polynomial

.am x C am

1 /X

m 1

C


C a0

lies in AX and vanishes on x. As it has degree < m, all of its coefficients must lie in
p. In particular, am x C am 1 2 p. If am is a unit, then x is integral over A, as required.
Otherwise, am 2 p and am 1 is a unit (because otherwise all coefficients of H lie in p);
hence am 1 2 pB, which is contradiction because pB  q.
L EMMA 17.15. Let B be an integral domain containing a polynomial ring AX and integral
over it. Then B is not quasi-finite over A at any prime ideal q.
P ROOF. Let q be a prime ideal of B, and let p D q \ A. If B is quasi-finite over A at q, then
q is both maximal and minimal among the prime ideals lying over p. We shall assume that q
is maximal and prove that it cant then be minimal.
Suppose first that A is integrally closed, and let r D q \ AX . If r were not maximal
among the prime ideals of AX lying over p, then the going-up theorem (7.6) would imply
that q is not either. Therefore r is maximal among the prime over p, and it follows that its
image xr in .p/X is maximal. In particular, xr 0, and so r strictly contains the prime ideal
pAX in AX . As A is integrally closed, AX is also (6.17), and the going down theorem
(7.11) shows that q strictly contains a prime ideal lying over pAX . Therefore, q is not
minimal among the prime ideals lying over p.
In the general case, we let B 0 denote the integral closure of B in its field of fractions.
Then B 0 contains the integral closure A0 of A, and is integral over A0 T . Let q0 be a prime
ideal of B 0 lying over q (which exists by 7.5), and let p0 D q0 \ A0 . As q is maximal among
the primes lying over p, q0 is maximal among those lying over p0 (apply 7.4 to B  B 0 ). But,
according to the preceding paragraph, q0 is not minimal, which implies that q is not minimal
(apply 7.4 again).
L EMMA 17.16. Let A  Ax  B be rings such that B is integral over Ax and A is
integrally closed in B. If there exists a monic polynomial F 2 AX such that F .x/B  Ax,
then Ax D B.
P ROOF. Let b 2 B be arbitrary. By assumption F .x/b 2 Ax, and so F .x/b D G.x/ for
some polynomial G in AX . As F is monic, we can divide F into G to get
G D QF C R;

deg R < deg F;

Q; R 2 AX :

Now
F .x/b D G.x/ D Q.x/F .x/ C R.x/:
For c D b

Q.x/;
F .x/c D R.x/:

(47)

17

QUASI-FINITE ALGEBRAS AND ZARISKIS MAIN THEOREM.

80

To show that b 2 Ax, it suffices to show that c 2 A, and for this it suffices to show that c is
integral over A.
Let A0 be the image of A in Bc . As deg R < deg F , the equality (47) shows that x=1, as
an element of Bc , is integral over the subring A0c . As B is integral over Ax, this implies that
Bc is integral over A0c . In particular, c=1 is integral over A0c , and so it satisfies an equation
whose coefficients we can assume to have a common denominator c M :
.c=1/m C

a1
.c=1/m
cM

C


C

am
D 0;
cM

ai 2 A,

(equality in Bc ). Therefore
c M Cm C a1 c m

C


C am

is an element of B whose image in Bc is zero, and so is killed by a power of c. This shows

that c is integral over A, as required.
Let B be a finite A-algebra. The conductor of B in A is
f.B=A/ D fa 2 A j aB  Ag:
This is an ideal of both A and B. In fact, it is the largest ideal in A that is also an ideal
in B, because every element a of such an ideal has the property that aB  A. For any
multiplicative subset S of A,
f.S

B=S

A/ D S

f.B=A/:

(48)

L EMMA 17.17. Let A  Ax  B be rings such that B is finite over Ax and A is integrally
closed in B. If B is quasi-finite over A at a prime ideal q, then Bp D Ap with p D q \ A.
P ROOF. Let f D f.B=Ax/, so
f D f 2 Ax j B  Axg:
We first consider the case that f 6 q. Let r D q \ Ax. For any u 2 f X q, we have
Axu D Bu , and so Lemma 17.8 shows that Ax is quasi-finite over A at r.23 Now Lemma
17.14 shows that Axp D Ap . But B is finite over Ax, and therefore Bp is finite over
Axp D Ap . As A is integrally closed in B, Ap is integrally closed in Bp , and therefore
Ap D Bp , as required.
It remains to consider the case f  q. We choose a prime ideal n  q of B minimal
among those containing f. Let t denote the image of x in the ring B=n, and let m D n \ A.
Now
A=m  .A=m/t  B=n,
and B=n is integral over .A=m/t . As B is quasi-finite over A at q, the quotient B=n is
quasi-finite over A=m at q=n. Now Lemma 17.15 implies that t is algebraic over A=m. We
shall complete the proof by obtaining a contradiction, which will show that this case doesnt
occur.
After making an extension of scalars A ! Am , we may assume that A is a local ring
with maximal ideal m. Let n0 D n \ Ax. Because t is algebraic over A=m, the integral
23 Here

we follow Hochster. Raynaud simply states that Ax is quasi-finite over A at r.

18

DIMENSION THEORY FOR FINITELY GENERATED k-ALGEBRAS

81

domain Ax=n0 is a finite A=m-algebra, and hence a field (see 1). Therefore, n0 is maximal
in Ax, and it follows from (7.3) that n is maximal in B. Thus B=n is a field.
Because t is algebraic over A=m, there exists a monic polynomial F in AX such
that F .x/ 2 n. But n is minimal among the prime ideals of B containing f, and so nBn is
minimal among the prime ideals of Bn containing fn . In fact, nBn is the only prime ideal
containing fn , and so nBn is the radical of fn . Therefore, there exists an integer r > 0 such
that .F .x//r 2 fn , and a y 2 B X n such that yF .x/r 2 f.
We therefore have yF .x/r B  Ax. On applying Lemma 17.16 with A  Ax  B 0 ,
0
B D AxyB; and F 0 D F r , we deduce that B 0 D Ax and therefore that yB  Ax.
Hence y 2 f  n, which contradicts the definition of y.
P ROOF OF P ROPOSITION 17.10
We use induction on the number n of generators of the A-algebra C . If n D 0, then B is
integral over A, and so B D A. Assume that n > 0 and that the proposition has been proved
when C is generated by n 1 elements.
Write C D Ax1 ; : : : ; xn , and let A0 be the integral closure of Ax1 ; : : : ; xn 1 in B.
Then
A0  A0 xn  B;
and B is finite over A0 xn . The ring B is finite over A0 xn and it is quasi-finite over A at q,
and so B is quasi-finite over A0 at q (by 17.7). From Lemma 17.17 we deduce that A0p0 D Bp0
with p0 D A0 \ q.
As A0 is integral over Ax1 ; : : : ; xn 1 , it is a union of its finite subalgebras,
[
A0 D
A0i ; A0i finite over Ax1 ; : : : ; xn 1 :
i

Let p0i D q \ A0i D p0 \ A0i . As B is finitely generated over Ax1 ; : : : ; xn

homomorphism
.A0i /p0i ! Bp0i

1 ,

the canonical

is an isomorphism for all sufficiently large i . For such an i , we have a fortiori that
.A0i /p0i ' Bq ;
and so A0i is quasi-finite over A at p0i . On applying the induction hypothesis to A; Ax1 ; : : : ; xn
and A0i , we deduce that
Ap ' .A0i /p ' .A0i /p0i ;
and consequently that Ap ' Bp . This completes the proof of Proposition 17.13 and hence of
Theorem 17.10.

18

Dimension theory for finitely generated k-algebras

Except in the final subsection, A is an integral domain containing a field k and finitely
generated as a k-algebra. We define the transcendence degree of A over k, tr degk A, to be
the transcendence degree over k of the field of fractions F .A/ of A (see 9 of my notes
Fields and Galois Theory). Thus A has transcendence degree d if it contains an algebraically
independent set of d elements, but no larger set (ibid. 8.12).

1 ,

18

DIMENSION THEORY FOR FINITELY GENERATED k-ALGEBRAS

82

P ROPOSITION 18.1. For any linear forms `1 ; : : : ; `m in X1 ; : : : ; Xn , the quotient ring

kX1 ; : : : ; Xn =.`1 ; : : : ; `m /
is an integral domain of transcendence degree equal to the dimension of the subspace of k n
defined by the equations
`i D 0; i D 1; : : : ; m:
P ROOF. This follows from the more precise statement:
Let c be an ideal in kX1 ; : : : ; Xn generated by linearly independent linear
forms `1 ; : : : ; `r , and let Xi1 ; : : : ; Xin r be such that
f`1 ; : : : ; `r ; Xi1 ; : : : ; Xin r g
is a basis for the linear forms in X1 ; : : : ; Xn . Then
kX1 ; : : : ; Xn =c ' kXi1 ; : : : ; Xin r :
This is obvious if the forms `i are X1 ; : : : ; Xr . In the general case, because fX1 ; : : : ; Xn g
and f`1 ; : : : ; `r ; Xi1 ; : : : ; Xin r g are both bases for the linear forms, each element of one set
can be expressed as a linear combination of the elements of the other. Therefore,
kX1 ; : : : ; Xn D k`1 ; : : : ; `r ; Xi1 ; : : : ; Xin r ;
and so
kX1 ; : : : ; Xn =c D k`1 ; : : : ; `r ; Xi1 ; : : : ; Xin r =c
' kXi1 ; : : : ; Xin r :
P ROPOSITION 18.2. For any irreducible polynomial f in kX1 ; : : : ; Xn , the quotient ring
kX1 ; : : : ; Xn =.f / has transcendence degree n 1.
P ROOF. Let
kx1 ; : : : ; xn D kX1 ; : : : ; Xn =.f /;

xi D Xi C .f /;

and let k.x1 ; : : : ; xn / be the field of fractions of kx1 ; : : : ; xn . Since f is not zero, some Xi ,
say, Xn , occurs in it. Then Xn occurs in every nonzero multiple of f , and so no nonzero
polynomial in X1 ; : : : ; Xn 1 belongs to .f /. This means that x1 ; : : : ; xn 1 are algebraically
independent. On the other hand, xn is algebraic over k.x1 ; : : : ; xn 1 /, and so fx1 ; : : : ; xn 1 g
is a transcendence basis for k.x1 ; : : : ; xn / over k.
P ROPOSITION 18.3. For every nonzero prime ideal p in a k-algebra A,
tr degk .A=p/ < tr degk .A/:
P ROOF. We may suppose that
A D kX1 ; : : : ; Xn =a D kx1 ; : : : ; xn :
For f 2 A, let fx denote the image of f in A=p, so that A=p D kx
x1 ; : : : ; xn . Let d D
tr degk A=p, and number the Xi so that x1 ; : : : ; xd are algebraically independent (for a proof

18

DIMENSION THEORY FOR FINITELY GENERATED k-ALGEBRAS

83

that this is possible, see 8.9 of my notes Fields and Galois Theory). I shall show that, for
any nonzero f 2 p, the d C 1 elements x1 ; : : : ; xd ; f are algebraically independent, which
shows that tr degk A  d C 1.
Suppose otherwise. Then there is a nontrivial algebraic relation, which we can write
a0 .x1 ; : : : ; xd /f m C a1 .x1 ; : : : ; xd /f m

C


C am .x1 ; : : : ; xd / D 0;

with ai 2 kX1 ; : : : ; Xd and a0 0. Because A is an integral domain, we can cancel a

power of f if necessary to make am .x1 ; : : : ; xd / nonzero. On applying the homomorphism
A ! A=p to the above equality, we find that
am .x
x1 ; : : : ; xd / D 0;
which contradicts the algebraic independence of x1 ; : : : ; xd .
P ROPOSITION 18.4. Let A be a polynomial ring. If p is a prime ideal in A such that
tr degk A=p D tr degk A 1, then p D .f / for some f 2 A.
P ROOF. The ideal p is nonzero because otherwise A and A=p would have the same transcendence degree. Therefore p contains a nonzero polynomial, and even an irreducible
polynomial f , because it is prime. According to (4.1), the ideal .f / is prime. If .f / p,
then
18.3
18.2
tr degk A=p > tr degk A=.f / D tr degk A 1;
which contradicts the hypothesis.
T HEOREM 18.5. Let f 2 A be neither zero nor a unit, and let p be a prime ideal that is
minimal among those containing .f /; then
tr degk A=p D tr degk A

1:

We first need a lemma.

L EMMA 18.6. Let A be an integrally closed integral domain, and let L be a finite extension
of the field of fractions K of A. If 2 L is integral over A, then NmL=K 2 A, and
divides NmL=K in the ring A.
P ROOF. Let X r C ar

1X

r 1 C


C a
0

be the minimum polynomial of over K. Then

n

r divides the degree n of L=K, and NmL=K ./ D a0r (see 5.40 of my notes Fields and
Galois Theory). Moreover, a0 lies in A by (6.11). From the equation
0 D . r

C ar

r 2

C


C a1 / C a0

we see that divides a0 in A, and therefore it also divides NmL=K .

P ROOF ( OF T HEOREM 18.5). Write rad.f / as an irredundant intersection of prime ideals
rad.f / D p1 \ : : : \ pr (see 14.9). Then V .a/ D V .p1 / [


[ V .pr / is S
the decomposition
of V .a/ into its irreducible components. There exists an m0 2 V .p1 / X i 2 V .pi / and an
S
open neighbourhood D.h/ of m0 disjoint from i 2 V .pi /. The ring Ah (resp. Ah =S 1 p)
is an integral domain with the same transcendance degree as A (resp. A=p) in fact, with

18

DIMENSION THEORY FOR FINITELY GENERATED k-ALGEBRAS

84

the same field of fractions. In Ah , rad. f1 / D rad.f /e D pe1 . Therefore, after replacing A
with Ah , we may suppose that rad.f / is prime, say, equal to p.
According to the Noether normalization theorem (8.1), there exist algebraically independent elements x1 ; : : : ; xd in A such that A is a finite kx1 ; : : : ; xd -algebra. Note that
def
d D tr degk A. According to the lemma, f0 D Nm.f / lies in kx1 ; : : : ; xd , and we shall
show that p \ kx1 ; : : : ; xd D rad.f0 /. Therefore, the homomorphism
kx1 ; : : : ; xd =rad.f0 / ! A=p
is injective. As it is also finite, this implies that
18.2

tr degk A=p D tr degk kx1 ; : : : ; xd =rad.f0 / D d

1;

as required.
By assumption A is finite (hence integral) over its subring kx1 ; : : : ; xd . The lemma
shows that f divides f0 in A, and so f0 2 .f /  p. Hence .f0 /  p \ kx1 ; : : : ; xd , which
implies
rad.f0 /  p \ kx1 ; : : : ; xd
because p is radical. For the reverse inclusion, let g 2 p \ kx1 ; : : : ; xd . Then g 2 rad.f /,
and so g m D f h for some h 2 A, m 2 N. Taking norms, we find that
g me D Nm.f h/ D f0
Nm.h/ 2 .f0 /;
where e is the degree of the extension of the fields of fractions, which proves the claim.
C OROLLARY 18.7. Let p be a minimal nonzero prime ideal in A; then tr degk .A=p/ D
tr degk .A/ 1.
P ROOF. Let f be a nonzero element of p. Then f is not a unit, and p is minimal among the
prime ideals containing f .
T HEOREM 18.8. The length d of any maximal (i.e., nonrefinable) chain of distinct prime
ideals
pd  pd 1 


 p0
(49)
in A is tr degk .A/. In particular, every maximal ideal of A has height tr degk .A/, and so the
Krull dimension of A is equal to tr degk .A/.
P ROOF. From (18.7), we find that
tr degk .A/ D tr degk .A=p1 / C 1 D


D tr degk .A=pd / C d:
But pd is maximal, and so A=pd is a finite field extension of k. In particular, tr degk .A=pd / D
0.
E XAMPLE 18.9. Let f .X; Y / and g.X; Y / be nonconstant polynomials with no common
factor. Then kX; Y =.f / has Krull dimension 1, and so kX; Y =.f; g/ has dimension zero.
E XAMPLE 18.10. We classify the prime ideals p in A D kX; Y . If A=p has dimension
2, then p D .0/. If A=p has dimension 1, then p D .f / for some irreducible polynomial
f of A (by 18.4). Finally, if A=p has dimension zero, then p is maximal. Thus, when k
is algebraically closed, the prime ideals in kX; Y are exactly the ideals .0/, .f / (with f
irreducible), and .X a; Y b/ (with a; b 2 k).

18

DIMENSION THEORY FOR FINITELY GENERATED k-ALGEBRAS

85

R EMARK 18.11. Let A be a finitely generated k-algebra (not necessarily an integral domain). Every maximal chain of prime ideals in A ending in fixed prime ideal p has length
tr degk .A=p/, and so the Krull dimension of A is max.tr degk .A=p// where p runs over the
minimal prime ideals of A. In the next section, we show that a noetherian ring has only
finitely many minimal prime ideals, and so the Krull dimension of A is finite.
If x1 ; : : : ; xm is an algebraically independent set of elements of A such that A is a finite
kx1 ; : : : ; xm -algebra, then dim A D m.
R EMARK 18.12. Let A be a discrete valuation ring A with maximal ideal ./. Then AX
is a noetherian integral domain of Krull dimension 2, and .X 1/ is a maximal ideal in
AX of height 1 (cf. 15.14).

A short proof that the Krull dimension equals the transcendence degree
The following proof shortens that in Coquand and Lombardi, Amer. Math. Monthly 112
(2005), no. 9, 826829.
Let A be an arbitrary commutative ring. Let x 2 A, and let Sfxg denote the multiplicative
subset of A consisting of the elements of the form
x n .1

ax/;

n 2 N;

a 2 A:

The boundary Afxg of A at x is defined to be the ring of fractions Sfxg1 A.

We write dim.A/ for the Krull dimension of A.
P ROPOSITION 18.13. Let A be a ring and let n 2 N. Then
dim.A/  n for all x 2 A, dim.Afxg /  n

1:

P ROOF. Recall (5.4) that Spec.S 1 A/ ' fp 2 Spec.A/ j p \ S D ;g. We shall need the
following statements.
(a) For every x 2 A and maximal ideal m  A, m \ Sfxg ;. Indeed, if x 2 m, then
x 2 m \ S ; otherwise x is invertible modulo m, and so there exists an a 2 A such that
1 ax 2 m.
(b) Let m be a maximal ideal, and let p be a prime ideal contained in m; for every
x 2 m X p, m \ Sfxg D ;. Indeed, if x n .1 ax/ 2 p, then 1 ax 2 p (as x p/; hence
1 ax 2 m, and so 1 2 m, which is a contradiction.
Statement (a) shows that every chain of prime ideals beginning with a maximal ideal is
shortened when passing from A to Afxg , while statement (b) shows that a maximal chain of
length n is shortened only to n 1 when x is chosen appropriately. From this, the proposition
is follows.
P ROPOSITION 18.14. Let k  F  E be fields. Then
tr degk E D tr degk F C tr degF E:
P ROOF. More precisely, if B and C are transcendence bases for F=k and E=F respectively,
then B [ C is a transcendence basis for E=k. This is easy to check (see, for example,
Jacobson, Lectures in Abstract Algebra III, 1964, Exercise 3, p.156).

19

PRIMARY DECOMPOSITIONS

86

P ROPOSITION 18.15. Let A be an integral domain with field of fractions F .A/, and let k
be a subfield of A. Then
tr degk F .A/  dim.A/:
P ROOF. If tr degk F .A/ D 1, there is nothing to prove, and so we assume that tr degk F .A/ D
n 2 N. We argue by induction on n. We can replace k with its algebraic closure in A without
changing tr degk F .A/. Let x 2 A. If x k, then it is transcendental over k, and so
tr degk.x/ F .A/ D n

by (18.14); since k.x/  Afxg , this implies (by induction) that dim.Afxg /  n 1. If x 2 k,
then 0 D 1 x 1 x 2 Sfxg , and so Afxg D 0; again dim.Afxg /  n 1. Now (18.13) shows
that dim.A/  n.
C OROLLARY 18.16. The polynomial ring kX1 ; : : : ; Xn has Krull dimension n.
P ROOF. The existence of the sequence of prime ideals
.X1 ; : : : ; Xn /  .X1 ; : : : ; Xn

1/




 .X1 /  .0/

shows that kX1 ; : : : ; Xn has Krull dimension at least n. Now(18.15) completes the proof.
T HEOREM 18.17. Let A be an integral domain containing a field k and finitely generated
as a k-algebra. Then
tr degk F .A/ D dim.A/:
P ROOF. According to the Noether normalization theorem (8.1), A is integral over a polynomial subring kx1 ; : : : ; xn . Clearly n Dtr degk F .A/. From the going-up theorem (7.7),
dim.A/  dim.kx1 ; : : : ; xn / D n, and so dim.A/ D n (18.15).

19

Primary decompositions

D EFINITION 19.1. An ideal q in A is primary if it is A and

ab 2 q, b q H) an 2 q for some n  1:
Thus, a proper ideal q in A is primary if and only if every zero-divisor in A=q is nilpotent.
Therefore, a radical ideal is primary if and only if it is prime, and an ideal .m/ in Z is primary
if and only if m is a power of a prime.
P ROPOSITION 19.2. The radical of a primary ideal q is a prime ideal containing q, and
it is contained in every other prime ideal containing q (i.e., it is the smallest prime ideal
containing p).
P ROOF. Suppose that ab 2 rad.q/ but b rad.q/. Some power, say am b m , of ab lies xin q,
but b m q, and so .am /n 2 q for some n. Hence, a 2 rad.q/. Therefore rad.q/ is prime.
Let p be a second prime ideal containing q, and let a 2 rad.q/. For some n, an 2 q  p,
which implies that a 2 p. Therefore p  rad.q/.
When q is a primary ideal and p is its radical, we say that q is p-primary. Note that this
means that if ab 2 q, then either a 2 q or b 2 p (or both).

19

PRIMARY DECOMPOSITIONS

87

P ROPOSITION 19.3. Every ideal q whose radical is a maximal ideal m is primary (in fact,
m-primary); in particular, every power of a maximal ideal m is m-primary.
P ROOF. Every prime ideal containing q contains its radical m, and therefore equals m. This
shows that A=q is local with maximal ideal m=q. Therefore, every element of A=q is either
a unit, and hence is not a zero-divisor, or it lies in m=q, and hence is nilpotent.
P ROPOSITION 19.4. Let 'W A ! B be a homomorphism of rings. If q is a p-primary ideal
def
in B, then qc D ' 1 .q/ is a pc -primary ideal in A.
P ROOF. The map A=qc ! B=q is injective, and so every zero-divisor in A=qc is nilpotent.
This shows that qc is primary, and it remains to show that rad.qc / D pc . But
2.11

rad.qc / D rad.q/c D pc
as claimed.
L EMMA 19.5. Let q and p be ideals in A such that
(a) q  p  rad.q/ and
(b) ab 2 q H) a 2 p or b 2 q.
Then p is a prime ideal and q is p-primary.
P ROOF. Clearly q is primary, hence rad.q/-primary, and rad.q/ is prime. By assumption
p  rad.q/, and it remains to show that they are equal. Let a 2 rad.q/, and let n be the
smallest positive integer such that an 2 q. If n D 1, then a 2 q  p; on the other hand, if
n > 1, then an D aan 1 2 q and an 1 q, and so a 2 p by (b).
P ROPOSITION 19.6. A finite intersection of p-primary ideals is p-primary.
P ROOF. Let q1 ; : : : ; qr be p-primary, and let q D q1 \ : : : \ qr . We show that the pair of
ideals q  p satisfies the conditions of (19.5).
T
Let a 2 p. Then some power of a, say, ani , lies in qi , and amax.ni / 2 qi D q. Therefore
p  rad.q/.
Let ab 2 q, so ab 2 qi all i . If a p, then b 2 qi all i , and so b 2 q.
The minimal prime ideals of an ideal a are the minimal elements of the set of prime
ideals containing a.
D EFINITION 19.7. A primary decomposition of an ideal a is a finite set of primary ideals
whose intersection is a. Such a decomposition S of a is minimal if
(a) the prime ideals rad.q/, q 2 S , are distinct, and
T
(b) no element of S can be omitted, i.e., for no q 2 S does q  fq0 j q0 2 S , q0 qg.
If a admits a primary decomposition, then it admits a minimal primary decomposition,
because Proposition 19.6 can be used to combine primary ideals with the same radical, and
any q that fails (b) can simply be omitted. The prime ideals occurring as the radical of an
ideal in a minimal primary decomposition of a are said to belong to a.
P ROPOSITION 19.8. Suppose that a D q1 \


\ qn where qi is pi -primary for i D 1; : : : ; n.
Then the minimal prime ideals of a are the minimal elements of the set fp1 ; : : : ; pn g.

19

PRIMARY DECOMPOSITIONS

88

P ROOF. Let p be a prime ideal containing a. Then p is a prime ideal containing q1


qn ,

and so p contains one of the ideals qi (2.1b). Now (19.2) shows that p contains pi :
In particular, if a admits a primary decomposition, then it has only finitely many minimal
prime ideals, and so its nilradical is a finite intersection of prime ideals (which is always the
case for noetherian rings, see 14.9).
For an ideal a in A and an element x 2 A, we let
.aW x/ D fa 2 A j ax 2 ag:
It is again an ideal in A, which contains a, and equals A if x 2 a.
L EMMA 19.9. Let q be a p-primary ideal and let x 2 A X q. Then .qW x/ is p-primary.
P ROOF. Let a 2 .qW x/; then ax 2 q and x q, and so a 2 p. Therefore p  .qW x/  q. On
taking radicals, we find that rad.qW x/ D p.
Let ab 2 .qW x/, so that abx 2 q. If a p, then bx 2 q, and so b 2 .qW x/. Therefore,
.qW x/ is primary, and hence p-primary.
T HEOREM 19.10. Let a D q1 \ : : : \ qn be a minimal primary decomposition of a, and let
pi D rad.qi /. Then
fp1 ; : : : ; pn g D frad.aW x/ j x 2 A;

rad.aW x/ primeg.

In particular, the set fp1 ; : : : ; pn g is independent of the choice of the minimal primary
decomposition.
P ROOF. For any a 2 A,
T
T
.aW a/ D . qi W a/ D .qi W a/;
and so
T
T
rad.aW a/ D rad . .qi W a// D rad.qi W a/:
Now rad.qi W a/ D pi or A according as a qi or a 2 qi (19.9), and so
\
rad.aW a/ D
pi .

(50)

i such that aqi

If rad.aW a/ is prime, then it contains one of the pi (2.1), and hence equals it, i.e.,
rad.aW a/ 2 fp1 ; : : : ; pn g.
T
On the other hand, for each i , there exists an a 2 j i qj X qi because the decomposition
is minimal, and (50) shows that rad.aW a/ D pi .
An ideal a is said to be irreducible if it cannot be expressed as the intersection of two
strictly large ideals, i.e., if
a D b \ c (b, c ideals) H) a D b or a D c:
T HEOREM 19.11. In a noetherian ring A, every ideal admits a primary decomposition.
More precisely:
(a) Every ideal in A can be expressed as a finite intersection of irreducible ideals.

19

PRIMARY DECOMPOSITIONS

89

(b) Every irreducible ideal in A is primary.

P ROOF. (a) Let S be the set of ideals for which (a) fails. If S is empty, then (a) is true.
Otherwise, it contains a maximal element a. Then a itself is not irreducible, and so a D b \ c
with b and c properly containing a. As a is maximal in S, both b and c can be expressed as
finite intersections of irreducible ideals, but then so can a.
def
(b) Let a be irreducible in A, and consider the quotient ring A0 D A=a: Let a be a
zero-divisor in A0 , say, ab D 0 with b 0. We have to show that a is nilpotent. As A0 is
noetherian, the chain of ideals
..0/W a/  ..0/W a2 / 



becomes constant, say,
..0/W am / D ..0/W amC1 // D


:
Let c 2 .b/ \ .am /. Because c 2 .b/, ca D 0, and because c 2 .am /, c D dam for some
d 2 A. But
.dam /a D 0 ) d 2 .0W amC1 / D .0W am / ) c D 0;
and so .b/ \ .am / D .0/. Because a is irreducible, the zero ideal in A0 is irreducible, and it
follows that am D 0.
A p-primary ideal q in a noetherian ring contains a power of p by Proposition 3.16. The
next result proves a converse when p is maximal.
P ROPOSITION 19.12. Let m be a maximal ideal of a noetherian ring. Any proper ideal a of
A that contains a power of a maximal ideal m is m-primary.
P ROOF. Suppose that mr  a, and let p be a prime ideal belonging to a. Then mr  a  p,
so that m  p, which implies that m D p. Thus m is the only prime ideal belonging to a,
which means that a is m-primary.
E XAMPLE 19.13. We give an example of a power of a prime ideal p that is not p-primary.
Let
A D kX; Y; Z=.Y 2 XZ/ D kx; y; z:
The ideal .X; Y / in kX; Y; Z is prime and contains .Y 2 XZ/, and so the ideal p D .x; y/
in A is prime. Now xz D y 2 2 p2 , but one checks easily that x p2 and z p, and so p2 is
not p-primary.
T
R EMARK 19.14. Let a be an ideal in a noetherian ring, and let b D n1 an . We give
another proof that ab D b (see p. 13). Let
ab D q1 \ : : : \ qs ;

rad.qi / D pi ;

be a minimal primary decomposition of ab. We shall show that b  ab by showing that

b  qi for each i .
If there exists a b 2 b X qi , then
ab  ab  qi ,
from which it follows that a  pi . We know that pri  qi for some r (see 3.16), and so
\
bD
an  ar  pri  qi ,
which is a contradiction. This completes the proof.

19

PRIMARY DECOMPOSITIONS

90

Primary decompositions for modules

Let M be a module over a ring A. The statements for modules below can be proved as for
ideals, or deduced from them by considering the ring A M (see 2.14).
For a submodule N of M , let
.N W M / D fa 2 A j aM  N g.
It is an ideal in A. Let
rM .N / D rad..N W M // D fa 2 A j an M  N for some n  0g:
An element a of A is a zero divisor of M if ax D 0 for some nonzero x 2 M , and it
is nilpotent on M if an M D 0 for some n. A submodule Q of M is primary if every zero
divisor of M=Q is nilpotent on M=Q.
P ROPOSITION 19.15. If Q is a primary submodule of M , then .QW M / is a primary ideal,
and so rM .Q/ is a prime ideal p. We say that Q is p-primary in M .
For simplicity, we now assume that A is noetherian and that M is finitely generated.
A prime ideal of A is an associated prime ideal of M if it is the annihilator ann.x/ of
an element of M . We write Ass.M / for the set of associated prime ideals of M .
P ROPOSITION 19.16. A submodule Q of M is primary if and only if Ass.M=Q/ consists
of a single element p, in which case p D rM .Q/.
P ROPOSITION 19.17. A finite intersection of p-primary submodules is p-primary.
A primary decomposition of a submodule N is a finite set of primary submodules whose
intersection is N . A primary decomposition S is minimal if
(a) the prime ideals rM .Q/, Q 2 S , are distinct, and
T
(b) no element of S can be omitted, i.e., for no Q 2 S does Q  fQ0 j Q0 2 S , Q0 Qg.
If N admits a primary decomposition, then it admits a minimal primary decomposition,
because (19.17) can be used to combine submodules with the same p, and any Q that fails
(b) can simply be omitted.
A submodule of M is irreducible it cannot be expressed as the intersection of two strictly
larger submodules.
T HEOREM 19.18. Every submodule of M (as above) admits a primary decomposition.
More precisely:
(a) Every submodule of M can be expressed as a finite intersection of irreducible submodules.
(b) Every irreducible submodule in M is primary.
T HEOREM 19.19. Let N be a submodule of M . Let N D Q1 \ : : : \ Qn be a minimal
primary decomposition of N , and let pi D rM .Qi /. Then
fp1 ; : : : ; pn g D Ass.M=N /.
In particular, the set fp1 ; : : : ; pn g is independent of the choice of the minimal primary
decomposition. Its elements are called the prime ideals belonging to N (in M /.

20

20

DEDEKIND DOMAINS

91

Dedekind domains

Discrete valuation rings

It follows from the elementary theory of principal ideal domains that the following conditions
on a principal ideal domain A are equivalent:
(a) A has exactly one nonzero prime ideal;
(b) A has exactly one prime element up to associates;
(c) A is local and is not a field.
A ring satisfying these conditions is called a discrete valuation ring.
E XAMPLE 20.1. The ring Z.p/ D f m
n 2 Q j n not divisible by pg is a discrete valuation ring
with .p/ as its unique nonzero prime ideal. The units in Z.p/ are the nonzero elements m=n
with neither m nor n divisible by p, and the prime elements are those of the form unitp.
def

In a discrete valuation ring A with prime element , nonzero elements of A can be

expressed uniquely as u m with u a unit and m  0 (and m > 0 unless the element is a unit).
Every nonzero ideal in A is of the form . m / for a unique m 2 N. Thus, if a is an ideal in
A and p denotes the (unique) maximal ideal of A, then a D pm for a well-defined integer
m  0.
Recall that, for an A-module M and an m 2 M , the annihilator of m
ann.m/ D fa 2 A j am D 0g:
It is an ideal in A, which is proper if m 0. Suppose that A is a discrete valuation ring,
and let c be a nonzero element of A. Let M D A=.c/. What is the annihilator of a nonzero
element b C .c/ of M ? Fix a prime element  of A, and let c D u m , b D v n with u and
v units. Then n < m (else b C .c/ D 0 in M ), and
ann.b C .c// D . m

/:

Thus, a b for which ann.b C .c// is maximal, is of the form v m 1 , and for this choice
ann.b C .c// is a prime ideal generated by bc . We shall exploit these observations in the
proof of the next proposition, which gives a criterion for a ring to be a discrete valuation
ring.
P ROPOSITION 20.2. An integral domain A is a discrete valuation ring if and only if
(a) A is Noetherian,
(b) A is integrally closed, and
(c) A has exactly one nonzero prime ideal.
P ROOF. The necessity of the three conditions is obvious, and so let A be an integral domain
satisfying (a), (b), and (c). We have to show that every ideal in A is principal. As a first step,
we prove that the nonzero prime ideal is principal. Note that (c) implies that A is a local ring.
def
Choose an element c 2 A, c 0, c unit, and consider the A-module M D A=.c/. For
each nonzero element m of M ,
ann.m/ D fa 2 A j am D 0g

20

DEDEKIND DOMAINS

92

is a proper ideal in A. Because A is Noetherian, we can choose an m so that ann.m/ is

maximal among these ideals. Write m D b C .c/ and p D ann.b C .c//. Note that c 2 p, and
so p 0, and that
p D fa 2 A j cjabg:
I claim that p is prime. If not there exist elements x, y 2 A such that xy 2 p but neither x
nor y 2 p. Then yb C .c/ is a nonzero element of M because y p. Consider ann.yb C .c//.
Obviously it contains p and it contains x, but this contradicts the maximality of p among
ideals of the form ann.m/. Hence p is prime.
I claim that bc A. Otherwise b D c
bc 2 .c/, and m D 0 (in M ).
I claim that bc 2 A, and p D . bc /. By definition, pb  .c/, and so p
bc  A, and it is an
ideal in A. If p
bc  p, then bc is integral over A (by 6.1, since p is finitely generated), and
so bc 2 A (because of condition (b)), but we know bc A. Thus p
bc D A (by (c)), and this
implies that p D . bc /:
Let  D bc , so that p D ./. Let a be a proper ideal of A, and consider the sequence
a  a

 a




:

If a r D a r 1 for some r, then  1 .a r / D a r , and  1 is integral over A (by

6.1), and so lies in A this is impossible ( is not a unit in A). Therefore the sequence
is strictly increasing, and (again because A is Noetherian) it cant be contained in A. Let
m be the smallest integer such that a m  A but a m 1 A. Then a m p, and so
a m D A. Hence a D . m /:

Dedekind domains
D EFINITION 20.3. A Dedekind domain is an integral domain A, not equal to a field, such
that
(a) A is Noetherian,
(b) A is integrally closed, and
(c) every nonzero prime ideal is maximal (i.e., A has Krull dimension 1).
Thus Proposition 20.2 says that a local integral domain is a Dedekind domain if and only
if it is a discrete valuation ring.
P ROPOSITION 20.4. Let A be a Dedekind domain, and let S be a multiplicative subset of
A. Then S 1 A is either a Dedekind domain or a field.
P ROOF. Condition (c) says that there is no containment relation between nonzero prime
ideals of A. If this condition holds for A, then (5.4) shows that it holds for S 1 A. Conditions
(a) and (b) follow from the next lemma.
P ROPOSITION 20.5. Let A be an integral domain, and let S be a multiplicative subset of A.
(a) If A is Noetherian, then so also is S 1 A:
(b) If A is integrally closed, then so also is S 1 A:
P ROOF. (a) Let a be an ideal in S 1 A. Then a D S
by any (finite) set of generators for a \ A:

1 .a \ A/ (see 5.4),

and so a is generated

20

DEDEKIND DOMAINS

93

(b) Let be an element of the field of fractions of A (D field of fractions of S

is integral over S 1 A. Then
m C a1 m

C


C am D 0, some ai 2 S

1 A/

that

A:

For each i , there exists an si 2 S such that si ai 2 A. Set s D s1


sm 2 S, and multiply

through the equation by s m W
.s/m C sa1 .s/m

C


C s m am D 0:

This equation shows that s is integral over A, and so lies in A. Hence D .s/=s 2 S
(See also 6.15.)

1 A.

C OROLLARY 20.6. For any nonzero prime ideal p in a Dedekind domain A, the localization
Ap is a discrete valuation ring.
P ROOF. We saw in (5.7) that Ap is local, and the proposition implies that it is Dedekind.
The main result concerning Dedekind domains is the following.
T HEOREM 20.7. Every proper nonzero ideal a in a Dedekind domain can be written in the
form
a D pr11


prss
with the pi distinct prime ideals and the ri > 0; the ideals pi are exactly the prime ideals
containing a, and the exponents ri are uniquely determined.
P ROOF. The primary ideals in a Dedekind domain are exactly the powers of prime ideals,
and so this follows from the preceding section. (For an elementary proof, see my notes on
algebraic number theory.)
R EMARK 20.8. Note that
ri > 0 aApi Api a  pi :
C OROLLARY 20.9. Let a and b be ideals in A; then
a  b aAp  bAp
for all nonzero prime ideals p of A. In particular, a D b if and only if aAp D bAp for all p.
P ROOF. The necessity is obvious. For the sufficiency, factor a and b
a D pr11


prmm ;

b D ps11


psmm ;

ri ; si  0:

Then aApi D pri i Api and aApi D pri i Api

aApi  bApi ri  si ;
(recall that Api is a discrete valuation ring) and ri  si all i implies a  b.
C OROLLARY 20.10. Let A be an integral domain with only finitely many prime ideals;
then A is a Dedekind domain if and only if it is a principal ideal domain.

20

DEDEKIND DOMAINS

94

P ROOF. Assume A is a Dedekind domain. After (20.7), to show that A is principal, it

suffices to show that the prime ideals are principal. Let p1 ; : : : ; pm be these ideals. Choose
an element x1 2 p1 X p21 . According to the Chinese Remainder Theorem (2.13), there is an
element x 2 A such that
x  x1

mod p21 ;

x1

mod pi ;

i 1:

Now the ideals p1 and .x/ generate the same ideals in Api for all i , and so they are equal in
A (by 20.9).
C OROLLARY 20.11. Let a  b 0 be two ideals in a Dedekind domain; then a D b C .a/
for some a 2 A:
P ROOF. Let b D pr11


prmm and a D ps11


psmm with ri ; sj  0. Because b  a, si  ri for
all i . For 1  i  m, choose an xi 2 A such that xi 2 psi i , xi psi i C1 . By the Chinese
Remainder Theorem, there is an a 2 A such that
a  xi

mod pri i , for all i:

Now one sees that b C .a/ D a by looking at the ideals they generate in Ap for all p:
C OROLLARY 20.12. Let a be an ideal in a Dedekind domain, and let a be any nonzero
element of a; then there exists a b 2 a such that a D .a; b/:
P ROOF. Apply (20.11) to a  .a/:
C OROLLARY 20.13. Let a be a nonzero ideal in a Dedekind domain; then there exists a
nonzero ideal a in A such that aa is principal. Moreover, a can be chosen to be relatively
prime to any particular ideal c, and it can be chosen so that aa D .a/ with a any particular
element of a (but not both).
P ROOF. Let a 2 a, a 0; then a  .a/, and so we have
.a/ D pr11


prmm and a D ps11


psmm ;

si  ri :

If a D pr11 s1


prmm sm , then aa D .a/:

We now show that a can be chosen to be prime to c. We have a  ac, and so (by 20.11)
there exists an a 2 a such that a D ac C .a/. As a  .a/, we have .a/ D a
a for some ideal
a (by the above argument); now, ac C aa D a, and so c C a D A. (Otherwise c C a  p
some prime ideal, and ac C aa D a.c C a /  ap a:/
In basic graduate algebra courses, it is shown that
A a principal ideal domain ) A is a unique factorization domain.
The converse is false because, for example, kX; Y is a unique factorization domain in
which the ideal .X; Y / is not principal, but it is true for Dedekind domains.
P ROPOSITION 20.14. A Dedekind domain that is a unique factorization domain is a principal ideal domain.

20

DEDEKIND DOMAINS

95

P ROOF. In a unique factorization domain, an irreducible element  can divide a product

bc only if  divides b or c (write bc D q and express each of b, c, and q as a product of
irreducible elements). This means that ./ is a prime ideal.
Now let A be a Dedekind domain with unique factorization. It suffices to show that each
nonzero prime ideal p of A is principal. Let a be a nonzero element of p. Then a factors into
a product of irreducible elements (see 4.4) and, because p is prime, it will contain one of
these irreducible factors . Now p  ./  .0/, and, because ./ is a nonzero prime ideal,
it is maximal, and so equals p.

Modules over Dedekind domains.

The structure theorem for finitely generated modules over principal ideal domains has an
interesting extension to modules over Dedekind domains. Throughout this subsection, A is a
Dedekind domain.
First, note that a finitely generated torsion-free A-module M need not be free. For
example, every fractional ideal is finitely generated and torsion-free but it is free if and only
if it is principal. Thus the best we can hope for is the following.
T HEOREM 20.15. Let A be a Dedekind domain.
(a) Every finitely generated torsion-free A-module M is isomorphic to a direct sum of
fractional ideals,
M  a1


am :
(b) Two finitely generated torsion-free A-modules
1


am and N  b1



QM  aQ
bn are isomorphic if and only if m D n and ai  bi modulo principal ideals.
Hence,
M  a1


am  A


A a1


am :
Moreover, two fractional ideals a and b of A are isomorphic as A-modules if and only if they
define the same element of the class group of A.
P ROOF. (a) Let A be a Dedekind domain, and let M be finitely generated torsion-free
A-module. Then Ap M is free, hence projective, for every nonzero prime ideal p in A
(because Ap is principal ideal domain), and so M is projective (12.5). From a surjective
homomorphism Ar ! M , we get a homomorphism M ! Ar whose composite with some
projection Ar ! A will be nonzero, and hence have image a nonzero ideal a in A. As
a is projective, there exists a section to the map M  a, and so M  a M1 for some
submodule M1 of M . Now M1 is projective because it is a direct summand of a projective
module, and so we can repeat the argument with M1 . This process ends because M is
noetherian.
(b) Omitted.
The rank of a module M over an integral domain R is the dimension of K R M as a
K-vector space, where K is the field of fractions of R. Clearly the rank of M  a1


am
is m:
These remarks show that the set of isomorphism classes of finitely generated torsion-free
A-modules of rank 1 can be identified with the class group of A. Multiplication of elements
in Cl(A) corresponds to the formation of tensor product of modules. The Grothendieck
group of the category of finitely generated A-modules is Cl.A/ Z.

21

DIMENSION THEORY FOR NOETHERIAN RINGS

96

T HEOREM 20.16 (I NVARIANT FACTOR THEOREM ). Let M  N be finitely generated torsionfree A-modules of the same rank m. Then there exist elements e1 ; :::; em of M , fractional
ideals a1 ; :::; am , and integral ideals b1  b2  ::.  bm such that
M D a1 e1


am em ;

N D a1 b1 e1


am bm em :

P ROOF. Omitted.
The ideals b1 , b2 , ..., bm are uniquely determined by the pair M  N , and are called the
invariant factors of N in M:
The last theorem also yields a description of finitely generated torsion A-modules.
A SIDE 20.17. The Jordan-Holder and Krull-Schmidt theorems fail for finitely generated projective
modules over non-principal Dedekind domains. For example, suppose that A has a nonprincipal ideal
a of order 2 in the class group. Then a a  A A, contradicting both theorems.

21

Dimension theory for noetherian rings

Let A be a noetherian ring and let p be a prime ideal in A. Let Ap D S 1 A where S D A X p.

We begin by studying extension and contraction of ideals with respect to the homomorphism
def
A ! Ap (cf. 2.10). Recall (5.7) that Ap is a local ring with maximal ideal pe D pAp . The
ideal

ec
pn
D fa 2 A j sa 2 pn for some s 2 Sg
is called the nth symbolic power of p, and is denoted p.n/ . If m is maximal, then m.n/ D mn
(see 5.8).
L EMMA 21.1. The ideal p.n/ is p-primary.
P ROOF. According to Proposition 19.3, the ideal .pe /n is pe -primary. Hence (see 19.4),
..pe /n /c is .pe /c -primary. But pec D p (see 5.4), and
2.11

...pe /n /c D ..pn /e /c D p.n/ :

def

(51)

L EMMA 21.2. Consider ideals a  p0  p with p0 prime. If p0 is a minimal prime ideal of a,

then p0e is a minimal prime ideal of ae (extension relative to A ! Ap ).
P ROOF. If not, there exists a prime ideal p00 p0e such that p0e  p00  ae . Now, by (5.4),
p0 D p0ec and p00c p0ec , and so
p0 D p0ec p00c  aec  a
contradicts the minimality of p0 .
T HEOREM 21.3 (K RULL S PRINCIPAL IDEAL THEOREM ). Let A be a noetherian ring. For
any nonunit b 2 A, the height of a minimal prime ideal p of .b/ is at most one.

21

DIMENSION THEORY FOR NOETHERIAN RINGS

97

P ROOF. Consider A ! Ap . According to Lemma 21.2, pe is a minimal prime ideal of

.b/e D . b1 /, and (5.4) shows that the theorem for Ap  pe  . b1 / implies it for A  p  .b/.
Therefore, we may replace A with Ap , and so assume that A is a noetherian local ring with
maximal ideal p.
Suppose that p properly contains a prime ideal p1 : we have to show that p1  p2 H)
p1 D p2 .
.r/
Let p1 be the rth symbolic power of p1 . The only prime ideal of the ring A=.b/ is
p=.b/, and so A=.b/ is artinian (apply 16.6). Therefore the descending chain of ideals






.1/
.2/
.3/
p1 C .b/ =.b/  p1 C .b/ =.b/  p1 C .b/ =.b/ 



eventually becomes constant: there exists an s such that
.s/

.sC1/

p1 C .b/ D p1

.sC2/

C .b/ D p1

C .b/ D


:

(52)

We claim that, for any m  s,

.m/

p1

.m/

 .b/p1

.mC1/

C p1

(53)

.m/

Let x 2 p1 . Then
.m/ (52)

x 2 .b/ C p1

.mC1/

D .b/ C p1

.mC1/

.m/

and so x D ab C x 0 with a 2 A and x 0 2 p1

. As p1 is p1 -primary (see 21.1) and
.m/
.m/
.m/
.mC1/
0
ab D x x 2 p1 but b p1 , we have that a 2 p1 . Now x D ab C x 0 2 .b/p1 C p1
as claimed.
We next show that, for any m  s,
.m/

.mC1/

D p1
.


.m/ .mC1/
.m/ .mC1/
.m/ .mC1/
As b 2 p, (53) shows that p1 =p1
D p
p1 =p1
, and so p1 =p1
D 0 by
Nakayamas lemma (3.9).
Now
.s/
.sC1/
.sC2/
ps1  p1 D p1
D p1
D



T
.m/
and so ps1  ms p1 . Note that
p1

\
ms

.m/ (51)

p1

\
ms

\
..pe1 /m /c D .

ms

3.15

.pe1 /m /c D .0/c ;

and so for any x 2 ps1 , there exists an a 2 A X p1 such that ax D 0. Let x 2 p1 ; then ax s D 0
for some a 2 A X p1  A X p2 , and so x 2 p2 (because p2 is prime). We have shown that
p1 D p2 , as required.
C OROLLARY 21.4. A noetherian integral domain A is a unique factorization domain if
every prime ideal of height 1 is principal.
P ROOF. After (4.1) and (4.3), it suffices to show that every irreducible element a of A is
prime. Let p be minimal among the prime ideals containing .a/. According to the principal
ideal theorem (21.3), p has height 1, and so is principal, say p D .b/. As .a/  .b/, b divides
a, and so a D bunit. Hence .a/ D .b/ D p, and p is prime.

21

DIMENSION THEORY FOR NOETHERIAN RINGS

98

In order to extend Theorem 21.7 to non principal ideals, we shall need a lemma.
L EMMA 21.5. Let p be a prime ideal in a noetherian ring A, and let S be a finite set of
prime ideals in A, none of which contains p. If there exists a chain of distinct prime ideals
p  pd




 p0 ;

then there exists such a chain with p1 not contained in any ideal in S.
P ROOF. We first prove this in the special case that the chain has length 2. Suppose that
p  p1  p0 are distinct prime ideals and that p is not contained in any prime ideal in S.
According to Proposition 2.8, there exists an element
S
a 2 p X .p0 [ fp0 2 S g/:
As p contains .a/ C p0 , it also contains a minimal prime ideal p01 of .a/ C p0 . Now p01 =p0
is a minimal prime ideal of the principal ideal ..a/ C p0 / =p0 in A=p0 , and so has height 1,
whereas the chain p=p0  p1 =p0  p0 =p0 shows that p=p0 has height at least 2. Therefore
p  p01  p0 are distinct primes, and p01 S because it contains a. This completes the proof
of the special case.
Now consider the general case. On applying the special case to p  pd 1  pd 2 , we
see that there exists a chain of distinct prime ideals p  p0d 1  pd 2 such that p0d 1 is not
contained in any ideal in S. Then on applying the special case to p0d 1  pd 2  pd 1 , we
we see that there exists a chain of distinct prime ideals p  p0d 1  p0d 2  pd 2 such that
p0d 2 is not contained in any ideal in S . Repeat the argument until the proof is complete.
T HEOREM 21.6. Let A be a noetherian ring. For any proper ideal a D .a1 ; : : : ; am /, the
height of a minimal prime ideal of a is at most m.
P ROOF. For m D 1, this was just proved. Thus, we may suppose that m  2 and that the
theorem has been proved for ideals generated by m 1 elements. Let p be a minimal prime
ideal of a, and let p01 ; : : : ; p0t be the minimal prime ideals of .a2 ; : : : ; am /. Each p0i has height
at most m 1. If p is contained in one of the p0i , it will have height  m 1, and so we may
suppose that it isnt.
Let p have height d . We have to show that d  m. According to the lemma, there exists
a chain of distinct prime ideals
p D pd  pd




 p0 ;

d  1;

with p1 not contained in any p0i , and so Proposition 2.8 shows that there exists a
S
b 2 p1 X riD1 p0i :
We next show that p is a minimal prime ideal of .b; a2 ; : : : ; am /. Certainly p contains a
minimal prime ideal p0 of this ideal. As p0  .a2 ; : : : ; am /, p contains one of the p0i s, but, by
construction, it cannot equal it. If p p0 , then
p  p0  p0i
def
def
are distinct ideals, which shows that x
p D p=.a2 ; : : : ; am / has height at least 2 in Ax D
A=.a2 ; : : : ; am /. But x
p is a minimal ideal in Ax of the principal ideal .a1 ; : : : ; an /=.a2 ; : : : ; an /,
which contradicts Theorem 21.3. Hence p is minimal, as claimed.

21

DIMENSION THEORY FOR NOETHERIAN RINGS

99

But now p=.b/ is a minimal prime ideal of .b; a2 ; : : : ; am / in R=.b/, and so the height
of p=.b/ is at most m 1 (by induction). The prime ideals
p=.b/ D pd =.b/  pd
are distinct, and so d

1m

1 =.b/




 p1 =.b/

1. This completes the proof that d D m.

The height of an ideal a in a noetherian ring is the minimum height of a prime ideal
containing it,
ht.a/ D min ht.p/:
pa, p prime

The theorem shows that ht.a/ is finite.

The following provides a (strong) converse to Theorem 21.6.
T HEOREM 21.7. Let A be a noetherian ring, and let a be a proper ideal of A of height r.
Then there exist r elements a1 ; : : : ; ar of a such that, for each i  r, .a1 ; : : : ; ai / has height
i.
P ROOF. If r D 0, then we take the empty set of ai s. Thus, suppose that r  1. There are
only finitely many prime ideals of height 0, because such an ideal is a minimal prime ideal
of .0/, and none of these ideals can contain a because it has height  1. Proposition 2.8
shows that there exists an
S
a1 2 a X fprime ideals of height 0g:
By construction, .a1 / has height at least 1, and so Theorem 21.3 shows it has height exactly
1.
This completes the proof when r D 1, and so suppose that r  2. There are only finitely
many prime ideals of height 1 containing .a1 / because such an ideal is a minimal prime
ideal of .a1 /, and none of these ideals can contain a because it has height  2. Choose
S
a2 2 a X fprime ideals of height 1 containing .a1 /g:
By construction, .a1 ; a2 / has height at least 2, and so Theorem 21.6 shows that it has height
exactly 2.
This completes the proof when r D 2, and when r > 2 we can continue in this fashion
until it is complete.
C OROLLARY 21.8. Every prime ideal of height r in a noetherian ring arises as a minimal
prime ideal for an ideal generated by r elements.
P ROOF. According to the theorem, an ideal a of height r contains an ideal .a1 ; : : : ; ar / of
height r. If a is prime, then it is a minimal ideal of .a1 ; : : : ; ar /.
C OROLLARY 21.9. Let A be a commutative noetherian ring, and let a be an ideal in A that
can be generated by n elements. For any prime ideal p in A containing a,
ht.p=a/  ht.p/ ht.p=a/ C n:

22

REGULAR LOCAL RINGS

100

P ROOF. The first inequality follows immediately from the correspondence between ideals
in A and in A=a.
def
Denote the quotient map A ! A0 D A=a by a 7! a0 . Let ht.p=a/ D d . Then there
exist elements a1 ; : : : ; ad in A such that p=a is a minimal prime ideal of .a10 ; : : : ; ad0 /. Let
b1 ; : : : ; bn generate a. Then p is a minimal prime ideal of .a1 ; : : : ; ad ; b1 ; : : : ; bn /, and hence
has height  d C n.
We now use dimension theory to prove a stronger version of generic flatness (11.16).
T HEOREM 21.10 (G ENERIC FREENESS ). Let A be a noetherian integral domain, and let
B be a finitely generated A-algebra. For any finitely generated B-module M , there exists a
nonzero element a of A such that Ma is a free Aa -module.
P ROOF. Let F be the field of fractions of A. We prove the theorem by induction on the
Krull dimension of F A B, starting with the case of Krull dimension 1. Recall that this
means that F A B D 0, and so a1B D 0 for some nonzero a 2 A. Then Ma D 0, and so
the theorem is trivially true (Ma is the free Aa -module generated by the empty set).
In the general case, an argument as in (11.16) shows that, after replacing A, B, and M
with Aa , Ba , and Ma for a suitable a 2 A, we may suppose that the map B ! F A B
is injective we identify B with its image. The Noether normalization theorem (8.1)
shows that there exist algebraically independent elements x1 ; : : : ; xm of F A B such that
F A B is a finite F x1 ; : : : ; xm -algebra. As in the proof of (11.16), there exists a nonzero
a 2 A such that Ba is a finite Aa x1 ; : : : ; xm -algebra. Hence Ma is a finitely generated
Aa x1 ; : : : ; xm -module.
As any extension of free modules is free24 , Proposition 3.5 shows that it suffices to
prove the theorem for Ma D Aa x1 ; : : : ; xm =p for some prime ideal p in Aa x1 ; : : : ; xm . If
p D 0, then Ma is free over Aa (with basis the monomials in the xi ). Otherwise, F A
.Aa x1 ; : : : ; xm =p/ has Krull dimension less than that of F A B, and so we can apply the
induction hypothesis.
C OROLLARY 21.11. Let A be a noetherian ring, and let 'W A ! B be a finitely generated
A-algebra. If U is a nonempty open subset of Spec.B/, then ' a .U / contains a nonempty
open subset of its closure in Spec.A/.
P ROOF. We may replace A with its image in B, and B with Bf for some f such that
D.f /  U . Then we have to show that the image of ' a W Spec.B/ ! Spec.A/ contains a
nonempty open subset of Spec.A/. According to (21.10), there exists an a 2 A such that Ba
is a nonzero free Aa -module. For any prime ideal p of A not containing a, B A A=p '
Ba Aa A=p 0. As B A A=p is nonzero, it contains a prime ideal, but the prime ideals
in B A A=p correspond to prime ideals q in B such that q \ A D p. Therefore the image of
' a contains D.a/.

22

Regular local rings

Throughout this section, A is a noetherian local ring with maximal ideal m and residue field
k. The Krull dimension d of A is equal to the height of m, and
(21.6)

(3.11)

ht.m/  minimum number of generators of m D dimk .m=m2 /:

24 If

def

M 0 is a submodule of M such that M 00 D M=M 0 is free, then M  M 0 M 00 .

22

REGULAR LOCAL RINGS

101

When equality holds, the ring A is said to be regular. In other words, dimk .m=m2 /  d ,
and equality holds exactly when the ring is regular.
For example, when A has dimension zero, it is regular if and only if its maximal ideal
can be generated by the empty set, and so is zero. This means that A is a field; in particular,
it is an integral domain. The main result of this section is that all regular rings are integral
domains.
L EMMA 22.1. Let A be a noetherian local ring with maximal ideal m, and let c 2 m X m2 .
def
Denote the quotient map A ! A0 D A=.c/ by a 7! a0 . Then
dimk m=m2 D dimk m0 =m02 C 1
def

where m0 D m=.c/ is the maximal ideal of A0 .

P ROOF. Let e1 ; : : : ; en be elements of m such that fe10 ; : : : ; en0 g is a k-linear basis for m0 =m02 .
We shall show that fe1 ; : : : ; en ; cg is a basis for m=m2 .
As e10 ; : : : ; en0 span m0 =m02 , they generate the ideal m0 (see 3.11), and so m D .e1 ; : : : ; en /C
.c/, which implies that fe1 ; : : : ; en ; cg spans m=m2 .
Suppose that a1 ; : : : ; anC1 are elements of A such that
a1 e1 C


C an en C anC1 c  0 mod m2 .

(54)

Then
a10 e10 C


C an0 en0  0 mod m02 ,
and so a10 ; : : : ; an0 2 m0 . It follows that a1 ; : : : ; an 2 m. Now (54) shows that anC1 c 2 m2 .
If anC1 m, then it is a unit in A, and c 2 m2 , which contradicts its definition. Therefore,
anC1 2 m, and the relation (54) is the trivial one.
P ROPOSITION 22.2. If A is regular, then so also is A=.a/ for any a 2 m X m2 ; moreover,
dim A D dim A=.a/ C 1.
P ROOF. With the usual notations, (21.9) shows that
ht.m0 /  ht.m/  ht.m0 / C 1:
Therefore
dimk .m0 =m02 /  ht.m0 /  ht.m/

1 D dimk .m=m2 /

1 D dimk .m0 =m02 /:

Equalities must hold throughout, which proves that A0 is regular with dimension dim A

1.

T HEOREM 22.3. Every regular noetherian local ring is an integral domain.

P ROOF. Let A be a regular local ring of dimension d . We have already noted that the
statement is true when d D 0.
We next prove that A is an integral domain if it contains distinct ideals a  p with a D .a/
principal and p prime. Let b 2 p, and suppose that b 2 an D .an / for some n  1. Then
b D an c for some c 2 A. As a is not in the prime ideal p, we must have that c 2 p  a, and
T
3.15
so b 2 anC1 . Continuing in this fashion, we see that b 2 n an D f0g. Therefore p D f0g,
and so A is an integral domain.

23

FLATNESS AND FIBRES

102

We now assume d  1, and proceed by induction on d . Let a 2 m X m2 . As A=.a/ is

regular of dimension d 1, it is an integral domain, and so .a/ is a prime ideal. If it has
height 1, then the last paragraph shows that A is an integral domain. Thus, we may suppose
that, for all a 2 m X m2 , the prime ideal .a/ has height 0, and so is a minimal prime ideal
of A. Let S be the set of allS
minimal prime ideals of A recall
S (19) that S is finite. We
have shown that m X m2  fp j p 2 S g, and so m  m2 [ fp j p 2 S g. It follows from
Proposition 2.8 that either m  m2 (and hence m D 0) or m is a minimal prime ideal of A,
but both of these statements contradict the assumption that d  1:
C OROLLARY 22.4. A regular noetherian local ring of dimension 1 is a principal ideal
domain (with a single nonzero prime ideal).
P ROOF. Let A be a regular local ring of dimension 1 with maximal ideal m, and let a be
a nonzero proper ideal in A. The conditions imply that m is principal, say m D .t /. The
radical of a is m because m is the only prime ideal containing a, and so a  mr for some
r (by 3.16). The ring A=mr is local and artinian, and so a D .t s / C mr for some s  1 (by
16.8). This implies that a D .t s / by Nakayamas lemma (3.9).
T HEOREM 22.5. Let A be a regular noetherian local ring.
(a) For any prime ideal p in A, the ring Ap is regular.
(b) The ring A is a unique factorization domain (hence is integrally closed).
P ROOF. Omitted for the moment.
The best proof uses homological methods. See May, RegularLocal.pdf or Matsumura
1986 19.3, 20.3.
D EFINITION 22.6. Let .A; m/ be a noetherian local ring of dimension d . A system of
parameters of A is a set of elements fa1 ; : : : ; ad g such that .a1 ; : : : ; ad /  mn for some n.
If .a1 ; : : : ; ad / D m, then fa1 ; : : : ; ad g is called a regular system of parameters.
In other words, fa1 ; : : : ; ad g is a system of parameters if the ideal .a1 ; : : : ; ad / is mprimary. A system of parameters always exists, and a regular system of parameters exists if
and only if A is regular.

23

Flatness and fibres

Recall that, for a prime ideal p in a ring A, the field of fractions of A=p is denoted .p/. For
example, for a maximal ideal m, .m/ D A=m; more generally, .p/ D Ap =pAp .
Let 'W A ! B be a homomorphism of rings. We say that the going-down theorem holds
for ' if the statement (7.12) holds with qi \ A interpreted as qci :
q1

qm

qn
pi D qci :

p1

pm

pn

T HEOREM 23.1. Let 'W A ! B be a homomorphism of noetherian rings. Let q be a prime

ideal of B, and let p D qc .

23

FLATNESS AND FIBRES

103

(a) We have dim Bq  dim Ap C dim.Bq .p//:
(b) If the going-down theorem holds for ', then equality holds in (a).
P ROOF. The statement depends only on the homomorphism of local rings Ap ! Bq defined
by '. Thus, we can replace A and B with Ap and Bq , and q and p with the maximal ideals
n D qBq and m D pAp . Then the inequality becomes
dim.B/  dim.A/ C dim.B=mB/:
(a) Let fa1 ; : : : ; ar g be a system of parameters for A, so that
mn  .a1 ; : : : ; ar /:
for some n. Let b1 ; : : : ; bs be elements of B whose images in B=mB form a system or
parameters for B=mB, so that
0

nn  .b1 ; : : : ; bs / C mB
for some n0 . Now

nn n  .b1 ; : : : ; bs / C .a1 ; : : : ; ar /B
and so fa1 ; : : : ; ar ; b1 ; : : : bs g generates an n-primary ideal in B. Hence
dim.B/  r C s D dim.A/ C dim.B=mB/:
(b) Let m D dim.B=mB/, and let
n D q0 


 qm
be a chain of distinct prime ideals in B containing mB. Clearly qci D m for all i. Let
m0 D dim A, and let
m D p0 


 pm0
be a chain of distinct prime ideals in A. By the going-down theorem, there exists a chain of
ideals
qm 


 qmCm0
such that qcmCi D pi for all i . The existence of the chain
q0 


 qmCm0
of distinct prime ideals in B shows that dim.B/  m0 C m D dim A C dim.B=mB/.
T HEOREM 23.2. The going-down theorem holds for every flat homomorphism 'W A ! B.
P ROOF. Let p0  p be prime ideals in A, and let q be a prime ideal in B such that qc D p.
We have to show that there exists a prime ideal q0  q in B such that q0c D p0 . Because ' is
flat, Ap ! Bq is faithfully flat (11.13), and so there exists a prime ideal in Bq contracting to
p0 Ap in Ap (11.14). The contraction of this ideal to B has the required properties.
C OROLLARY 23.3. Let 'W A ! B be a homomorphism of rings, and let q be a prime ideal
of B. If ' is flat, then
ht.q/ D ht.p/ C dim.Bq .p//;

p D qc :

23

FLATNESS AND FIBRES

104

P ROOF. According to the theorem, ' satisfies the going-down theorem, and so we can apply
(23.1).
C OROLLARY 23.4. Let A be a noetherian ring, and let 'W A ! B be a homomorphism of
rings. If ' is flat of finite type, then the map ' a W Spec.B/ ! Spec.A/ is open.
P ROOF. According to the theorem, ' satisfies the going-down theorem, and so we can apply
(14.16).
Let 'W A ! B be a homomorphism of rings such that all maximal ideals in A have the
same height and similarly for B. If ' is flat and spm.'/ is surjective, then (23.3) says that
dim.B/ D dim.A/ C dim.B A .m//
for all maximal ideals of A. In other words, the dimension of the fibre
spm.B/ ! spm.A/
over m 2 spm.A/ is dim.spm.B// dim.spm.A//:
Corollary 23.3 has a converse.
T HEOREM 23.5. Let 'W A ! B be a local homomorphism of noetherian local rings, and let
m be the maximal ideal of A. If A is regular, B is Cohen-Macaulay, and
dim.B/ D dim.A/ C dim.B .m//;
then ' is flat.
P ROOF. Matsumura 1986, 23.1.
We dont define notion of being Cohen-Macaulay here (see ibid. p.134), but merely list
some of its properties.
23.6. A noetherian ring A is Cohen-Macaulay if and only if Am is Cohen-Macaulay for
every maximal ideal m of A (this is part of the definition).
23.7. Zero-dimensional and reduced one-dimensional noetherian rings are Cohen-Macaulay
(ibid. p.139).
23.8. Regular noetherian rings are Cohen-Macaulay (ibid. p.137).
23.9. Let 'W A ! B be a flat local homomorphism of noetherian local rings, and let m be
the maximal ideal of A. Then B is Cohen-Macaulay if and only if both A and B A .m/
are Cohen-Macaulay (ibid. p.181).
P ROPOSITION 23.10. Let 'W A ! B be a finite homomorphism noetherian rings with A
regular. Then ' is flat if and only if B is Cohen-Macaulay.
P ROOF. Note that dim.B .m// is zero-dimensional, hence Cohen-Macaulay, for every
maximal ideal m of A (23.7), and that ht.n/ D ht.nc / for every maximal ideal n of B. If ' is
flat, then B is Cohen-Macaulay by (23.9). Conversely, if B is Cohen-Macaulay, then ' is
flat by (23.5).
A SIDE 23.11. In contrast to the going-down theorem, the going-up theorem fails for flat homomorphisms it even fails for Z ! ZX (see 7.8).

24

COMPLETIONS

105

Exercises
E XERCISE 23.12. Show that the only flat surjective homomorphisms from a noetherian
ring are the projection maps A1  A2 ! A1 .

24

Completions

Let A be a ring and a an ideal in A. For any A-module, we get an inverse system of quotient
maps
M=aM
M=a2 M



M=an M




y of M :
whose limit we define to be the a-adic completion M
def
y D
M
lim M=an M:

For example, the a-adic completion of A is

def
Ay D lim A=an .

We now explain why this is called the completion. Let M be an A-module. A filtration
on M is a sequence of submodules
M D M0 


 Mn 


:
L EMMA 24.1. Let .Mn /n2N be a filtration on an A-module M . There is a unique topology
on M such that, for each x 2 M , the set fx C Mn j n 2 Ng is a fundamental system of
y of M relative to this topology is canonically
neighbourhoods for x. The completion M
isomorphic to lim M=Mn .
y consists of the
P ROOF. The first statement is obvious. For the second, recall that M
equivalence classes of Cauchy sequences in M . Let .mn /n2N be a Cauchy sequence. For
each n, the image of mi in M=Mn becomes constant for large i let m
x n denote the constant
value. The family .m
x n /n2N depends only on the equivalence class of the Cauchy sequence
.mn /n2N , and
y ! lim M=Mn
.mn / 7! .m
x n /W M
is an isomorphism.
Let A be a ring and let a be an ideal in A. A filtration .Mn /n2N on an A-module M is
an a-filtration if aMn  MnC1 for all n. An a-filtration is stable if aMn D MnC1 for all
sufficiently large n.
L EMMA 24.2. Any two stable a-filtrations on an A-module M define the same topology on
M.
P ROOF. It suffices to show that a stable a-filtration .Mn /n2N defines the a-adic topology
on M . As aMn  MnC1 for all n, we have that an M  Mn for all n. For some n0 ,
aMn D MnC1 for all n  n0 , and so MnCn0 D an Mn0  an M .

24

COMPLETIONS

106

L EMMA 24.3 (A RTIN -R EES ). If A is noetherian and M is finitely generated, then, for any
A-submodule M 0 of M , the filtration .M 0 \ an M /n2N on M 0 is a stable a-filtration.
P ROOF. Omitted for the moment.
P ROPOSITION 24.4. For every noetherian ring A and ideal a, the functor M
on finitely generated A-modules.

y is exact
M

P ROOF. Let
0 ! M 0 ! M ! M 00 ! 0
be an exact sequence of A-modules. For each n, the sequence
0 ! M 0 \ an M ! an M ! an M 00 ! 0
is exact, and so
0 ! M 0 =.M 0 \ an M / ! M=an M ! M 00 =an M 00 ! 0
is exact. On passing to the inverse limit, we obtain an exact sequence
y !M
y 00 ! 0,
0 ! lim M 0 =.M 0 \ an M / ! M
n

but the last three lemmas show that lim M 0 =.M 0 \ an M / is the a-adic completion of M 0 .
n

P ROPOSITION 24.5. For every ideal a in a noetherian ring A and finitely generated Amodule M , the homomorphism
y
a m 7! amW Ay A M ! M
is an isomorphism.
y
P ROOF. In other words, when A is noetherian, the functors M
Ay M and M
M
agree on finitely generated A-modules M . This is obvious for M D A, and it follows for
finitely generated free A-module because both functors take finite direct sums to direct sums.
Choose a surjective homomorphism Am ! M , and let N be its kernel. The exact sequence
0 ! N ! Am ! M ! 0
gives rise to a exact commutative diagram
Ay A N
a

Ny

Aym
'

Aym

Ay A M

y
M

Because the middle vertical arrow is an isomorphism, the arrow b is surjective. But M
is arbitrary, and so the arrow a is also surjective, which implies that the arrow b is an
isomorphism.

SOLUTIONS TO THE EXERCISES.

107

P ROPOSITION 24.6. For every noetherian ring A and ideal a, the a-adic completion Ay of A
is a flat A-algebra.
P ROOF. It follows from (24.4) and (24.5) that Ay A is exact on finitely generated Amodules, but this implies that it is exact on all A-modules.
A SIDE 24.7. Let m be a maximal ideal of a ring A, and let A ! Ay denote the m-adic completion
y m
y n is the m-adic completion of A=mn , but A=mn is discrete, and so
of A. Then A=mn ! A=
n
n
y
y m
y is an isomorphism. Similarly, Am =mnm ! Aym =m
y nm ' A=
y n is an isomorphism. On
A=m ! A=m
combining these statements, we obtain a conceptual proof of (5.8).

Sections to be added.
25.
26.
27.
28.
29.
30.
31.

Henselian rings.
Graded rings.
Hilbert polynomials.
Homological methods.
Regular local rings revisited.
Connections with geometry.
Computational commutative algebra.

Solutions to the exercises.

1.1. For n D 1, use that a nonzero polynomial in one variable has only finitely many
roots (which follows from unique factorization, for example). Now suppose
n > 1, and
P
assume the statement for polynomials in  n 1 symbols. Write f D gi Xni with each
gi 2 kX1 ; : : : ; Xn 1 . If f is not the zero polynomial, then some gi is not the zero polynomial, and there exist .a1 ; : : : ; an 1 / 2 k n 1 such that f .a1 ; : : : ; an 1 ; Xn / is not the zero
polynomial. Now, by the degree-one case, there exists a b such that f .a1 ; : : : ; an 1 ; b/ 0.
P
6.20. Let f D bi T m i , bi 2 B. If the coefficients bi of f are integral over A, then they
are integral over AT
P (as elements of BT ). Certainly T is integral over AT , and so this
implies that f D bi T i is integral over A (see 6.5).
11.17.Q The set spm.A
fi / consists of the maximal ideals
`
Q in A not containing fi , and
spm. i Afi / D i spm.Afi /. Therefore the map spm. i Afi / ! spm.A/ is surjective
if and only if .f1 ; : : : ; fm / D A. Now apply (11.13). For the second statement, it is only
a question of showing that the
Qsequence in (11.8) becomes the sequence in (11.17) when
i W A ! B is taken to be A ! i Afi .
15.15. (a) Let B be a countable local domain, and number its elements b1 ; b2 ; : : :. Consider
the homomorphism A ! B sending Xi to bi . It is surjective, and its kernel is a prime ideal
p of A. The ideal p is not an intersection of maximal ideals because the only maximal ideal
of A containing p is the inverse image of the maximal ideal in B.
(b) Let f be a nonzero element of A, say, f D f .X1 ; : : : ; Xn /. Choose a1 ; : : : ; an 2 Q
such that f .a1 ; : : : ; an / 0 (Exercise 1.1). The kernel of the homomorphism A ! Q
sending Xi to ai for i  n and Xi to 0 for i > n is a maximal ideal in A not containing f .
23.12. Consider surjective homomorphism A ! A=a. The set V .a/ is closed in spec.A/ (by
definition of the topology on spec.A/). If A ! A=a is flat, then V .a/ is also open. Therefore

REFERENCES

108

A D A1  A2 and a is of the form b  A2 with b an ideal in A1 such that V .b/ D spec.A1 /.

On tensoring
0 ! b  A2 ! A1  A2 ! A1 =b ! 0
with A1 =b we get an exact sequence
id

0 ! b=b2 ! A1 =b ! A1 =b ! 0:
Therefore b D b2 , but b is contained in all prime ideals of A1 , and so this implies that b D 0
(Nakayamas lemma, 3.9).

References
ATIYAH , M. F. AND M ACDONALD , I. G. 1969. Introduction to commutative algebra. AddisonWesley Publishing Co., Reading, Mass.-London-Don Mills, Ont.
ements de mathematique. Hermann; Masson, Paris.
B OURBAKI , N. AC. Alg`ebre Commutative. El
Chap. IIV Masson 1985; Chap. VVII Hermann 1975; Chap. VIII-IX Masson 1983; Chap. X
Masson 1998.
K RULL , W. 1938. Dimensionstheorie in stellenringen. J. Reine Angew. Math. 179:204226.
M ATSUMURA , H. 1986. Commutative ring theory, volume 8 of Cambridge Studies in Advanced
Mathematics. Cambridge University Press, Cambridge.
NAGATA , M. 1962. Local rings. Interscience Tracts in Pure and Applied Mathematics, No. 13.
Interscience Publishers, New York-London.
N ORTHCOTT, D. G. 1953. Ideal theory. Cambridge Tracts in Mathematics and Mathematical Physics,
No. 42. Cambridge, at the University Press.
R AYNAUD , M. 1970. Anneaux locaux henseliens. Lecture Notes in Mathematics, Vol. 169. SpringerVerlag, Berlin.

Index
A A,

9
algebra, 3
finite, 3
finitely generated, 3
finitely presented, 3
symmetric, 42
tensor, 42
annihilator, 10
axiom of dependent choice, 9
belong to, 87
belonging, 90
boundary, 85
coefficient
leading, 11
components
irreducible, 65
conductor, 80
constructible, 66
content of a polynomial, 16
contraction
of an ideal, 7
Cramers rule, 24
Dedekind domain, 92
degree
of a polynomial, 17
total, 17
dimension
Krull, 12
directed, 35
discrete valuation ring, 91
domain
unique factorization, 14

homomorphism
finite, 3
finite type, 3
of algebras, 3
ideal, 3
generated by a subset, 4
irreducible, 88
maximal, 4
minimal prime, 87
primary, 86
prime, 4
principal, 4
radical, 5
idempotent, 61
trivial, 61
identity element, 2
integral closure, 26
integral domain, 3
integrally closed, 27
normal, 27
irreducible, 90
.p/, 4
lemma
Gausss, 16
Nakayamas, 11
Zariskis, 57
limit
direct, 35
map

element
integral over a ring, 24
irreducible, 14
prime, 14
extension
of an ideal, 7

bilinear, 38
module
artinian, 73
faithful, 3
finitely presented, 50
noetherian, 9
monomial, 17
multiplicative subset, 5
saturated, 24

faithfully flat, 44
flat, 44

nilpotent, 5, 90
nilradical, 5

generalization, 67
generate
an algebra, 3

orthogonal idempotents, 61
complete set of, 61

height, 99
of a prime ideal, 12

p-primary, 90
polynomial
monic, 24

109

INDEX
primitive, 16
primary, 86, 90
primary decomposition, 90
minimal, 87, 90
primary, 87
prime ideal
associated, 90
radical
Jacobson, 6
of an ideal, 5
relations
between generators, 50
relatively prime, 8
ring
artinian, 73
Jacobson, 68
local, 6
noetherian, 9
normal, 29
reduced, 5
regular local, 101
set
directed, 35
spec, 20
specialization, 67
spectrum, 69
spm.A/, 47
subring, 3
symbolic power, 96
system
direct, 35
system of parameters, 102
regular, 102
tensor product
of algebras, 40
of modules, 38
theorem
Chinese remainder, 8
generic flatness, 47
generic freeness, 100
going-up, 31
Hilbert basis, 11
invariant factor, 95
Krull intersection, 13
Krulls principal ideal, 96
modules over Dedekind domain, 95
Noether normalization, 33
Nullstellensatz, 59
strong Nullstellensatz, 59
unique factorization of ideals, 93
topological space
irreducible , 64

110
noetherian, 64
quasi-compact, 64
topology
Zariski, 61
unit, 2
zero divisor, 90