Matsumura: Commutative Algebra

Daniel Murfet
October 5, 2006
These notes closely follow Matsumuras book [Mat80] on commutative algebra. Proofs are
the ones given there, sometimes with slightly more detail. Our focus is on the results needed in
algebraic geometry, so some topics in the book do not occur here or are not treated in their full
depth. In particular material the reader can nd in the more elementary [AM69] is often omitted.
References on dimension theory are usually to Robert Ashs webnotes since the author prefers this
approach to that of [AM69].
Contents
1 General Rings 1
2 Flatness 6
2.1 Faithful Flatness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Going-up and Going-down . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3 Constructible Sets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3 Associated Primes 15
4 Dimension 15
4.1 Homomorphism and Dimension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
5 Depth 18
5.1 Cohen-Macaulay Rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
6 Normal and Regular Rings 32
6.1 Classical Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
6.2 Homological Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
6.3 Koszul Complexes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
6.4 Unique Factorisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
1 General Rings
Throughout these notes all rings are commutative, and unless otherwise specied all modules are
left modules. A local ring A is a commutative ring with a single maximal ideal (we do not require
A to be noetherian).
Lemma 1 (Nakayama). Let A be a ring, M a nitely generated A-module and I an ideal of A.
Suppose that IM = M. Then there exists an element a A of the form a = 1 +x, x I such that
aM = 0. If moreover I is contained in the Jacobson radical, then M = 0.
Corollary 2. Let A be a ring, M an A-module, N and N

submodules of M and I an ideal of

A. Suppose that M = N + IN

, and that either (a) I is nilpotent or (b) I is contained in the

Jacobson radical and N

is nitely generated. Then M = N.

1
Proof. In case (a) we have M/N = I(M/N) = I
2
(M/N) = = 0. In (b) apply Nakayamas
Lemma to M/N.
In particular let (A, m, k) be a local ring and M an A-module. Suppose that either m is
nilpotent or M is nitely generated. Then a subset G of M generates M i. its image in
M/mM = M
A
k generates M
A
k as a k-vector space. In fact if N is submodule generated by
G, and if the image of G generates M
A
k, then M = N +mM whence M = N by the Corollary.
Since M
A
k is a nitely generated vector space over the eld k, it has a nite basis, and if we
take an arbitrary preimage of each element this collection generates M. A set of elements which
becomes a basis in M/mM (and therefore generates M) is called a minimal basis. If M is a nitely
generated free A-module, then it is clear that
rank
A
M = rank
k
(M/mM)
In fact, of rank
A
M = n 1 and {x
1
, . . . , x
n
} is a basis of M, then {x
1
+ mM, . . . , x
n
+ mM} is
a basis of the k-module M/mM. Or equivalently, the x
i
1 are a basis of the k-module M k.
Let A be a ring and : Z A. The kernel is (n) for some integer n 0 which we call the
characteristic of A. The characteristic of a eld is either 0 or a prime number, and if A is local
the characteristic ch(A) is either 0 or a power of a prime number (m is a primary ideal and the
contraction of primary ideals are primary, and 0 and (p
n
) are the only primary ideals in Z).
Lemma 3. Let A be an integral domain with quotient eld K, all localisations of A can be viewed
as subrings of K and in this sense A =

m
A
m
where the intersection is over all maximal ideals.
Proof. Given x K we put D = {a A| ax A}, we call D the ideal of denominators of x. The
element x is in A i. D = A and x A
p
i. D p. Therefore if x / A, there exists a maximal
ideal m such that D p and x / A
m
for this m.
Lemma 4. Let A be a ring and S T multiplicatively closed subsets. Then
(a) There is a canonical isomorphism of S
1
A-algebras T
1
A

= T
1
(S
1
A) dened by a/t
(a/1)/(t/1).
(b) If M is an A-module then there is a canonical isomorphism of S
1
A-modules T
1
M

=
T
1
(S
1
M) dened by m/t (m/1)/(t/1).
Proof. (a) Just using the universal property of localisation we can see T
1
A

= T
1
(S
1
A) as
S
1
A-algebras via the map a/t (a/1)/(t/1). (b) is also easily checked.
Lemma 5. Let A be an integral domain with quotient eld K and B a subring of K containing
A. If Q is the quotient eld of B then there is a canonical isomorphism of B-algebras K

= Q.
If : A B is a ring isomorphism and S A is multiplicatively closed (denote also by S the
image in B) then there is an isomorphism of rings S
1
A

= S
1
B making the following diagram
commute
S
1
A

S
1
B
A

Lemma 6. Let A be a ring, S A a multiplicatively closed subset and p a prime ideal with
p S = . Let B = S
1
A. Then there is a canonical ring isomorphism B
pB

= A
p
.
Proof. A B B
pB
sends elements of A not in p to units, so we have an induced ring
morphism A
p
B
pB
dened by a/s (a/1)/(s/1) and it is easy to check this is an isomorphism.
2
Let : A B be a morphism of rings and I an ideal of A. The extended ideal IB consists
of sums

(a
i
)b
i
with a
i
I, b
i
B. Consider the exact sequence of A-modules
0 I A A/I 0
Tensoring with B gives an exact sequence of B-modules
I
A
B A
A
B (A/I)
A
B 0
The image of I
A
B in B

= A
A
B is simply IB. So there is an isomorphism of B-modules
B/IB

= (A/I)
A
B dened by b +IB 1 b. In fact, this is an isomorphism of rings as well.
Of course, for any two A-algebras E, F twisting gives a ring isomorphism E
A
F

= F
A
E.
Lemma 7. Let : A B be a morphism of rings, S a multiplicatively closed subset of A and
set T = (S). Then for any B-module M there is a canonical isomorphism of S
1
A-modules
natural in M
: S
1
M T
1
M
(m/s) = m/(s)
In particular there is a canonical isomorphism of S
1
A-algebras S
1
B

= T
1
B.
Proof. One checks easily that is a well-dened isomorphism of S
1
A-modules. In the case
M = B the S
1
A-module S
1
B becomes a ring in the obvious way, and preserves this ring
structure.
In particular, let S be a multiplicatively closed subset of a ring A, let I be an ideal of A and
let T denote the image of S in A/I. Then there is a canonical isomorphism of rings
T
1
(A/I)

= A/I
A
S
1
A

= S
1
A/I(S
1
A)
(a +I)/(s +I) a/s +I(S
1
A)
Denition 1. A ring A is catenary if for each pair of prime ideals q p the height of the prime
ideal p/q in A/q is nite and is equal to the length of any maximal chain of prime ideals between p
and q. Clearly the catenary property is stable under isomorphism, and any quotient of a catenary
ring is catenary. If S A is a multiplicatively closed subset and A is catenary, then so is S
1
A.
Lemma 8. Let A be a ring. Then the following are equivalent:
(i) A is catenary;
(ii) A
p
is catenary for every prime ideal p;
(iii) A
m
is catenary for every maximal ideal m.
Proof. The implications (i) (ii) (iii) are obvious. (iii) (i) If q p are primes, nd a
maximal ideal m containing p and pass to the catenary ring A
m
to see that the required property
is satised for q, p.
Lemma 9. Let A be a noetherian ring. Then A is catenary if for every pair of prime ideals q p
we have ht.(p/q) = ht.p ht.q.
Proof. Since A is noetherian, all involved heights are nite. Suppose A satises the condition and
let q p be prime ideals. Obviously ht.(p/q) is nite, and there is at least one maximal chain
between p and q with length ht.(p/q). Let
q = q
0
q
1
q
n
= p
be a maximal chain of length n. Then by assumption 1 = ht.(q
i
/q
i1
) = ht.q
i
ht.q
i1
for
1 i n. Hence ht.p = ht.q +n, so n = ht.(p/q), as required.
3
Denition 2. A ring A is universally catenary if A is noetherian and every nitely generated
A-algebra is catenary. Equivalently, a noetherian ring A is universally catenary if A is catenary
and A[x
1
, . . . , x
n
] is catenary for n 1.
Lemma 10. Let A be a ring and S A a multiplicatively closed subset. Then there is a canonical
ring isomorphism S
1
(A[x])

= (S
1
A)[x]. In particular if q is a prime ideal of A[x] and p = qA
then A[x]
q

= A
p
[x]
qA
p
[x]
.
Proof. The ring morphism A S
1
A induced A[x] (S
1
A)[x] which sends elements of
S A[x] to units. So there is an induced ring morphism : S
1
(A[x]) (S
1
A)[x] dened by

_
a
0
+a
1
x + +a
n
x
n
s
_
=
a
0
s
+
a
1
s
x + +
a
n
s
x
n
This is easily checked to be an isomorphism. In the second claim, there is an isomorphism
A
p
[x]

= A[x]
p
, where the second ring denotes (Ap)
1
(A[x]), and qA
p
[x] denotes the prime ideal
of A
p
[x] corresponding to qA[x]
p
. Using the isomorphism it is clear that qA
p
[x] A
p
= pA
p
.
Using Lemma 6, there is clearly an isomorphism of rings A[x]
q

= A
p
[x]
qA
p
[x]
.
If R is a ring and M an R-module, then let Z(M) denote the set of zero-divisors in M. That
is, all elements r R with rm = 0 for some nonzero m M.
Lemma 11. Let R be a nonzero reduced noetherian ring. Then Z(R) =

i
p
i
, with the union
being taken over all minimal prime ideals p
i
.
Proof. Since R is reduced,

i
p
i
= 0. If ab = 0 with b = 0, then b / p
j
for some j, and therefore
a p
j

i
p
i
. The reverse inclusion follows from the fact that no minimal prime can contain a
regular element (since otherwise by Krulls PID Theorem it would have height 1).
Lemma 12. Let R be a nonzero reduced noetherian ring. Assume that every element of R is
either a unit or a zero-divisor. Then dim(R) = 0.
Proof. Let p
1
, . . . , p
n
be the minimal primes of R. Then by Lemma 11, Z(R) = p
1
p
n
.
Let p be a prime ideal. Since p is proper, p Z(R) and therefore p p
i
for some i. Since p
i
is
minimal, p = p
i
, so the p
i
are the only primes in R. Since these all have height zero, it is clear
that dim(R) = 0.
Lemma 13. Let R be a reduced ring, p a minimal prime ideal of R. Then R
p
is a eld.
Proof. If p = 0 this is trivial, so assume p = 0. Since pR
p
is the only prime ideal in R
p
, it is also
the nilradical. So if x p then tx
n
= 0 for some t / p and n > 0. But this implies that tx is
nilpotent, and therefore zero since R is reduced. Therefore pR
p
= 0 and R
p
is a eld.
Let A
1
, . . . , A
n
be rings. Let A be the product ring A =

n
i=1
A
i
. Ideals of A are in bijection
with sequences I
1
, . . . , I
n
with I
i
an ideal of A
i
. This sequence corresponds to
I
1
I
n
This bijection identies the prime ideals of A with sequences I
1
, . . . , I
n
in which every I
i
= A
i
except for a single I
j
which is a prime ideal of A
j
. So the primes look like
A
1
p
i
A
n
for some i and some prime ideal p
i
of A
i
. Given i and a prime ideal p
i
of A
i
, let p be the prime
ideal A
1
p
i
A
n
. Then the projection of rings A A
i
gives rise to a ring morphism
A
p
(A
i
)
p
i
(a
1
, . . . , a
i
, . . . , a
n
)/(b
1
, . . . , b
i
, . . . , b
n
) a
i
/b
i
4
It is easy to check that this is an isomorphism. An orthogonal set of idempotents in a ring A is a
set e
1
, . . . , e
r
with 1 = e
1
+ + e
r
, e
2
i
= e
i
and e
i
e
j
= 0 for i = j. If A =

n
i=1
A
i
is a product
of rings, then the elements e
1
= (1, 0, . . . , 0), . . . , e
n
= (0, . . . , 0, 1) are clearly such a set.
Conversely if e
1
, . . . , e
r
is an orthogonal set of idempotents in a ring A, then the ideal e
i
A
becomes a ring with identity e
i
. The map
A e
1
A e
r
A
a (e
1
a, . . . , e
r
a)
is a ring isomorphism.
Proposition 14. Any nonzero artinian ring A is a nite direct product of local artinian rings.
Proof. See [Eis95] Corollary 2.16. This shows that there is a nite list of maximal ideals m
1
, . . . , m
n
(allowing repeats) and a ring isomorphism A

n
i=1
A
m
i
dened by a (a/1, . . . , a/1).
Proposition 15. Let : A B be a surjective morphism of rings, M an A-module and
p SpecB. There is a canonical morphism of B
p
-modules natural in M
: Hom
A
(B, M)
p
Hom
A

1
p
(B
p
, M

1
p
)
(u/s)(b/t) = u(b)/
1
(st)
If A is noetherian, this is an isomorphism.
Proof. Let a morphism of A-modules u : B M, s, t B\ p and b B be given. Choose k A
with (k) = st. We claim the fraction u(b)/k M

1
p
doesnt depend on the choice of k. If we
have (l) = st also, then
ku(b) = u(kb) = u((k)b) = u((l)b) = u(lb) = lu(b)
so u(b)/l = u(b)/k, as claimed. Throughout the proof, given x B we write
1
(x) for an
arbitrary element in the inverse image of x. One checks the result does not depend on this choice.
We can now dene a morphism of B
p
-modules (u/s)(b/t) = u(b)/
1
(st) which one checks is
well-dened and natural in M.
Now assume that A is noetherian. In showing that is an isomorphism, we may as well assume
is the canonical projection A A/a for some ideal a. In that case the prime ideal p is q/a for
some prime q of A containing a, and if we set S = A \ q and T = (S) we have by Lemma 7 an
isomorphism
Hom
A
(B, M)
p
= T
1
Hom
A
(A/a, M)

= S
1
Hom
A
(A/a, M)

= Hom
S
1
A
(S
1
(A/a), S
1
M)

= Hom
S
1
A
(T
1
(A/a), S
1
M)
= Hom
A

1
p
(B
p
, M

1
p
)
We have use the fact that A is noetherian to see that A/a is nitely presented, so we have the
second isomorphism in the above sequence. One checks easily that this isomorphism agrees with
, completing the proof.
Remark 1. The right adjoint Hom
A
(B, ) to the restriction of scalars functor exists for any
morphism of rings : A B, but as we have just seen, this functor is not local unless the
ring morphism is surjective. This explains why the right adjoint f
!
to the direct image functor in
algebraic geometry essentially only exists for closed immersions.
5
2 Flatness
Denition 3. Let A be a ring and M an A-module. We say M is at if the functor
A
M :
AMod AMod is exact (equivalently M
A
is exact). Equivalently M is at if whenever
we have an injective morphism of modules N N

the morphism N
A
M N

A
M is
injective. This property is stable under isomorphism.
We say M is faithfully at if a morphism N N

is injective if and only if N

A
M
N

A
M is injective. This property is also stable under isomorphism. An A-algebra A B is
at if B is a at A-module and we say A B is a at morphism.
Example 1. Nonzero free modules are faithfully at.
Lemma 16. We have the following fundamental properties of atness:
Transitivity: If : A B is a at morphism of rings and N a at B-module, then N is
also at over A.
Change of Base: If : A B is a morphism of rings and M is a at A-module, then
M
A
B is a at B-module.
Localisation: If A is a ring and S a multiplicatively closed subset, then S
1
A is at over A.
Proof. The second and third claims are done in our Atiyah & Macdonald notes. To prove the rst
claim, let M M

be a monomorphism of A-modules and consider the following commutative

diagram of abelian groups
M
A
N

A
N

M
A
(B
B
N)

A
(B
B
N)

(M
A
B)
B
N

(M

A
B)
B
N
Since B is a at A-module and N is a at B-module the bottom row is injective, hence so is the
top row.
Lemma 17. Let : A B be a morphism of rings and N a B-module which is at over A. If
S is a multiplicatively closed subset of B, then S
1
N is at over A. In particular any localisation
of a at A-module is at.
Proof. If M M

is a monomorphism of A-modules then we have a commutative diagram

M
A
S
1
N

A
S
1
N

(M
A
N)
B
S
1
B

(M

A
N)
B
S
1
B
The bottom row is clearly injective, and hence so is the top row, which shows that S
1
N is at
over A.
Lemma 18. Let A be a ring and M, N at A-modules. Then M
A
N is also at over A.
Lemma 19. Let : A B be a at morphism of rings and S a multiplicatively closed subset
of A. Then T = (S) is a multiplicatively closed subset of B and T
1
B is at over S
1
A.
Proof. This follows from Lemma 7 and stability of atness under base change.
6
Lemma 20. Let A B be a morphism of rings. Then the functor
A
B : AMod BMod
preserves projectives.
Proof. The functor
A
B is left adjoint to the restriction of scalars functor. This latter functor
is clearly exact, so since any functor with an exact right adjoint must preserve projectives, P
A
B
is a projective B-module for any projective A-module P.
Lemma 21. Let A B be a at morphism of rings. If I is an injective B-module then it is
also an injective A-module.
Proof. The restriction of scalars functor has an exact left adjoint
A
B : AMod BMod,
and therefore preserves injectives.
Lemma 22. Let : A B be a at morphism of rings, and let M, N be A-modules. Then there
is an isomorphism of B-modules Tor
A
i
(M, N)
A
B

= Tor
B
i
(M
A
B, N
A
B). If A is noetherian
and M nitely generated over A, there is an isomorphism of B-modules Ext
i
A
(M, N)
A
B

=
Ext
i
B
(M
A
B, N
A
B).
Proof. Let X : X
1
X
0
M 0 be a projective resolution of the A-module M.
Since B is at, the sequence
X
A
B : X
1

A
B X
0

A
B M
A
B 0
is a projective resolution of M
A
B. The chain complex of B-modules (X
A
B)
B
(B
A
N)
is isomorphic to (X
A
N)
A
B. The exact functor
A
B commutes with taking homology so
there is an isomorphism of B-modules Tor
A
i
(M, N)
A
B

= Tor
B
i
(M
A
B, N
A
B), as required.
If A is noetherian and M nitely generated we can assume that the X
i
are nite free A-modules.
Then Ext
i
A
(M, N) is the i-cohomology module of the sequence
0 Hom(X
0
, N) Hom(X
1
, N) Hom(X
2
, N)
Since tensoring with B is exact, Ext
i
A
(M, N)
A
B is isomorphic as a B-module to the i-th
cohomology of the following sequence
0 Hom(X
0
, N)
A
B Hom(X
1
, N)
A
B
After a bit of work, we see that this cochain complex is isomorphic to Hom
B
(X
A
B, N
A
B),
and the i-th cohomology of this complex is Ext
i
B
(M
A
B, N
A
B), as required.
In particular for a ring A and prime ideal p A we have isomorphisms of A
p
-modules for i 0
Tor
A
p
i
(M
p
, N
p
)

= Tor
A
i
(M, N)
p
Ext
i
A
p
(M
p
, N
p
)

= Ext
i
A
(M, N)
p
the latter being valid for A noetherian and M nitely generated.
Lemma 23. Let A be a ring and M an A-module. Then the following are equivalent
(i) M is a at A-module;
(ii) M
p
is a at A
p
-module for each prime ideal p;
(iii) M
m
is a at A
m
-module for each maximal ideal m.
Proof. See [AM69] or any book on commutative algebra.
Proposition 24. Let (A, m, k) be a local ring and M an A-module. Suppose that either m is
nilpotent or M is nitely generated over A. Then M is free M is projective M is at.
7
Proof. It suces to show that if M is at then it is free. We prove that any minimal basis of M
is a basis of M. If M/mM = 0 then M = 0 and M is trivially free. Otherwise it suces to show
that if x
1
, . . . , x
n
M are elements whose images in M/mM = M
A
k are linearly independent
over k, then they are linearly independent over A. We use induction on n. For n = 1 let ax = 0.
Then there exist y
1
, . . . , y
r
M and b
1
, . . . , b
r
A such that ab
i
= 0 for all i and x =

b
i
y
i
.
Since x +mM = 0 not all b
i
are in m. Suppose b
1
/ m. Then b
1
is a unit in A and ab
1
= 0, hence
a = 0.
Suppose n > 1 and

n
i=1
a
i
x
i
= 0. Then there exist y
1
, . . . , y
r
M and b
ij
A(1 j r)
such that x
i
=

j
b
ij
y
j
and

i
a
i
b
ij
= 0. Since x
n
/ mM we have b
nj
/ m for at least one j.
Since a
1
b
1j
+ +a
n
b
nj
= 0 and b
nj
is a unit, we have
a
n
=
n1

i=1
c
i
a
i
c
i
= b
ij
/b
nj
Then
0 =
n

i=1
a
i
x
i
= a
1
(x
1
+c
1
x
n
) + +a
n1
(x
n1
+c
n1
x
n
)
Since the residues of x
1
+c
1
x
n
, . . . , x
n1
+c
n1
x
n
are linearly independent over k, by the inductive
hypothesis we get a
1
= = a
n1
= 0 and a
n
=

c
i
a
i
= 0.
Corollary 25. Let A be a ring and M a nitely generated A-module. Then the following are
equivalent
(i) M is a at A-module;
(ii) M
p
is a free A
p
-module for each prime ideal p;
(iii) M
m
is a free A
m
-module for each maximal ideal m.
Proof. This is immediate from the previous two results.
Proposition 26. Let A be a ring and M a nitely presented A-module. Then M is at if and
only if it is projective.
Proof. See Stenstrom Chapter 1, Corollary 11.5.
Corollary 27. Let A be a noetherian ring, M a nitely generated A-module. Then the following
conditions are equivalent
(i) M is projective;
(ii) M is at;
(ii) M
p
is a free A
p
-module for each prime ideal p;
(iii) M
m
is a free A
m
-module for each maximal ideal m.
Proof. Since A is noetherian, M is nitely presented, so (i) (ii) is an immediate consequence
of Proposition 26. The rest of the proof follows from Corollary 25.
Lemma 28. Let A B be a at morphism of rings, and let I, J be ideals of A. Then (I J)B =
IB JB and (I : J)B = (IB : JB) if J is nitely generated.
Proof. Consider the exact sequence of A-modules
I J A A/I A/J
Tensoring with B we get an exact sequence
(I J)
A
B = (I J)B B B/IB B/JB
8
This means (I J)B = IB JB. For the second claim, suppose rstly that J is a principal ideal
aA and use the exact sequence
(I : aA)
i

A
f

A/I
where i is the injection and f(x) = ax+I. Tensoring with B we get the formula (I : a)B = (IB : a).
In the general case, if J = (a
1
, . . . , a
n
) we have (I : J) =

i
(I : a
i
) so that
(I : J)B =

(I : a
i
)B =

(IB : a
i
) = (IB : JB)
Example 2. Let A = k[x, y] be a polynomial ring over a eld k and put B = A/(x)

= k[y]. Then
B is not at over A since y A is regular but is not regular on B. Let I = (x +y) and J = (y).
Then I J = (xy +y
2
) and IB = JB = yB, (I J)B = y
2
B = IB JB.
Example 3. Let k be a eld, put A = k[x, y] and let K be the quotient eld of A. Let B be the
subring k[x, y/x] of K (i.e. the k-subalgebra generated by x and z = y/x). Then A B K. Let
I = xA, J = yA. Then I J = xyA and (I J)B = x
2
zB, IB JB = xzB so B is not at over
A. The map SpecB SpecA corresponding to A B is the projection to the (x, y)-plane
of the surface F : xz = y in (x, y, z)-space. Note F contains the whole z-axis so it does not look
at over the (x, y)-plane.
Proposition 29. Let : A B be a morphism of rings, M an A-module and N a B-module.
Then for every p SpecB there is a canonical isomorphism of B
p
-modules natural in both variables
: M
pA

A
pA
N
p
(M
A
N)
p
(m/s n/t) = (mn)/(s)t
Proof. Fix p SpecB and q = pA. There is a canonical ring morphism A
q
B
p
and we make
N
p
into an A
q
-module using this morphism. One checks that the following map is well-dened
and A
q
-bilinear
: M
q
N
p
(M
A
N)
p
(m/s, n/t) = (mn)/(s)t
We show that in fact this is a tensor product of A
q
-modules. Let Z be an abelian group and
: M
q
N
p
Z an A
q
-bilinear map.
M
q
N
p

Z
(M
A
N)
p

(1)
We have to dene a morphism of abelian groups unique making this diagram commute. For
s / p we dene an A-bilinear morphism

s
: M N Z by

s
(m, n) = (m/1, n/s). This
induces a morphism of abelian groups

s
: M
A
N Z

s
(mb) = (m/1, b/s)
We make some observations about these morphisms
Suppose w/s = w

/s

in (M
A
N)
p
, with say w =

i
m
i
n
i
, w

i
m

i
n

i
and t / p
such that ts

w = tsw

. That is,

i
m
i
ts

n
i
=

i
m

i
tsn

i
. Applying

tss
to both sides
of this equality gives

s
(w) =

s
(w

).
9
For w/s, w

/s

(M
A
N)
p
we have

s
(w) +

s
(w

) =

ss
(s

w +sw

).
It follows that (w/s) =

s
(w) gives a well-dened morphism of abelian groups : (M
A
N)
p

Z which is clearly unique making (1) commute. By uniqueness of the tensor product there is an
induced isomorphism of abelian groups : M
q

A
q
N
p
(M
A
N)
p
with (m/s n/t) =
(m n)/(s)t. One checks that this is a morphism of B
p
-modules. The inverse is dened by
(mn)/t m/1 n/t. Naturality in both variables is easily checked.
Corollary 30. Let : A B be a morphism of rings, M an A-module and p SpecB. Then
there is a canonical isomorphism of B
p
-modules M
pA

A
pA
B
p
(M
A
B)
p
natural in M.
We will not actually use the next result in these notes, so the reader not familiar with homo-
logical -functors can safely skip it. Alternatively one can provide a proof by following the one
given in Matsumura (the proof we give is more elegant, provided you know about -functors).
Proposition 31. Let : A B be a morphism of rings, M an A-module and N a B-module.
Then for every p SpecB and i 0 there is a canonical isomorphism of B
p
-modules natural in
M

i
: Tor
A
i
(N, M)
p
Tor
A
pA
i
(N
p
, M
pA
)
Proof. Fix p SpecB and a B-module N and set q = p A. Then N is a B-A-bimodule and N
p
is a B
p
-A
q
-bimodule so by (TOR,Section 5.1) the abelian group Tor
A
i
(N, M) acquires a canonical
B-module structure, and Tor
A
q
i
(N
p
, M
q
) acquires a canonical B
p
-module structure for any A-
module M and i 0. Using (TOR,Lemma 14) and (DF,Denition 23) we have two homological
-functors between AMod and B
p
Mod
{Tor
A
i
(N, )
p
}
i0
, {Tor
A
q
i
(N
p
, ()
q
)}
i0
For i > 0 these functors all vanish on free A-modules, so by (DF,Theorem 74) both -functors are
universal. For i = 0 we have the canonical natural equivalence of Proposition 29

0
: Tor
A
0
(N, )
p

= (N
A
)
p

= N
p

A
q
()
q

= Tor
A
q
0
(N
p
, ()
q
)
By universality this lifts to a canonical isomorphism of homological -functors . In particular
for each i 0 we have a canonical natural equivalence
i
: Tor
A
i
(N, )
p
Tor
A
q
i
(N
p
, ()
q
), as
required.
We know from Lemma 23 that atness is a local property. We are now ready to show that
relative atness (i.e. atness with respect to a morphism of rings) is also local. This is particularly
important in algebraic geometry. The reader who skipped Proposition 31 will also have to skip
the implication (iii) (i) in the next result, but this will not aect their ability to read the rest
of these notes.
Corollary 32. Let A B be a morphism of rings and N a B-module. Then the following
conditions are equivalent
(i) N is at over A.
(ii) N
p
is at over A
pA
for all prime ideals p of B.
(iii) N
m
is at over A
mA
for all maximal ideals m of B.
Proof. (i) (ii) If N is at over A then N
pA
is at over A
pA
for any prime p of B. By an
argument similar to the one given in Lemma 19 we see that N
p
is isomorphic as a B
pA
-module
to a localisation of N
pA
. Applying Lemma 17 to the ring morphism A
pA
B
pA
we see that
N
p
is at over A
pA
, as required. (ii) (iii) is trivial. (iii) (i) For every A-module M and
maximal ideal m of B we have by Proposition 31
Tor
A
1
(N, M)
m

= Tor
A
mA
1
(N
m
, M
mA
) = 0
since by assumption N
m
is at over A
mA
. Therefore Tor
A
1
(N, M) = 0 for every A-module M,
which implies that N is at over A and completes the proof.
10
Lemma 33. Let A B be a morphism of rings. Then the following conditions are equivalent
(i) B is at over A;
(ii) B
p
is at over A
pA
for all prime ideals p of B;
(iii) B
m
is at over A
mA
for all maximal ideals m of B.
2.1 Faithful Flatness
Theorem 34. Let A be a ring and M an A-module. The following conditions are equivalent:
(i) M is faithfully at over A;
(ii) M is at over A, and for any nonzero A-module N we have N
A
M = 0;
(iii) M is at over A, and for any maximal ideal m of A we have mM = M.
Proof. (i) (ii) Let N be an A-module and : N 0 the zero map. Then if M is faithfully
at and N
A
M = 0 we have
A
M = 0 which means that is injective and therefore N = 0.
(ii) (iii) Since A/m = 0 we have (A/m)
A
M = M/mM = 0 by hypothesis. (iii) (ii)
Let N be a nonzero A-module and pick 0 = x N. Let : A N be 1 x. If I = Ker
then there is an injective morphism of modules A/I N. Let m be a maximal ideal containing
I. Then M mM IM so (A/I)
A
M = M/IM = 0. Since M is at the morphism
(A/I)
A
M N
A
M is injective so N
A
M = 0. (ii) (i) Let : N N

be a morphism
of modules with kernel K N. If N
A
M N

A
M is injective then K
A
M = 0, which
is only possible if K = 0.
Corollary 35. Let A and B be local rings, and : A B a local morphism of rings. Let M be
a nonzero nitely generated B-module. Then
M is at over A M is faithfully at over A
In particular, B is at over A if and only if it is faithfully at over A.
Proof. Let m and n be the maximal ideals of A and B, respectively. Then mM nM since is
local, and nM = M by Nakayama, so the assertion follows from the Theorem.
Lemma 36. We have the following fundamental properties of atness:
Transitivity: If : A B is a faithfully at morphism of rings and N a faithfully at
B-module, then N is a faithfully at A-module.
Change of Base: If : A B is a morphism of rings and M is a faithfully at A-module,
then M
A
B is a faithfully at B-module.
Descent: If : A B is a ring morphism and M is a faithfully at B-module which is
also faithfully at over A, then B is faithfully at over A.
Proof. The diagram in the proof of transitivity for atness makes it clear that faithful atness is
also transitive. Similarly the atness under base change proof in our Atiyah & Macdonald notes
shows that faithful atness is also stable under base change. The descent property is also easily
checked.
Proposition 37. Let : A B be a faithfully at morphism of rings. Then
(i) For any A-module N, the map N N
A
B dened by x x1 is injective. In particular
is injective and A can be viewed as a subring of B.
(ii) For any ideal I of A we have IB A = I.
(iii) The map : Spec(B) Spec(A) is surjective.
11
(iv) If B is noetherian then so is A.
Proof. (i) Let 0 = x N. Then 0 = Ax N and since B is at we see that Ax
A
B is isomorphic
to the submodule (x 1)B of N
A
B. It follows from Theorem 34 that x 1 = 0.
(ii) By change of base, B
A
(A/I) = B/IB is faithfully at over A/I. Now the assertion
follows from (i). For (iii) let p Spec(A). The ring B
p
= B
A
A
p
is faithfully at over A
p
so
by (ii) pB
p
= B
p
. Take a maximal ideal m of B
p
containing pB
p
. Then m A
p
pA
p
, therefore
m A
p
= pA
p
since pA
p
is maximal. Putting q = m B, we get
q A = (m B) A = m A = (m A
p
) A = pA
p
A = p
as required. (iv) Follows immediately from (ii).
Theorem 38. Let : A B be a morphism of rings. The following conditions are equivalent.
(1) is faithfully at;
(2) is at, and : Spec(B) Spec(A) is surjective;
(3) is at, and for any maximal ideal m of A there is a maximal ideal n of B lying over m.
Proof. (1) (2) was proved above. (2) (3) By assumption there exists q Spec(B) with
qA = m. If n is any maximal ideal of B containing q then nA = m as m is maximal. (3) (1)
The existence of n implies mB = B, so B is faithfully at over A by Theorem 34.
Denition 4. In algebraic geometry we say a morphism of schemes f : X Y is at if the
local morphisms O
Y,f(x)
O
X,x
are at for all x X. We say the morphism is faithfully at if
it is at and surjective.
Lemma 39. Let A be a ring and B a faithfully at A-algebra. Let M be an A-module. Then
(i) M is at (resp. faithfully at) over A M
A
B is so over B,
(ii) If A is local and M nitely generated over A we have M is A-free M
A
B is B-free.
Proof. (i) Let N N

be a morphism of A-modules. Both claims follow from commutativity of

the following diagram
(N
A
M)
A
B

(N

A
M)
A
B

N
A
(M
A
B)

A
(M
A
B)

N
A
(B
B
(M
A
B))

A
(B
B
(M
A
B))

(N
A
B)
B
(M
A
B)

(N

A
B)
B
(M
A
B)
(ii) The functor
A
B preserves coproducts, so the implication () is trivial. () follows from
(i) because, under the hypothesis, freeness of M is equivalent to atness as we saw in Proposition
24.
12
2.2 Going-up and Going-down
Denition 5. Let : A B be a morphism of rings. We say that the going-up theorem holds
for if the following condition is satised:
(GU) For any p, p

Spec(A) such that p p

and for any prime q Spec(B) lying

over p, there exists q

Spec(B) lying over p

such that q q

.
Similarly we say that the going-down theorem holds for if the following condition is satised:
(GD) For any p, p

Spec(A) such that p p

and for any prime q

Spec(B) lying
over p

, there exists q Spec(B) lying over p such that q q

.
Lemma 40. The condition (GD) is equivalent to the following condition (GD): For any p
Spec(A) and any minimal prime overideal q of pB we have q A = p.
Proof. (GD) (GD) Clearly q A p. If this inclusion is proper then by (GD) there exists a
prime q
1
of B with q
1
q and q
1
A = p, contradicting minimality of q. (GD) (GD) Suppose
primes p p

of A are given and q

A = p

. We can shrink q

to a prime q minimal among all

prime ideals containing pB, and by assumption q A = p, which completes the proof.
Let a be any proper radical ideal in a noetherian ring B. Then a is the intersection of all
its minimal primes p
1
, . . . , p
n
and the closed irreducible sets V (p
1
) V (a) are the irreducible
components of the closed set V (a) in the noetherian space Spec(B).
Let : A B a morphism of rings, put X = Spec(A), Y = Spec(B) and let : Y X
the corresponding morphism of ane schemes, and suppose B is noetherian. Then (GD

) can be
formulated geometrically as follows: let p X, put X

= V (p) X and let Y

be an arbitrary
irreducible component of
1
(X

) (which we assume is nonempty). Then maps Y

generically
onto X

in the sense that the generic point of Y

is mapped to the generic point p of X

.
Theorem 41. Let : A B be a at morphism of rings. Then the going-down theorem holds
for .
Proof. Let p

p be prime ideals of A and let q be a prime ideal of B lying over p. Then B

q
is
at over A
p
by Lemma 33, hence faithfully at since A
p
B
q
is local. Therefore Spec(B
q
)
Spec(A
p
) is surjective. Let q

be a prime ideal of B
q
lying over p

A
p
. Then q

= q

B is a prime
ideal of B lying over p

and contained in q.
Theorem 42. Let B be a ring and A a subring over which B is integral. Then
(i) The canonical map Spec(B) Spec(A) is surjective.
(ii) If two prime ideals q q

lie over the same prime ideal of A then they are equal.
(iii) The going-up theorem holds for A B.
(iv) If A is a local ring and m its maximal ideal, then the prime ideals of B lying over m are
precisely the maximal ideals of B.
(v) If A and B are integral domains and A is integrally closed, then the going-down theorem
holds for A B.
Proof. See [AM69] or [Mat80] Theorem 5.
2.3 Constructible Sets
Denition 6. A topological space X is noetherian if the descending chain condition holds for
the closed sets in X. The spectrum Spec(A) of a noetherian ring A is noetherian. If a space
is covered by a nite number of noetherian subspaces then it is noetherian. Any subspace of a
noetherian space is noetherian. A noetherian space is quasi-compact. In a noetherian space X
any nonempty closed set Z is uniquely decomposed into a nite number of irreducible closed sets
Z = Z
1
Z
n
such that Z
i
Z
j
for i = j. The Z
i
are called the irreducible components of Z.
13
Lemma 43 (Noetherian Induction). Let X be a noetherian topological space, and P a property
of closed subsets of X. Assume that for any closed subset Y of X, if P holds for every proper
closed subset of Y , then P holds for Y (in particular P holds for the empty set). Then P holds
for X.
Proof. Suppose that P does not hold for X, and let Z be the set of all proper closed subsets
of X which do not satisfy P. Then since X is noetherian Z has a minimal element Y . Since
Y is minimal, every proper closed subset of Y must satisfy P, and therefore Y satises P,
contradicting the fact that Y Z.
Lemma 44. Let X be a noetherian topological space, and P a property of general subsets of X.
Assume that for any subset Y of X, if P holds for every proper subset Y

of Y with Y

Y , then
P holds for Y (in particular P holds for the empty set). Then P holds for X.
Proof. Suppose that P does not hold for X, and let Z be the set of all closures Q of proper subsets
Q of X with Q X and P not holding for Q. Let Q be a minimal element of Z. If Q

is any
proper subset of Q with Q

Q then Q

must satisfy P, otherwise Q

would contradict minimality

of Q in Z. But by assumption this implies that Q satises P, which is a contradiction.
Denition 7. Let X be a topological space and Z a subset of X. We say Z is locally closed in
X if it satises the following equivalent properties
(i) Every point z Z has an open neighborhood U in X such that U Z is closed in U.
(ii) Z is the intersection of an open set in X and a closed set in X.
(iii) Z is an open subset of its closure.
Denition 8. Let X be a noetherian space. We say a subset Z of X is a constructible set in X
if it is a nite union of locally closed sets in X, so Z =

m
i=1
(U
i
F
i
) with U
i
open and F
i
closed.
The set F of all constructible subsets of X is the smallest collection of subsets of X containing all
the open sets which is closed with respect to the formation of nite intersections and complements.
It follows that all open and closed sets are constructible, and F is also closed under nite unions.
We say that a subset Z is pro-constructible (resp. ind-constructible) if it is the intersection
(resp. union) of an arbitrary collection of constructible sets in X.
Proposition 45. Let X be a noetherian space and Z a subset of X. Then Z is constructible if
and only if the following condition is satised.
() For each irreducible closed set X
0
in X, either X
0
Z is not dense in X
0
, or X
0
Z
contains a nonempty open set of X
0
.
Proof. Assume that Z is constructible and Z X
0
nonempty. Then we can write X
0
Z =

m
i=1
U
i
F
i
for U
i
open in X, F
i
closed and irreducible in X (by taking irreducible components)
and U
i
F
i
nonempty for all i. Then U
i
F
i
= F
i
since F
i
is irreducible, therefore X
0
Z =

i
F
i
.
If X
0
Z is dense in X
0
, we have X
0
=

i
F
i
so that some F
i
, say F
1
, is equal to X
0
. Then
U
1
X
0
= U
1
F
1
is a nonempty open subset of X
0
contained in X
0
Z.
Next we prove the converse. We say that a subset T of X has the property P if whenever a
subset Z of T satises () it is constructible. We need to show that X has the property P, for
which we use the form of noetherian induction given in Lemma 44. Suppose that Y is a subset
of X with P holding for every proper subset Y

of Y with Y

Y . We need to show that P

holds for Y . Let Z be a nonempty subset of Y satisfying (), and let Z = F
1
. . . F
r
be the
decomposition of Z into irreducible components. Since Z = Z F
1
Z F
r
we have
F
1
= F
1
Z = F
1

_
Z F
1
. . . Z F
r
_
= (F
1
Z F
1
) (F
1
Z F
r
)
Since F
1
is irreducible and not contained in any other F
i
we must have F
1
= Z F
1
, so F
1
Z
is dense in F
1
, whence by () there exists a proper closed subset F

of F
1
such that F
1
\ F

Z.
Then, putting F

= F

F
2
F
r
we have Z = (F
1
\ F

) (Z F

). The set F
1
\ F

is locally
14
closed in X, so to complete the proof it suces to show that Z F

is constructible in X. Since
Z F

Z Y , by the inductive hypothesis P holds for ZF

, so it suces to show that

Z F

satises (). If X
0
is irreducible and Z F

X
0
= X
0
, the closed set F

must contain
X
0
and so Z F

X
0
= Z X
0
, which contains a nonempty open subset of X
0
since Z satises
(), and clearly Z X
0
is dense in X
0
.
Lemma 46. Let : A B be a morphism of rings and f : Spec(B) Spec(A) the corre-
sponding morphism of schemes. Then f dominant if and only if Ker nil(A). In particular if
A is reduced, the f dominant if and only if is injective.
Proof. Let X = Spec(A) and Y = Spec(B). The closure f(Y ) is the closed set V (I) dened by
the ideal I =

pY

1
p =
1

pY
p, which is
1
(nil(B)). Clearly Ker I. Suppose that
f(Y ) is dense in X. Then V (I) = X, whence I = nil(A) and so Ker nil(A). Conversely,
suppose Ker nil(A). Then it is clear that I =
1
(nil(B)) = nil(A), which means that
f(Y ) = V (I) = X.
3 Associated Primes
This material can be found in [AM69] Chapter 11, webnotes of Robert Ash or in [Mat80] itself.
There is not much relevant to add here, apart from a few small comments.
Lemma 47. Let A be a ring and M an A-module. Let a be an ideal in A that is maximal among
all annihilators of nonzero elements of M. Then a is prime.
Proof. Say a = Ann(x). Given ab a we must show that a a or b a. Assume a / a.
Then ax = 0. We note that Ann(ax) a. By hypothesis it cannot properly be larger. Hence
Ann(ax) = a. Now b annihilates ax; hence b a.
Lemma 48. Let A be a noetherian ring and M an A-module. If 0 = a M then Ann(a) is
contained in an associated prime of M.
Proposition 49. Let A be a noetherian ring and M a nonzero nitely generated A-module. A
maximal ideal m is an associated prime of M if and only if no element of m is regular on M.
Proof. One implication is obvious. If x m is not regular on M, say x Ann(b) for some nonzero
b, then x is contained in an associated prime of M. Thus m is contained in the nite union of the
associated primes of M, and since m is maximal it must be one of them.
Proposition 50. Let A be a nonzero noetherian ring, I an ideal, and M a nonzero nitely
generated A-module. If there exist elements x, y I with x regular on A and y regular on M, then
there exists an element of I regular on both A and M.
Proof. Let p
1
, . . . , p
n
be the associated primes of A and q
1
, . . . , q
m
the associated primes of M.
By assumption I is not contained in any of these primes. But if no element of I is regular on both
A and M, then I is contained in the union p
1
p
n
q
1
q
m
, and therefore contained
in one of these primes, which is a contradiction.
4 Dimension
This is covered in [AM69], so we restrict ourselves here to mentioning some of the major points.
Recall that an ideal q R in a ring is primary if it is proper and if whenever xy q we have
either x q or y
n
q for some n > 0. Then the radical of q is a prime ideal p, and we say q is
a p-primary ideal. If a is an ideal and b a is p-primary, then in the ring R/a the ideal b/a is
p/a-primary. A minimal primary decomposition of an ideal b is an expression
b = q
1
q
n
15
where
j=i
q
j
q
i
for all i, and the primes p
i
= r(q
i
) are all distinct. If a is an ideal contained in
b, then
b/a = q
1
/a q
n
/a
is a minimal primary decomposition of b/a in A/a.
Let A be a nonzero ring. Recall that dimension of an A-module M is the Krull dimension of
the ring A/Ann(M) and is dened for all modules M (1 if M = 0). The rank is dened for free
A-modules, and is the common size of any basis (0 if M = 0). Throughout these notes dim(M)
will denote the dimension, not the rank.
Denition 9. Let (A, m, k) be a noetherian local ring of dimension d. An ideal of denition is an
m-primary ideal. Recall that the dimension of A is the size of the smallest collection of elements
of A which generates an m-primary ideal. Recall that rank
k
(m/m
2
) is equal to the size of the
smallest set of generators for m as an ideal, so always d rank
k
(m/m
2
).
A system of parameters is a set of d elements generating an m-primary ideal. If d = rank
k
(m/m
2
),
or equivalently there is a system of parameters generating m, we say that A is a regular local ring
and we call such a system of parameters a regular system of parameters.
Proposition 51. Let (A, m) be a noetherian local ring of dimension d 1 and let x
1
, . . . , x
d
be
a system of parameters of A. Then
dim(A/(x
1
, . . . , x
i
)) = d i = dim(A) i
for each 1 i d.
Proof. Put A = A/(x
1
, . . . , x
i
). If i = d then the zero ideal in A is an ideal of denition, so clearly
dim(A) = 0. If 1 i < d then dim(A) d i since x
i+1
, . . . , x
d
generate an ideal of denition of
A. Let dim(A) = p. If p = 0 then (x
1
, . . . , x
i
) must be an ideal of denition, contradicting i < d.
So p 1, and if y
1
, . . . , y
p
is a system of parameters of A, then x
1
, . . . , x
i
, y
1
, . . . , y
p
generate an
ideal of denition of A, so that p +i d. That is, p d i.
Denition 10. Let A be a nonzero ring and I a proper ideal. The height of I, denoted ht.I, is
the minimum of the heights of the prime ideals containing I
ht.I = inf{ht.p | p I}
This is a number in {0, 1, 2, . . . , }. Equivalently we can take the inmum over the heights of
primes minimal over I. Clearly ht.0 = 0 and if I J are proper ideals then it is clear that
ht.I ht.J. If I is a prime ideal then ht.I is the usual height of a prime ideal. If A is a noetherian
ring then ht.I < for every proper ideal I, since A
p
is a local noetherian ring and ht.p = dim(A
p
).
Lemma 52. Let A be a nonzero ring and I a proper ideal. Then we have
ht.I = inf{ht.IA
m
| m is a maximal ideal and I m}
Lemma 53. Let A be a noetherian ring and suppose we have an exact sequence
0 M

M M

0
in which M

, M, M

are nonzero and nitely generated. Then dimM = max{dimM

, dimM

}.
Proof. We know that Supp(M) = Supp(M

) Supp(M

) and for all three modules the dimension

is the supremum of the coheights of prime ideals in the support. So the result is straightforward
to check.
16
4.1 Homomorphism and Dimension
Let : A B be a morphism of rings. If p Spec(A) then put (p) = A
p
/pA
p
. Let B
p
denote
the ring T
1
B where T = (Ap). There is an isomorphism of A-algebras B
p

= B
A
A
p
. There
is a commutative diagram of rings
B

R
R
R
R
R
R
R
R
R
R
R
R
R
R
R
R B
p

B
p
/pB
p

B
A
(p)
The vertical isomorphism is dened by b/(s) +pB
p
b (1/s +pA
p
). We call Spec(B
A
(p))
the bre over p of the map : Spec(B) Spec(A). Since primes of B
p
/pB
p
clearly correspond
to primes q of B with q A = p, it is easy to see that the ring morphism B B
A
(p) gives
rise to a continuous map Spec(B
A
(p)) SpecB which gives a homeomorphism between

1
(p) and Spec(B
A
(p)). See [AM69] Chapter 3.
Lemma 54. Let q be a prime ideal of B with q A = p and let P be the corresponding prime
ideal of B
A
(p). Then there is an isomorphism of A-algebras
B
q

A
(p)

= (B
A
(p))
P
b/t (a/s + pA
p
) (b (a/1 + pA
p
))/(t (s/1 + pA
p
))
Proof. It is not dicult to see that there is an isomorphism of rings B
q

= (B
p
)
qB
p
dened by
b/t (b/1)/(t/1). Consider the prime ideal qB
p
/pB
p
. We know that there is a ring isomorphism
(B
A
(p))
P

= (B
p
/pB
p
)
qB
p
/pB
p

= (B
p
)
qB
p
/p(B
p
)
qB
p

= B
q
/pB
q
by the comments following Lemma 7. It is not hard to check there is a ring isomorphism B
q
/pB
q

=
B
q

A
(p) dened by b/t +pB
q
b/t 1 (the inverse of b/t (a/s +pA
p
) is b(a)/t(s) +pB
q
).
So by denition of P there is an isomorphism of rings (B
A
(p))
P

= B
q

A
(p), and this is
clearly an isomorphism of A-algebras.
In particular if : A B is a ring morphism, p Spec(A) and q Spec(B) such that
q A = p, then there is an isomorphism of rings (B/pB)
q/pB

= B
q
/pB
q
, so we have ht.(q/pB) =
dim(B
q

A
(p)).
Theorem 55. Let : A B be a morphism of noetherian rings. Let q Spec(B) and put
p = q A. Then
(1) ht.q ht.p +ht.(q/pB). In other words dim(B
q
) dim(A
p
) +dim(B
q

A
(p)).
(2) We have equality in (1) if the going-down theorem holds for (in particular if is at).
(3) If : Spec(B) Spec(A) is surjective and if the going-down theorem holds, then we have
dim(B) dim(A) and ht.I = ht.(IB) for any proper ideal I of A.
Proof. (1) Replacing A by A
p
and B by B
q
we may suppose that (A, p) and (B, q) are noetherian
local rings such that q A = p, and we must show that dim(B) dim(A) + dim(B/pB). Let I
be a p-primary ideal of A. Then p
n
I for some n > 0, so p
n
B IB pB. Thus the ideals pB
and IB have the same radical, and so by denition dim(B/pB) = dim(B/IB). If dimA = 0 then
we can take I = 0 and the result is trivial. So assume dimA = r 1 and let I = (a
1
, . . . , a
r
) for
a system of parameters a
1
, . . . , a
r
. If dim(B/IB) = 0 then IB is an q-primary ideal of B and so
dim(B) r, as required. Otherwise if dim(B/IB) = s 1 let b
1
+ IB, . . . , b
s
+ IB be a system
of parameters of B/IB. Then b
1
, . . . , b
s
, a
1
, . . . , a
r
generate an ideal of denition of B. Hence
dim(B) r +s.
(2) We use the same notation as above. If ht.(q/pB) = s 0 then there exists a prime chain
of length s, q = q
0
q
1
q
s
such that q
s
pB. As p = q A q
i
A p all the q
i
17
lie over p. If ht.p = r 0 then there exists a prime chain p = p
0
p
1
p
r
in A, and by
going-down there exists a prime chain q
s
= t
0
t
1
t
r
of B such that t
i
A = p
i
. Then
q = q
0
q
s
t
1
t
r
is a prime chain of length r +s, therefore ht.q r +s.
(3) (i) follows from (2) since dim(A) = sup{ht.p | p Spec(A)}. (ii) Since is surjective IB
is a proper ideal. Let q be a minimal prime over IB such that ht.q = ht.(IB) and put p = q A.
Then ht.(q/pB) = 0, so by (2) we nd that ht.(IB) = ht.q = ht.p ht.I. For the reverse
inclusion, let p be a minimal prime ideal over I such that ht.p = ht.I and take a prime q of B
lying over p. Replacing q if necessary, we may assume that q is a minimal prime ideal over pB.
Then ht.I = ht.p = ht.q ht.(IB).
Theorem 56. Let A be a nonzero subring of B, and suppose that B is integral over A. Then
(1) dim(A) = dim(B).
(2) Let q Spec(B) and set p = q A. Then we have coht.p = coht.q and ht.q ht.p.
(3) If the going-down theorem holds between A and B, then for any ideal J of B with J A = A
we have ht.J = ht.(J A).
Proof. (1) By Theorem 42 the going-up theorem holds for A B and Spec(B) Spec(A) is
surjective, so we can lift any chain of prime ideals p
0
p
1
p
n
in A to a chain of prime
ideals q
0
q
n
in B. On the other hand, if q q

are prime ideals of B and q A = q

A,
then q = q

, so any chain of prime ideals in B restricts to a chain of the same length in A. Hence
dim(A) = dim(B).
(2) Since B/q is integral over A/p it is clear from (1) that coht.p = coht.q. If q = q
0
q
n
is a chain of prime ideals in B then intersecting with A gives a chain of length n descending from
p. Hence ht.q ht.p.
(3) Given the going-down theorem, it is clear that ht.q = ht.p in (2). Let J be a proper ideal of
B with J A = A and let q be such that ht.J = ht.q. Then ht.(J A) ht.(qA) = ht.q = ht.J.
On the other hand, B/J is integral over B/J A, so every prime ideal p of A containing J A
can be lifted to a prime ideal q of B containing J. In particular we can lift a prime ideal p with
ht.(J A) = ht.p, to see that ht.J ht.q = ht.p = ht.(J A), as required.
5 Depth
Denition 11. Let A be a ring, M an A-module and a
1
, . . . , a
r
a sequence of elements of A.
We say a
1
, . . . , a
r
is an M-regular sequence (or simply M-sequence) if the following conditions are
satised:
(1) For each 2 i r, a
i
is regular on M/(a
1
, . . . , a
i1
)M and a
1
is regular on M.
(2) M = (a
1
, . . . , a
n
)M.
If a
1
, . . . , a
r
is an M-regular sequence then so is a
1
, . . . , a
i
for any i r. When all a
i
belong to an
ideal I we say a
1
, . . . , a
r
is an M-regular sequence in I. If, moreover, there is no b I such that
a
1
, . . . , a
r
, b is M-regular, then a
1
, . . . , a
r
is said to be a maximal M-regular sequence in I. Notice
that the notion of M-regular depends on the order of the elements in the sequence. If M, N are
isomorphic A-modules then a sequence is regular on M i. it is regular on N.
Lemma 57. A sequence a
1
, . . . , a
r
with r 2 is M-regular if and only if a
1
is regular on M
and a
2
, . . . , a
r
is an M/a
1
M-regular sequence. If the sequence a
1
, . . . , a
r
is a maximal M-regular
sequence in I then a
2
, . . . , a
r
is a maximal M/a
1
M-regular sequence in I.
18
Proof. The key point is that for ideals a b there is a canonical isomorphism of A-modules
M/bM

= N/bN where N = M/aM. If a
1
, . . . , a
r
is M-regular then a
1
is regular on M, a
2
is
regular on N = M/a
1
M and for 3 i r, a
i
is regular on
M/(a
1
, . . . , a
i1
)M

= N/(a
2
, . . . , a
i1
)N
Hence a
2
, . . . , a
r
is an N-regular sequence. The converse follows from the same argument.
More generally if a
1
, . . . , a
r
is an M-regular sequence and we set N = M/(a
1
, . . . , a
r
)M, and
if b
1
, . . . , b
s
is an N-regular sequence, then a
1
, . . . , a
r
, b
1
, . . . , b
s
is an M-regular sequence.
Lemma 58. If a
1
, . . . , a
r
is an A-regular sequence and M is a at A-module, then a
1
, . . . , a
r
is
also M-regular provided (a
1
, . . . , a
r
)M = M.
Proof. Left multiplication by a
1
denes a monomorphism A A since a
1
is A-regular. Tensoring
with M and using the fact that M is at we see that left multiplication by a
1
also gives a
monomorphism M M, as required. Similarly tensoring with the monomorphism a
2
: A/a
1

A/a
1
we get a monomorphism M/a
1
M M/a
1
M, and so on.
Lemma 59. Let A be a ring and M an A-module. Given an integer n 1, a sequence a
1
, . . . , a
r
is M-regular if and only if it is M
n
-regular.
Proof. Suppose the sequence a
1
, . . . , a
r
is M-regular. We prove it is M
n
-regular by induction on r.
The case r = 1 is trivial, so assume r > 1. By the inductive hypothesis the sequence a
1
, . . . , a
r1
is
M
n
-regular. Let L = (a
1
, . . . , a
r1
)M. Then (a
1
, . . . , a
r1
)M
n
= L
n
and there is an isomorphism
of A-modules M
n
/L
n
= (M/L)
n
. So we need only show that a
r
is regular on (M/L)
n
. Since by
assumption it is regular on M/L, this is not hard to check. Clearly (a
1
, . . . , a
r
)M
n
= M
n
, so the
sequence a
1
, . . . , a
r
is M
n
-regular, as required. The converse is similarly checked.
Lemma 60. Let A be a nonzero ring, M an A-module and a
1
, . . . , a
r
A. If a
1
, . . . , a
r
A
m
is
M
m
-regular for every maximal ideal m of A then the sequence a
1
, . . . , a
r
is M-regular.
Proof. This follows from the fact that given an A-module M an element a A is regular on M if
and only if its image in A
m
is regular on M
m
for every maximal ideal m of A.
Lemma 61. Suppose that a
1
, . . . , a
r
is M-regular and a
1

1
+ + a
r

r
= 0 for
i
M. Then

i
(a
1
, . . . , a
r
)M for all i.
Proof. By induction on r. For r = 1, a
1

1
= 0 implies that
1
= 0. Let r > 1. Since a
r
is regular
on M/(a
1
, . . . , a
r1
)M we have
r
=

r1
i=1
a
i

i
, so

r1
i=1
a
i
(
i
+ a
r

i
) = 0. By the inductive
hypothesis for i < r we have
i
+a
r

i
(a
1
, . . . , a
r1
)M so that
i
(a
1
, . . . , a
r
)M.
Theorem 62. Let A be a ring and M an A-module, and let a
1
, . . . , a
r
be an M-regular sequence.
Then for every sequence n
1
, . . . , n
r
of integers > 0 the sequence a
n
1
1
, . . . , a
n
r
r
is M-regular.
Proof. Suppose we can prove the following statement
() Given an integer n > 0, an A-module M and any M-regular sequence a
1
, . . . , a
r
the sequence a
n
1
, a
2
, . . . , a
r
is also M-regular.
We prove the rest of the Theorem by induction on r. For r = 1 this follows immediately from ().
Let r > 1 and suppose a
1
, . . . , a
r
is M-regular. Then by () a
n
1
1
, a
2
, . . . , a
r
is M-regular. Hence
a
2
, . . . , a
r
is M/a
n
1
1
M-regular. By the inductive hypothesis a
n
2
2
, . . . , a
n
r
r
is M/a
n
1
1
M-regular and
therefore a
n
1
1
, . . . , a
n
r
r
is M-regular by Lemma 57.
So it only remains to prove (), which we do by induction on n. The case n = 1 is trivial, so let
n > 1 be given, along with an A-module M and an M-regular sequence a
1
, . . . , a
r
. By the inductive
hypothesis a
n1
1
, a
2
, . . . , a
r
is M-regular and clearly a
n
1
is regular on M. Since (a
n
1
, a
2
, . . . , a
r
)
(a
n1
1
, a
2
, . . . , a
r
) it is clear that M = (a
n
1
, a
2
, . . . , a
r
)M. Let i > 1 and assume that a
n
1
, a
2
, . . . , a
i1
is an M-regular sequence. We need to show that a
i
is regular on M/(a
n
1
, a
2
, . . . , a
i1
)M. Suppose
19
that a
i
= a
n
1

1
+a
2

2
+ +a
i1

i1
. Then = a
n1
1

1
+a
2

2
+ +a
i1

i1
by the inductive
hypothesis. So
a
n1
1
(a
1

1
a
i

1
) +a
2
(
2
a
i

2
) + +a
i1
(
i1
a
i

i1
) = 0
Hence a
1

1
a
i

1
(a
n1
1
, a
2
, . . . , a
i1
)M by Lemma 61. It follows that a
i

1
(a
1
, a
2
, . . . , a
i1
)M,
hence
1
(a
1
, . . . , a
i1
)M and so (a
n
1
, a
2
, . . . , a
i1
)M, as required. This proves () and
therefore completes the proof.
Let A be a ring. There is an isomorphism of A[x
1
, . . . , x
n
]-modules M
A
A[x
1
, . . . , x
n
]

=
M[x
1
, . . . , x
n
] where the latter module consists of polynomials in x
1
, . . . , x
n
with coecients in M
(see our Polynomial Ring notes). For any f(x
1
, . . . , x
n
) M[x
1
, . . . , x
n
] and tuple (a
1
, . . . , a
n
)
A
n
we can dene an element of M
f(a
1
, . . . , a
n
) =

1
1
a

n
n
f()
For an element r R and h M[x
1
, . . . , x
n
]
(f +h)(a
1
, . . . , a
n
) = f(a
1
, . . . , a
n
) +h(a
1
, . . . , a
n
)
(r f)(a
1
, . . . , a
n
) = r f(a
1
, . . . , a
n
)
For an ideal I in R the R-submodule IM[x
1
, . . . , x
n
] consists of all polynomials whose coecients
are in the R-submodule IM M.
Let us review the denition of the associated graded ring and modules. Let A be a ring and I
an ideal of A. Then the abelian group
gr
I
(A) = A/I I/I
2
I
2
/I
3

becomes a graded ring in a fairly obvious way. For an A-module M we have the graded gr
I
(A)-
module
gr
I
(M) = M/IM IM/I
2
M I
2
M/I
3
M
If A is noetherian and M is a nitely generated A-module, then gr
I
(A) is a noetherian ring and
if gr
I
(M) is a nitely generated gr
I
(A)-module.
Given elements a
1
, . . . , a
n
A and I = (a
1
, . . . , a
n
), we dene a morphism of abelian groups
: M[x
1
, . . . , x
n
] gr
I
(M) as follows: if f is homogenous of degree m 0, dene (f) to be the
image of f(a
1
, . . . , a
n
) in I
m
M/I
m+1
M. This denes a morphism of groups M[x
1
, . . . , x
n
]
m

I
m
M/I
m+1
M and together these dene the morphism of groups . Since maps IM[x
1
, . . . , x
n
]
to zero it induces a morphism of abelian groups : (M/IM)[x
1
, . . . , x
n
] gr
I
(M), and
Proposition 63. Let A be a ring and M an A-module. Let a
1
, . . . , a
n
A and set I =
(a
1
, . . . , a
n
). Then the following conditions are equivalent
(a) For every m > 0 and for every homogenous polynomial f(x
1
, . . . , x
n
) M[x
1
, . . . , x
n
] of
degree m such that f(a
1
, . . . , a
n
) I
m+1
M, we have f IM[x
1
, . . . , x
n
].
(b) If f(x
1
, . . . , x
n
) M[x
1
, . . . , x
n
] is homogenous and f(a
1
, . . . , a
n
) = 0 then the coecients
of f are in IM.
(c) The morphism of abelian groups : (M/IM)[x
1
, . . . , x
n
] gr
I
(M) dened by mapping
a homogenous polynomial f(x
1
, . . . , x
n
) of degree m to f(a
1
, . . . , a
n
) I
m
M/I
m+1
M is an
isomorphism.
Proof. It is easy to see that (a) (c) and (a) (b). To show (b) (a) let f M[x
1
, . . . , x
n
] be
a homogenous polynomial of degree m > 0 and suppose f(a
1
, . . . , a
n
) I
m+1
M. Any element of
I
m+1
M can be written as a sum of terms of the form a

1
1
a

n
n
m with

i
= m+1. By shifting
one of the a
i
across we can write f(a
1
, . . . , a
n
) = g(a
1
, . . . , a
n
) for a homogenous polynomial
g M[x
1
, . . . , x
n
] of degree m all of whose coecients belong to IM. Hence (f g)(a
1
, . . . , a
n
) = 0
so by (b) the coecients of f g belong to IM, and this implies implies that the coecients of f
also belong to IM, as required.
20
Denition 12. Let A be a ring and M an A-module. A sequence a
1
, . . . , a
n
A is M-quasiregular
if it satises the equivalent conditions of the Proposition. Obviously this concept does not depend
on the order of the elements. But a
1
, . . . , a
i
for i < n need not be M-quasiregular.
Recall that for an A-module M, a submodule N M and x A the notation (N : x) means
{m M| xm N}. This is a submodule of M. If A is a ring, I an ideal and M an A-module,
recall that M is separated in the I-adic topology when

n
I
n
M = 0.
Theorem 64. Let A be a ring, M a nonzero A-module, a
1
, . . . , a
n
A and I = (a
1
, . . . , a
n
).
Then
(i) If a
1
, . . . , a
n
is M-quasiregular and x A is such that (IM : x) = IM, then (I
m
M : x) =
I
m
M for all m > 0.
(ii) If a
1
, . . . , a
n
is M-regular then it is M-quasiregular.
(iii) If M, M/a
1
M, M/(a
1
, a
2
)M, . . . , M/(a
1
, . . . , a
n1
)M are separated in the I-adic topology,
then the converse of (ii) is also true.
Proof. (i) By induction on m, with the case m = 1 true by assumption. Suppose m > 1 and
x I
m
M. By the inductive hypothesis I
m1
M. Hence there exists a homogenous polynomial
f M[x
1
, . . . , x
n
] of degree m1 such that = f(a
1
, . . . , a
n
). Since x = xf(a
1
, . . . , a
n
) I
m
M
the coecients of f are in (IM : x) = IM. Therefore = f(a
1
, . . . , a
n
) I
m
M.
(ii) By induction on n. For n = 1 this is easy to check. Let n > 1 and suppose a
1
, . . . , a
n
is M-
regular. Then by the induction hypothesis a
1
, . . . , a
n1
is M-quasiregular. Let f M[x
1
, . . . , x
n
]
be homogenous of degree m > 0 such that f(a
1
, . . . , a
n
) = 0. We prove that f IM[x
1
, . . . , x
n
]
by induction on m (the case m = 0 being trivial). Write
f(x
1
, . . . , x
n
) = g(x
1
, . . . , x
n1
) +x
n
h(x
1
, . . . , x
n
)
Then g and h are homogenous of degrees m and m1 respectively. By (i) we have
h(a
1
, . . . , a
n
) ((a
1
, . . . , a
n1
)
m
M : a
n
) = (a
1
, . . . , a
n1
)
m
M I
m
M
Since by assumption a
1
, . . . , a
n
is regular on M, so a
n
is regular on M/(a
1
, . . . , a
n1
)M and
hence ((a
1
, . . . , a
n1
)M : a
n
) = (a
1
, . . . , a
n1
)M. So by the induction hypothesis on m we have
h IM[x
1
, . . . , x
n
] (by the argument of Proposition 63). Since h(a
1
, . . . , a
n
) (a
1
, . . . , a
n1
)
m
M
there exists H M[x
1
, . . . , x
n1
] which is homogenous of degree m such that h(a
1
, . . . , a
n
) =
H(a
1
, . . . , a
n1
). Putting
G(x
1
, . . . , x
n1
) = g(x
1
, . . . , x
n1
) +a
n
H(x
1
, . . . , x
n1
)
we have G(a
1
, . . . , a
n1
) = 0, so by the inductive hypothesis on n we have G IM[x
1
, . . . , x
n
],
hence g IM[x
1
, . . . , x
n
] and so f IM[x
1
, . . . , x
n
].
(iii) By induction on n 1. Assume that a
1
, . . . , a
n
is M-quasiregular and the modules
M, M/a
1
M, . . . , M/(a
1
, . . . , a
n1
)M are all separated in the I-adic topology. If a
1
= 0 then
IM, hence =

a
i

i
and

a
1
a
i

i
= 0, hence
i
IM and so I
2
M. In this way we
see that

t
I
t
M = 0. Thus a
1
is regular on M, and this also takes care of the case n = 1
since M = IM by the separation condition. So assume n > 1. By Lemma 57 it suces to show
that a
2
, . . . , a
n
is an N-regular sequence, where N = M/a
1
M. Since there is an isomorphism of
A-modules for 2 i n 1
M/(a
1
, . . . , a
i
)M

= N/(a
2
, . . . , a
i
)N
The modules N, N/a
2
N, . . . , N/(a
2
, . . . , a
n1
)N are separated in the I-adic topology. So by the
inductive hypothesis it suces to show that the sequence a
2
, . . . , a
n
is N-quasiregular.
It suces to show that if f(x
2
, . . . , x
n
) M[x
2
, . . . , x
n
] is homogenous of degree m 1 with
f(a
2
, . . . , a
n
) a
1
M then the coecients of f belong to IM. Put f(a
2
, . . . , a
n
) = a
1
. We claim
21
that I
m1
M. Let 0 i m1 be the largest integer with I
i
M. Then = g(a
1
, . . . , a
n
)
for some homogenous polynomial of degree i, and
f(a
2
, . . . , a
n
) = a
1
g(a
1
, . . . , a
n
) (2)
If i < m1 then g IM[x
1
, . . . , x
n
] and so I
i+1
M, which is a contradiction. Hence i = m1
and so I
m1
M. Again using (2) we see that f(x
2
, . . . , x
n
)x
1
g(x
1
, . . . , x
n
) IM[x
1
, . . . , x
n
].
Since f does not involve x
1
we have f IM[x
1
, . . . , x
n
], as required.
The theorem shows that, under the assumptions of (iii) any permutation of an M-regular
sequence is M-regular.
Corollary 65. Let A be a noetherian ring, M a nitely generated A-module and let a
1
, . . . , a
n
be contained in the Jacobson radical of A. Then a
1
, . . . , a
n
is M-regular if and only if it is M-
quasiregular. In particular if a
1
, . . . , a
n
is M-regular so is any permutation of the sequence.
Proof. From [AM69] we know that for any ideal I contained in the Jacobson radical, the I-adic
topology on any nitely generated A-module is separated.
If A is a ring and M an A-module, then any M-regular sequence a
1
, . . . , a
n
A gives rise to
a strictly increasing chain of submodules a
1
M, (a
1
, a
2
)M, . . . , (a
1
, . . . , a
n
)M. Hence the chain of
ideals (a
1
), (a
1
, a
2
), . . . , (a
1
, . . . , a
n
) must also be strictly increasing.
Lemma 66. Let A be a noetherian ring and M an A-module. Any M-regular sequence a
1
, . . . , a
n
in an ideal I can be extended to a maximal M-regular sequence in I.
Proof. If a
1
, . . . , a
n
is not maximal in I, we can nd a
n+1
I such that a
1
, . . . , a
n
, a
n+1
is an
M-regular sequence. Either this process terminates at a maximal M-regular sequence in I, or it
produces a strictly ascending chain of ideals
(a
1
) (a
1
, a
2
) (a
1
, a
2
, a
3
)
Since A is noetherian, we can exclude this latter possibility.
Theorem 67. Let A be a noetherian ring, M a nitely generated A-module and I an ideal of A
with IM = M. Let n > 0 be an integer. Then the following are equivalent:
(1) Ext
i
A
(N, M) = 0 for i < n and every nitely generated A-module N with Supp(N) V (I).
(2) Ext
i
A
(A/I, M) = 0 for i < n.
(3) There exists a nitely generated A-module N with Supp(N) = V (I) and Ext
i
A
(N, M) = 0
for i < n.
(4) There exists an M-regular sequence a
1
, . . . , a
n
of length n in I.
Proof. (1) (2) (3) is trivial. With I xed we show that (3) (4) for every nitely generated
module M with IM = M by induction on n. We have 0 = Ext
0
A
(N, M)

= Hom
A
(N, M). Since
M is nitely generated and nonzero, the set of associated primes of M is nite and nonempty. If
no elements of I are M-regular, then I is contained in the union of these associated primes, and
hence I p for some p Ass(M) (see [AM69] for details). By denition there is a monomorphism
of A-modules : A/p M. There is an isomorphism of A-modules
(A/p)
p

= A/p
A
A
p

= A
p
/pA
p
= k
It is not hard to check this is an isomorphism of A
p
-modules as well. Since
p
is a monomorphism
and k = 0 it follows that Hom
A
p
(k, M
p
) = 0. Since p V (I) = Supp(N) we have N
p
= 0 and
so the k-module N
p
/pN
p
= 0 is nonzero and therefore free, so Hom
k
(N
p
/pN
p
, k) = 0. Since
k

= (A/p)
p
as A
p
-modules it follows that Hom
A
(N, A/p)
p

= Hom
A
p
(N
p
, (A/p)
p
) = 0. Since A/p
is isomorphic to a submodule of M it follows that Hom
A
(N, M) = 0, which is a contradiction,
22
therefore there exists an M-regular element a
1
I, which takes care of the case n = 1. If n > 1
then put M
1
= M/a
1
M. From the exact sequence
0

M
a
1

M

M
1

0 (3)
we get the long exact sequence
Ext
i
A
(N, M) Ext
i
A
(N, M
1
) Ext
i+1
A
(N, M)
which shows that Ext
i
A
(N, M
1
) = 0 for 0 i < n 1. By the inductive hypothesis on n there
exists an M
1
-regular sequence a
2
, . . . , a
n
in I. The sequence a
1
, . . . , a
n
is then an M-regular
sequence in I.
(4) (1) By induction on n with I xed. For n = 1 we have a
1
I regular on M and so (3)
gives an exact sequence of R-modules
0

Hom
A
(N, M)
a
1

Hom
A
(N, M)
Where a
1
denotes left multiplication by a
1
. Since Supp(N) = V (Ann(N)) V (I) it fol-
lows that I r(Ann(N)), and so a
r
1
N = 0 for some r > 0. It follows that a
r
1
annihilates
Hom
A
(N, M) as well, but since the action of a
r
1
on Hom
A
(N, M) gives an injective map it follows
that Hom
A
(N, M) = 0. Now assume n > 1 and put M
1
= M/a
1
M. Then a
2
, . . . , a
n
is an
M
1
-regular sequence, so by the inductive hypothesis Ext
i
A
(N, M
1
) = 0 for i < n 1. So the long
exact sequence corresponding to (3) gives an exact sequence for 0 i < n
0

Ext
i
A
(N, M)
a
1

Ext
i
A
(N, M)
Here a
1
denotes left multiplication by a
1
, which is equal to Ext
i
A
(, M) = Ext
i
(N, ) where ,
are the endomorphisms given by left multiplication by a
1
on N, M respectively. Assume that a
r
1
annihilates N with r > 0. Then
r
is the zero map, so Ext
i
A
(, M)
r
= 0 and so a
r
1
also annihilates
Ext
i
A
(N, M). Since the a
1
is regular on this module, it follows that Ext
i
A
(N, M) = 0 for i < n,
as required.
Corollary 68. Let A be a noetherian ring, M a nitely generated A-module, and I an ideal of
A with IM = M. If a
1
, . . . , a
n
a maximal M-regular sequence in I, then Ext
i
A
(A/I, M) = 0 for
i < n and Ext
n
A
(A/I, M) = 0.
Proof. We already know that Ext
i
A
(A/I, M) = 0 for i < n, so with I xed we prove by induction
on n that Ext
n
A
(A/I, M) = 0 for any nitely generated module M with IM = M admitting a
maximal M-regular sequence of length n. For n = 1 we have a
1
I regular on M and an exact
sequence (3) where M
1
= M/a
1
M. Part of the corresponding long exact sequence is
Ext
0
A
(A/I, M) Ext
0
A
(A/I, M
1
) Ext
1
A
(A/I, M)
We know from the Theorem 67 that Ext
0
A
(A/I, M) = 0, so it suces to show that we have
Hom
A
(A/I, M
1
) = 0. But if Hom
A
(A/I, M
1
) = 0 then it follows from the proof of (3) (4)
above that there would be b I regular on M
1
, so a
1
, b is an M-regular sequence. This is a
contradiction since the sequence a
1
was maximal, so we conclude that Ext
1
A
(A/I, M) = 0.
Now assume n > 1 and let a
1
, . . . , a
n
be a maximal M-regular sequence in I. Then a
2
, . . . , a
n
is a maximal M
1
-regular sequence in I, so by the inductive hypothesis Ext
n1
A
(A/I, M
1
) = 0. So
from the long exact sequence for (3) we conclude that Ext
n
A
(A/I, M) = 0 also.
It follows that under the conditions of the Corollary every maximal M-regular sequence in I
has a common length, and you can nd this length by looking at the sequence of abelian groups
Hom
A
(A/I, M), Ext
1
A
(A/I, M), Ext
2
A
(A/I, M), . . . , Ext
n
A
(A/I, M), . . .
If there are M-regular sequences in I, then this sequence will start o with n 1 zero groups,
where n 1 is the common length of every maximal M-regular sequence. The nth group will
23
be nonzero, and we cant necessarily say anything about the rest of the sequence. Notice that
since any M-regular sequence can be extended to a maximal one, any M-regular sequence has
length n. There are no M-regular sequences in I if and only if the rst term of this sequence
is nonzero.
Denition 13. Let A be a noetherian ring, M a nitely generated A-module, and I an ideal of
A. If IM = M then we dene the I-depth of M to be
depth
I
(M) = inf{i | Ext
i
A
(A/I, M) = 0}
So depth
I
(M) = 0 if and only if there are no M-regular sequences in I, and otherwise it is the
common length of all maximal M-regular sequences in I, or equivalently the supremum of the
lengths of M-regular sequences in I. We dene depth
I
(M) = if IM = M. In particular
depth
I
(0) = . Isomorphic modules have the same I-depth. When (A, m) is a local ring we write
depth(M) or depth
A
M for depth
m
(M) and call it simply the depth of M. Thus depth(M) =
i. M = 0 and depth(M) = 0 i. m Ass(M).
Lemma 69. Let : A B be a surjective local morphism of local noetherian rings, and let M
be a nitely generated B-module. Then depth
A
(M) = depth
B
(M).
Proof. It is clear that depth
A
(M) = i. depth
B
(M) = , so we may as well assume both
depths are nite. Given a sequence of elements a
1
, . . . , a
n
m
A
it is clear that they are an
M-regular sequence i. the images (a
1
), . . . , (a
n
) m
B
are an M-regular sequence. Given an
M-regular sequence b
1
, . . . , b
n
in m
B
you can choose inverse images a
1
, . . . , a
n
m
A
and these
form an M-regular sequence. This makes it clear that depth
A
(M) = depth
B
(M).
Lemma 70. Let A be a noetherian ring and M a nitely generated A-module. Then for any ideal
I and integer n 1 we have depth
I
(M) = depth
I
(M
n
).
Proof. We have IM
n
= (IM)
n
so depth
I
(M) = if and only if depth
I
(M
n
) = . In the nite
case the result follows immediately from Lemma 59.
Lemma 71. Let A be a noetherian ring, M a nitely generated A-module and p a prime ideal.
Then depth
A
p
(M
p
) = 0 if and only if p Ass
A
(M).
Proof. We have depth
A
p
(M
p
) = 0 i. pA
p
Ass
A
p
(M
p
) which by [Ash] Chapter 1, Lemma 1.4.2
is i. p Ass
A
(M). So the associated primes are precisely those with depth
A
p
(M
p
) = 0.
Lemma 72. Let A be a noetherian ring, and M a nitely generated A-module. For any prime p
we have depth
A
p
(M
p
) depth
p
(M).
Proof. If depth
p
(M) = then pM = M, and this implies that (pA
p
)M
p
= M
p
so depth
A
p
(M) =
. If depth
A
p
(M
p
) = 0 then pA
p
Ass(M
p
) which can only occur if p Ass(M), and this implies
that Hom
A
(A/p, M) = 0, so depth
p
(M) = 0 (since we have already excluded the possibility of
pM = M). So we can reduce to the case where depth
A
p
(M
p
) = n with 0 < n < and pM = M.
We have seen earlier in notes that there is an isomorphism of groups for i 0
Ext
i
A
p
((A/p)
p
, M
p
)

= Ext
i
A
(A/p, M)
p
As an A
p
-module (A/p)
p

= A
p
/pA
p
and by assumption Ext
n
A
p
(A
p
/pA
p
, M
p
) = 0, so it follows
that Ext
n
A
(A/p, M) = 0 and hence depth
p
(M) n.
Denition 14. Let A be a noetherian ring and M a nitely generated A-module. Then we
dene the grade of M, denoted grade(M), to be depth
I
(A) where I is the ideal Ann(M). So
grade(M) = if and only if M = 0. Isomorphic modules have the same grade.
If I is an ideal of A then we call grade(A/I) = depth
I
(A) the grade of I and denote it by G(I).
So the grade of A is and the grade of any proper ideal I is the common length of the maximal
A-regular sequences in I (zero if none exist).
24
Lemma 73. Let A be a noetherian ring and M a nonzero nitely generated A-module. Then
grade(M) = inf{i | Ext
i
(M, A) = 0}
Proof. Put I = Ann(M). Since M and A/I are both nitely generated A-modules whose supports
are equal to V (I) it follows from Theorem 67 that for any n > 0 we have Ext
i
(A/I, A) = 0 for
all i < n if and only if Ext
i
(M, A) = 0 for all i < n. In particular Ext
0
(M, A) = 0 if and only if
Ext
0
(A/I, A) = 0. By denition
grade(M) = depth
I
(A) = inf{i | Ext
i
(A/I, M) = 0}
so the claim is straightforward to check.
The following result is a generalisation of Krulls Principal Ideal Theorem.
Lemma 74. Let A be a noetherian ring and a
1
, . . . , a
r
an A-regular sequence. Then every minimal
prime over (a
1
, . . . , a
r
) has height r, and in particular ht.(a
1
, . . . , a
r
) = r.
Proof. By assumption I = (a
1
, . . . , a
r
) is a proper ideal. If r = 1 then this is precisely Krulls
PID Theorem. For r > 1 we proceed by induction. If a
1
, . . . , a
r
is an A-regular sequence then
set J = (a
1
, . . . , a
r1
). Clearly a
r
+ J is a regular element of R/J which is not a unit, so every
minimal prime over (a
r
+J) in R/J has height 1. But these are precisely the primes in R minimal
over I. So if p is any prime ideal minimal over I there is a prime J q p with q minimal over
J. By the inductive hypothesis ht.q = r 1 so ht.p r. We know the height is r by another
result of Krull.
For any nonzero ring A the sequence x
1
, . . . , x
n
in A[x
1
, . . . , x
n
] is clearly a maximal A-regular
sequence. So in some sense regular sequences in a ring generalise the notion of independent
variables.
Lemma 75. Let A be a noetherian ring, M a nonzero nitely generated A-module and I a proper
ideal. Then grade(M) proj.dim.M and G(I) ht.I.
Proof. For a nonzero module M the projective dimension is the largest i 0 for which there
exists a module N with Ext
i
(M, N) = 0. So clearly grade(M) proj.dim.M. The second claim
is trivial if G(I) = 0 and otherwise G(I) is the length r of a maximal A-regular sequence a
1
, . . . , a
r
in I. But then r = ht.(a
1
, . . . , a
r
) ht.I, so the proof is complete.
Proposition 76. Let A be a noetherian ring, M, N nitely generated A-modules with M nonzero,
and suppose that grade(M) = k and proj.dim.N = < k. Then
Ext
i
A
(M, N) = 0 (0 i < k )
Proof. Induction on . If = 1 then this is trivial. If = 0 then is a direct summand of a free
module. Since our assertion holds for A by denition, it holds for N also. If > 0 take an exact
sequence 0 N

L N 0 with L free. Then proj.dim.N

= 1 and our assertion

is proved by induction.
Lemma 77 (Ischebeck). Let (A, m) be a noetherian local ring and let M, N be nonzero nitely
generated A-modules. Suppose that depth(M) = k, dim(N) = r. Then
Ext
i
A
(N, M) = 0 (0 i < k r)
Proof. By induction on integers r with r < k (we assume k > 0). If r = 0 then Supp(N) = {m}
so the assertion follows from Theorem 67. Let r > 0. First we prove the result in the case where
N = A/p for a prime ideal p. We can pick x m\ p and then the following sequence is exact
0

N
x

N

N

0
25
where N

= A/(p + Ax) has dimension < r. Then using the induction hypothesis we get exact
sequences of A-modules
0 = Ext
i
(N

, M)

Ext
i
A
(N, M)
x

Ext
i
A
(N, M)

Ext
i+1
A
(N

, M) = 0
for 0 i < k r, and so Ext
i
A
(N, M) = 0 by Nakayama, since the module Ext
i
A
(N, M) is nitely
generated (see our Ext notes). This proves the result for modules of the form N = A/p.
For general N we use know from [Ash] Chapter 1, Theorem 1.5.10 that there is a chain of
modules 0 = N
0
N
s
= N such that for 1 j s we have an isomorphism of A-modules
N
j
/N
j1

= A/p
j
where the p
j
are prime ideals of A. Lemma 53 shows that dimN
1
dimN
2

dimN = r, so since N
1

= A/p
1
we have already shown the result holds for N
1
. Consider
the exact sequence
0 N
1
N
2
A/p
2
0
By Lemma 53 we know that r dimA/p
2
, so the result holds for A/p
2
and the following piece of
the long exact Ext sequence shows that the result is true for N
2
as well
Ext
i
A
(A/p
2
, M) Ext
i
A
(N
2
, M) Ext
i
A
(N
1
, M)
Proceeding in this way proves the result for all N
j
and hence for N, completing the proof.
Theorem 78. Let (A, m) be a noetherian local ring and let M be a nonzero nitely generated
A-module. Then depth(M) dim(A/p) for every p Ass(M).
Proof. If p Ass(M) then Hom
A
(A/p, m) = 0, hence depth(M) dim(A/p) by Lemma 77.
Lemma 79. Let A be a ring and let E, F be nitely generated A-modules. Then Supp(E
A
F) =
Supp(E) Supp(F).
Proof. See [AM69] Chapter 3, Exercise 19.
The Dimension Theorem for modules (see [AM69] Chapter 11) shows that for a nonzero nitely
generated module M over a noetherian local ring A, the dimension of M is zero i. M is of nite
length, and otherwise dim(M) is the smallest r 1 for which there exists elements a
1
, . . . , a
r
m
with M/(a
1
, . . . , a
n
)M of nite length.
Proposition 80. Let A be a noetherian local ring and M a nitely generated A-module. Let
a
1
, . . . , a
r
be an M-regular sequence. Then
dimM/(a
1
, . . . , a
r
)M = dimM r
In particular if a
1
, . . . , a
r
is an A-regular sequence, then the dimension of the ring A/(a
1
, . . . , a
r
)
is dimAr.
Proof. Let N = M/(a
1
, . . . , a
r
)M. Then N is a nonzero nitely generated A-module, so if
k = dim(N) then 0 k < . If k = 0 then it is clear from the preceding comments that
dimM/(a
1
, . . . , a
r
)M dimMr. If k 1 and b
1
, . . . , b
k
m are elements such that the module
N/(b
1
, . . . , b
k
)N

= M/(a
1
, . . . , a
r
, b
1
, . . . , b
k
)M
is of nite length, then since the a
i
all belong to m we conclude that dim(M) r +k. Hence we at
least have the inequality dimM/(a
1
, . . . , a
n
)M dim(M) r. On the other hand, suppose f m
is an M-regular element. We have Supp(M/fM) = Supp(M) Supp(A/fA) = Supp(M) V (f)
by Lemma 79, and f is not in any minimal element of Supp(M) since these coincide with the
minimal elements of Ass(M), and f is regular on M. Since
dimM = sup{coht.p | p Supp(M)}
it follows easily that dim(M/fM) < dimM. Proceeding by induction on r we see that
dimM/(a
1
, . . . , a
r
)M dimM r
as required.
26
Corollary 81. Let (A, m) be a noetherian local ring and M a nonzero nitely generated A-module.
Then depthM dimM.
Proof. This is trivial if depthM = 0. Otherwise let a
1
, . . . , a
r
be a maximal M-regular sequence
in m, so depthM = r. Then we know from Proposition 80 that r = dimMdimM/(a
1
, . . . , a
r
)M,
so of course r dimM.
Lemma 82. Let A be a noetherian ring, M a nitely generated A-module and I an ideal. Let
a
1
, . . . , a
r
be an M-regular sequence in I and assume IM = M. Then
depth
I
(M/(a
1
, . . . , a
r
)M) = depth
I
(M) r
Proof. Let N = M/(a
1
, . . . , a
r
)M. It is clear that IM = M i. IN = N so both depths are
nite. If depth
I
(N) = 0 then the sequence a
1
, . . . , a
r
must be a maximal M-regular sequence in
I, so depth
I
(M) = r and we are done. Otherwise let b
1
, . . . , b
s
be a maximal N-regular sequence
in I. Then a
1
, . . . , a
r
, b
1
, . . . , b
s
is a maximal M-regular sequence in I, so depth
I
(M) = r + s =
r +depth
I
(N), as required.
Lemma 83. Let A be a noetherian local ring, a
1
, . . . , a
r
an A-regular sequence. If I = (a
1
, . . . , a
r
)
then
depth
A/I
(A/I) = depth
A
(A) r
Proof. A sequence b
1
, . . . , b
s
m is A/I-regular i. b
1
+ I, . . . , b
s
+ I m/I is A/I-regular, so
it is clear that depth
A/I
(A/I) = depth
A
(A/I). By Lemma 82, depth
A
(A/I) = depth
A
(A) r, as
required.
Proposition 84. Let A be a noetherian ring, M a nitely generated A-module and I a proper
ideal. Then
depth
I
(M) = inf{depthM
p
| p V (I)}
Proof. Let n denote the value of the right hand side. If n = 0 then depthM
p
= 0 for some p I
and then I p Ass(M), since pA
p
Ass(M
p
) implies p Ass(M). Thus depth
I
(M) = 0,
since there can be no M-regular sequences in p. If 0 < n < then I is not contained in any
associated prime of M, and so it is not contained in their union, which is the set of elements not
regular on M. Hence there exists a I regular on M. Moreover IM = M since otherwise we
would have (pA
p
)M
p
= M
p
and hence depthM
p
= for any p I, which would contradict the
fact that n < . Put M

= M/aM. Then for any p I with M

p
= 0 the element a/1 A
p
is an
M
p
-regular sequence in pA
p
, so
depthM

p
= depthM
p
/aM
p
= depthM
p
1
and depth
I
(M

) = depth
I
(M)1 by the Lemma 82. Therefore our assertion is proved by induction
on n.
If n = then M
p
= 0 for all p I. If IM = M then Supp(M/IM) is nonempty, since
Ass(M/IM) Supp(M/IM) and Ass(M/IM) = i. M/IM = 0. If p Supp(M/IM) =
Supp(M) V (I) then (M/IM)
p
= 0 and so M
p
/IM
p
= 0, which is a contradiction. Hence
IM = M and therefore depth
I
(M) = .
5.1 Cohen-Macaulay Rings
Denition 15. Let (A, m) be a noetherian local ring and M a nitely generated A-module. We
know that depthM dimM provided M is nonzero. We say that M is Cohen-Macaulay if M = 0
or if depthM = dimM. If the noetherian local ring A is Cohen-Macaulay as an A-module then
we call A a Cohen-Macaulay ring. So a noetherian local ring is Cohen-Macaulay if its dimension
is equal to the common length of the maximal A-regular sequences in m. The Cohen-Macaulay
property is stable under isomorphisms of modules and rings.
27
Example 4. Let A be a noetherian local ring. If dim(A) = 0 then A is Cohen-Macaulay, since
m is an associated prime of A and therefore no element of m is regular. If dim(A) = d 1 then
A is Cohen-Macaulay if and only if there is an A-regular sequence in m of length d.
Recall that for a module M over a noetherian ring A, the elements of Ass(M) which are
not minimal are called the embedded primes of M. Since a noetherian ring has descending chain
condition on prime ideals, every associated prime of M contains a minimal associated prime.
Theorem 85. Let (A, m) be a noetherian local ring and M a nitely generated A-module. Then
(i) If M is a Cohen-Macaulay module and p Ass(M), then we have depthM = dim(A/p).
Consequently M has no embedded primes.
(ii) If a
1
, . . . , a
r
is an M-regular sequence in m and M

= M/(a
1
, . . . , a
r
)M then M is Cohen-
Macaulay M

is Cohen-Macaulay.
(iii) If M is Cohen-Macaulay, then for every p Spec(A) the A
p
-module M
p
is Cohen-Macaulay
and if M
p
= 0 we have depth
p
(M) = depth
A
p
(M
p
).
Proof. (i) Since Ass(M) = , M is nonzero and so depthM = dimM. Since p Supp(M) we
have p Ann(M) and therefore dimM dim(A/p) and dim(A/p) depthM by Theorem 78. If
p Ass(M) were an embedded prime, there would be a minimal prime q Ass(M) with q p.
But since coht.p = coht.q are both nite this is impossible.
(ii) By Nakayama we have M = 0 i. M

= 0. Suppose M = 0. Then dimM

= dimMr by
Proposition 80 and depthM

= depthM r by Lemma 82.

(iii) We may assume that M
p
= 0. Hence p Ann(M). We know that dimM
p
depth
A
p
M
p

depth
p
(M). So we will prove depth
p
(M) = dimM
p
by induction on depth
p
(M). If depth
p
(M) = 0
then no element of p is regular on M, so by the usual argument p is contained in some p

Ass(M).
But Ann(M) p p

and the associated primes of M are the minimal primes over the ideal
Ann(M) by (i). Hence p = p

, and so p is a minimal element of Supp(M). The dimension of M

p
is the length of the longest chain in Supp(M
p
). If p
0
A
p
p
s
A
p
= pA
p
is a chain of length
s = dimM
p
then p
0
A
p
is minimal and therefore p
0
Ass(M). It follows that p
0
= p and so s = 0,
as required.
Now suppose depth
p
(M) > 0, take an M-regular element a p and set M
1
= M/aM. The
element a/1 A
p
is then M
p
-regular. Therefore we have
dim(M
1
)
p
= dimM
p
/aM
p
= dimM
p
1
and depth
p
(M
1
) = depth
p
(M) 1. Since M
1
is Cohen-Macaulay by (ii), by the inductive hypoth-
esis we have dim(M
1
)
p
= depth
p
(M
1
), which completes the proof.
Corollary 86. Let (A, m) be a noetherian local ring and a
1
, . . . , a
r
an A-regular sequence in
m. Let A

be the ring A/(a

1
, . . . , a
r
). Then A is a Cohen-Macaulay ring if and only if A

is a
Cohen-Macaulay ring.
Proof. Let I = (a
1
, . . . , a
r
). It suces to show that A

is a Cohen-Macaulay ring if and only if it

is a Cohen-Macaulay module over A. The dimension of A

as an A-module, the Krull dimension

of A

and the dimension of A

as a module over itself are all equal. So it suces to observe that

a sequence b
1
, . . . , b
s
m is an A

-regular sequence i. b
1
+ I, . . . , b
s
+ I m/I is an A

-regular
sequence, so depth
A
A

= depth
A
A

.
Corollary 87. Let A be a Cohen-Macaulay local ring and p a prime ideal. Then A
p
is a Cohen-
Macaulay local ring and ht.p = dimA
p
= depth
p
(A).
Proof. This all follows immediately from Theorem 85. In the statement, by dimA
p
we mean the
Krull dimension of the ring.
Lemma 88. Let A be a noetherian ring, I a proper ideal and a I a regular element. Then
ht.I/(a) = ht.I 1.
28
Proof. The minimal primes over the ideal I/(a) of A/(a) correspond to the minimal primes over I,
and we know from [Ash] Chapter 5, Corollary 5.4.8 that for any prime p containing a, ht.p/(a) =
ht.p 1, so the proof is straightforward.
Lemma 89. Let A be a Cohen-Macaulay local ring and I a proper ideal with ht.I = r 1. Then
we can choose a
1
, . . . , a
r
I in such a way that ht.(a
1
, . . . , a
i
) = i for 1 i r.
Proof. We claim that there exists a regular element a I. Otherwise, if every element of I was a
zero divisor on A, then I would be contained in the union of the nite number of primes in Ass(A),
and hence contained in some p Ass(M). By Theorem 85 (i) these primes are all minimal, so
I p implies ht.I = 0, a contradiction.
Now we proceed by induction on r. For r = 1 let a I be regular. It follows from Krulls
PID Theorem that ht.(a) = 1. Now assume r > 1 and let a I be regular. Then by Corollary 86
the ring A

= A/(a) is Cohen-Macaulay, and by Lemma 88, ht.I/(a) = r 1, so by the inductive

hypothesis there are a
1
, . . . , a
r1
I with ht.(a, a
1
, . . . , a
i
)/(a) = i for 1 i r 1. Hence
ht.(a, a
1
, . . . , a
i
) = i + 1
for 1 i r 1, as required.
Theorem 90. Let (A, m) be a Cohen-Macaulay local ring. Then
(i) For every proper ideal I of A we have
ht.I = depth
I
(A) = G(I)
ht.I +dim(A/I) = dimA
(ii) A is catenary.
(iii) For every sequence a
1
, . . . , a
r
m the following conditions are equivalent
(1) The sequence a
1
, . . . , a
r
is A-regular.
(2) ht.(a
1
, . . . , a
r
) = i for 1 i r.
(3) ht.(a
1
, . . . , a
r
) = r.
(4) There is a system of parameters of A containing {a
1
, . . . , a
r
}.
Proof. (iii) (1) (2) is immediate by Lemma 74. (2) (3) is trivial. (3) (4) If dimA = r 1
then (a
1
, . . . , a
r
) must be m-primary, so this is trivial. If dimA > r then m is not minimal over
(a
1
, . . . , a
r
), so we can take a
r+1
m which is not in any minimal prime ideal of (a
1
, . . . , a
r
).
Then by construction ht.(a
1
, . . . , a
r+1
) r +1, and therefore ht.(a
1
, . . . , a
r+1
) = r +1 by Krulls
Theorem. Continuing in this way we produce the desired system of parameters. Note that these
implications are true for any noetherian local ring. (4) (1) It suces to show that every system
of parameters x
1
, . . . , x
n
of a Cohen-Macaulay ring A is an A-regular sequence, which we do by
induction on n. Let I = (x
1
, . . . , x
n
) and put A

= A/(x
1
). If n = 1 and (x
1
) is m-primary then
it suces to show that x
1
is regular. If not, then x
1
p for some p Ass(A), which implies
that m = p is a minimal prime over 0 (since by Theorem 85 every prime of Ass(M) is minimal),
contradicting the fact that dimA = 1. Now assume n > 1. Since A is Cohen-Macaulay the
dimensions dim(A/p) for p Ass(A) all agree, and hence they are all equal to n = dimA. For
any p Ass(A) the ideal p +I is m-primary since
r(I + p) = r(r(I) +r(p)) = r(m + p) = m
Thus p +I/p is an m/p-primary ideal in the ring A/p, which has dimension n, so p +I/p cannot
be generated by fewer than n elements. This shows that x
1
/ p for any p Ass(A), and therefore
x
1
is A-regular. Put A

= A/(x
1
). By Corollary 86, A

is a Cohen-Macaulay ring, and it has

dimension n1 by Proposition 80. The images of x
2
, . . . , x
n
in A

form a system of parameters for

A

. Thus the residues x

2
+(x
1
), . . . , x
n
+(x
1
) form an A

-regular sequence (A

as an A

-module) by
29
the inductive hypothesis, and therefore x
2
, . . . , x
n
is an A

-regular sequence (A

as an A-module).
Hence x
1
, . . . , x
n
is an A-regular sequence, and we are done.
(i) Let I be a proper ideal of A. If ht.I = 0 then there is a prime p minimal over I with
ht.p = 0. Since p is minimal over 0, we have p Ass(A) and every element of I annihilates
some nonzero element of A. Therefore no A-regular sequence can exist in I, and G(I) = 0. Now
assume ht.I = r with r 1. Using Lemma 89 we produce a
1
, . . . , a
r
I with ht.(a
1
, . . . , a
i
) = i
for 1 i r. Then the sequence a
1
, . . . , a
r
is A-regular by (iii). Hence r G(I). Conversely,
if b
1
, . . . , b
s
is an A-regular sequence in I then ht.(b
1
, . . . , b
s
) = s ht.I by Lemma 74. Hence
ht.I = G(I).
We rst prove the second formula for prime ideals p. Put dimA = depthA = n and ht.p = r.
If r = 0 then dim(A/p) = depthA = n by Theorem 85 (i). If r 1 then since A
p
is a Cohen-
Macaulay local ring and ht.p = dimA
p
= depth
p
(A) we can nd an A-regular sequence a
1
, . . . , a
r
in p. Then A/(a
1
, . . . , a
r
) is a Cohen-Macaulay ring of dimension nr, and p is a minimal prime
of (a
1
, . . . , a
r
). Therefore dim(A/p) = n r by Theorem 85 (i), so the result is proved for prime
ideals. Now let I be an arbitrary proper ideal with ht.I = r. We have
dim(A/I) = sup{dim(A/p) | p V (I)}
= sup{dimAht.p | p V (I)}
There exists a prime ideal p minimal over I with ht.p = r, so it is clear that dim(A/I) = dimAr,
as required.
(ii) If q p are prime ideals of A, then since A
p
is Cohen-Macaulay we have dimA
p
=
ht.qA
p
+dimA
p
/qA
p
, i.e. ht.p ht.q = ht.(p/q). Therefore A is catenary.
Denition 16. We say a noetherian ring A is Cohen-Macaulay if A
p
is a Cohen-Macaulay local
ring for every prime ideal of A. A local noetherian ring is Cohen-Macaulay in this new sense i.
it is Cohen-Macaulay in the original sense. The Cohen-Macaulay property is stable under ring
isomorphism.
Lemma 91. Let A B be nonzero noetherian rings with B integral over A and suppose that B
is a at A-module. If A is Cohen-Macaulay then so is B.
Proof. Let q be a prime ideal of B and let p = q A. By Lemma 33, B
q
is at over A
p
and
so using Lemma 58 it follows that depth
B
q
(B
q
) depth
A
p
(A
p
) = dim(A
p
). By Theorem 56 we
have dim(B
q
) dim(A
p
), and hence depth
B
q
(B
q
) dim(B
q
), which shows that B
q
is Cohen-
Macaulay.
Denition 17. Let A be a noetherian ring and I a proper ideal, and let Ass
A
(A/I) = {p
1
, . . . , p
s
}
be the associated primes of I. We say that I is unmixed if ht.p
i
= ht.I for all i. In that case all the
p
i
are minimal, and A/I has no embedded primes. We say that the unmixedness theorem holds in
A if the following is true: for r 0 if I is a proper ideal of height r generated by r elements, then
I is unmixed. Note that such an ideal is unmixed if and only if A/I has no embedded primes, and
for r = 0 the condition means that A has no embedded primes.
Lemma 92. Let A be a noetherian ring. If the unmixedness theorem holds in A
m
for every
maximal ideal m, then the unmixedness theorem holds in A.
Proof. Let I be a proper ideal of height r generated by r elements with r 0, and let I = q
1
q
n
be a minimal primary decomposition with q
i
being p
i
-primary for 1 i n. Assume that one of
these associated primes, say p
1
, is an embedded prime of I, and let m be a maximal ideal containing
p
1
. Arrange the q
i
so that the primes p
1
, . . . , p
s
are contained in m whereas p
s+1
, , p
n
are not.
Then by [AM69] Proposition 4.9 the following is a minimal primary decomposition of the ideal
IA
m
IA
m
= q
1
A
m
q
s
A
m
So {p
1
A
m
, . . . , p
s
A
m
} are the associated primes of IA
m
. Since p
1
is embedded, there is some
1 i s with p
i
p
1
, and therefore p
i
A
m
p
1
A
m
. But this is a contradiction, since IA
m
has height r, is generated by r elements, and the unmixedness theorem holds in A
m
. So the
unmixedness theorem must hold in A.
30
Lemma 93. Let A be a noetherian ring and assume that the unmixedness theorem holds in A. If
a A is regular then the unmixedness theorem holds in A/(a).
Proof. Let I be a proper ideal of A containing a, and supppose the ideal I/(a) has height r and
is generated by r elements in A/(a). By Lemma 88 the ideal I has height r + 1 and is clearly
generated by r+1 elements in A. Therefore I is unmixed. If {p
1
, . . . , p
n
} are the associated primes
of I then the associated primes of I/(a) are {p
1
/(a), . . . , p
n
/(a)}. Since ht.p/(a) = ht.p 1 =
ht.I 1 = ht.I/(a) it follows that I/(a) is unmixed, as required.
Lemma 94. Let A be a noetherian ring and assume that the unmixedness theorem holds in
A. Then if I is a proper ideal with ht.I = r 1 we can choose a
1
, . . . , a
r
I such that
ht.(a
1
, . . . , a
i
) = i for 1 i r.
Proof. The proof is the same as Lemma 89 except we use the fact that 0 has no embedded primes
to show I contains a regular element, and we use Lemma 93.
Theorem 95. Let A be a noetherian ring. Then A is Cohen-Macaulay if and only if the unmixed-
ness theorem holds in A.
Proof. Suppose the unmixedness theorem holds in A and let p be a prime ideal of height r 0.
We know that r = dimA
p
depth(A
p
) depth
p
A by Lemma 72. If r = 0 then no regular
element can exist in p, so depth
p
A = 0 and consequently dimA
p
= 0 = depth(A
p
) so A
p
is Cohen-
Macaulay. If r 1 then by Lemma 94 we can nd a
1
, . . . , a
r
p such that ht.(a
1
, . . . , a
i
) = i for
1 i r. The ideal (a
1
, . . . , a
i
) is unmixed by assumption, so a
i+1
lies in no associated primes
of A/(a
1
, . . . , a
i
). Thus a
1
, . . . , a
r
is an A-regular sequence in p, so depth
p
A r and consequently
dimA
p
= r = depth(A
p
), so again A
p
is Cohen-Macaulay. Hence A is a Cohen-Macaulay ring.
Conversely, suppose A is Cohen-Macaulay. It suces to show that the unmixedness theorem
holds in A
m
for all maximal m, so we can reduce to the case where A is a Cohen-Macaulay local
ring. We know from Theorem 85 that 0 is unmixed. Let (a
1
, . . . , a
r
) be an ideal of height r > 0.
Then a
1
, . . . , a
r
is an A-regular sequence by Theorem 90, hence A/(a
1
, . . . , a
r
) is Cohen-Macaulay
and so (a
1
, . . . , a
r
) is unmixed.
Corollary 96. A noetherian ring A is Cohen-Macaualy if and only if A
m
is a Cohen-Macaulay
local ring for every maximal ideal m.
Proof. This follows immediately from Theorem 95 and Lemma 92.
Corollary 97. Let A be a Cohen-Macaulay ring. If a
1
, . . . , a
r
A are such that ht.(a
1
, . . . , a
i
) = i
for 1 i r then a
1
, . . . , a
r
is an A-regular sequence.
Theorem 98. Let A be a Cohen-Macaulay ring. Then the polynomial ring A[x
1
, . . . , x
n
] is also
Cohen-Macaulay. Hence any Cohen-Macaulay ring is universally catenary.
Proof. It is enough to consider the case n = 1. Let q be a prime ideal of B = A[x] and put
p = q A. We have to show that B
q
is Cohen-Macaulay. It follows from Lemma 10 that B
q
is
isomorphic to A
p
[x]
qA
p
[x]
where qA
p
[x] is a prime ideal of A
p
[x] contracting to pA
p
. Since A
p
is Cohen-Macaulay we can reduce to showing B
q
is Cohen-Macaulay in the case where A is a
Cohen-Macaulay local ring and p = q A is the maximal ideal. Then B/pB

= k[x] where k is a
eld. Therefore we have either q = pB or q = pB+fB where f B = A[x] is a monic polynomial
of positive degree. By Theorem 55 we have (Krull dimensions)
dim(B
q
) = dim(A) +ht.(q/pB)
If q = pB then this implies that dim(B
q
) = dim(A). So to show B
q
is Cohen-Macaulay it suces
to show that depth
B
q
(B
q
) dimA. If dimA = 0 this is trivial, so assume dimA = r 1 and let
a
1
, . . . , a
r
be an A-regular sequence. As B is at over A, so is B
q
, and therefore a
1
, . . . , a
r
is also
a B
q
-regular sequence by Lemma 58. It is then not dicult to check that the images of the a
i
in
B
q
form a B
q
-regular sequence, so depth
B
q
(B
q
) r, as required.
31
If q = pB + fB then dim(B
q
) = dim(A) + 1 (since every nonzero prime in k[x] has height
1), and so it suces to show that depth
B
q
(B
q
) dim(A) + 1. If dimA = 0 then since f is
monic it is clearly regular in B and therefore also in B
q
, which shows that depth
B
q
(B
q
) 1. If
dimA = r 1 let a
1
, . . . , a
r
be an A-regular sequence. Since f is monic it follows that f is regular
on B/(a
1
, . . . , a
r
)B. Therefore a
1
, . . . , a
r
, f is a B-regular sequence. Applying Lemma 58 we see
that this sequence is also B
q
-regular, and therefore the images in B
q
form a B
q
-regular sequence.
This shows that depth
B
q
(B
q
) r + 1, as required.
It follows from Lemma 8 and Theorem 90 that any Cohen-Macaulay ring is catenary. Therefore
if A is Cohen-Macaulay, A[x
1
, . . . , x
n
] is catenary for n 1, and so any Cohen-Macaulay ring is
universally catenary.
Corollary 99. If k is a eld then k[x
1
, . . . , x
n
] is Cohen-Macaulay and therefore universally
catenary for n 1.
6 Normal and Regular Rings
6.1 Classical Theory
Denition 18. We say that an integral domain A is normal if it is integrally closed in its quotient
eld. The property of being normal is stable under ring isomorphism. If an integral domain A is
normal, then so is S
1
A for any multiplicatively closed subset S of A not containing zero.
Proposition 100. Let A be an integral domain. Then the following are equivalent:
(i) A is normal;
(ii) A
p
is normal for every prime ideal p;
(iii) A
m
is normal for every maximal ideal m.
Proof. See [AM69] Proposition 5.13.
Denition 19. Let A be an integral domain with quotient eld K. An element u K is almost
integral over A if there exists a nonzero element a A such that au
n
A for all n > 0.
Lemma 101. If u K is integral over A then it is almost integral over A. The elements of K
almost integral over A form a subring of K containing the integral closure of A. If A is noetherian
then u K is integral if and only if it is almost integral.
Proof. It is clear that any element of A is almost integral over A. Let u = b/t K with b, t A
nonzero be integral over A, and let
u
n
+a
1
u
n1
+ +a
1
u +a
0
= 0
be an equation of integral dependence. We claim that t
n
u
m
A for any m > 0. If m n this is
trivial, and if m > n then we can write u
m
as an A-linear combination of strictly smaller powers
of u, so t
n
u
m
A in this case as well. It is easy to check that the almost integral elements form
a subring of K.
Now assume that A is noetherian, and let u be almost integral over A. If a is nonzero and
au
n
A for n 1 then A[u] is a submodule of the nitely generated A-module a
1
A, whence
A[u] itself is nitely generated over A and so u is integral over A.
Denition 20. We say that an integral domain A is completely normal if every element u K
which is almost integral over A belongs to A. Clearly a completely normal domain is normal, and
for a noetherian ring domain normality and complete normality coincide. The property of being
completely normal is stable under ring isomorphism.
Example 5. Any eld is completely normal, and if k is a eld then the domain k[x
1
, . . . , x
n
] is
completely normal, since it is noetherian and normal.
32
Denition 21. We say that a ring A is normal if A
p
is a normal domain for every prime ideal p.
An integral domain is normal in this new sense i. it is normal in the original sense. The property
of being normal is stable under ring isomorphism.
Lemma 102. Let A be a ring and suppose that A
p
is a domain for every prime ideal p. Then A
is reduced. In particular a normal ring is reduced.
Proof. Let a A be nilpotent. For any prime ideal p we have a/1 = 0 in A
p
so ta = 0 for some
t / p. Hence Ann(a) cannot be a proper ideal, and so a = 0.
Lemma 103. Let A
1
, . . . , A
n
be normal domains. Then A
1
A
n
is a normal ring.
Proof. Let A = A
1
A
n
. A prime ideal p of A is A
1
p
i
A
n
for some 1 i n
and prime ideal p
i
of A
i
. Moreover A
p

= (A
i
)
p
i
, which by assumption is a normal domain. Hence
A is a normal ring.
Proposition 104. Let A be a completely normal domain. Then a polynomial ring A[x
1
, . . . , x
n
]
is also completely normal. In particular k[x
1
, . . . , x
n
] is completely normal for any eld k.
Proof. It is enough to treat the case n = 1. Let K denote the quotient eld of A. Then the
canonical injective ring morphism A[x] K[x] induces an isomorphism between the quotient
eld of A[x] and K(x), the quotient eld of K[x], so we consider all our rings as subrings of K(x).
Let 0 = u K(x) be almost integral over A[x]. Since A[x] K[x] and K[x] is completely normal,
the element u must belong to K[x]. Write
u =
r
x
r
+
r+1
x
r+1
+ +
d
x
d
for some r 0 and
r
= 0. Let f(x) = b
s
x
s
+b
s+1
x
s+1
+ +b
t
x
t
A[x] with b
s
= 0 be such that
fu
n
A[x] for all n > 0. Then b
s

n
r
A for all n so that
r
A. Then u
r
x
r
=
r+1
x
r+1
+
is almost integral over A[x], so we get
r+1
A as before, and so on. Therefore u A[x].
Proposition 105. Let A be a normal ring. Then A[x
1
, . . . , x
n
] is normal.
Proof. It suces to consider the case n = 1. Let q be a prime ideal of A[x] and let p = q A.
Then A[x]
q
is a localisation of A
p
[x] at a prime ideal, and A
p
is a normal domain. So we reduce
to the case where A is a normal domain with quotient eld K. As before we identify the quotient
eld of A[x] with K(x), the quotient eld of K[x]. We have to prove that A[x] is integrally closed
in K(x). Let u = p(x)/q(x) with p, q A[x] be a nonzero element of K(x) which is integral over
A[x]. Let
u
d
+f
1
(x)u
d1
+ +f
d
(x) = 0 f
i
A[x]
be an equation of integral dependence. In order to prove that u A[x], consider the subring A
0
of A generated by 1 and the coecients of p, q and all the f
i
. Identify A
0
, A
0
[x] and the quotient
eld of A
0
[x] with subrings of K(x). Then u is integral over A
0
[x]. The proof of Proposition 104
shows that u belongs to K[x], and moreover
u =
r
x
r
+ +
d
x
d
where each coecient
i
K is almost integral over A
0
. As A
0
is noetherian,
i
is integral over
A
0
and therefore integral over A. Therefore
i
A, which is what we wanted.
Let A be a ring and I an ideal with

n=1
I
n
= 0. Then for each nonzero a A there is
an integer n 0 such that a I
n
and a / I
n+1
. We then write n = ord(a) (or ord
I
(a))
and call it the order of a with respect to I. We have ord(a + b) min{ord(a), ord(b)} and
ord(ab) ord(a) + ord(b). Put A

= gr
I
(A) =

n0
I
n
/I
n+1
. For an element a of A with
ord(a) = n, we call the sequence in A

with a single a in I
n
/I
n+1
the leading form of a and
denote it by a

. Clearly a

= 0. We dene 0

= 0. The map a a

is in general not additive

or multiplicative, but for nonzero a, b if a

= 0 (i.e. if ord(ab) = ord(a) +ord(b)) then we have

(ab)

= a

and if ord(a) = ord(b) and a

+b

= 0 then we have (a +b)

= a

+b

.
33
Theorem 106 (Krull). Let A be a nonzero ring, I an ideal and gr
I
(A) the associated graded
ring. Then
(1) If

n=1
I
n
= 0 and gr
I
(A) is a domain, so is A.
(2) Suppose that A is noetherian and that I is contained in the Jacobson radical of A. Then if
gr
I
(A) is a normal domain, so is A.
Proof. We denote the ring gr
I
(A) by A

for convenience. (1) Let a, b be nonzero elements of A.

Then a

= 0 and b

= 0, hence a

= 0 and therefore ab = 0.
(2) Since I is contained in the Jacobson radical it is immediate that

n=1
I
n
= 0 (see [AM69]
Corollary 10.19) and so by (1) the ring A is a domain. Let K be the quotient eld of A and
suppose we are given nonzero a, b A with a/b integral over A. We have to prove that a bA.
The A-module A/bA is separated in the I-adic topology by Corollary 10.19 of A & M. In other
words
bA =

n=1
(bA+I
n
)
Therefore it suces to prove the following for every n 1:
() For nonzero a, b A with a/b integral over A, if a bA+I
n1
then a bA+I
n
.
Suppose that a bA + I
n1
for some n 1. Then a = br + a

with r A and a

I
n1
, and
a

/b = a/b r is integral over A. If a

= 0 then a = br and we are done. Otherwise we can reduce

to proving () in the case where a I
n1
.
So we are given an integer n 1, nonzero a, b with a I
n1
and a/b integral over A, and we
have to show that a bA + I
n
. Since a/b is almost integral over A there exists nonzero c A
such that ca
m
b
m
A for all m > 0. Since A

is a domain the map a a

is multiplicative, hence
we have c

(a

)
m
(b

)
m
A

for all m, and since A

is noetherian (see Proposition 10.22 of A &

M) and normal we have a

. Therefore we can nd d A with a

= b

. If a I
n
then
we would be done, so suppose a / I
n
and therefore ord(a) = n 1. Since a

= b

the residue
of a bd in I
n1
/I
n
is zero, and therefore a bd I
n
. Hence a bA+I
n
, as required.
Denition 22. Let (A, m, k) be a noetherian local ring of dimension d. Recall that the ring A
is said to be regular if m can be generated by d elements, or equivalently if rank
k
m/m
2
= d.
Regularity is stable under ring isomorphism.
Recall that if k is a eld, a graded k-algebra is a k-algebra R which is also a graded ring in
such a way that the graded pieces R
d
are k-submodules for every d 0. A morphism of graded
k-algebras is a morphism of graded rings which is also a morphism of k-modules.
Theorem 107. Let (A, m, k) be a noetherian local ring of dimension d. Then A is regular if
and only if the graded ring gr
m
(A) =

m
n
/m
n+1
is isomorphic as a graded k-algebra to the
polynomial ring k[x
1
, . . . , x
d
].
Proof. The rst summand in gr
m
(A) is the eld k = A/m, so this ring becomes a graded k-algebra
in a canonical way. For d = 0 we interpret the statement as saying A is regular i. gr
m
(A) is
isomorphic as a graded k-algebra to k itself. See the section in [AM69] on regular local rings for
the proof.
Theorem 108. Let A be a regular local ring of dimension d. Then
(1) A is a normal domain.
(2) A is a Cohen-Macaulay local ring.
If d 1 and {a
1
, . . . , a
d
} is a regular system of parameters, then
(3) a
1
, . . . , a
d
is an A-regular sequence.
34
(4) p
i
= (a
1
, . . . , a
i
) is a prime ideal of height i for each 1 i d and A/p
i
is a regular local
ring of dimension d i.
(5) Conversely, if I is a proper ideal of A such that A/I is regular and has dimension d i
for some 1 i d, then there exists a regular system of parameters {y
1
, . . . , y
d
} such that
I = (y
1
, . . . , y
i
). In particular I is prime.
Proof. (1) Follows from Theorems 106 and 107.
(2) If d = 0 this is trivial, and if d 1 this follows from (3) below.
(3) From the proof of [AM69] Theorem 11.22 we know that there is an isomorphism of graded k-
algebras : k[x
1
, . . . , x
d
] gr
m
(A) dened by x
i
a
i
m/m
2
. If f(x
1
, . . . , x
d
) is homogenous
of degree m 0 then (f) is the element

1
1
a

n
n
f() of m
m
/m
m+1
. So agrees with the
morphism of abelian groups dened in Proposition 63 (c). Thus a
1
, . . . , a
d
is an A-quasiregular
sequence. It then follows from Corollary 65 that a
1
, . . . , a
d
is an A-regular seqence.
(4) We have dim(A/p
i
) = d i for 1 i d by Proposition 51, and hence ht.p
i
= i by (2)
and Theorem 90 (i). The ring A/p
d
is a eld, and therefore trivially a regular local ring of the
correct dimension. If i < d then the maximal ideal m/p
i
of A/p
i
is generated by d i elements
x
i+1
, . . . , x
d
. Therefore A/p
i
is regular, and hence p
i
is prime by (1).
(5) Let A = A/I and put m = m/I. Then we can identify k with A/m and there is clearly an
isomorphism of k-modules
m
2
/(m
2
+I)

= m/m
2
So we have
d i = rank
k
m/m
2
= rank
k
m/(m
2
+I)
Since I m the A-module (m
2
+ I)/m
2
is canonically a k-module, and we have a short exact
sequence of k-modules
0 (m
2
+I)/m
2
m/m
2
m/(m
2
+I) 0
Consequently d i = rank
k
m/m
2
rank
k
(m
2
+ I)/m
2
, and therefore rank
k
(m
2
+ I)/m
2
= i.
Thus we can choose i elements y
1
, . . . , y
i
of I which span m
2
+ I mod m
2
over k, and d i
elements y
i+1
, . . . , y
d
of m which, together with y
1
, . . . , y
i
, span m mod m
2
over k (if i = d then
the original y
1
, . . . , y
i
will do). Then {y
1
, . . . , y
d
} is a regular system of parameters of A, so that
(y
1
, . . . , y
i
) = p is a prime ideal of height i by (4). Since p I and dim(A/I) = dim(A/p) = di,
we must have I = p.
Let A be an integral domain with quotient eld K. A fractional ideal is an A-submodule of K.
If M, N are two fractional ideals then so is M N = {

m
i
n
i
| m
i
M, n
i
N}. This product is
associative, commutative and M A = M for any fractional ideal M. For any nonzero ideal a of A
we put a
1
= {x K| xa A}. Then a
1
is a fractional ideal and we have A a
1
. Moreover
a a
1
A is an ideal of A.
Lemma 109. Let A be a noetherian domain with quotient eld K. Let a be a nonzero element
of A and p Ass
A
(A/(a)). Then p
1
= A.
Proof. By denition of associated primes there is b / (a) with p = ((a) : b). Then (b/a)p A and
b/a / A.
Lemma 110. Let (A, m) be a noetherian local domain such that m = 0 and mm
1
= A. Then m
is a principal ideal, and so A is regular of dimension 1.
Proof. By assumption we have dimA 1. By [AM69] Proposition 8.6 it follows that m = m
2
.
Take a mm
2
. Then am
1
A, and if am
1
m then (a) = am
1
m m
2
, contradicting the
choice of a. Since am
1
is an ideal we must have am
1
= A, that is, (a) = am
1
m = m. Using
the dimension theory of noetherian local rings we see that dimA 1 and therefore A is regular
of dimension 1.
Theorem 111. Let (A, m) be a noetherian local ring of dimension 1. Then A is regular i. it is
normal.
35
Proof. If A is regular then it is a normal domain by Theorem 108. Now suppose that A is normal
(hence a domain since A

= A
m
). By Lemma 110 to show A is regular it suces to show that
mm
1
= A. Assume the contrary. Then mm
1
is a proper ideal, and since 1 m
1
we have
m mm
1
, hence mm
1
= m. Let a
1
, . . . , a
n
be generators for m (since dimA 1 we can assume
all a
i
= 0) and let a m
1
. Since aa
i
m for all i, we have coecients r
ij
A, 1 i, j n and
equations aa
i
= r
i1
a
1
+ +r
in
a
n
. Collecting terms we have:
0 = (r
11
a)a
1
+ +r
1n
a
n
0 = r
21
a
1
+ (r
21
a)a
2
+ +r
2n
a
n
.
.
.
0 = r
n1
a
1
+ + (r
nn
a)a
n
The determinant of the coecient matrix B = (r
ij
ij a) must satisfy detB a
i
= 0 and thus
detB = 0 since A is a domain. This gives an equation of integral dependence of a over A, whence
m
1
= A since A is integrally closed. But since dimA = 1 we have m Ass(A/(b)) for any
nonzero b m so that m
1
= A by Lemma 109. Thus mm
1
= A cannot occur.
Theorem 112. Let A be a noetherian normal domain. Then as subrings of the quotient eld K
of A we have
A =

htp=1
A
p
Moreover any nonzero proper principal ideal in A is unmixed, and if dim(A) 2 then A is
Cohen-Macaulay.
Proof. Suppose 0 = a is a nonunit of A and p Ass(A/(a)). We claim that htp = 1. Replacing
A by A
p
we may assume that A is local with maximal ideal p (since pA
p
= ((a/1) : (b/1)). Then
we have p
1
= A by Lemma 109. If htp > 1 then pp
1
= A, since otherwise we can run the proof
of Theorem 111 and obtain a contradiction (in that proof we only use dimA = 1 to show that
m = 0 and that m Ass(A/(b)) for some nonzero b m). But then Lemma 110 implies that A is
regular of dimension 1, contradicting the fact that htp > 1. Hence htp = 1, which shows that the
ideal (a) is unmixed.
Now suppose x A
p
for all primes of height 1 and write x = a/b. We need to show that x A,
so we can assume that b is not a unit and a / (b). The ideal ((b) : a) is the annihilator of the
nonzero element a +(b) of A/(b). The set of annihilators of nonzero elements of A/(b) containing
((b) : a) has a maximal element since A is noetherian, and by Lemma 47 this maximal element
is a prime ideal p = ((b) : h) for some h / (b). By denition p Ass(A/(b)) and thus htp = 1.
Since a/b A
p
we have a/b = c/s for some s / p. Then sa = bc (b) so s ((b) : a) p, which
is a contradiction. Hence we must have had a (b) and thus x A to begin with.
Now suppose that A is a noetherian normal domain with dim(A) 2. By Theorem 95 it
is enough to show that the unmixedness theorem holds in A. Since A is a domain it is clear
that 0 has no embedded primes, and we have just shown that every proper principal ideal of
height 1 is unmixed. If I = (a
1
, a
2
) is a proper ideal of height 2, then every associated prime
p of I has ht.p 2, but also ht.p 2 since dim(A) = 2. Therefore I is unmixed and A is
Cohen-Macaulay.
Denition 23. Let A be a nonzero noetherian ring. Consider the following conditions about A
for k 0:
(S
k
) For every prime p of A we have depth(A
p
) inf{k, ht.p}.
(R
k
) For every prime p of A, if ht.p k then A
p
is regular.
The condition (S
0
) is trivial, and for every k 1 we have (S
k
) (S
k1
) and (R
k
) (R
k1
).
For a nonzero noetherian ring A we can express (S
k
) dierently as follows: for every prime
p, if ht.p k then depth(A
p
) ht.p and otherwise depth(A
p
) k. We introduce the following
auxiliary condition for k 1
36
(T
k
) For every prime p of A, if ht.p k then depth(A
p
) k.
It is not hard to see that for k 1, the condition (S
k
) is equivalent to (T
i
) being satised for all
1 i k.
Proposition 113. Let A be a nonzero noetherian ring. Then
(S
1
) Ass(A) has no embedded primes every prime p with ht.p 1 contains a
regular element.
(S
2
) (S
1
) and Ass(A/fA) has no embedded primes for any regular nonunit f A.
The ring A is Cohen-Macaulay i it satises (S
k
) for all k 0.
Proof. For a noetherian ring A with prime ideal p, we have depth(A
p
) = 0 i. pA
p
Ass(A
p
)
which by [Ash] Chapter 1, Lemma 1.4.2 is i. p Ass(A). So the associated primes are precisely
those with depth(A
p
) = 0. A prime p Ass(A) is embedded i. ht.p 1, so saying that
Ass(A) has no embedded primes is equivalent to saying that if p Spec(A) and ht.p 1 then
depth(A
p
) 1. Hence the rst two statements are equivalent. If Ass(A) has no embedded primes
and ht.p 1 then p must contain a regular element, since otherwise by [Ash] Chapter 1, Theorem
1.3.6, p is contained in an associated prime of A, and these all have height zero. Conversely, if
every prime of height 1 contains a regular element, then certainly no prime of height 1 can
be an associated prime of A, so Ass(A) has no embedded primes.
To prove the second statement, we assume A is a nonzero noetherian ring satisfying (S
1
), and
show that (S
2
) is equivalent to Ass(A/fA) having no embedded primes. Suppose A satises (S
2
)
and let a regular nonunit f be given. If p Ass(A/fA) then the following Lemma implies that
ht.p depth(A
p
) = 1, and p is a minimal prime i. ht.p = 1. So the condition (T
2
) shows that
Ass(A/fA) can have no embedded primes. Conversely, suppose p is a prime ideal with ht.p 2
not satisfying (T
2
). Since A has no embedded primes, this can only happen if depth(A
p
) = 1. But
then by the following Lemma, p Ass(A/fA) for some regular f A. Since ht.p 2, this is an
embedded prime, which is impossible.
If A is Cohen-Macaulay then ht.p = depth(A
p
) for every prime p, so clearly (S
k
) is satised
for k 0. Conversely if A satises every (S
k
) then by choosing k large enough we see that
depth(A
p
) ht.p for every prime p, and hence A is Cohen-Macaulay.
Lemma 114. Let A be a nonzero noetherian ring satisfying (S
1
). Then for a prime p the following
are equivalent
(i) depth(A
p
) = 1;
(ii) There exists a regular element f p with p Ass(A/fA).
If f p is regular and p Ass(A/fA) then p is a minimal prime of Ass(A/fA) if and only if
ht.p = 1.
Proof. Let f p be a regular element. Then f/1 A
p
is regular, and it is not hard to see there
is an isomorphism of A
p
-modules A
p
/fA
p

= (A/fA)
p
. Note also that
depth(A
p
/fA
p
) = depth(A
p
) 1 (4)
(i) (ii) Since ht.p = dim(A
p
) depth(A
p
) we have ht.p 1, and therefore since A satises (S
1
)
there is a regular element f p. The above shows that depth(A
p
/fA
p
) = depth((A/fA)
p
) = 0
and therefore by Lemma 71, p Ass(A/fA), as required. (ii) (i) follows from Lemma 71 and
(4). If (i) is satised, then the above proof shows that p is an associated prime of A/fA for any
regular f p.
Suppose p is a minimal prime of Ass(A/fA). Then by (i), depth(A
p
) = 1, and since p is
a minimal prime over fA it follows from Krulls PID Theorem that ht.p = 1. Conversely if
depth(A
p
) = ht.p = 1 then clearly p is minimal over fA.
37
Proposition 115. Let A be a nonzero noetherian ring. Then A is reduced i it satises (R
0
)
and (S
1
).
Proof. Suppose that A is reduced. Then Lemma 13 shows that A satises (R
0
). Suppose that A
does not satisfy (S
1
). Let p be an associated prime of A which is not minimal: so ht.p 1 and
p = Ann(b) for some nonzero b A. Then A
p
is a reduced noetherian ring in which every element
is either a unit or a zero-divisor, so by Lemma 12 we must have dim(A) = 0, which contradicts
the fact that ht.p 1. Therefore A must satisfy (S
1
).
Now suppose that A satises (R
0
) and (S
1
). Let a A be nonzero and nilpotent. By Lemma
47 there is an associated prime p Ass(A) with Ann(a) p. By (S
1
) we have ht.p = 0 and
therefore A
p
is a eld by (R
0
). Since a/1 A
p
is nilpotent we have ta = 0 for some t / p, which
is a contradiction. Hence A is reduced.
If A is a nonzero ring, the set S of all regular elements is a multiplicatively closed subset. Let
A denote the localisation S
1
A, which we call the total quotient ring of A. If A is a domain,
this is clearly the quotient eld.
Theorem 116 (Criterion of Normality). A nonzero noetherian ring A is normal if and only
if it satises (S
2
) and (R
1
).
Proof. Let A be a nonzero noetherian ring. Suppose rst that A is normal, and let p be a prime
ideal. Then A
p
is a eld for ht.p = 0 and regular for ht.p = 1 by Theorem 111, hence the condition
(R
1
) is satised. Since A is normal it is reduced, so it satises (S
1
) by Proposition 115. To show
A satises (S
2
) it suces by Proposition 113 to show that Ass(A/fA) has no embedded primes
for any regular nonunit f. Let f be a regular nonunit with associated primes
Ass(A/fA) = {p
1
, . . . , p
n
}
Suppose wlog that p = p
1
is an embedded prime, and that p
1
, . . . , p
i
are the associated primes
contained in p
1
. Since A
p
/fA
p

= (A/fA)
p
we have
Ass
A
p
(A
p
/fA
p
) = {pA
p
, p
2
A
p
, . . . , p
i
A
p
}
by [Ash] Chapter 1, Lemma 1.4.2. At least one of the p
i
is properly contained in p, so pA
p
is an
embedded prime of Ass(A
p
/fA
p
). But since A
p
is a noetherian normal domain, this contradicts
Theorem 112. Hence A satises (S
2
).
Next, suppose that A satises (S
2
) and (R
1
). Then it also satises (R
0
) and (S
1
), so it is
reduced. Let p
1
, . . . , p
r
be the minimal prime ideals of A. Then we have 0 = p
1
p
r
. Let
S be the set of all regular elements in A. Then by Proposition 113 the p
i
are precisely the prime
ideals of A avoiding S. Therefore A = S
1
A is an artinian ring, and since (S
1
A)
S
1
p
i

= A
p
i
,
Proposition 14 gives an isomorphism of rings
: A
s

j=1
A
q
j
, a/s (a/s, . . . , a/s)
where for each i, q
i
{p
i
, . . . , p
r
} (some of the p
i
may more than once or not at all among the
q
j
). Also note that by Lemma 13 for each j the ring K
j
= A
q
j
is a eld. For each j let T
j
be image of the ring morphism A K
j
. Then taking the product gives a subring

j
T
j
of

j
K
j
which contains the image of A under . Let e
1
, . . . , e
s
be the preimage in A of the tuples
(1, 0, . . . , 0), . . . , (0, . . . , 0, 1). These clearly form a family of orthogonal idempotents in A.
Suppose that we could show that A was integrally closed in A. For each j the element e
j
satises e
2
j
e
j
= 0, so e
j
A. We claim that identies the subrings A and

j
T
j
. It is enough to
show that maps the former subring onto the latter. If a
1
, . . . , a
s
A give a tuple (a
1
/1, . . . , a
s
/1)
of

j
T
j
, then since e
j
A we have e
1
a
1
+ + e
s
a
s
A, and since (e
1
) = (1, 0, . . . , 0) and
similarly for the other e
j
, it is clear that
(e
1
a
1
+ +e
s
a
s
) = (a
1
/1, . . . , a
s
/1)
38
as required. Since A is integrally closed in A it is straightforward to check that each T
j
is
integrally closed in K
j
, and is therefore a normal domain. Hence A is isomorphic to a direct
product of normal domains, so A is a normal ring by Lemma 103.
So it only remains to show that A is integrally closed in A. Suppose we have an equation of
integral dependence in A
(a/b)
n
+c
1
(a/b)
n1
+ +c
n
= 0
where a, b and the c
i
are elements of A and b is A-regular. Then a
n
+

n
i=1
c
i
a
ni
b
i
= 0. We
want to prove that a bA, so we may assume b is a regular nonunit of A. To show that a bA
it suces to show that a
p
b
p
A
p
for every associated prime p of bA (here a
p
denotes a/1 A
p
).
Since bA is unmixed of height 1 by (S
2
), it suces to prove this for primes p with ht.p = 1. By
(R
1
) if ht.p = 1 then A
p
is regular and therefore by Theorem 108 a normal domain. But in the
quotient eld of A
p
we have
a
n
p
+
n

i=1
(c
i
)
p
a
ni
p
b
i
p
= 0
If b
p
= 0 then clearly a
p
= 0. Otherwise a
p
/b
p
is integral over A
p
, and so a
p
b
p
A
p
, as
required.
Corollary 117. A nonzero normal noetherian ring A is isomorphic to a nite direct product of
normal domains.
Theorem 118. Let A be a ring such that for every prime ideal p the localisation A
p
is regular.
Then the polynomial ring A[x
1
, . . . , x
n
] over A has the same property.
Proof. As in the proof of Theorem 98 we reduce to the case where (A, p) is a regular local ring,
n = 1 and q is a prime ideal of B = A[x] lying over p. And we have to prove that B
q
is regular.
We have q pB and B/pB

= k[x] where k is a eld. Therefore either q = pB or q = pB + fB
where f B = A[x] is a monic polynomial of positive degree. Put dim(A) = d 0. Then p is
generated by d elements, so if q = pB then q is generated by d elements, and by d + 1 elements
if q = pB + fB. From [Ash] Chapter 5 we know that ht.pB = ht.p = d (use Propositions 5.6.3
and 5.4.3). On the other hand if q = pB +fB then by Krulls Theorem ht.q d +1, and since q
contains p properly, we must have ht.q = d + 1. This shows that B
q
is regular.
Corollary 119. If k is a eld then k[x
1
, . . . , x
n
]
p
is a regular local ring for every prime ideal p
of k[x
1
, . . . , x
n
].
6.2 Homological Theory
The following results are proved in our Dimension notes.
Proposition 120. Let A be a ring, M an A-module. Then
(i) M is projective i. Ext
1
A
(M, N) = 0 for all A-modules N.
(ii) M is injective i. Ext
1
A
(N, M) = 0 for all A-modules N.
(iii) M is injective i. Ext
1
A
(A/I, M) = 0 for all ideals I of A.
(iv) M is at i. Tor
A
1
(A/I, M) = 0 for all nitely generated ideals I.
(v) M is at i. Tor
A
1
(N, M) = 0 for all nitely generated A-modules N.
So injectivity is characterised by vanishing of Ext
1
A
(, M), and we can restrict consideration
to ideal quotients in the rst variable. Flatness is characterised by vanishing of Tor
1
A
(, M) (or
equivalently, Tor
1
A
(M, )) and we can restrict consideration to ideal quotients or nitely generated
modules. The next result shows that the projectivity condition can also be restricted to a special
class of modules:
39
Lemma 121. Let A be a noetherian ring and M a nitely generated A-module. Then M is
projective if and only if Ext
1
A
(M, N) = 0 for all nitely generated A-modules N.
Proof. Take an exact sequence 0 R F M 0 with F nitely generated and
free. Then R is nitely generated, so by assumption Ext
1
A
(M, R) = 0. Thus the sequence
Hom(F, R) Hom(R, R) 0 is exact. It follows that R F is a coretraction, so that M
is a direct summand of a free module.
If A is a nonzero ring, then the global dimension of A, denoted gl.dim(A), is the largest
integer n 0 for which there exists modules M, N with Ext
n
A
(M, N) = 0. The Tor dimension
of A, denoted tor.dim(A), is the largest integer n 0 for which there exists modules M, N with
Tor
A
n
(M, N) = 0. We know from our Dimension notes that
gl.dim(A) = sup{proj.dim.M| M AMod}
= sup{inj.dim.M| M AMod}
= sup{proj.dim.A/I | I a left ideal of A}
and
tor.dim(A) = sup{flat.dim.M| M AMod}
= sup{flat.dim.A/I | I is a left ideal of A}
Proposition 122. Let A be a noetherian ring. Then tor.dim(A) = gl.dim(A) and for every
nitely generated A-module M, flat.dim.M = proj.dim.M.
Proof. See our Dimension notes.
Lemma 123. Let (A, m, k) be a noetherian local ring, and let M be a nitely generated A-module.
Then for n 0
proj.dim.M n Tor
A
n+1
(M, k) = 0
In particular, if M is nonzero then proj.dim.M is the largest n 0 such that Tor
A
n
(M, k) = 0.
Proof. This is trivial if M = 0, so assume M is nonzero. Since flat.dim.M proj.dim.M the
implication is clear. We prove the converse by induction on n. Let m = rank
k
(M/mM). Then
m 1 since M is nonzero, and by Nakayama we can nd elements {u
1
, . . . , u
m
} which generate M
as an A-module and map to a k-basis in M/mM. Let : A
m
M be induced by the elements
u
i
, and let K be the kernel of , which is nitely generated since A is noetherian. So we have an
exact sequence
0 K A
m
M 0
It follows that proj.dim.M proj.dim.K + 1. If n > 0 then using the long exact Tor sequence
we see that Tor
A
n+1
(M, k)

= Tor
A
n
(K, k), which proves the inductive step. So it only remains to
consider the case n = 0. Then by assumption Tor
A
1
(M, k) = 0 so the top row in the following
commutative diagram of A-modules is exact
0

K
A
k

A
m

A
k
1

M
A
k

0
0

K/mK

k
m
M/mM

0
By construction k
m
M/mM is the morphism of k-modules corresponding to the basis dened
by the u
i
, so it is an isomorphism. Hence K/mK = 0, so K = 0 by Nakayamas Lemma. Hence
M

= A
m
and so proj.dim.M = 0.
40
Remark 2. Let A be a ring and M an A-module. By localising any nite projective resolution
of M, we deduce that proj.dim
A
p
M
p
proj.dim
A
M for any prime ideal p. Given an A
p
-module
N we have N

= N
p
as A
p
-modules and it follows that gl.dim(A
p
) gl.dim(A).
Lemma 124. Let A be a nonzero noetherian ring and M a nitely generated A-module. Then
(i) proj.dim.M = sup{proj.dim
A
m
M
m
| m a maximal ideal of A}
(ii) For n 0, proj.dim.M n if and only if Tor
A
n+1
(M, A/m) = 0 for every maximal ideal m.
(iii) For every maximal ideal m, gl.dim(A
m
) gl.dim(A). Moreover
gl.dim(A) = sup{gl.dim(A
m
) | m a maximal ideal of A}
Proof. (i) This is trivial if M = 0, so assume M is nonzero. For any module N and maximal ideal
m we know from Lemma 22 that there is an isomorphism of A
m
-modules for n 0
Ext
n
A
(M, N)
m

= Ext
n
A
m
(M
m
, N
m
)
The module Ext
n
A
(M, N) is nonzero if and only if some Ext
n
A
m
(M
m
, N
m
) is nonzero, and proj.dim.M
is the largest integer n 0 for which there exists a module N with Ext
n
A
(M, N) = 0, so the claim
is easily checked.
(ii) Let n 0. Then by (i), proj.dim.M n if and only if proj.dim
A
m
M
m
n for every
maximal ideal m. Since A
m
/mA
m

= (A/m)
m
as A
m
-modules, we can use Lemma 22 and Lemma
123 to see that this if and only if for every maximal ideal m
0 = Tor
A
m
n+1
(M
m
, A
m
/mA
m
)

= Tor
A
n+1
(M, A/m)
m
If m, n are distinct maximal ideals, then (A/m)
n
= 0, so Tor
A
n+1
(M, A/m) = 0 if and only if
Tor
A
n+1
(M, A/m)
m
= 0, which completes the proof.
(iii) For any maximal ideal m and A
m
-module N, there is an isomorphism of A
m
-modules
N

= N
m
, so using (i) and the fact that gl.dim(A) = sup{proj.dim.M} the various claims are easy
to check.
Theorem 125. Let (A, m, k) be a noetherian local ring. Then for n 0
gl.dim(A) n Tor
A
n+1
(k, k) = 0
Consequently, we have gl.dim(A) = proj.dim
A
(k).
Proof. Since tor.dim(A) = gl.dim(A) the implication is immediate. If Tor
A
n+1
(k, k) = 0 then
proj.dim
A
(k) n by Lemma 123. Hence Tor
A
n+1
(M, k) = 0 for all modules M, so by Lemma 123,
proj.dim.M n for every nitely generated module M. Hence gl.dim(A) n. Using Lemma 123
again we see that gl.dim(A) = proj.dim
A
(k).
Proposition 126. Let (A, m, k) be a noetherian local ring and M a nonzero nitely generated
A-module. If proj.dim.M = r < and if x m is M-regular, then proj.dim(M/xM) = r + 1.
Proof. By assumption the following sequence of A-modules is exact
0

M
x

M

M/xM

0
Therefore the sequence 0 Tor
A
i
(M/xM, k) 0 is exact and so Tor
A
i
(M/xM, k) = 0 for
i > r + 1. The following sequence of A-modules is also exact
0 = Tor
A
r+1
(M, k)

Tor
A
r+1
(M/xM, k)

Tor
A
r
(M, k)
x

Tor
A
r
(M, k)
where x denotes left multiplication by x, which is equal to Tor
A
r
(x, k) and also Tor
A
r
(M, x) (see our
Tor notes). Since k = A/m is annihilated by x, so is Tor
A
r
(M, k). Therefore Tor
A
r+1
(M/xM, k)

=
Tor
A
r
(M, k) = 0 and hence proj.dim(M/xM) = r + 1 by Lemma 123.
41
Corollary 127. Let (A, m, k) be a noetherian local ring, M a nonzero nitely generated A-module
and a
1
, . . . , a
s
an M-regular sequence. If proj.dim.M = r < then proj.dim(M/(a
1
, . . . , a
s
)) =
r +s.
Proof. Since A is local and (a
1
, . . . , a
s
)M = M we have a
i
m for each i. We proceed by induction
on s. The case s = 1 was handled by Proposition 126. If s > 1 then set N = M/(a
1
, . . . , a
s1
)M.
Then a
s
m is N-regular, and by the inductive hypothesis proj.dim.N = r + s 1 < . So by
the case s = 1, proj.dim(N/a
s
N) = r +s, and N/a
s
N

= M/(a
1
, . . . , a
s
)M, so we are done.
Theorem 128. Let (A, m, k) be a regular local ring of dimension d. Then gl.dim(A) = d.
Proof. If d = 0 then A is a eld, and trivially gl.dim(A) = 0. Otherwise let {a
1
, . . . , a
d
} be a
regular system of parameters. Then the sequence a
1
, . . . , a
d
is A-regular by Theorem 108 and k =
A/(a
1
, . . . , a
d
) so proj.dim.k = d by Corollary 127. Theorem 125 implies that gl.dim(A) = d.
Among many other things, Theorem 128 allows us to give a much stronger version of Lemma
121 for regular local rings.
Corollary 129. Let (A, m, k) be a regular local ring of dimension d and M a nitely generated
A-module. Then
(i) M is projective if and only if Ext
i
(M, A) = 0 for i > 0.
(ii) For n 0 we have proj.dim.M n if and only if Ext
i
(M, A) = 0 for i > n.
Proof. If M = 0 the result is trivial, so assume otherwise. (i) Suppose that Ext
i
(M, A) = 0 for all
i > 0. Since Ext
i
(M, ) is additive, it follows that Ext
i
(M, ) vanishes on nite free A-modules
for i > 0. We show for 1 j d + 1 that Ext
j
(M, N) = 0 for every nitely generated A-module
N (we may assume d 1 since otherwise M is trivially projective).
Theorem 128 implies that proj.dim.M d and therefore Ext
d+1
(M, ) = 0, so this is at
least true for j = d + 1. Suppose that Ext
j
(M, ) vanishes on nitely generated modules, and
let N be a nitely generated A-module. We can nd a short exact sequence of nitely generated
A-modules 0 R F N 0 with F a nite free A-module. Since Ext
j1
(M, F) = 0
and Ext
j
(M, R) = 0 by the inductive hypothesis, it follows from the long exact sequence that
Ext
j1
(M, N) = 0, as required. The case j = 1 implies that M is projective, using Lemma 121.
(ii) The case n = 0 is (i), so assume n 1. If proj.dim.M n then by denition Ext
i
(M, ) =
0 for i > n, so this direction is trivial. For the converse, suppose that Ext
i
(M, A) = 0 for i > n.
We can construct an exact sequence
0 K P
n1
P
0
M 0
with K nitely generated and the P
i
nitely generated projectives. It suces to show that K
is projective. But by dimension shifting we have Ext
i
(K, A)

= Ext
i+n
(M, A) = 0 for i > 0.
Therefore by (i), K is projective and the proof is complete.
Corollary 130 (Hilberts Syzygy Theorem). Let A = k[x
1
, . . . , x
n
] be a polynomial ring over
a eld k. Then gl.dim(A) = n.
Proof. See our Dimension notes for another proof. By Theorem 118 every local ring of A is regular.
So if m is a maximal ideal then A
m
is regular of global dimension ht.m by Theorem 128. So by
Lemma 124 (iii), gl.dim(A) is the supremum of the heights of the maximal ideals in A, which is
clearly dim(A) = n.
Theorem 131. Let (A, m, k) be a noetherian local ring, and M a nonzero nitely generated A-
module. If proj.dim(M) < then
proj.dim(M) +depth(M) = depth(A)
42
Proof. By induction on depth(A). Let proj.dim(M) = n 0. If depth(A) = 0 then m Ass(A).
This implies that there is a short exact sequence of A-modules
0 k A C 0
Thus we have an exact sequence
0 Tor
A
n+1
(M, C) Tor
A
n
(M, k) Tor
A
n
(M, A)
By Proposition 122, flat.dim(M) = n, so Tor
A
n+1
(M, C) = 0. But if n 1 then Tor
A
n
(M, A) =
0 and Lemma 123 yields Tor
A
n+1
(M, C)

= Tor
A
n
(M, k) = 0, which is a contradiction. Hence
proj.dim(M) = 0. This means that M is projective and hence free by Proposition 24. Thus also
depth(M) = 0 by Lemma 70, which completes the proof in the case depth(A) = 0.
Now we x a ring A with depth(A) > 0 and proceed by induction on depth(M). First suppose
that depth(M) = 0. Then m Ass(M), say m = Ann(y) with 0 = y M. Since depth(A) > 0
we can nd a regular element x m. Find an exact sequence
0

K

A
m

M

0
It follows from Lemma 70 that M cannot be free, and hence by Proposition 24 cannot be projective
either. Thus proj.dim(M) = proj.dim(K)+1. Choose u A
m
with (u) = y. Clearly m (K : u)
and therefore xu K. Since x is regular on A
m
and u / K it follows that xu / xK. But
m (xK : xu), so m Ass(K/xK) and consequently depth(K/xK) = 0. Since K is a submodule
of a free module, x is regular on K. By the third Change of Rings theorem for projective dimension
(see our Dimension notes)
proj.dim
A/x
(K/xK) = proj.dim
A
(K) = proj.dim
A
(M) 1
By Lemma 83, depth
A/x
(A/x) = depth
A
(A) 1, so using the inductive hypothesis (on A)
depth
A
(A) = 1 +depth
A/x
(A/x)
= 1 +depth
A/x
(K/xK) +proj.dim
A/x
(K/xK)
= proj.dim
A
(M)
Finally, we consider the case depth(M) > 0. Let x m be regular on M. By Lemma 82 we have
depth(M/xM) = depth(M) 1 and by Proposition 126, proj.dim(M/xM) = proj.dim(M) + 1.
Using the inductive hypothesis (for M) we have
depth(A) = depth(M/xM) +proj.dim(M/xM)
= depth(M) 1 +proj.dim(M) + 1
= depth(M) +proj.dim(M)
which completes the proof.
Remark 3. If A is a regular local ring of dimension d, then by Theorem 128 the global dimension
of A is d, and for any A-module M we have proj.dim.M d. We can now answer the question:
how big is the dierence d proj.dim.M?
Corollary 132. Let A be a regular local ring of dimension d, and M a nonzero nitely generated
A-module. Then proj.dim(M) +depth(M) = d.
Remark 4. With the notation of Corollary 132 the integer proj.dim(M) measures how projec-
tive the module M is. To be precise, the closer proj.dim(M) is to zero the more projective M is.
Using the Corollary, we can rephrase this by saying that the projectivity of M is measured by the
largest number of independent variables in M. The module M admits d independent variables
if and only if it is projective.
43
6.3 Koszul Complexes
Throughout this section let A be a nonzero ring. In this section a complex will mean a positive
chain complex in AMod (notation of our Derived Functor notes). This is a sequence of A-modules
and module morphisms {M
n
, d
n
: M
n
M
n1
}
nZ
with M
n
= 0 for n < 0 and d
n1
d
n
= 0 for
all n. Visually


M
n
d
n

M
n1
d
n1

d
1

M
0
d
0

0


We denote the complex by M and dierentials d
n
by d where no confusion is likely. Let C denote
the abelian category of all positive chain complexes in AMod (this is an abelian subcategory
of the category ChAMod of all chain complexes). If L is a complex then for k 0 let L[1]
denote the complex obtained by shifting the objects and dierentials one position left. That is,
L[1]
n
= L
n1
. Clearly if : L

is a morphism of complexes then [1]

n
=
n1
denes a
morphism of complexes [1] : L[1] L

[1]. This denes an exact functor T : C C, and

clearly T
k
shifts k positions left for k 1. If M is an A-module, then we consider it as a complex
concentrated in degree 0 and denote this complex also by M.
If L and M are two complexes, we dene a chain complex L M by
(L M)
n
=

p+q=n
L
p

A
M
q
= (L
0

A
M
n
) (L
1

A
M
n1
) (L
n

A
M
0
)
If x y is an element of one of these summands, then by abuse of notation we also use x y
to denote the image in (L M)
n
. For n 1 and integers p, q 0 with p + q = n we induce a
morphism
p,q
: L
p

A
M
q
(L M)
n1
of A-modules out of the tensor product using the
following formula

p,q
(x y) =
_

_
d
L
(x) y + (1)
p
x d
M
(y) p > 0, q > 0
d
L
(x) y q = 0
(1)
p
x d
M
(y) p = 0
Together these dene a morphism of A-modules d : (L M)
n
(L M)
n1
. It is easy to
check that this makes L M into a complex of A-modules. Given morphisms of complexes
: L L

and : M M

we obtain for each pair of integers p, q 0 a morphism of

A-modules
p

q
: L
p

A
M
q
L

A
M

q
, and these give rise to a morphism of complexes
: L M L

( )
n
= (
0

0
) (
1

1
) (
n

n
)
So the tensor product denes a covariant functor : C C C which is additive in each
variable. That is, for any complex L the partial functors L and L are additive.
Proposition 133. For complexes L, M, N there is a canonical isomorphism

L,M,N
: (L M) N L (M N)
which is natural in all three variables.
44
Proof. For n 0 we have an isomorphism of A-modules
((L M) N)
n
=

p+q=n
(L M)
p

A
N
q
=

p+q=n
_

r+s=p
L
r

A
M
s
_

A
N
q

r+s+q=n
(L
r

A
M
s
)
A
N
q

r+s+q=n
L
r

A
(M
s

A
N
q
)

p+q=n
L
p

A
_

r+s=q
M
r

A
N
s
_
= (L (M N))
n
Given integers with r +s +q = n and elements x L
r
, y M
s
, z N
q
we have x y (LM)
p
and this isomorphism sends (x y) z ((LM) N)
n
to x (y z) in (L(MN))
n
. It is
straightforward to check that this is an isomorphism of complexes natural in all three variables.
Proposition 134. For any complex L the functors L and L are naturally equivalent
and both are right exact. The functor A is naturally equivalent to the identity functor, and
A[1] is naturally equivalent to T.
Proof. To show that L and L are naturally equivalent, the only subtle point is that for
p, q 0 if : L
p
M
q

= M
q
L
p
is the canonical isomorphism, then we use the isomorphism
(1)
pq
in dening (LM)
n

= (M L)
n
. Suppose we have a short exact sequence 0 A
B C 0 in C. Then for every j 0 the sequence of A-modules 0 A
j
B
j

C
j
0 is exact, and therefore
L
i
A
j
L
i
B
j
L
i
C
j
0
is also exact for any i 0. Coproducts are exact in AMod so for any n 0 the following sequence
is also exact

i+j=n
L
i
A
j

i+j=n
L
i
B
j

i+j=n
L
i
C
j
0
But this is (L A)
n
(L B)
n
(L C)
n
0, so the sequence L A L B
L C 0 is pointwise exact and therefore exact. Consider A as a complex concentrated in
degree 0. For a complex M the natural isomorphism M

= A M is given pointwise by the
isomorphism M
n

= AM
n
. There is also a natural isomorphism AM

= M given pointwise by
AM
n

= M
n
. It is not hard to check that this is the same as M

= AM followed by the twist
A M

= M A. The complex A[1] M is isomorphic to M[1] but we have to be careful,
since the signs of the dierentials in A[1] M are the opposite of those in M[1], so we use the
isomorphism M[1]
n
= M
n1

= A M
n1
given by (1)
n+1
where : M
n1

= A M
n1
is
canonical. This isomorphism is clearly natural in M.
On the other hand, there is a natural isomorphism M[1]

= M A[1] given pointwise by
M[1]
n

= M
n1
A, with no sign problems. In fact this isomorphism is M[1]

= A[1] M
followed by the twist A[1] M

= M A[1].
In our Module Theory notes we dene the exterior algebra M associated to any A-module
M. It is a graded A-algebra, and if M is free of rank n 1 with basis {x
1
, . . . , x
n
} then for
0 p n,
p
M is free of rank
_
n
p
_
with basis x
i
1
x
i
p
indexed by strictly ascending
sequences i
1
< < i
p
in the set {1, . . . , n}. For p > n we have
p
M = 0.
45
Denition 24. Fix n 1 and let F = A
n
be the canonical free A-module of rank n, with
canonical basis x
1
, . . . , x
n
. Suppose we are given elements a
1
, . . . , a
n
A. We dene a complex
of A-modules called the Koszul complex, and denoted K(a
1
, . . . , a
n
)

d
p+1

p
F
d
p

d
3

2
F
d
2

1
F
d
1

0
F

0
We identify
1
F with F and
0
F with A. These modules become zero beyond
n
F. The map d
1
is dened by d
1
(x
i
) = a
i
. For p 2 with
p
F = 0 we dene
d
p
(x
i
1
x
i
p
) =
p

r=1
(1)
r1
a
i
r
(x
i
1
x
i
r
x
i
p
)
where x
i
r
indicates that we have omitted x
i
r
. All other morphisms are zero. It is not hard to
check that d
p
d
p+1
= 0 for all p 1, so this is actually a complex.
Denition 25. Let a
1
, . . . , a
n
A. If C is a chain complex, then we denote by C(a
1
, . . . , x
n
)
the tensor product C K(a
1
, . . . , x
n
). If M is an A-module then we consider it is as a complex
concentrated in degree 0 and denote by K(a
1
, . . . , a
n
, M) the complex M K(a
1
, . . . , a
n
). This
is isomorphic to the complex


M
p
F



M
2
F

M
1
F

M
0
F

0
Example 6. If a
1
A then K(a
1
) is isomorphic to the complex


0

A
a
1

A

0
concentrated in degrees 0 and 1, where the morphism A A is left multiplication by a
1
. Then
H
0
(K(a
1
)) = A/a
1
A and H
1
(K(a
1
)) = Ann(a
1
) as A-modules.
Proposition 135. Let a
1
, . . . , a
n
A and a multiplicatively closed set S A be given. Then there
is a canonical isomorphism of complexes of S
1
A-modules S
1
K(a
1
, . . . , a
n
)

= K(a
1
/1, . . . , a
n
/1).
Proof. There is a canonical isomorphism of S
1
A-modules S
1
F

= (S
1
A)
n
identifying x
i
/1 with
the canonical ith basis element. Using (TES,Corollary 16) we have for each p 0 a canonical
isomorphism of S
1
A-modules
S
1
K(a
1
, . . . , a
n
)
p
= S
1
(
p

A
F)

=
p

S
1
A
S
1
F

=
p

S
1
A
G
n
= K(a
1
/1, . . . , a
n
/1)
p
where G = (S
1
A)
n
. Together these isomorphisms form an isomorphism of complexes of S
1
A-
modules, as required.
Proposition 136. Let a
1
, . . . , a
n+1
A with n 1. Then there is a canonical isomorphism
K(a
1
, . . . , a
n
) K(a
n+1
)

= K(a
1
, . . . , a
n+1
)
Proof. Let T = K(a
1
, . . . , a
n
) K(a
n+1
). Write F = A
n
and let {x
1
, . . . , x
n
} be the canonical
basis. Let G = A with canonical basis {x
n+1
}. Then T
0
=
0
F
0
G

= AA

= A and
T
1
= (
0
F
1
G) (
1
F
0
G)

=
1
G
1
F

= A
n+1
For p 2 we have
T
p
=

i+j=p

i
F
j
G

= (
p1
F
1
G) (
p
F
0
G)

=
p1
F
p
F
So for p > n + 1 we have T
p
= 0, and for p n + 1 the A-module T
p
is free of rank
_
n+1
p
_
.
So at least the modules T
p
are free of the same rank as K
p
(a
1
, . . . , a
n+1
). Let H = A
n+1
have
46
canonical basis e
1
, . . . , e
n+1
. The isomorphism
0
H

= T
0
sends 1 to 1 1. The isomorphism

1
H

= T
1
sends e
1
, . . . , e
n
to x
i
1 and e
n+1
to 1 1. For p 2 the action of isomorphism

p
H

= T
p
on a basis element e
i
1
e
i
p
is described in two cases: if i
p
n then use the basis
element (x
i
1
x
i
p
) 1 of
p
F
0
G, and otherwise if i
p
= n + 1 use the basis element
(x
i
1
x
i
p1
) 1 of
p1
F
1
G. One checks that these isomorphisms are compatible with
the dierentials.
For any a A we have an exact sequence of complexes
0 A K(a) A[1] 0
Let C be any complex. Tensoring with C and using the natural isomorphisms C A

= C and
C A[1]

= C[1] we have an exact sequence

0 C C(a) C[1] 0
For p Z we have H
p
(C[1]) = H
p1
(C), so the long exact homology sequence is


H
p+1
(C)

H
p+1
(C(a))

H
p
(C)

p

H
p
(C)




1

H
1
(C)

H
1
(C(a))

H
0
(C)

0

H
0
(C)

H
0
(C(a))

0
It is not dicult to check that the connecting morphism
p
is multiplication by (1)
p
a. Therefore
Lemma 137. If C is a complex with H
p
(C) = 0 for p > 0 then H
p
(C(a)) = 0 for p > 1 and there
is an exact sequence
0

H
1
(C(a))

H
0
(C)
a

H
0
(C)

H
0
(C(a))

0
If a is H
0
(C)-regular, then we have H
p
(C(a)) = 0 for all p > 0 and H
0
(C(a))

= H
0
(C)/aH
0
(C).
Theorem 138. Let A be a ring, M an A-module and a
1
, . . . , a
n
an M-regular sequence in A.
Then we have
H
p
(K(a
1
, . . . , a
n
, M)) = 0 (p > 0)
H
0
(K(a
1
, . . . , a
n
, M))

= M/(a
1
, . . . , a
n
)M
Proof. The last piece of Koszul complex K(a
1
, . . . , a
n
, M) is isomorphic to
M
n
M 0
where the last map is (m
1
, . . . , m
n
) (a
1
m
1
, . . . , a
n
m
n
). So clearly there is an isomorphism of A-
modules H
0
(K(a
1
, . . . , a
n
, M))

= M/(a
1
, . . . , a
n
)M. We prove the other claim by induction on n,
having already proven the case n = 1 in Lemma 137. Let C be the complex K(a
1
, . . . , a
n1
, M).
Then H
0
(C)

= M/(a
1
, . . . , a
n1
)M so that a
n
is H
0
(C)-regular. By the inductive hypothesis
H
p
(C) = 0 for p > 0 and therefore by Lemma 137, H
p
(C K(a
n
)) = 0 for p > 0. But by Lemma
136 and Proposition 133 there is an isomorphism CK(a
n
)

= K(a
1
, . . . , a
n
, M), which completes
the proof.
Remark 5. In other words, for an M-regular sequence a
1
, . . . , a
n
the corresponding Koszul com-
plex K(a
1
, . . . , a
n
, M) gives a canonical resolution of the A-module M/(a
1
, . . . , a
n
)M. That is,
the following sequence is exact
M
2
F M
1
F M
0
F M/(a
1
, . . . , a
n
)M 0
Taking M = A we see that the Koszul complex K(a
1
, . . . , a
n
) gives a free resolution of the A-
module A/(a
1
, . . . , a
n
). That is, the following sequence is exact
0
n
F
2
F
1
F
0
F A/(a
1
, . . . , a
n
) 0 (5)
In particular we observe that proj.dim
A
(A/(a
1
, . . . , a
n
)) n.
47
Lemma 139. Let A be a ring and a
1
, . . . , a
n
an A-regular sequence. Then for any A-module M
there is a canonical isomorphism of A-modules Ext
n
A
(A/(a
1
, . . . , a
n
), M)

= M/(a
1
, . . . , a
n
)M.
Proof. We have already observed that (5) is a projective resolution of A/(a
1
, . . . , a
n
). Taking
Hom
A
(, M) the end of the complex we are interested in is
Hom
A
(
n1
F, M) Hom
A
(
n
F, M) 0
Use the canonical bases to dene isomorphisms
n1
F

= A
n
and
n
F

= A. Then we have a
commutative diagram
Hom
A
(
n1
F, M)

Hom
A
(
n
F, M)

M
n

M
where (m
1
, . . . , m
n
) =

n
r=1
(1)
r1
a
r
m
r
. It is clear that Im = (a
1
, . . . , a
n
)M, so we have an
isomorphism of A-modules Ext
n
A
(A/(a
1
, . . . , a
n
), M)

= M/(a
1
, . . . , a
n
)M.
Denition 26. Let (A, m, k) be a local ring and u : M N a morphism of A-modules. We
say that u is minimal if u 1 : M k N k is an isomorphism. Clearly any isomorphism
M

= N is minimal.
Lemma 140. Let (A, m, k) be a local ring. Then
(i) Let u : M N be a morphism of nitely generated A-modules. Then u is minimal if and
only if it is surjective and Ker(u) mM.
(ii) If M is a nitely generated A-module then there is a minimal morphism u : F M with
F nite free and rank
A
F = rank
k
(M k).
Proof. (i) Suppose that u is minimal. Let N

be the image of M. Then since M/mM

= N/mN
we have N

+ mN = N and therefore N

= N by Nakayama, so u is surjective. It is clear that

Ker(u) mM. Conversely suppose that u is surjective and Ker(u) mM. Since u is surjective
it follows that u(mM) = mN. Therefore the morphism of A-modules M N N/mN
has kernel mM and so M/mM N/mN is an isomorphism, as required. (ii) If M = 0 then
this is trivial, since we can take F = 0. Otherwise let m
1
, . . . , m
n
be a minimal basis of M and
u : A
n
M the corresponding morphism. This is clearly minimal.
Let (A, m, k) be a noetherian local ring and M a nitely generated A-module. A free resolution
L :

L
i
d
i

L
i1


d
1

L
0
d
0

M

0
is called a minimal resolution if L
0
M is minimal, and L
i
Ker(d
i1
) is minimal for each
i 1. Since L
i+1
L
i
Ker(d
i1
) = 0 for all i 1 it follows that in the complex of
A-modules L k
L
i
k L
i1
k L
0
k 0
the dierentials are all zero. Therefore we have Tor
A
i
(M, k)

= L
i
k as k-modules for all i 0.
Since M, k are nitely generated, for i 0 the A-modules Tor
A
i
(M, k) and L
i
k are nitely
generated. Hence L
i
k is a nitely generated free k-module, which shows that L
i
is a nitely
generated A-module.
Proposition 141. Let (A, m, k) be a noetherian local ring and M a nitely generated A-module.
Then a minimal free resolution of M exists, and is unique up to a (non-canonical) isomorphism.
48
Proof. By Lemma 140 (ii) we can nd a minimal morphism d
0
: L
0
M with L
0
nite free of
rank rank
k
(Mk). Let K L
0
be the kernel of d
0
. Find a minimal morphism L
1
K with
L
1
nite free, and so on. This denes a minimal free resolution of M. To prove the uniqueness,
let : L M and

: L

M be two minimal free resolutions of M. We can lift the identity

1
M
to a morphism of chain complexes : L L

, so we have a commutative diagram

L
1

L
0

1

L

M
Since ,

are minimal, the map

0
1 : L
0
k L

0
k is an isomorphism of k-modules. In
particular we have
rank
A
L
0
= rank
k
(L
0
k) = rank
k
(L

0
k) = rank
A
L

0
So L
0
, L

0
are free of the same nite rank. We claim that
0
is an isomorphism. This is trivial
if L
0
= L

0
= 0, so assume they are both nonzero. Then
0
is described by a square matrix
T M
n
(A). If you take residues you get the matrix T

M
n
(k) of
0
1, which has nonzero
determinant since it is an isomorphism. But it is clear that det(T) +m = det(T

), so det(T) / m.
Therefore
0
itself is an isomorphism.
Since
0
is an isomorphism, so the induced morphism on the kernels Ker() Ker(

), and
we can repeat the same argument to see that
1
is an isomorphism, and similarly to show that
all the
i
are isomorphisms.
Lemma 142. Let (A, m, k) be a noetherian local ring and u : F G a morphism of nitely
generated free A-modules. Then u is minimal if and only if it is an isomorphism.
Denition 27. Let (A, m, k) be a noetherian local ring and M a nitely generated A-module.
Choose a minimal free resolution of M. Then for i 0 the integer b
i
= rank
A
L
i
0 is
called the i-th Betti number of M. It is independent of the chosen resolution, and moreover
rank
k
Tor
A
i
(M, k) = b
i
.
Example 7. Let (A, m, k) be a noetherian local ring and let M be a nitely generated A-module.
Then
(i) Proposition 141 shows that b
0
= rank
k
(M k).
(ii) If M = 0 then the zero complex is a minimal free resolution of M, so b
i
= 0 for i 0.
(iii) If M is at then Tor
A
i
(M, k) = 0 for all i 1, so b
i
= 0 for i 1. In particular this is true
if M is free or projective.
(iv) If M is free of nite rank s 1 then M k is a free k-module of rank s, so b
0
= s.
Lemma 143. Let (A, m, k) be a noetherian local ring and M a nitely generated A-module.
Suppose that we have two complexes L, F together with morphisms ,

such that the following

sequences are exact in the last two nonzero positions
L :

L
i
d
i

L
i1


d
1

L
0

M

0
F :

F
i
d

F
i1


d

F
0

M

0
Assume the following
(i) L is a minimal free resolution of M;
(ii) Each F
i
is a nitely generated free A-module;
49
(iii)

1 : F
0
k M k is injective;
(iv) For each i 0, d

i+1
(F
i+1
) mF
i
and the induced morphism F
i+1
/mF
i+1
mF
i
/m
2
F
i
is
an injection.
Then there exists a morphism of complexes f : F L lifting the identity 1
M
such that for f
i
maps F
i
isomorphically onto a direct summand of L
i
. Consequently we have
rank
A
F
i
rank
A
L
i
= rank
k
Tor
A
i
(M, k)
Proof. Both L, F are positive chain complexes, with F projective and L acyclic, so by our Derived
Functor notes there is a morphism f : F L of chain complexes giving a commutative diagram


F
0
f
0

0


L
0

M

0
We have to prove that for each i 0 the morphism f
i
: F
i
L
i
is a coretraction.We claim that
f
i
is a coretraction i. f
i
1 : F
i
k L
i
k is an injective morphism of k-modules. One
implication is clear. So assume that f
i
1 is injective. The claim is trivial if either of F
i
, L
i
are
zero, so assume they are both of nonzero nite rank. Pick bases for F
i
, L
i
(which are obviously
minimal bases), and use the fact that f
i
1 is a coretraction to dene a morphism : L
i
F
i
such that (f
i
) 1 : F
i
k F
i
k is the identity. By Lemma 142 it follows that f
i
is an
isomorphism, and therefore clearly f
i
is a coretraction.
We prove by induction that f
i
1 is injective for all i 0. By assumptions (i), (iii) it is clear
that f
0
1 is injective. We have the following commutative diagram
F
1

F
F
F
F
F
F
F
F
F
f
1

F
0
f
0

Ker

x
x
x
x
x
x
x
x
x

Ker

F
F
F
F
F
F
F
F
F
L
1

x
x
x
x
x
x
x
x
x
d
1

L
0

M
By assumption 1 is an isomorphism. So to show f
1
1 is injective, it suces to show that
1, 1 are injective, or equivalently that
1
(mKer) = mKer

and
1
(mKer

) = mF
1
.
Suppose that a F
1
and d

1
(a) mKer

. Since

1 is injective, we have mKer

m
2
F
0
.
Hence d

1
(a) m
2
F
0
and therefore by (iv) a mF
1
, as required.
Now suppose that a Ker

and f
0
(a) mKer. Let g be such that gf
0
= 1. Then f
0
(a)
m
2
L
0
and therefore a = gf
0
(a) g(m
2
L
0
) m
2
F
0
. Let b F
1
be such that d

1
(b) = a m
2
F
0
.
Then (iv) implies that b mF
1
and therefore a = (b) mKer

, as required. This shows that

f
1
1 is injective.
Suppose that f
i
1 is injective for some i 1. Then we show f
i+1
1 is injective using a
similar setup. We replace Ker

by Imd

i+1
(in the case i = 0 they are equal) and use (iv) to
show that Kerd

i
mF
i
and (i) to show that Kerd
i
mL
i
. The proof that 1 is injective is
straightforward. For 1, let a Imd

i+1
be such that f
i
(a) mKerd
i
. As before we nd that
f
i
(a) m
2
L
i
, and hence a = gf
i
(a) m
2
F
i
. Let b F
i+1
be such that a = d

i+1
(b). Then by (iv),
b mF
i+1
and therefore a mImd

i+1
, as required. This proves that f
i
is a coretraction for i 0,
and the rank claim follows from the fact that rank
A
F
i
= rank
k
(F
i
k) rank
k
(L
i
k).
50
Lemma 144. Let A be a ring with maximal ideal m. If s / m and a A, then sa m
k
implies
a m
k
for any k 1.
Theorem 145. Let (A, m, k) be a noetherian local ring and let s = rank
k
m/m
2
. Then we have
rank
k
Tor
A
i
(k, k)
_
s
i
_
0 i s
Here rank
k
Tor
A
i
(k, k) is the i-th Betti number of the A-module k.
Proof. We have Tor
A
0
(k, k)

= k as k-modules, so rank
k
Tor
A
0
(k, k) = rank
k
k = 1, which takes
care of the case s = 0. So assume that s 1 and let {a
1
, . . . , a
s
} be a minimal basis of m, with
associated Koszul complex F = K(a
1
, . . . , a
s
). The canonical morphism

: F
0

= A k gives a
complex exact in the last two nonzero places
F
i
F
i1
F
1
F
0
k 0
We claim this complex satises the conditions of Lemma 143. It clearly satises (ii) and (iii).
It only remains to check condition (iii). By the denition of d
p+1
: F
p+1
F
p
it is clear that
d
p+1
(F
p+1
) mF
p
for p 0. We also have to show that d
1
p+1
(m
2
F
p
) mF
p+1
. . This is trivial
if p + 1 > s, and also if p = 0 since {a
1
, . . . , a
s
} is a minimal basis. So assume 0 < p s 1.
Assume that
d
p+1
_
_

i
1
<<i
p+1
m
i
1
i
p+1
(x
i
1
x
i
p+1
)
_
_
=

i
1
<<i
p+1
p

r=1
(1)
r1
a
i
r
m
i
1
i
p+1
(x
i
1
x
i
r
x
i
p+1
) m
2
F
p
Then collecting terms, we obtain a number of equations of the form

(1)
e
t
a
t
m
t
m
2
where a
t
is one of the a
i
r
and m
t
one of the m
i
1
i
p+1
. Since the residues of the a
i
give a basis of m/m
2
over k, it follows that m
t
m, which completes the proof that F satises all the conditions of
Lemma 143. Choosing any minimal free resolution L of M, and applying Lemma 143 we see that
for 0 i s
_
s
i
_
= rank
A
F
i
rank
k
Tor
A
i
(k, k)
as required.
Theorem 146 (Serre). Let (A, m, k) be a noetherian local ring. Then A is regular if and only if
the global dimension of A is nite.
Proof. We have already proved one part in Theorem 128. So suppose that gl.dim(A) < .
Then Tor
A
s
(k, k) = 0 by Theorem 145, hence gl.dim(A) rank
k
m/m
2
since by Proposition
122 we have tor.dim(A) = gl.dim(A). On the other hand, it follows from Theorem 125 that
proj.dim(k) = gl.dim(A) < , so by Theorem 131 we have gl.dim(A) = proj.dim(k) = depth(A).
Therefore we get
dim(A) rank
k
m/m
2
gl.dim(A) = depth(A) dim(A)
Whence dim(A) = rank
k
m/m
2
, and A is regular.
Corollary 147. If A is a regular local ring then A
p
is regular for any p Spec(A).
Proof. Let M be a nonzero A
p
-module. Then considering M as an A-module, there is an exact
sequence of nite length n gl.dim(A) with all P
i
projective
0 P
n
P
0
M 0
51
Since A
p
is at the following sequence is also exact
0 (P
n
)
p
(P
0
)
p
M
p
0
The modules (P
i
)
p
are projective A
p
-modules, and M

= M
p
as A
p
-modules, so it follows that
gl.dim(A
p
) gl.dim(A) < .
Denition 28. A ring A is called a regular ring if A
p
is a regular local ring for every prime ideal
p of A. Note that A is not required to be noetherian. Regularity is stable under ring isomorphism.
A noetherian local ring A is regular in this sense if and only if it is regular in the normal sense.
It follows from Theorem 108 that any regular ring is normal, and a noetherian regular ring is
Cohen-Macaulay. It follows from Lemma 8 and Theorem 90 that a regular ring is catenary.
Lemma 148. A ring A is regular if and only if A
m
is regular for all maximal ideals m.
Proof. One implication is clear. For the other, given a prime ideal p, nd a maximal ideal m with
p m. Then A
p

= (A
m
)
pA
m
, so A
p
is a regular local ring.
Lemma 149. If A is a regular ring and S A is multiplicatively closed, then S
1
A is a regular
ring.
Lemma 150. If A is a regular ring, then so is A[x
1
, . . . , x
n
]. In particular k[x
1
, . . . , x
n
] is a
regular ring for any eld k.
Proof. This follows immediately from Theorem 118.
Theorem 151. Let A be a regular local ring which is a subring of a domain B, and suppose that
B is a nitely generated A-module. Then B is at (equivalently free) over A if and only if B is
Cohen-Macaulay. In particular, if B is regular then it is a free A-module.
Proof. Since B is a nitely generated A-module it is integral over A, and so by Lemma 91 if B is
at it is Cohen-Macaulay. Conversely, suppose that B is Cohen-Macaulay. If dim(A) = 0 then A
is a eld so B is trivially at, so throughout we may assume dim(A) 1. Since A is normal the
going-down theorem holds between A and B by Theorem 42, so by Theorem 55 (3) for any proper
ideal I of A, IB is proper and ht.I = ht.IB. We claim that depth
A
(A) = depth
A
(B). Notice that
depth
A
(B) is nite, since otherwise m Ass
A
(B) and hence dim(A) = 0.
Firstly we prove the inequality . Since A is regular it is Cohen-Macaulay, so depth
A
(A) =
dim(A). Set s = dim(A) and let {a
1
, . . . , a
s
} be a regular system of parameters. Then ht.(a
1
, . . . , a
i
)A =
i and therefore ht.(a
1
, . . . , a
i
)B = i for all 1 i s by Theorem 108. It follows from Corollary
97 that a
1
, . . . , a
s
is a B-regular sequence, and therefore depth
A
(B) s.
To prove the reverse inequality, set d = depth
A
(B) and let a
1
, . . . , a
d
m be a maximal B-
regular sequence. Then as elements of B the sequence a
1
, . . . , a
d
is B-regular, so by Lemma 74 we
have ht.(a
1
, . . . , a
d
) = d. But (a
1
, . . . , a
d
) mB and ht.mB = ht.m = dim(A), so d dim(A), as
required.
Since gl.dim(A) < we have proj.dim
A
(B) < , so we can apply Theorem 131 to see that
proj.dim
A
(B) + depth
A
(B) = depth
A
(A), so proj.dim
A
(B) = 0 and therefore B is projective.
Since A is local and B nitely generated, projective free at, so the proof is complete.
6.4 Unique Factorisation
Recall that if A is a ring, two elements p, q A are said to be associates if p = uq for some unit
u A. This is an equivalence relation on the elements of A.
Denition 29. Let A be an integral domain. An element of A is irreducible if it is a nonzero
nonunit which cannot be written as the product of two nonunits. An element p A is prime if it is
a nonzero nonunit with the property that if p|ab then p|a or p|b. Equivalently p is prime i. (p) is
a nonzero prime ideal. We say A is a unique factorisation domain if every nonzero nonunit a A
can be written essentially uniquely as a = up
1
p
r
where u is a unit and each p
i
is irreducible.
52
Essentially uniquely means that if a = vq
1
q
s
where v is a unit and the q
j
irreducible, then
r = s and after reordering (if necessary) q
i
is an associate of p
i
. The property of being a UFD is
stable under ring isomorphism.
Theorem 152. A noetherian domain A is a UFD if and only if every prime ideal of height 1 is
principal.
Lemma 153. Let A be a noetherian domain and let x A be prime. Then A is a UFD if and
only if A
x
is.
Proof. By assumption (x) is a prime ideal of height 1. If p is a prime ideal of height 1 then either
x p, in which case p = (x), or x / p, and these primes are in bijection with the primes of A
x
. So
using Theorem 152 it is clear that if A is a UFD so is A
x
. Suppose that A
x
is a UFD and let p be
a prime ideal of height 1 in A. We can assume that x / p. Let a p be such that pA
x
= a/1A
x
.
By [AM69] Corollary 10.18 we have
i
(x
i
) = 0, so if x|a there is a largest integer n 1 with x
n
|a.
Write a = cx
n
. Since x / p we have c p, so by replacing a with c we can assume pA
x
= a/1A
x
with a / (x). Then it is clear that p = (a), as required.
Denition 30. Let R be an integral domain. If M is a torsion-free R-module then the rank of
M is the maximum number of linearly independent elements in M, rank(M) {0, 1, . . . , }.
Proposition 154. Let R be an integral domain and M a torsion-free R-module. If T R
is multiplicatively closed, then T
1
M is a torsion-free T
1
R-module and rank
T
1
R
(T
1
M) =
rank
R
(M).
Proof. If rank(M) = 0 this is trivial, so assume M is nonzero. It is clear that T
1
M is torsion-free.
If rank
R
(M) = r and x
1
, . . . , x
r
M are linearly independent, then x
1
/1, . . . , x
r
/1 T
1
M are
linearly independent over T
1
R. Similarly if x
1
/s
1
, . . . , x
n
/s
n
T
1
M are linearly independent
in T
1
M, then x
1
, . . . , x
n
are linearly independent in M. So the result is clear.
Corollary 155. Let R be an integral domain with quotient eld K. Then
(i) If M is a torsion-free R-module, rank
R
(M) = dim
K
(M K).
(ii) If M, N are two torsion-free R-modules of nite rank, then rank
R
(M N) = rank
R
(M) +
rank
R
(N).
In particular if M is a free R-module then the rank just dened is equal to the normal free rank,
and we can write rank(M) without confusion.
Let R be a noetherian domain and suppose a
1
, . . . , a
n
R are linearly independent elements
which do not generate R. Then a
1
, . . . , a
r
is an R-regular sequence, so by Lemma 74 the ideals
(a
1
, . . . , a
i
) have height i for 1 i n. So it follows immediately that
Lemma 156. Let R be a noetherian domain and I an ideal. Then rank(I) ht.I.
Lemma 157. Let R be a domain and M a nitely generated projective R-module of rank 1. Then

i
M = 0 for i > 1.
Proof. By localisation. If p is a prime ideal then M
p
is a nitely generated projective module over
the local ring R
p
, so M
p
is free of rank 1 and M
p

= R
p
. Hence for i > 1
(
i
M)
p

=
i
M
p

=
i
R
q
= 0
as required.
Theorem 158 (Auslander-Buchsbaum). A regular local ring (A, m) is UFD.
53
Proof. We use induction on dimA. If dimA = 0 then A is a eld, and if dimA = 1 then A is a
principal ideal domain. Suppose dimA > 1 and let a
1
, . . . , a
d
be a regular system of parameters.
Then x = a
1
is prime by Theorem 108, so it suces by Lemma 153 to show that A
x
is UFD.
Let q be a prime ideal of height 1 in A
x
and put p = q A, so q = pA
x
. By Theorem 128,
gl.dim.A = dimA < , so we can produce an exact sequence of A-modules with all F
i
nitely
generated free
0 F
n
F
n1
F
0
p 0 (6)
Maximal ideals of A
x
correspond to primes of A maximal among those not containing x. These
primes must all be properly contained in m, so if PA
x
is a maximal ideal then ht.P < dimA.
Therefore (A
x
)
PA
x

= A
P
is UFD by the inductive assumption, and so q(A
x
)
n
is either principal
or zero for every maximal n of A
x
. Then by Lemma 124 we have proj.dim
A
x
(q) = 0 and therefore
q is projective. Localising (6) with respect to S = {1, x, x
2
, . . .} we see that the following sequence
of A
x
-modules is exact
0 F

n
F

n1
F

0
q 0 (7)
where F

i
= F
i
A
x
are nitely generated and free over A
x
. If we decompose (7) into short exact
sequences
0 K

0
F

0
q 0
0 K

1
F

1
K

0
0
.
.
.
0 F

n
F

n1
K

n2
0
(8)
then the rst sequence splits since q is projective. Hence K

0
must be projective, and in this way
we show that all the sequences split, and all the K

i
are projective. It follows that

i even
F

i odd
F

i
q
Thus, we have nite free A
x
-modules F, G such that F

= Gq. Since A
x
is a noetherian domain
and q a nonzero ideal of height 1, it follows from Lemma 156 that rank(q) = 1. If rank(G) = r
then rank(F) = r + 1.
So to show q is principal and complete the proof, it suces to show that q is free. But by our
notes on Tensor, Symmetric and Exterior algebras we have
A
x

=
r+1

F

=
r+1

(Gq)

i+j=r+1
(
i
G) (
j
q)

= (
r+1
G
0
q) (
r
G
1
q)

= q
Since
r+1
G = 0,
r
G

= A
x
and
i
q = 0 for i > 1 by Lemma 157.
References
[AM69] M. F. Atiyah and I. G. Macdonald, Introduction to commutative algebra, Addison-Wesley
Publishing Co., Reading, Mass.-London-Don Mills, Ont., 1969. MR MR0242802 (39
#4129)
[Ash] Robert Ash, A course in commutative algebra, http://www.math.uiuc.edu/r-
ash/ComAlg.html.
[Eis95] David Eisenbud, Commutative algebra, Graduate Texts in Mathematics, vol. 150,
Springer-Verlag, New York, 1995, With a view toward algebraic geometry. MR
MR1322960 (97a:13001)
[Mat80] Hideyuki Matsumura, Commutative algebra, second ed., Mathematics Lecture Note Se-
ries, vol. 56, Benjamin/Cummings Publishing Co., Inc., Reading, Mass., 1980. MR
MR575344 (82i:13003)
54