MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA

THOMAS J. HAINES
1. Lecture 1
1.1. What is this course about? The foundations of dierential geometry (=
study of manifolds) rely on analysis in several variables as local machinery: many
global theorems about manifolds are reduced down to statements about what hap-
pens in a local neighborhood, and then anaylsis is brought in to solve the local
problem.
Analogously, algebraic geometry uses commutative algebraic as its local ma-
chinery. Our goal is to study commutative algebra and some topics in algebraic
geometry in a parallel manner. For a (somewhat) complete list of topics we plan
to cover, see the course syllabus on the course web-page.
1.2. References. See the course syllabus for a list of books you might want to
consult. There is no required text, as these lecture notes should serve as a text.
They will be written up in real time as the course progresses. Of course, I will
be grateful if you point out any typos you nd.
In addition, each time you are the rst person to point out any real mathematical
inaccuracy (not a typo), I will pay you 10 dollars!!
1.3. Conventions. Unless otherwise indicated in specic instances, all rings in
this course are commutative with identity element, denoted by 1 or sometimes by
e. We will assume familiarity with the notions of homomorphism, ideal, kernels,
quotients, modules, etc. (at least).
We will use Zorns lemma (which is equivalent to the axiom of choice): Let S,
be any non-empty partially ordered set. A chain T in S is a subset T S such
that x, y T implies x y or y x holds. If S, is such that every chain has an
upper bound in S (an element s S with t s for all t T), then S contains at
least one maximal element.
1.4. Correspondence between ideals and homomorphisms. We call any sur-
jective homomorphism A B a quotient. We say the quotients f
1
: A B
1
and
f
2
: A B
2
are equivalent if there exists a ring isomorphism : B
1
B
2
satisfying
f
1
= f
2
.
The terminology is justied because any surjective homomorphism f : A B is
clearly equivalent to the canonical quotient A A/ker(f).
Proposition 1.4.1. (1) There is an order-preserving correspondence
ideals I A equivalence classes of quotients A B.
The correspondence sends an ideal I to the equivalence class of the canonical
quotient A A/I, and the quotient f : A B to the ideal ker(f) A.
Date: Fall 2005.
1
2 THOMAS J. HAINES
(2) Fix an ideal I A. There is an order-preserving correspondence
ideals J A containing I ideals of A/I,
given by: send an ideal J I to its image J in A/I, and send an ideal
J

A/I to its pre-image under the canonical map A A/I.

1.5. Prime and maximal ideals. A domain is a ring A with the property: 1 ,= 0
and if x, y A and xy = 0, then x = 0 or y = 0. Examples are the integers Z, and
any ring of polynomial functions over a eld.
An ideal p A is prime if it is proper (p ,= A) and xy p implies x p or y p.
Thus, p is prime if and only if A/p is a domain.
An ideal m A is maximal if m ,= A and there is no ideal I satisfying m _ I _ A.
Equivalently, m is maximal if and only A/m is a eld. To see this, check that any
ring R having only (0) and R as ideals is a eld. Now m is maximal if and only if
A/m has no ideals other than (0) and A/m (Prop. 1.4.1), so the result follows on
taking R = A/m in the previous statement.
In particular, every maximal ideal is prime.
Proposition 1.5.1. Maximal ideals exist in any ring A with 1 ,= 0.
Proof. This is a standard application of Zorns lemma. Let S be the set of all
proper ideals in A, ordered by inclusion. Let T = I

A
be a chain of proper
ideals. Then the union
A
I

is an ideal which is an upper bound of T in S.

Hence by Zorns lemma S has maximal elements, and this is what we claimed.
Let us dene Spec(A) to be the set of all prime ideals of A, and Spec
m
(A) to be
the subset consisting of all maximal ideals. These are some of the main objects of
study in this course. The nomenclature spectrum comes from functional analysis,
and will be explained later on. Also, pretty soon we will give the set Spec(A) the
structure of a topological space and discuss the foundations of algebraic geometry...
1.6. Operations of contraction and extension. Fix a homomorphism : A
B. For an ideal I A dene its extension I
e
B to the ideal generated by the
image (I); equivalently, I
e
=
J
J where J B ranges over all ideals containing
the set (I).
Dually, for an ideal J B dene the contraction J
c
:=
1
(J), an ideal in A.
Note that J prime J
c
prime, so contraction gives a map of sets Spec(B)
Spec(A).
On the other hand, contraction does not preserve maximality: consider the con-
traction of J = (0) under the inclusion Z . Therefore, a homomorphism
: A B does not always induce a map of sets Spec
m
(B) Spec
m
(A).
As we will see later on, there is a natural situation where does induce a map
Spec
m
(B) Spec
m
(A): this happens if A, B happen to be nitely generated al-
gebras over a eld. This is quite important and is a consequence of Hilberts
Nullstellensatz, one of the rst important theorems we will cover.
1.7. Nilradical. Dene the nilradical of A by
rad(A) := f A [ f
n
= 0, for some n 1.
Check that rad(A) really is an ideal. Elements f satisfying the condition f
n
= 0
for some n 1 are called nilpotent.
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 3
Counterexample: For a non-commutative ring, it is no longer always true that
the sum of two nilpotent elements is nilpotent. The elements
_
0 1
0 0
_
and
_
0 0
1 0
_
,
in the ring M
2
(R) over a ring R with 1 ,= 0, are nilpotent, but their sum
_
0 1
1 0
_
is
not.
Lemma 1.7.1.
rad(A) =

pSpec(A)
p.
Proof. The inclusion is clear from the denition of prime ideal. For the reverse
inclusion, suppose f A is not nilpotent, i.e., suppose f
n
,= 0 for every n 1.
Let = I [ f
n
/ I, n 1. This set is non-empty (it contains the ideal
I = (0)) and this set has a maximal element (Zorn). Call it p. We claim that p is
prime (and this is enough to prove ). If not, choose x, y / p such that xy p.
Since p +(x) _ p and p +(y) _ p, we have f
n
p +(x) and f
m
p +(y) for some
positive integers n, m. But then f
n+m
p, a contradiction.
1.8. Radical of an ideal. Dene r(I) = f A [ f
n
I, for some n 1. Often,
we denote r(I) =

I. Check that

I is an ideal. Note that rad(A) =

_
(0). Also,
it is easy to check the following fact:
(1.8.1)

I =

pI
p.
Here p ranges over prime ideals containing I.
1.9. Jacobson radical. Dene the ideal rad
m
(A) =

mSpec
m
(A)
m.
Proposition 1.9.1.
rad
m
(A) = x A [ 1 xy is a unit for all y A.
Proof. : Say x rad
m
(A). If y is such that 1 xy is not a unit, then 1 xy m,
for some maximal ideal m. But then 1 m, which is nonsense.
: If x / m for some m, then (x) +m = A. But then 1 = z +xy, for some z m
and y A. So 1 xy is not a unit.
Exercise 1.9.2. Prove the following statements.
(i) r(r(I)) = r(I);
(ii) rad(A/rad(A)) = 0;
(iii) rad
m
(A/rad
m
(A)) = 0.
We call an ideal I radical if r(I) = I. So, (i) shows that the radical ideals are
precisely those of the form r(I), for some ideal I.
There exist ideals which are not radical. Consider (X
2
) C[X], and note that
r(X
2
) = (X).
Both rad(A) and rad
m
(A) have some meaning in algebraic geometry, which we
will return to shortly. Also, we will see that radical ideals play an important role
too.
4 THOMAS J. HAINES
1.10. Modules. Let M be an abelian group. Then the ring of group endomor-
phisms of M, denoted End(M), is a ring (in general non-commutative). Giving M
the structure of an A-module is precisely the same thing as giving a ring homomor-
phism
A End(M).
We have correspondences as in Prop. 1.4.1
submodules N M quotients M M

and
submodules N

M containing N submodules of M/N.

Also, we have the fundamental isomorphisms of A-modules
(i) If N, N

M, then
N +N

=
N

N N

;
(ii) If N

N M, then
M/N

N/N

= M/N.
1.11. NAK Lemmas. These lemmas are collectively called the Nakayama (or
Nakayama-Azumaya-Krull) lemmas. They concern nitely-generated A-modules.
We say M is nitely generated (abbrev. f.g.) if M is a quotient of the free A-
module A
n
for some positive integer n. Equivalently, there exist elements m
1
, . . . , m
n

M such that every element m M can be expressed in the form m = a
1
m
1
+ +
a
n
m
n
, for elements a
i
A. (Note the expression is always unique if and only if
A
n
= M, in which case we say M is nitely-generated and free.)
If I A is an ideal, dene IM M as the set of all nite linear combinations
IM = a
1
m
1
+ +a
r
m
r
[ a
i
I, m
i
M, i.
Check that IM is an A-submodule of M, which is the smallest submodule contain-
ing all the elements of form am, where a I, m M.
Proposition 1.11.1 (NAK). If M is f.g. and I rad
m
(A), then IM = M
M = 0.
Proof. Suppose M ,= 0 and choose a minimal set of generators m
1
, . . . , m
n
, for a
positive integer n. Using M = IM, write m
1
= a
1
m
1
+ +a
n
m
n
, for elements a
i

I rad
m
(A). Observe that the element (1a
1
)m
1
is contained in Am
2
+ +Am
n
,
and since 1 a
1
is a unit in A, so is the element m
1
. This means that m
2
, . . . , m
n
generate M, violating the minimality and giving us a contradiction of the hypothesis
M ,= 0.
Corollary 1.11.2. Suppose I rad
m
(A). If N M is a submodule, and M is
f.g., then M = N +IM M = N.
Proof. Apply Prop. 1.11.1 to M/N.
Now we specialize to the case where A is a local ring. Recall that (A, m) is local
if m is the unique maximal ideal of A. In this case A m = A

, i.e. the units

in A are precisely the elements outside of m. Conversely, if A has a ideal I such
that A I A

, then A, I is local. Indeed, let m

be a maximal ideal that is not

contained in I, and choose x m

I. This is impossible since x / I x A

.
Thus I is maximal, and is the unique such.
Examples of local rings
Any eld
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 5
The p-adic numbers Z
p
(well come back to these)
Power series rings k[[X]], where k is a eld (ditto).
Of course, for local rings rad
m
(A) = m, so the NAK lemma becomes even simpler.
Here is a consequence:
Corollary 1.11.3. Suppose M is f.g. over a local ring (A, m), and write k :=
A/m for the residue eld. If x
1
, . . . , x
n
generate M/mM as a k-vector space, then
x
1
, . . . , x
n
generate M as an A-module.
Proof. Appy Cor. 1.11.2 to N = Ax
1
+ +Ax
n
and I = m.
2. Lecture 2
2.1. Improved NAK lemma. For a f.g. A-module M we can use the follow-
ing determinant trick (essentially the Cayley-Hamilton theorem generalized from
elds to commutative rings):
Lemma 2.1.1 (Cayley-Hamilton). Let be an A-module endomorphism of M such
that (M) IM, for an ideal I A. Then satises an equation of the form

r
+a
r1

r1
+ +a
0
= 0,
where a
i
I for all i.
Proof. Let x
1
, . . . , x
r
generate M. We may write (x
i
) =

j
a
ij
x
j
, for elements
a
ij
I. Thus for all i

j
(
ij
a
ij
)x
j
= 0,
where
ij
is the Kronecker delta. Multiplying the matrix (
ij
a
ij
) on the left by its
adjoint, we get det(
ij
a
ij
) annihilates each x
i
, hence is the zero endomorphism
of M. Expanding out the determinant gives the desired equation.
Remark. We used the Cramers Rule adj(X) X = det(X) I
n
for any nn matrix
X over a commutative ring A. This can be deduced from the case where A is a
eld. Indeed, the formula is equivalent to n
2
polynomial relations in the entries
of X. It is enough to prove these relations hold in the polynomial ring Z[X
ij
]
in n
2
indeterminates X
ij
, and those relations follow in turn from the relations in
the rational function eld (X
ij
). This kind of trick is quite common to prove
statements for commutative rings which are already known to hold over elds. For
instance, use it to do the following exercise.
Exercise 2.1.2. Let A be a commutative ring. Show that for X, Y M
n
(A),
det(XY ) = det(X)det(Y ). Deduce from this and Cramers rule that X has an
inverse in M
n
(A) if and only if det(X) A

.
Corollary 2.1.3 (Improved NAK). If M is f.g. and IM = M, then there exists
a A with a 1mod I, and aM = 0.
Proof. Take = id in Lemma 2.1.1, and note that a := 1+a
r1
+ +a
0
works.
Note that this corollary gives another proof of Prop. 1.11.1: I rad
m
(A) means
that a A

, and so aM = 0 implies M = 0.
6 THOMAS J. HAINES
2.2. Some applications of NAK. Here we give two quick applications of the
NAK lemmas.
1st application.
Proposition 2.2.1. Suppose f : M M is a surjective A-module endomorphism
of a f.g. A-module M. Then f is injective, hence is an automorphism.
Proof. Using f we dene on M the structure of an A[X]-module by setting X m =
f(m). By Improved NAK applied to A[X] and I = (X) there exists Y A[X],
such that (1 + Y X)M = 0. Now let u ker(f). We have 0 = (1 + Y X)(u) =
u +Y f(u) = u. Hence ker(f) = 0, as desired.
The following related result is actually proved using a dierent argument. (If
you are not already familiar with Noetherian rings, we will return to these again
later.)
Exercise 2.2.2. Suppose A is a Noetherian ring. Then any surjective ring homo-
morphism f : A A is injective, hence an automorphism.
The following exercise can be proved using the proposition.
Exercise 2.2.3. Let A be a commutative ring, and suppose that as A-modules,
A
n
= A
m
. Prove that n = m.
2nd application.
Recall that an A-module P is projective if it has the following property: let
f : M N be a surjective morphism, and let : P N be any morphism; then
there exists a morphism : P M such that f = . In other words, the
natural map Hom
A
(P, M) Hom
A
(P, N) induced by f is surjective.
It is easy to prove that P free P projective. Also, it is easy to show the
following result.
Proposition 2.2.4. P is projective if and only if it is a direct summand of a free
module.
If you havent seen these statements before, you should try to prove them your-
self, but you can also look them up in N. Jacobsons book, Basic Algebra II (or in
pretty much any book on basic algebra).
We have the following sharper result when (A, m) is local, our second application
of the NAK lemma.
Proposition 2.2.5. Let M be a f.g. projective module over a local ring (A, m).
Then M is free.
Proof. This result actually holds without the assumption f.g. see [Mat2], Thm.
2.5. We shall not need it in that generality.
Choose a minimal generating set m
1
, . . . , m
n
for M, and dene the surjective
map : F = A
n
M by (a
1
, . . . , a
n
) a
1
m
1
+ + a
n
m
n
. Let K := ker().
The minimal basis property shows that

i
a
i
m
i
= 0 a
i
m, i.
Thus K mF. Because M is projective, there exists : M F such that
F = K (M), and it follows that K = mK. Since K is a quotient of F, it is also
f.g. over A, hence by NAK, K = 0. This shows F

= M, so M is free.
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 7
2.3. Special kinds of rings. A Euclidean domain is a ring where a division al-
gorithm holds (I am not going to make this precise). Examples are Z, and k[X],
where k is any eld.
A PID is a domain wherein every ideal is principal, i.e., generated by a single
element.
A UFD is a ring wherein every non-zero, non-unit element can be written as a
unit times a product of irreducible elements, in an essentially unique way. Again,
I am not going to make this precise.
The following implications hold: Euclidean PID UFD. Further, if A is a
UFD, then A[X] is also (Gauss lemma); but A[X] need not be Euclidean (resp.
PID) even if A is. Can you give some examples showing what goes wrong?
2.4. Classifying the prime/max ideals in ring. Consider the ring C[X]. This is
a Euclidean domain, hence as above it is a PID hence a UFD. Hence, C[X
1
, . . . , X
n
]
is also a UFD for any n 1. This gives rise to some natural questions:
What are the prime/maximal ideals in C[X
1
, . . . , X
n
]?
What are the irreducible elements in the UFD C[X
1
, . . . , X
n
]?
Consider again the case C[X]. The non-zero prime ideals are generated by the
irreducible polynomials. By the fundamental theorem of algebra, these are precisely
those of the form X, where C. Therefore, as a set, we have an identication
Spec
m
C[X] = C.
To fully understand what we can say about C[X
1
, . . . , X
n
], we need algebraic
geometry.
2.5. Maximal ideals in C[X
1
, . . . , X
n
] rst step. Let = (
1
, . . . ,
n
) C
n
.
Evaluation at this point, i.e. the map f f(
1
, . . . ,
n
) C, gives us a surjective
homomorphism
ev

: C[X
1
, . . . , X
n
] C.
The kernel is a maximal ideal, call it m

.
Claim: m

= (X
1

1
, . . . , X
n

n
).
Proof. The inclusion is clear. If f m

, then write it as a polynomial in

X
n

n
, with coecients in C[X
1
, . . . , X
n1
]. The constant (i.e. deg
Xn
= 0) term
is a polynomial in X
1
, . . . , X
n1
vanishing at (
1
, . . . ,
n1
). By induction, that
constant term is in (X
1

1
, . . . , X
n1

n1
), so were done.
Deeper fact well soon show (from Hilberts Nullstellensatz): All maximal
ideals of C[X
1
, . . . , X
n
] are of the form m

. Hence, like for C[X], we will have an

identication
Spec
m
C[X
1
, . . . , X
n
] = C
n
.
2.6. Zariski topology. Let A be a ring. We are going to put a topology on the
set SpecA (well put the subspace topology on the subset Spec
m
A). From the above
remarks, this will actually dene a new and interesting topology on the familiar set
C
n
, which is very dierent from the standard metric topology.
8 THOMAS J. HAINES
3. Lecture 3
3.1. Denition of Zariski topology. Recall that to dene a topology on a set X
is to specify a collection U of subsets of X (called open) satisfying the following
axioms:
, X U
Let I be any index set. If for all i I, U
i
U, then
iI
U
i
U
If U, V U, then U V U.
To determine the topology, it is enough to specify the closed sets, which by
denition are the complements of the open sets. (I leave it to you to formulate the
axioms the closed sets must verify see below.) That is how we will dene the
topology on Spec(A).
Namely, for any subset E A, dene
V (E) := p Spec(A) [ p E.
Now we show that the subsets V (E) Spec(A) are the closed sets in a topology,
which we call the Zariski topology on Spec(A). Parts (ii)-(iv) of the following
lemma accomplish this. The other parts are also useful.
Lemma 3.1.1. The following properties hold.
(i) V (E) = V (E), where E is the ideal generated by E.
(ii) V () = V (0) = Spec(A), and V (1) = .
(iii) V (I) V (J) = V (I J) = V (IJ), where IJ is the ideal generated by the
set of products xy with x I and y J.
(iv) Let I be any index set. Then
iI
V (a
i
) = V (

i
a
i
) = V (
i
a
i
).
(v) I J = V (J) V (I).
(vi) V (I) = V (r(I)).
(vii) V (I) V (J) r(J) r(I).
Note that (ii)-(iv) show we get a topology, whereas (i) and (v)-(vii) show that
I V (I) gives an order-reversing bijective correspondence
radical ideals in A closed subsets in Spec(A).
Proof. Parts (i),(ii), and (v) are clear. Parts (vi) and (vii) follow using (1.8.1).
Part (iv) is also easy from the denitions.
Let us prove (iii). The inclusions V (I) V (J) V (I J) V (IJ) are easy: use
(v) applied to the inclusions I J I (resp. I J J) and IJ I J. Now to
prove V (IJ) V (I) V (J), assume p V (IJ), i.e., p IJ. If p _ I and p _ J,
then there exist x I p and y J p; but note that xy IJ p. This is
nonsense since p is prime.
3.2. Some further remarks about the Zariski topology. The following re-
marks help us get a grip on the strange properties of the Zariski topology.
If p is a prime ideal, it is also a point in the topological space Spec(A). When we
think of it as a point, we often write it as p
x
(the symbol x is often used to denote
a point in a space, thus the subscript reminds us to think of the ideal as a point in
the space). With this notation, the following equation describes the closure of the
point p
x
:
(3.2.1) p
x
= V (p
x
).
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 9
Let us prove this. The closure is the intersection of all closed sets containing p
x
,
that is, the closure is

Ip
V (I) = V (

Ip
I) = V (p).
This is striking: a point in our space Spec(A) is not usually a closed set! In fact,
it follows that
(3.2.2) p
x
is a closed point if and only if p
x
is a maximal ideal.
The space Spec(A) is not Hausdor, but is T
0
: for any two distinct points x, y,
there exists an open U containing x but not y, or vice-versa. (Prove this!)
If A is a domain, then Spec(A) = 0 (and the ideal (0) is called the generic
point : it is a single point, but it is actually dense in the whole space!).
Spec(A) is compact: any cover by open subsets has a nite sub-covering.
Proof. Suppose Spec(A) =

iI
U
i
, where U
i
is the open complement of a closed
set, call it V (a
i
). Taking complements, we nd

i
V (a
i
) = = V (1)
= V (

i
a
i
) = V (1)
= r(

i
a
i
) = (1)
= 1

i
a
i
.
Thus, on renumbering, we may assume 1

r
i=1
a
i
, which in turn entails

r
i=1
V (a
i
) = ,
i.e. U
1
, . . . , U
r
cover Spec(A).
Exercise 3.2.1. At this point, it is instructive to work through exercises 15-21,
Chapter 1, of the book by Atiyah-Macdonald.
3.3. Integral extensions. Recall that our immediate goal is to classify all the
maximal ideals in a polynomial ring such as C[X
1
, X
n
]. We will do this us-
ing Hilberts Nullstellensatz. Our approach to that theorem is to rst prove
Noethers Normalization Theorem. That requires us to rst explain the basic
facts about integral extensions of rings.
Suppose A B is a subring. We say b B is integral over A if b satises a
monic polynomial of the form b
n
+ a
n1
b
n1
+ + a
0
= 0, with a
i
A for all
i = 0, . . . , n 1.
Proposition 3.3.1. The following properties are equivalent.
(a) b B is integral over A;
(b) b C B, for some subring C containing A, which is nitely generated
as an A-module.
10 THOMAS J. HAINES
Proof. (a) = (b): Take C = A[b], the subring generated by Aand b. By induction
on r (the case r = 0 following from (a)), check that b
n+r
A + Ab + + Ab
n1
,
for all r 0. This proves that C = A[b] = A + Ab + + Ab
n1
, hence is f.g. as
an A-module.
(b) = (a): Apply Lemma 2.1.1 with M = C, I = A, and = multiplication by b.

For a subring A B, dene

A := b B [ b is integral over A. This set

A is
called the integral closure of A in B.
Corollary 3.3.2.

A is a subring of B.
Proof. Let x, y

A. We need to show xy, x y

A. Let A[x, y] be the subring of
B generated by A and the elements x, y. Using Proposition 3.3.1 we see that A[x]
is f.g. as an A-module, and A[x, y] is f.g. as an A[x]-module. Then the subring
A[x, y] is f.g. as an A-module. Since A[x, y] contains xy and x y, Proposition
3.3.1 implies that these elements are integral over A, and were done.
Remark: Note that a similar argument shows: if the elements x
1
, . . . , x
n
B are
integral over A, then the ring A[x
1
, . . . , x
n
] is f.g. as an A-module.
Exercise 3.3.3. Consider the complex numbers = e
2i/3
and = e
2i/4
. These
are both integral over Z. Find the minimal polynomial for +, i.e., the minimal-
degree monic polynomial F(X) with Z-coecients such that F( +) = 0.
Corollary 3.3.4. Consider ring inclusions A B C. If C is integral over B
and B is integral over A, then C is integral over A.
Proof. Let c C, and suppose it satises a polynomial relation of form c
n
+
b
n1
c
n1
+ + b
0
= 0. Then A[b
0
, . . . , b
n1
, c] is a f.g. A-module (check this
you will need to invoke the remark above). Thus c is integral over A by Proposition
3.3.1.
In particular, we see that

A =

A.
We shall prove much more about integral extensions later. But to nish our
preparations for Noether Normalization, we content ourselves with just one
more thing.
Lemma 3.3.5. Suppose A B are domains, with B integral over A. Then A is a
eld if and only if B is a eld.
Proof. (): Assume b ,= 0 and suppose b
n
+a
n1
b
n1
+ +a
0
= 0 is a minimal
degree monic polynomial satised by b. Then a
0
,= 0, so a
1
0
A. But then
b
1
= a
1
0
(b
n1
+a
n1
b
n2
+ +a
1
) B,
which shows that B is a eld.
(): Assume a ,= 0. Then a
1
B implies that there exists a relation, with all

i
A, of form
a
n
+
n1
a
n+1
+ +
0
= 0.
Multiplying this by a
n1
, we deduce that a
1
A, and so A is a eld.
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 11
3.4. Aside: Beginning facts about integrally closed domains. We now pause
a moment to briey discuss integrally closed domains. Assume A is a domain, with
eld of fractions K. In this case

A K is called simply the integral closure of
A (in its fraction eld). We say A is integrally closed (or normal) if

A = A. The
following lemma provides lots of examples of integrally closed domains.
Lemma 3.4.1. Any UFD is integrally closed.
Proof. Any element in K

may be written in the form

a
b
for elements a, b A0.
By cancelling common irreducible factors, we may assume that a and b have no
factors in common.
Now the integrality condition yields, for some elements
i
A, the equation
(
a
b
)
n
+
n1
(
a
b
)
n1
+ +
0
= 0,
which implies after clearing denominators
a
n
+
n1
a
n1
b + +
0
b
n
= 0.
But then any irreducible factor dividing b also divides a
n
and hence also a, a
contradiction. So b is a unit and
a
b
A.
Question 1: Do there exist integrally closed domains which are not UFDs? The
answer is YES; we shall show later that if A is a Noetherian domain, then A is
a UFD if and only if it is integrally closed, and every fractional ideal is principal.
Well come back to this, but for the moment let me highlight one consequence: for
Dedekind domains (which are automatically integrally closed), we have UFD
every fractional ideal is principal PID. This is important in number theory.
Question 2: When is a domain integrally closed? We will give one complete answer
to this question in this course (but there are other, more useful answers). Here is
an interesting result we will prove later. Assume k is a eld, and char(k) ,= 2. Let
f(X
1
, . . . , X
n
) k[X
1
, . . . , X
n
], and suppose f is not a square. Then
k[X
1
, . . . , X
n
, Z]/(Z
2
f) is integrally closed f is square-free.
Recall that f is square-free means that it is not divisible by the square of any
irreducible element.
3.5. Algebraic independence. Let k be any eld, and let R be a k-algebra. By
denition, this means that R is a ring containing k. More generally we can dene
the notion of an A-algebra for any commutative ring A. This is just a ring B
together with a ring homomorphism f : A B. Then we may dene on B the
structure of an A-module by setting a b := f(a)b, where the multiplication on
the right is that in B. The A-module structure on B is compatible with the ring
structure on B in an obvious sense (formulate it!).
Return now to k R. We say elements u
1
, . . . , u
n
R are algebraically indepen-
dent over k if there is no non-zero polynomial F(U
1
, . . . , U
n
) in the polynomial ring
k[U
1
, . . . , U
n
] such that F(u
1
, . . . , u
n
) = 0. In that case, the map U
i
u
i
induces
an isomorphism of rings k[U
1
, . . . U
r
]

= k[u
1
, . . . , u
n
].
12 THOMAS J. HAINES
3.6. Noether Normalization. From now on, we will often use the following ter-
minology: Suppose B is an A-algebra. We say B is module nite, or simply nite
over A if B is nite-generated as an A-module. We say B is a f.g. A-algebra if
B can be written as a quotient ring of A[Y
1
, . . . , Y
r
], for some nite number of
variables Y
1
, . . . , Y
r
. In that case, we often write B = A[y
1
, . . . , y
r
], where here the
y
i
are the images of the Y
i
under the quotient map.
Theorem 3.6.1 (Noether Normalization). Let k be a eld, and suppose R is a f.g.
k-algebra, R = k[u
1
, . . . , u
n
]. Then there exist algebraically independent elements
x
1
, . . . , x
t
(with t n), such that R is module nite over k[x
1
, . . . , x
t
].
Moreover, t < n unless u
1
, . . . , u
n
are algebraically independent.
I learned the following proof from Mel Hochster many years ago.
Proof. We use induction on n. If n = 0, there is nothing to prove. Suppose n 1
and assume the result is true for algebras generated by n1 elements. If u
1
, . . . , u
n
are algebraically independent, there is nothing to prove. So WLOG
1
there exists
F ,= 0 with F(u
1
, . . . , u
n
) = 0. WLOG, U
n
occurs in F.
Choose a positive integer N with N > deg(F).
We have R = k[v
1
, . . . , v
n
], where by denition v
i
:= u
i
u
N
i
n
for 1 i n 1,
and v
n
:= u
n
. Dene new indeterminates V
1
, . . . , V
n
, and dene G k[V
1
, . . . , V
n
]
by
G(V
1
, . . . , V
n
) = F(V
1
+V
N
n
, . . . , V
n1
+V
N
n1
n
, V
n
).
Note that G(v
1
, . . . , v
n
) = 0.
Claim: G = (non-zero scalar) (monic in V
n
with coecients in k[V
1
, . . . , V
n1
]).
Once we establish the claim, we will know that R is module nite over k[v
1
, . . . , v
n1
],
which by induction is module nite over k[x
1
, . . . , x
t
], where t n 1. Thus we
will be done.
Proof of Claim: Letting stand for the n-tuple of non-negative integers (
1
, . . . ,
n
),
we write
F =

U
1
1
U
n
n
.
Here we let range over the n-tuples with

,= 0. Thus
G =

(V
1
+V
N
n
)
1
(V
n1
+V
N
n1
n
)
n1
V
n
n
=

[V
()
n
+ lower terms in V
n
with coes in k[V
1
, . . . , V
n1
]],
where () :=
n
+
1
N +
2
N
2
+ +
n1
N
n1
.
Choose now a

such that (

) is maximal. Then the highest deg

Vn
term that
appears anywhere is

V
(

)
n
.
This term cant be cancelled; in fact there is only one

which maximizes the

function . Why? Because of the uniqueness of base N expansions! This completes
the proof of the claim.
1
Without Loss Of Generality.
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 13
3.7. Aside: Geometric meaning of Noether Normalization. For those of
you who already know some algebraic geometry, the following is the geometric
reformulation of Theorem 3.6.1.
At this point we can give preliminary denitions of ane variety and ane
scheme. Let A be any ring. Then we will call the topological space Spec(A) an
ane algebraic scheme. Now let R be a f.g. k-algebra. Then we will call the
topological space Spec
m
(R) an ane algebraic variety over k. In both cases, the
complete denition of scheme/variety will be the topological space endowed with
some extra structure, namely a sheaf of rings on it (stay tuned for more...).
Theorem 3.7.1. Let X = Spec
m
(R) be an ane variety over a eld k. Let A
t
k
denote t-dimensional ane space, i.e. A
t
k
:= Spec(k[X
1
, . . . , X
t
]). Then there is a
nite surjective morphism of algebraic varieties
X A
t
k
where t = dim
k
(X).
The theorem states that every ane algebraic variety is almost an ane space
k
t
.
We will dene algebraic varieties and all the necessary concepts we need to
understand these statements later.
3.8. Hilbert Nullstellensatz. Now we apply this to get
Corollary 3.8.1 (Nullstellensatz - weak form). (1) Let R be a f.g. k-algebra.
Assume R is a eld. Then R is a nite eld extension of k.
(2) If k = k (i.e. k is algebraically closed), then moreover R = k.
Proof. For part (1), Theorem 3.6.1 says that R is module nite over a domain of
the form k[x
1
, . . . , x
t
]. The latter must be a eld, by Lemma 3.3.5. But then t = 0
(why?), and thus R is module nite over the eld k, as desired.
For part (2), note that k = k implies that there are no non-trivial nite extensions
of k, so that R = k is forced.
3.9. Maximal ideals of C[X
1
, . . . , X
n
] - nal step. The eld C is algebraically
closed, so we may apply the above corollary to prove that every maximal ideal is
of the form m

, for some = (
1
, . . . ,
n
) C
n
. Let m be a maximal ideal, and
let R := C[X
1
, . . . , X
n
]/m. Then by the above Corollary 3.8.1, we know that the
inclusion C R is actually an isomorphism. Dene
i
to the be complex number
which is the image of X
i
under the homomorphism
C[X
1
, . . . , X
n
] R

= C.
The above map can be identied with the evaluation map ev

, and thus m = m

.
3.10. Further consequences of the Nullstellensatz.
Corollary 3.10.1. Let : R S be a homomorphism of f.g. k-algebras. Let
m Spec
m
(S). Then m
c
=
1
(m) Spec
m
(R).
In particular, the map

: Spec(S) Spec(R) given by p p

c
, takes Spec
m
(S)
into Spec
m
(R).
14 THOMAS J. HAINES
Proof. The map induces an inclusion R/m
c
S/m of k-algebras. Since S/m is a
eld, the Nullstellensatz implies it is a nite extension of k, and thus it is necessarily
module nite over R/m
c
. But then this latter domain is itself a eld, by Lemma
3.3.5, and thus m
c
is maximal.
Exercise 3.10.2. Show that the map

is continuous (see Atiyah-Macdonald,

Chapter 1, #21 (i)).
Question: If k = k, we can now identify
Spec
m
(k[X
1
, . . . , X
n
]) = k
n
.
The Zariski topology on the left hand side thus gives us a new topology on k
n
.
What does this topology look like? That is the subject we will look at next.
4. Lecture 4
4.1. Algebraic Zeros Theorem. Let k be a eld, k an algebraic closure of k. Let
k[X
1
, . . . , X
n
] be a subset. We call = (
1
, . . . ,
n
) k
n
an algebraic zero of
if f() = 0 for all f .
Theorem 4.1.1 (Algebraic zeros theorem). Write k[X
1
, . . . , X
n
] = k[X] for short.
(i) If has no algebraic zeros, then = (1) = k[X].
(ii) If f k[X] vanishes at every algebraic zero of , then f r().
Proof. (i). Write I := . If 1 / I, then I m, for a maximal ideal m. Since
k[X]/m is a nite extension of k (Corollary 3.8.1), there is an embedding of elds
k[X]
m
k.
Let
i
:= image of X
i
. But then all elements of m hence also I vanish at , a
contradiction. Thus, I = (1).
(ii). Inside k[X, Y ] consider 1Y f(X). This set has no algebraic zeros (why?).
So by part (i) there exist functions Q(X, Y ), g
i
(X, Y ) k[X, Y ] and h
i
(X) for
i = 1, . . . , r such that
r

i=1
g
i
(X, Y )h
i
(X) +Q(X, Y )(1 Y f(X)) = 1.
Specializing Y = f(X)
1
, we get

i
g
i
(X, f(X)
1
)h
i
(X) = 1.
Now we multiply by some high power f
N
(X) to clear the denominators to nd that
f
N

i
h
i
k[X] ,
as desired.
Remark. Note that (i) says: if f k[X
1
, . . . , X
n
] is not a unit, then it has at least
one zero k
n
. You already knew this for n = 1: any polynomial in k[X] which
is not a unit, has a zero in the eld k.
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 15
4.2. Consequences of Nullstellensatz and Algebraic Zeros Theorem. 1st
application.
Proposition 4.2.1. Let R be a nitely generated k-algebra, and I R an ideal.
Then
r(I) =

max m I
m.
In particular,
(1) For any prime ideal p, we have p =
mp
m.
(2) rad(R) = rad
m
(R).
Proof. Write R = k[u
1
, . . . , u
n
]; there is a surjection
: k[X
1
, . . . , X
n
] R
given by sending X
i
u
i
for all i. By the correspondence of ideals between I
and R with those between J := I
c
and k[X, . . . , X
n
] (under which prime/max
ideals correspond to prime/max ideals), it is enough to prove the proposition for
R = k[X
1
, . . . , X
n
]. But this case will follow from Theorem 4.1.1, (ii). We start by
verifying the following: if
f

max m J
m,
then f vanishes at every algebraic zero of J. Lets check this statement. If is an
algebraic zero of J, then J is in the kernel of
ev

: k[X
1
, . . . , X
n
] k
and this kernel is itself a maximal ideal m, since the image of ev

is a domain which
is an integral extension of k (being contained in k) hence by Lemma 3.3.5 is a eld.
Since f m, we see that f vanishes at , as desired. Thus by (ii) of Theorem 4.1.1,
we see f r(J).
The conclusion of Proposition 4.2.1 is not true in general for all commutative
rings.
Exercise 4.2.2. (1) Find a domain A and a proper ideal I A such that
r(I) _

mI
m.
(2) Consider a local k-algebra (A, m). If A is nitely generated as a k-algebra,
what can you say about the prime ideals of A?
(3) Suppose (A, m) is a local k-algebra. Show that the following are equivalent:
(i) A is a f.g. k-algebra;
(ii) A is an Artin ring and A/m is a nite extension of k;
(iii) A is Artin and each m
n
/m
n+1
is nite-dimensional as a k-vector space;
(iv) A is nite dimensional as a k-vector space.
Hint: You might want to make use of the material from Chapter 8 of Atiyah-
Macdonald.
2nd application.
Proposition 4.2.3. Let R be a f.g. k-algebra. Then Spec
m
(R) Spec(R) is dense.
16 THOMAS J. HAINES
Proof. There is a basis of open subsets of Spec(R) given by the subsets
D(f) := q [ f / q.
[Aside: to check D(f)
fR
indeed form a basis for a topology, we need to check
that if q Spec(R) V (I), then there exists a D(f) with q D(f) Spec(R)
V (I). But just take f to be any element in I q.]
We will show: if p D(f), then there exists a maximal ideal m with
m p;
m D(f).
But this is obvious from Prop. 4.2.1, (1). Clearly this also shows Spec
m
(R) is dense
in Spec(R).
Exercise 4.2.4. Find a ring A such that Spec
m
(A) is not dense in Spec(A).
4.3. Closed subsets in k
n
and radical ideals. In this subsection we assume
k = k, and let R = k[X
1
, . . . , X
n
].
We have established an identication of sets Spec
m
(R) = k
n
(the same proof we
gave for k = C works, as we only used the property that C is algebraically closed).
The left hand side is given the Zariski topology: more precisely, the subspace
topology it inherits from the Zariski topology on Spec(R). What does this mean
concretely? First we note that under the identication Spec
m
(R) = k
n
, we have:
Z k
n
is closed Z = Spec
m
(R) V (I), for some (radical) ideal I R
Z = k
n
[ f() = 0, f I =: Z(I).
This leads us to dene
For a subset Y k
n
, let J(Y ) := f R [ f(y) = 0, y Y ;
For an ideal I R, let Z(I) := k
n
[ f() = 0, f I.
Note that J(Y ) is always a radical ideal, and Z(I) is always a closed subset.
Theorem 4.3.1 (Classical Nullstellensatz 1st form). Let I R be any ideal, and
let Y k
n
be any subset, with Zariski-closure Y . Then we have:
(a) J(Z(I)) = r(I).
(b) Z(J(Y )) = Y .
In particular, I Z(I) gives an order-reversing bijection
radical ideals in R Zariski closed subsets in k
n
,
with inverse Z J(Z).
Proof. (a): The Algebraic zeros theorem (ii) gives the non-trival inclusion .
(b): Clearly Z(J(Y )) Y , hence Z(J(Y )) Y .
By the Lemma below, it is enough to show that equality holds after we apply
J(). But then the equality we want can be derived using part (a) (using that J(Y )
is radical):
JZ(J(Y )) = J(Y ) = J(Y ).
Here, the last equality holds because polynomial functions are continuous as func-
tions k
n
k (check this!see Exercise below).
Lemma 4.3.2. Suppose for Zariski closed subsets Y
1
Y
2
we have J(Y
1
) = J(Y
2
).
Then Y
1
= Y
2
.
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 17
Proof. Suppose not. Then there is a point Y
1
Y
2
. There is a principal open
subset D(f) = x k
n
[ f(x) ,= 0, such that D(f), but D(f) Y
2
= . Then
we see that f vanishes on Y
2
, i.e. f J(Y
2
). Since the latter ideal is also J(Y
1
) by
hypothesis, we see that f also vanishes on Y
1
, hence on , a contradiction.
Exercise 4.3.3. Give k and k
n
the Zariski topologies. Show that all polynomials
are continuous as functions k
n
k.
4.4. Examples. We can draw some pictures of Z(I) for various ideals I C[X, Y, Z].
X
2
Y
2
Z = 0: saddle point at origin.
X
4
+ (Y
2
X
2
)Z
2
= 0: gure-8 cones along Z-axis emanating from origin.
X
2
+Z
3
= 0: tent draped over Y -axis through origin.
I = (XY, Y Z), i.e., both XY = 0 and Y Z = 0: union of the Y -axis and the
XZ-plane.
To see this last example, note that Z(I) = Z(XY ) Z(Y Z), and Z(XY ) is
the union of the Y Z-plane and the XZ-plane. Similarly, Z(Y Z) is the union of
the XZ-plane and the XY -plane. Hence the intersection Z(I) is the union of the
Y -axis and the XZ-plane.
5. Lecture 5
5.1. Classical Nullstellensatz for reduced f.g k-algebras. We call a ring A
reduced provided rad(A) = 0; in other words, A has no non-zero nilpotent elements.
For example, if J k[X] = k[X
1
, . . . , X
n
] is a radical ideal, then the quotient
R = k[X]/J is reduced.
Now we can give a more complete version of the classical Nulltstellensatz, this
time for arbitrary reduced f.g. k-algebras in place of k[X] = k[X
1
, . . . , X
n
]. Let R
and J be as in the previous paragraph. Again assume k = k. Let V := Z(J) k
n
,
a Zariski closed subset. As before, the statement should concern radical ideals of
R and Zariski-closed subsets in V .
We can extend our previous denitions of the maps Z() and J() as follows.
Via contraction, the ideals I R correspond bijectively to ideals I
c
k[X] which
contain J, and radical ideals correspond to radical ideals. Therefore, we may set
Z(I) := Z(I
c
) k
n
. Note that I
c
J implies (by Theorem 4.3.1) that Z(I)
Z(J) =: V . Thus I R gives us a Zariski-closed subset of k
n
which is contained
in the Zariski-closed subset V k
n
.
In the reverse direction, any Zariski-closed subset Y V is also Zariski-closed
as a subset of k
n
, and J(Y ) is a radical ideal of k[X] which contains J(V ) = J, by
Theorem 4.3.1. We can therefore regard J(Y ) as a radical ideal of R.
The two operations I Z(I) and Y J(Y ) are mutually inverse (just use
(a),(b) of Theorem 4.3.1 to see this). We have proved:
Theorem 5.1.1 (Classical Nullstellensatz nal form). Let J k[X] be a radical
ideal and R := k[X]/J. The rule I Z(I) gives an order-reversing bijection
radical ideals in R Zariski closed subsets in V ,
with inverse Z J(Z).
5.2. Remarks on irreducible sets and dimension. What is the dimension of
an arbitrary topological space X? Here we give one reasonable denition that works
in algebraic geometry (but not in classical geometry), using the notion of irreducible
subset.
18 THOMAS J. HAINES
We call a a topological space X irreducible provided the following equivalent
conditions are satised:
Any two non-empty open subsets in X intersect;
Every non-empty open subset in X is dense in X;
X is not the union of two proper closed subsets.
Note that this concept is not very interesting for Hausdor spaces: a non-empty
Hausdor space is irreducible if and only if it consists of a single point.
If Y X is a subset, we give it the subspace topology, and then we say Y is
irreducible, if it is irreducible once it is given that topology.
We now dene the Krull dimension of X to be
Krulldim(X) = supn N [ closed irreducible subsets Y
0
_ Y
1
_ _ Y
n
.
N.B. In this denition, we require that the subsets Y
i
are non-empty (the empty
set is irreducible and we dont want to allow it in a chain).
Note that Krulldim(Haussdor space) = 0. However, this notion of dimension
works well for algebraic geometry, as we shall shortly see.
Note that any irreducible subset Y X is contained in a maximal irreducible
subset, which is closed. Why? We need to notice two things:
(i) Zorns lemma applied to = irred. Y

Y shows that this collection

possesses maximal elements (for a chain Y

, observe that

is irre-
ducible);
(ii) The closure of an irreducible set is irreducible (assume Y is irreducible; if
Y = F
1
F
2
with F
i
closed proper subsets of Y , then Y is the union of
the closed and proper subsets Y F
i
, violating our assumption that Y is
irreducible).
Thus we may speak of the maximal irreducible subsets (which are closed) in X;
we call them the irreducible components of X.
What are the irreducible components of X = Spec(A), for a ring A?
Proposition 5.2.1. Let A ,= 0 be a ring.
(i) The non-empty closed irreducible subsets are those of the form V (p), where
p is a prime ideal.
(ii) The irreducible components are the V (p) for p a minimal prime ideal.
In particular, the fact that irreducible components exist for Spec(A) implies that
minimal prime ideals exist in any ring A ,= 0 (compare with Atiyah-Macdonald,
Ch. 1, Ex. 8).
Proof. (i): Let I be a radical ideal. We need to show that V (I) is irreducible
i I is prime. If I is not prime, then choose x, y / I such that xy I. Let
a := I + (x) _ I, and b := I + (y) _ I. Then we see that ab I, and so r(ab) I
and V (I) V (ab) = V (a) V (b). So V (I) = (V (I) V (a)) (V (I) V (b)), a
union of proper closed subsets. This shows that V (I) is not irreducible.
Conversely, assume V (I) = V (a) V (b) is a union of proper closed subsets, for
radical ideals a, b. Then a b is still radical, and so the equality V (I) = V (a b)
shows that I = a b. But this shows that I is not prime. Indeed, since I _ a, b,
we have elements x a I and y b I. But then xy a b = I, which means
I is not prime.
(ii): This follows using (i) and the fact that V (p) V (q) p q.
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 19
Remarks: (1) From previous work, we know that p
x
= V (p). This means that
the point p
x
is dense in the closed irreducible set V (p). If p is minimal, we call p
x
the generic point of the irreducible component V (p).
(2) In part (i) above, we actually proved the following fact (see Atiyah-Macdonald,
Ch.1, Exer.19): Spec(A) is irreducible i the nilradical of A is prime. Indeed, take
I = r((0)), and note that V (I) = Spec(A).
The above considerations lead us to give the following denition of dimension
(sometimes called Krull dimension) for a ring A ,= 0:
dim(A) = supn N [ prime ideals p
0
_ p
1
_ _ p
n
.
It is not at all obvious that dim(A) < , and indeed sometimes dim(A) =
, even if A is assumed to be Noetherian (Nagatas example)! However, if A is
Noetherian and local, then it turns out that its dimension is always nite. It is also
nite for certain nice rings, such as polynomial rings. Furthermore, the notion is
intuitively correct because, as we shall see, dimC[X
1
, . . . , X
n
] = n. One of the
major parts of this course will be dimension theory of Noetherian local rings.
5.3. Localization denitions. We now return to pure algebra for a while. We
need to develop some technical tools to help us prove more about integral ring ex-
tensions, which will also help us work toward giving the denitions in the statement
of Theorem 3.7.1.
Let A ,= 0 be a ring, and let S A be a subset. We call S a multiplicative
subset provided 1 S and x, y S xy S.
Given A, S, we will dene a new ring S
1
A and a homomorphism
A S
1
A
which satises a certain universal property.
Let S
1
A be the set of equivalence classes of all formal quotients
a
s
, for a A,
s S. We say
a
s

a

if and only if t S such that t(s

a sa

) = 0. Check that
this is an equivalence relation. Sometimes we denote the equivalence classes using
the symbol [
a
s
].
Proposition 5.3.1. Let S A be a multiplicative subset.
(1) S
1
A is a ring with homomorphism can : A S
1
A given by a [
a
1
].
(2) (Universal property): If : A B is any ring homomorphism such that
(S) B

, then there is a unique homomorphism

: S
1
A B such
that

can = .
The homomorphism

is dened by

[
a
s
] = (a)(s)
1
.
Proof. This is standard and easy to prove, the main point being to show that
the various ring operations and ring homomorphism are well-dened. See Atiyah-
Macdonald, Ch. 3 for details. In brief, the identity element is [
1
1
], and we add and
multiply elements by the rules
[
a
s
][
a

] = [
aa

ss

]
[
a
s
] + [
a

] = [
s

a +sa

ss

].

20 THOMAS J. HAINES
Examples: (1) If A is a domain and S = A0, then S
1
A = Frac(A).
(2) If p is prime, then S := Ap is a multiplicative subset; denote S
1
A simply
by A
p
.
(3) If f A, let S := f
n
, n 0. Then S
1
A = A
f
= A[
1
f
].
5.4. Localization of modules. The same construction works for modules: let
S A be as above, and let M be an A-module. Then we can dene in a parallel
way S
1
M S
1
A-Mod. The S
1
A-module structure is dened using the rule
[
a
s
][
m
t
] = [
am
st
].
Addition is dened as in S
1
A. I leave it to you to check that the operations are
well-dened, and S
1
M really is an S
1
A-module.
The map M S
1
M is a functor in an obvious way (well discuss functors soon,
so dont worry if you dont know what this means). Just note that an A-module
homomorphism f : M M

induces a well-dened S
1
A-module homomorphism
S
1
M S
1
M

given by
S
1
f([
m
s
]) = [
f(m)
s
].
Lemma 5.4.1. The functor M S
1
M is exact, i.e. it takes exact sequences in
A-Mod into exact sequences in S
1
A-Mod.
Proof. Suppose
M

M
g

is exact. Note that

m
s
ker(S
1
g) t S such that tg(m) = 0
t, m

such that f(m

) = tm, i.e. f(
m

st
) =
m
s

m
s
im(S
1
f).

Lemma 5.4.2. There is an isomorphism S

1
A
A
M

= S
1
M as S
1
A-modules,
given by
a
s
m
am
s
.
Proof. The map is clearly surjective. We prove injectivity: every element on the
LHS
2
is of form
1
s
m (check this!), and
1
s
m 0 implies that t S such that
tm = 0, and in that case
1
s
m =
1
st
tm = 0.

Recall that an A-module N is at provided that the following holds:

Whenever M

M M

is an exact sequence of A-modules, then the induced

sequence N
A
M

N
A
M N
A
M

is also exact. In other words, the

functor N
A
is exact.
The following exercise provides lots of examples of at modules.
Exercise 5.4.3. Any free module is at. Any projective module is a direct sum-
mand of a free module, hence is also at.
2
Left Hand Side
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 21
Corollary 5.4.4. S
1
A is a at A-module.
Proof. The corollary follows from the two preceding lemmas.
Now we can produce a lot of at modules which are not projective (so also not
free). The simplest example is , a at Z-module which is not projective (since its
not free, and over Z, free projective).
6. Lecture 6
6.1. Further properties of localization. In the statements below, M
i
, N, P, etc.
are all A-modules, and = means that there is a canonical isomorphism of S
1
A-
modules.
Lemma 6.1.1. The following properties of localization hold.
(i) S
1
(M
1
M
2
) = S
1
M
1
S
1
M
2
.
(ii) S
1
(N P) = S
1
N S
1
P.
(iii) S
1
(M/N) = S
1
M/S
1
N.
(iv) S
1
(rad(A)) = rad(S
1
A).
(v) S
1
(M
A
N) = S
1
M
S
1
A
S
1
N. In particular, for a prime ideal p,
we have (M
A
N)
p
= M
p

Ap
N
p
.
Proof. In statements (ii) and (iii), we understand N, P as submodules of a module
M; by exactness of S
1
(), we can regard S
1
N, S
1
P are submodules in S
1
M.
In statement (v), the isomorphism is given by
mn
s

m
1

n
s
=
m
s

n
1
with inverse
m
s

n
t

mn
st
.
The remaining statements are easy to check.
6.2. Local properties. What are local properties? They are properties of an A-
module M that hold i they hold for all the localizations M
p
. Why call them
local? Because the ring A
p
turns out to be a local ring (see Prop. 6.4.1 below).
Lemma 6.2.1. For an A-module M, the following statements are equivalent:
(1) M = 0.
(2) M
p
= 0 for all p Spec(A).
(3) M
m
= 0 for all m Spec
m
(A).
Proof. (3) = (1): For x M, dene Ann(x) = a A [ ax = 0, an ideal in
A. If x M and x ,= 0, then Ann(x) ,= (1), and so Ann(x) m for some maximal
ideal m. But then
x
1
,= 0 in M
m
, so the latter is not zero.
Lemma 6.2.2. For an A-module homomorphism : M N, the following are
equivalent:
(1) is injective (resp. surjective, bijective).
(2)
p
is injective (resp. surjective, bijective) for all p Spec(A).
(3)
m
is injective (resp. surjective, bijective) for all m Spec
m
(A).
Proof. For each p, the following sequence is exact, by Lemma 5.4.1;
0 ker()
p
M
p
N
p
coker()
p
0.
Now use Lemma 6.2.1.
22 THOMAS J. HAINES
These lemmas help us prove the following lemma (atness is a local property).
Lemma 6.2.3. For an A-module M, the following are equivalent:
(1) M is A-at.
(2) M
p
is A
p
-at for every p Spec(A).
(3) M
m
is A
m
-at for every m Spec
m
(A).
Proof. (1) = (2) follows from a more general fact: Let f : A B be a
homomorphism; then M is A-at implies that B
A
M is B-at. This follows
(check this!) from the fact that for any B-module N we have an isomorphism of
B-modules
N
B
(B
A
M)

= N
A
M.
On the RHS, we are viewing N as an A-module, via the homomorphism f : A B.
See Atiyah-Macdonald Ch.2 for details. The point is that we may dene the A-
module structure with the rule a n := f(a)n.
(3) = (1): Its enough to show: if N P in A-Mod, then M
A
N M
A
P
as well. But now using Lemmas 5.4.1, 6.1.1, 6.2.2, we see
N P = N
m
P
m
, m
= M
m

Am
N
m
M
m

Am
P
m
, m
= (M
A
N)
m
(M
A
P)
m
, m
= M
A
N M
A
P.

6.3. Exercises on integrality. At this point we pause to test our understanding

a little bit. The following two lemmas are exercises that I leave to you. (They can
of course be found in the standard texts.)
Lemma 6.3.1. Let A, S be as above. If A B is an integral extension, then so is
S
1
A S
1
B. More generally, if C is the integral closure of A in B, then S
1
C
is the integral closure of S
1
A in S
1
B.
In particular, if p Spec(A) and A B is integral, then so is A
p
B
p
.
Lemma 6.3.2. If A B is an integral extension, q Spec(B) and p := q
c

Spec(A), then p is maximal i q is maximal.

Exercise 6.3.3. Prove the two lemmas above.
Using the lemmas, we have normality is a local property. Note that if A is a
domain with fraction eld K, then every localization A
p
is also a domain, with the
same fraction eld.
Lemma 6.3.4. Let A be a domain with fraction eld K. The following are equiv-
alent:
(i) A is normal.
(ii) A
p
is normal for every p Spec(A).
(iii) A
m
is normal for every m Spec
m
(A).
Proof. Consider the inclusion f : A

A K, where

A is the integral closure of
A in K. Note that (i) holds i f is surjective. Similarly, since the exercise above
shows that (

A)
p
=

(A
p
), (ii) (resp. (iii)) holds i f
p
(resp. f
m
) is surjective for
every p (resp. m). Now the lemma follows from Lemma 6.2.2.
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 23
6.4. Extending and contracting ideals along A S
1
A.
Proposition 6.4.1. Let A, S be as in the previous section.
(1) Every ideal in S
1
A is an extended ideal, hence of the form S
1
a, for some
ideal a A.
(2) The rule S
1
p p gives a bijective correspondence between the prime ideals
in S
1
A and the prime ideals in A which are disjoint from S.
In particular, taking S = A p, we see Spec(A
p
) q Spec(A) [ q p. Thus
(A
p
, pA
p
) is a local ring.
Proof. (1): Let b S
1
A be an ideal, and suppose
x
s
b. Then
x
1
b, hence
x b
c
and so
x
s
b
ce
. But then b b
ce
b (the latter inclusion being automatic),
hence b = b
ce
.
(2): If q S
1
A is prime, then so is q
c
(and the latter clearly doesnt meet
S otherwise q would contain a unit in S
1
A). Moreover that fact that q is an
extended ideal implies that q = S
1
q
c
. (See Atiyah-Macdonald, Prop. 1.17.)
On the other hand, suppose p A is prime. Then S
1
p S
1
A is prime i
S
1
A/S
1
p ,= 0 and is a domain. Let S denote the image of S in A/p. Then
S
1
A/S
1
p

= (S)
1
(A/p). The latter ring is either zero or is a non-zero ring
contained in the eld of fractions of A/p, hence is a domain. Hence S
1
p is either
the unit ideal, or is prime. The former holds i S p ,= .
It follows that every prime ideal p which does not meet S gives rise to a prime
ideal S
1
p, and moreoever (S
1
p)
c
= p. The inclusion is clear, so let us prove
. Suppose x belongs to the left hand side. Then there exists p p and s S such
that x/1 = p/s, and thus there exists t S with t(sxp) = 0. But then (ts)x p.
Since S p = , this means that x p, as desired.
Putting these remarks together, the proposition is now proved.
6.5. Krull-Cohen-Seidenberg Theorems. The following ultra-slick treatment
of these theorems is taken from lectures of R. Swan. These theorems can be proved
from what we have established about integral extensions, using localization as a
tool. This is what is done in Atiyah-Macdonald. The point here is to show how
localization may be avoided, and in fact the proof we will give is about as elementary
as can be expected.
Theorem 6.5.1. Let A B be an integral extension. Let p A be a prime ideal.
Then an ideal P of B is a prime ideal with P A = p if and only if P is maximal
among the ideals such that P A p.
Proof. The last condition says P is maximal with respect to P S = , where
S = A p. Such an ideal is automatically prime (check this!), so in either case P
will be prime.
Now suppose that P is prime and has P A = p, but is not maximal among the
ideals with P A p; then there exists an ideal Q _ P with Q A = p. Working
mod P we can assume P = p = 0, and that B is a domain. Suppose b Q, b ,= 0,
and let b
n
+a
n1
b
n1
+ +a
0
= 0 be an equation for b over A of least degree n.
Then a
0
,= 0 since B is a domain. But then a
0
Q A = p, contradicting the fact
that p = 0.
Conversely, suppose that P is maximal with the property P A p. Working
mod P we can assume P = 0. We must show that p = 0. Suppose that there is an
element a p, a ,= 0. We claim that (Ba) A p, contradicting the maximality
24 THOMAS J. HAINES
of P. Indeed, suppose ba (Ba) A. Let b
n
+ a
n1
b
n1
+ + a
0
= 0 be an
equation for b over A. Then (ab)
n
+ aa
n1
(ab)
n1
+ + a
n
a
0
= 0, showing that
(ab)
n
Aa p, so that ab p.
The following three important results are immediate from Theorem 6.5.1.
Corollary 6.5.2 (Lying Over Theorem). If A B is an integral extension, then
Spec(B) Spec(A) is surjective.
When q
c
= p, we says that q lies over p.
Corollary 6.5.3. Let A B be an integral extension. Let q, q

Spec(B) such
that q q

and q
c
= (q

)
c
=: p. Then q = q

.
Geometrically, this says that if i : A B is an integral extension, then there
are no containments in the bers of i

: Spec(B) Spec(A).
Corollary 6.5.4 (Going Up Theorem). Let A B be an integral extension. If
p p

are prime ideals in A, and q is a prime ideal in B lying over p, then there
exists a prime ideal q

lying over p

, and with q q

.
Proof. Note that q A p

; choose an ideal q

which is maximal among those

which contain q and have q

A p

. By Theorem 6.5.1, we have q

is prime, and
q

A = p

.
7. Lecture 7
7.1. Dimension is invariant under formation of integral extensions.
Proposition 7.1.1. If A B is an integral extension, then dim(A) = dim(B).
Proof. If P
0
_ P
1
_ _ P
n
is a chain of prime ideals in B, then the chain
P
0
A _ P
1
A _ _ P
n
A has the same length, by Corollary 6.5.3. If
p
0
_ p
1
_ _ p
n
is a chain of prime ideals in A, then we can lift it to a chain
P
0
_ P
1
_ _ P
n
in B, using the Lying Over Theorem to lift p
0
and then using
the Going Up Theorem repeatedely to lift p
i
, for i > 0.
7.2. Going Down Theorem.
Theorem 7.2.1 (Going Down Theorem). Let A B be an integral extension.
Assume that A is a normal domain and that B is torsion-free as an A-module. Let
p Spec(A) and P Spec(B) with P A = p. Let q Spec(A) with q p. Then
there is a prime Q Spec(B) with Q P and Q A = q.
For the proof we need two lemmas.
Lemma 7.2.2. Let A be a normal domain with Frac(A) = K. Let f, g K[X] be
monic. If fg A[X], then f A[X].
Proof. The roots of f, being roots of fg are integral over A. Therefore so are the
coecients of f, but these are in K, hence in A since A is normal.
Lemma 7.2.3. Let A B be an integral extension and let I A be an ideal.
Let

I = x B [ x
n
+ a
n1
x
n1
+ + a
0
= 0, for some n and a
i
I. Then

I =

BI.
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 25
Proof. It is clear that

I

BI. For the converse, let x

n
=

i
b
i
c
i
with b
i
B
and c
i
I, for i = 1, . . . , r. Then C := B[b
1
, . . . , b
r
] is nite over A and x
n
C IC.
Now in Lemma 2.1.1, take M = C and = mult. by x
n
, to conclude that x satises
an equation of the required form, so that x

I.
Corollary 7.2.4. Let A B be an integral extension. Assume A is a normal
domain and B is torsion-free as an A-module. Let K = Frac(A), and I A be a
prime ideal. Let b B. Then b

BI i the minimal polynomial f of b over K

has the form f = X
n
+a
1
X
n1
+ +a
n
where all a
i
lie in I.
Proof. Suppose that b

BI. By Lemma 7.2.3, b

I, so that b is a root of a
polynomial of form h = X
k
+ q
1
X
k1
+ + q
k
with the q
i
I. Since f is the
minimal polynomial we have h = fg in K[X]. By Lemma 7.2.2, f and g lie in A[X].
Modulo I, fg = h = X
k
. Therefore, since I is prime, f = X
n
for some n k and
f has the form f = X
n
+a
1
X
n1
+ +a
n
, with the a
i
I.
Conversely, note that f(b) = 0 holds in K
A
B (by denition, f is the minimal
polynomial of b over K if it is the monic of least degree in K[X] such that f(b) = 0
in K
A
B). But then since B is torsion-free as an A-module, we have f(b) = 0 in
B as well, so that b

I. Then by Lemma 7.2.3, b

BI, as desired.
Proof of Theorem 7.2.1: It will suce to show that if S := (A q)(B P), then
Bq S = . Why? In that case we can choose Q Bq maximal with respect to
Q S = . Then it follows (check this!) that Q is a prime ideal contained in P,
such that Q A = q.
Suppose that s A q and t B P, and st Bq. By Corollary 7.2.4, the
minimal polynomial f of st over K is of the form f = X
n
+q
1
X
n1
+ +q
n
, with
the q
i
in q. Since s K 0, the minimal polynomial g of t over K has the form
g = X
n
+a
1
X
n1
+ +a
n
, where q
i
= s
i
a
i
for all i. By applying Corollary 7.2.4
(with I = A; note that that Corollary does hold true for I = A) to b = t, we see
that all the a
i
lie in A. Since s A q, we have a
i
q. By Corollary 7.2.4 again,
t

Bq. Since Bq P, it follows that t P, which is impossible.

7.3. Application of Going Down.
Theorem 7.3.1. Let A be a domain which is a f.g. k-algebra. Then all maximal
chains of prime ideals in A have length equal to tr.deg
k
A.
In particular, dim(A) = tr.deg
k
A.
First we need to review the notion of transcendence degree. If F k is a eld,
and B F is a subset, let k(B) F be the subeld of F generated by k and B
(= the smallest subeld that contains k, B). We say that B is a transcendence
basis provided that
The set B is algebraically independent over k, and
F is an algebraic extension of k(B).
Fact: Any extension of elds F/k has a transcendence basis B, and any two such
bases have the same cardinality.
Note that B = i F/k is algebraic. If F = k(B), we say F/k is purely
transcendental.
For the proof of the fact, see N. Jacobson, Basic Algebra II, section 8.12. We
then dene tr.deg
k
F = [B[, the cardinality of any transcendence basis. If A is a
k-algebra domain with fraction eld K, we dene tr.deg
k
A = tr.deg
k
K.
26 THOMAS J. HAINES
To prove the theorem, we need the following lemma. We say a prime P Spec(A)
has height n if
supk [ P
k
_ P
k1
_ P
0
= P = n.
(The number on the LHS is called simply ht(P).)
Lemma 7.3.2. Let A be a domain which is a f.g. k-algebra. Let P be a prime
ideal of height 1 in A (since A is a domain, this is just a minimal non-zero prime
ideal). Then tr.deg
k
A/P = tr.deg
k
A1.
Proof. Choose k[x
1
, . . . , x
t
] A as in the Noether Normalization theorem. Then
t = tr.deg
k
A. By Corollary 6.5.3, P k[x
1
, . . . , x
t
] ,= 0. Let f P k[x
1
, . . . , x
t
]
with f ,= 0. After using a substitution of variables as in the proof of Noether
Normalization, we may assume f is monic in x
t
(replace the x
i
with y
i
:= x
i
+x
mi
t
if i < t, and y
t
:= x
t
, for some large integers m
i
). Therefore k[x
1
, . . . , x
t
] is in-
tegral over k[x
1
, . . . , x
t1
, f], and we may replace x
t
by f. Then we can assume
x
t
p := P k[x
1
, . . . , x
t
]. If p ,= x
t
k[x
1
, . . . , x
t
], then the Going Down theo-
rem shows that we can nd Q _ P in A with Q k[x
1
, . . . , x
t
] = x
t
k[x
1
, . . . , x
t
],
which would contradict the fact that P has height 1. So p = x
t
k[x
1
, . . . , x
t
], so
k[x
1
, . . . , x
t1
] = k[x
1
, . . . , x
t
]/x
t
k[x
1
, . . . , x
t
] A/P. Since this is an integral
extension, tr.deg
k
A/P = t 1, as required.
Proof of Theorem 7.3.1: We use induction on d = tr.deg
k
A. If d = 0, then A is a
eld, which has dimension zero.
Suppose d > 0 and let 0 = P
0
_ P
1
_ _ P
n
be a maximal chain of prime
ideals in A. Then 0 = P
1
/P
1
_ P
2
/P
1
_ _ P
n
/P
1
is a maximal chain of prime
ideals in A

= A/P
1
. By Lemma 7.3.2, tr.deg
k
A

= d 1, and so n 1 = d 1 by
our induction hypothesis applied to A

.
7.4. Some counterexamples. In the theorem, it is essential that A is a domain.
What happens if we allow A to be more general? It is easy to see that if A =
k[X, Y ] k[Z], then we have a maximal chain of prime ideals in the rst factor
having length 2, and a maximal chain of primes ideals in the second factor having
length 1. In this case, the space Spec A is disconnected, and in fact is the disjoint
union Spec(k[X, Y ])

Spec(k[Z]) (see Atiyah-Macdonald, Ch. 1, Ex. 22), so it is

not surprising that it is made up of pieces with dierent dimensions.
How about if we avoid such silly examples by requiring A to be such that Spec A
is connected, but not necessarily irreducible? Can we still have irreducible com-
ponents that have dierent dimensions? The answer is yes, and an example has
already been provided. Namely, recall the ring A = k[X, Y, Z]/(XY, Y Z). As
we saw in Lecture 4, the corresponding variety is connected, a union of a line
and a plane, which clearly have dierent dimensions. Algebraically, note that
I = (Y )(X, Z) = (Y ) (X, Z), a radical ideal since (Y ) and (X, Z) are both prime.
It follows that (Y ) and (X, Z) are the only two minimal primes among those which
contain I, which corresponds to the fact that there are two irreducible components
in Spec(k[X, Y, Z]/I). Note that (Y ) is the bottom prime in a chain of length two,
and (X, Z) is the bottom prime in a chain of length one. This corresponds to the
statement that dim(V (Y )) = 2 and dim(V (X, Z)) = 1, as we already knew since
the rst is the XZ-plane, and the second is the Y -axis.
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 27
7.5. Returning to Geometric version of Noether Normalization. In par-
ticular, Theorem 7.3.1 says that dim(k[X
1
, . . . , X
n
]) = n, as we claimed earlier.
Moreover, in the geometric version of Noether normalization (Theorem 3.7.1), we
claimed that t = dim(A), where t was the total number of algebraically independent
elements x
1
, . . . , x
t
in k[x
1
, . . . , x
t
] A. This now follows from Lemma 7.1.1.
We have also justied that the map Spec A Spec(k[x
1
, . . . , x
t
]) is surjective.
We have not yet shown that the bers are nite sets, although that is true.
7.6. Hypersurfaces. The following statement appears in Hartshornes Algebraic
Geometry, Prop. 1.13. We will prove it later, using the Krull Hauptidealsatz and a
few other facts.
Proposition 7.6.1 (Codimension 1 subschemes in ane space). Then an irre-
ducible closed subset Y A
n
k
= Spec(k[X
1
, . . . , X
n
]) has dimension n 1 if and
only if Y = V (f), for some non-constant irreducible polynomial f k[X
1
, . . . , X
n
].
Now here is a very interesting exercise, which you should compare with the above
statement. How are they related?
Exercise 7.6.2. Suppose k is a eld which is NOT algebraically closed. Show that
if Z k
n
is a non-empty Zariski-closed subset (the set of zeros in k
n
of an ideal
I k[X
1
, . . . , X
n
]), then there is a single polynomial f k[X
1
, . . . , X
n
] such that
Z = Z(f), the set of zeros of f in k
n
.
Hint: It is enough to show that for any m 1, there is a polynomial in m
variables such that the only zero of in k
m
is (0, 0, . . . , 0). (Then, if Z = Z(I)
where I = (f
1
, . . . , f
m
), we can put f = (f
1
, . . . , f
m
).) Prove that exists, by rst
looking at the case m = 2.
8. Lecture 8
Categories and functors. Presheaves and sheaves. Stone-Cech theorem. Motiva-
tion for locally ringed spaces.
9. Lecture 9
Stalks, denition of locally ringed space. Examples. Denition of O
Spec(A)
, and
basic properties.
10. Lecture 10
Denition of ane scheme, and denition of scheme. Description of category of
(ane) algebraic varieties in scheme-theoretic language. Example (deformations).
DVRs; examples. Valuation rings. Basic properties.
11. Lecture 11
Recap of Noetherian rings/modules. Basic proposition. Proof E. Noethers the-
orem in invariant theory. Alternate proof of Nullstellensatz via basic proposition.
12. Lecture 12
We shall follow the treatment of associated primes and primary decompositions
from [Mat2], Chapter 2, 6. You can nd similar theorems (just for rings, not
modules), in Atiyah-Macdonald, Chapter 4.
28 THOMAS J. HAINES
12.1. Associated primes. Throughout, we assume A ,= 0. We say an ideal q A
is primary provided that A/q ,= 0 and every zero-divisor in A/q is nilpotent.
Equivalently, q ,= A, and
xy q = y q or x
n
q for some n 1.
If I A is an ideal, we say it has a primary decomposition if we can write
I = q
1
q
r
for some primary ideals q
i
. We shall prove that when A is
Noetherian, every ideal I A possesses a primary decomposition, and in that case
there are various uniqueness statements one can make. In fact, following [Mat2],
we shall actually prove analogous statements for all f.g. modules over A.
Fix an A-module M. Dene
Ass(M) = p Spec(A) [ p = ann(x), for some x M.
We call this the set of associated primes (to M). Note that p Ass(M) i M
contains a submodule isomorphic to A/p.
We say a A is a zero-divisor for M if x ,= 0 in M such that ax = 0.
We say a A is M-regular if it is not a zero-divisor for M.
Lemma 12.1.1. Let A be a Noetherian ring, and M ,= 0 an A-module.
(a) Every maximal element in the family T := ann(x) [ 0 ,= x M belongs
to Ass(M). In particular, Ass(M) ,= .
(b) zero-divisors for M =
pAss(M)
p.
Proof. First note that since A is Noetherian, any non-empty family of ideals (such
as T) possesses maximal elements.
(a): If ann(x) is a maximal element of T, then ann(x) is prime: abx = 0 and bx ,= 0
and ann(x) ann(bx) implies by maximality that ann(x) = ann(bx), hence that
ax = 0.
(b): The inclusion is clear. Lets prove . Suppose x ,= 0 and ax = 0. Then a
belongs to some maximal element of T, hence a belongs to the right hand side.
Lemma 12.1.2. If
0 M

M M

0
is an exact sequence of A-modules, then Ass(M) Ass(M

) Ass(M

).
Proof. Let p Ass(M). Then A/p

= N, for some submodule N M. Note that
p = ann(x) for any 0 ,= x N (since p is prime). So if N M

,= 0, there exists
0 ,= x

with p = ann(x

), so that p Ass(M

).
On the other hand, if NM

= 0, then N maps isomorphically onto its image in

M

, and so the latter contains a copy of A/p; hence in that case p Ass(M

).
Lemma 12.1.3. Let A be Noetherian, and M ,= 0 a f.g. A-module. Then there
exists a chain 0 = M
0
M
1
M
n
= M of submodules such that for each i,
M
i
/M
i1

= A/p
i
for some prime ideal p
i
.
Proof. Choose any p
1
Ass(M). Then M
1
exists with M
1

= A/p
1
. If M = M
1
,
we are done. If M
1
,= M, apply this to M/M
1
. Repeat to nd the desired chain.
It terminates at M in nitely many steps, since M is Noetherian.
For the next theorem, we need the notion of support Supp(M) of an A-module
M. By denition
Supp(M) = p Spec(A) [ M
p
,= 0.
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 29
Lemma 12.1.4. M a nite A-module = Supp(M) = V (ann(M)), a Zariski-
closed subset of Spec(A).
Proof. Write M = Am
1
+ +Am
n
. Fix p Spec(A). Then
M
p
,= 0 i with m
i
,= 0 in M
p
i with ann(m
i
) p
ann(M) =
i
ann(m
i
) p.
In the last , is clear. For , use the exercise below.
Exercise 12.1.5. If P is prime and P
n
i=1
a
i
for some ideals a
i
, then there
exists some j such that P a
j
.
Now we can state and prove the following fundamental result.
Theorem 12.1.6. Let A be Noetherian, and M a f.g. A-module. Then
(1) Ass(M) is a nite set.
(2) Ass(M) Supp(M).
(3) The minimal elements of Ass(M) and Supp(M) coincide.
Proof. (1): By Lemma 12.1.3 there is a chain 0 = M
0
M
1
M
n
= M such
that M
i
/M
i1
= A/p
i
. Now Lemma 12.1.2 (and induction) shows that
Ass(M)
i
Ass(A/p
i
) = p
i

i
.
This shows that Ass(M) is nite.
(2): If 0 A/p M is exact, then so is 0 A
p
/pA
p
M
p
, in which case
M
p
,= 0.
(3): First, we need a localyzing lemma for the behavior of Ass. In this proof, we
use the following notation: M
S
:= S
1
M, and A
S
:= S
1
A, for any multiplicative
set S A.
Lemma 12.1.7. (a) If N A
S
-mod, then Ass
A
S
(N) = Ass
A
(N) (via the
identication Spec(A
S
) Spec(A)).
(b) Suppose A is Noetherian and M A-mod. Then Ass(M
S
) = Ass(M)
Spec(A
S
).
In particular, if A is Noetherian, p Ass
A
(M) pA
p
Ass
Ap
(M
p
).
Proof. Note that the last statement is an immediate consequence of (a),(b).
(a): Let x N. We have ann
A
(x) = ann
A
S
(x) A. So P Ass
A
S
(N) = p :=
P A Ass(N).
Conversely, if p Ass
A
(N) and x N is such that p = ann
A
(x), then x ,= 0
hence p S = . Hence P = pA
S
is a prime ideal such that P = ann
A
S
(x).
(b): If p Ass(M) Spec(A
S
), then p S = and p = ann
A
(x), for some x M.
If (a/s)x = 0 then t S such that tax = 0; t / p, ta p a p. Hence
ann
A
S
(x) = pA
S
, and so pA
S
Ass(M
S
).
Conversely, if P Ass(M
S
), then WLOG P = ann
A
S
(x), for x M. If p :=
P A, we have P = pA
S
and p S = .
Claim: t S such that p = ann
A
(tx), hence p Ass(M) Spec(A
S
).
Proof of Claim: p is a f.g. ideal, say by f
1
, . . . , f
n
. Now f
i
x = 0 in M
S
implies that
t
i
S such that f
i
t
i
x = 0 in M. Take t = t
1
t
n
. This does the job, and the
30 THOMAS J. HAINES
claim is proved. To see why, note that
p ann
A
(tx) ann
A
S
(tx) A = ann
A
S
(x) A = p,
proving that p Ass(M) Spec(A
S
).
We have proved the lemma.
Now we nish the proof of the theorem by proving part (3). Its ETS
3
that a
minimal element of Supp(M) belongs to Ass(M). Let p be such an element.
Using Lemma 12.1.7, we have
0 ,= M
p
, = Ass(M
p
) = Ass(M) Spec(A
p
)
Supp(M) Spec(A
p
)
= p.
So p Ass(M
p
), and hence p Ass(M), as desired.
12.2. Consequences. Let A be a Noetherian ring, and M a f.g. A-module.
Let P
i

r
i=1
be the set of minimal elements of Supp(M) = V (ann(M)),
or equivalently the set of minimal elements of Ass(M) (the set is nite
since Ass(M) is nite). Then V (ann(M)) = V (P
1
) V (P
r
). In other
words, the V (P
i
)s are precisely the irreducible components of the closed
set V (ann(M)).
We call the primes P
i
here the isolated primes of M. We call the
remaining primes of Ass(M), the embedded primes of M.
Letting M = A/I, we see in particular that there are only nitely many
minimal prime ideals containing I. Furthermore, in this case we have
Ass(A/I) = P Spec(A) [ x Asuch that P = (I : x) .
Here for any subset J A, we dene the ideal (I : J) = a A [ aJ I.
Thus, Ass(A/I) is precisely the set
Ass(A/I) = the ideals (I : x), x A, which are prime.
Suppose A is reduced. Then Ass(A) is precisely the set of minimal primes
P
1
, . . . P
r
of A. Since every minimal prime ideal of A is associated (Theorem
12.1.6), we need only show that every associated prime is one of the P
i
s.
To prove this note that, A being reduced, we have a canonical inclusion
A =
A
P
1
P
r

i
A/P
i
,
and thus
Ass(A)
i
Ass(A/P
i
) = P
i

i
.
13. Lecture 13
13.1. Primary submodules. Let N M be a submodule of the A-module M.
We say N is primary if N ,= M and if the following property holds: if a A is a
zero-divisor of M/N, then a
_
ann(M/N). Equivalently, for all a A, x M,
we have
x / N and ax N = a

M N for some 1.
The primary submodules of M = A are precisely the primary ideals of A.
3
Enough To Show
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 31
The following theorem gives us a crucial characterization of primary submodules
as exactly those N for which Ass(M/N) is a singleton.
Theorem 13.1.1. Suppose M is a f.g. A-module, and N M is a submodule.
Then
N M is primary Ass(M/N) = P,
in which case I := ann(M/N) is primary and

I = P.
Corollary 13.1.2. I A is primary i ! P of the form P = (I : x), and in that
case

I = P.
The corollary is an immediate consequence of the theorem. Let us now prove
the theorem.
Proof. (): We have Supp(M/N) = V (P) = V (ann(M/N)), and so P =
_
ann(M/N).
Now a A is a zero-divisor for M/N implies a P (use e.g. the proof of Lemma
12.1.1 to see that a belongs to an associated prime). So N M is primary.
(): Conversely, if P Ass(M/N) then every a P is a zero-divisor for M/N,
and so (by assumption that N is primary) a
_
ann(M/N). So P
_
ann(M/N).
But ann(M/N) P (by denition of Ass(M/N)), hence P =
_
ann(M/N), and
Ass(M/N) consists of just one element, which is P =
_
ann(M/N).
Now that we have proved the equivalence , we must verify
Claim: In this case, I := ann(M/N) is primary.
Proof: Suppose a, b A, b / I, and ab I. Then ab(M/N) = 0, but b(M/N) ,= 0.
This implies that a is a zero-divisor for M/N, and thus (since N is primary) a
_
ann(M/N). Thus I is primary, as desired.
If Ass(M/N) = P, we say N is P-primary, or a primary submodule
belonging to P.
Corollary 13.1.3. If I A has

I = m, a maximal ideal of A, then I is m-

primary.
Proof. Its ETS that if P = (I : x) is prime, then P = m. But P = (I : x) I,
hence P =

I = m, which proves that P = m.

Example. Let k be a eld, and let A = k[X, Y, Z]/(XY Z
2
). Let x, y, z denote
the images of X, Y Z k[X, Y, Z] in A. Let p := (x, z) A. Note that
p is prime: A/p

= k[Y ];

_
p
2
= p;
p
2
is not primary: xy = z
2
p
2
, yet x / p
2
and y /
_
p
2
= p.
Hence

I being prime is not sucient to guarantee that I is primary.

13.2. Various denitions relating to primary decompositions. Our goal is
to investigate when a submodule N M can be written in the form N = N
1

N
r
, for some primary submodules N
i
M. We call such an expression a
primary decomposition of N. The following lemma says that we may as well
group together the N
i
s which belong to the same prime (i.e. if certain terms N
ij
all belong to P
i
, in the primary decomposition we denote the intersection
j
N
ij
simply by the symbol N
i
). In this way, we can assume that the N
i
s in a primary
decomposition belong to distinct prime ideals.
32 THOMAS J. HAINES
Lemma 13.2.1. If N, N

M are P-primary submodules, then so is N N

.
Proof. We have an inclusion
M
N N

M
N

M
N

,
and thus Ass(M/(N N

) Ass(M/N) Ass(M/N

) = P, which implies the

result.
The above will be exploited in proving a kind of uniqueness result for primary
decompositions. What about existence? This will be done using the following
notions.
We say N M is reducible if N = N
1
N
2
for submodules N
i
with N _ N
i
,
i = 1, 2. We say N is irreducible provided it is not reducible.
Lemma 13.2.2. Suppose M is Noetherian. Then any submodule N is an inter-
section of nitely many irreducible submodules.
Proof. Consider the family T := N M [ N has no such expression. We assume
T ,= and derive a contradiction.
Choose a maximal element N
0
T (using that M is Noetherian). Then N
0
is reducible, so we may write it as N
0
= N
1
N
2
, where N
0
_ N
i
, i = 1, 2.
By maximality each N
i
is an intersection of nitely many irreducible submodules;
hence so is N
0
. This is nonsense.
We say an expression N = N
1
N
r
is irredundant if no N
i
can be omitted.
That is, for each i, we have N
i
_
j=i
N
j
. We thus have the notion of an irredundant
primary decomposition N = N
1
N
r
, an irredundant expression in which each
N
i
is P
i
-primary, for a prime P
i
.
In an irredundant primary decomposition, if we group together the N
ij
s belong-
ing to the same prime according to Lemma 13.2.1, and write their intersection as a
single module, then we call the resulting expression a shortest primary decom-
position. It has the property that P
i
,= P
j
if i ,= j. In that case, N
i
is called
the P
i
-primary component of N (as we shall see below, sometimes N
i
is indeed
uniquely determined by P
i
and N).
The following theorem is our main result concerning the existence and uniqueness
of primary decompositions.
Theorem 13.2.3. Let A be Noetherian, and let M be a nite A-module.
(i) Any irreducible submodule is primary.
(ii) If N = N
1
N
r
, with Ass(M/N
i
) = P
i
, is an irredundant primary
decomposition of N _ M, then Ass(M/N) = P
1
, . . . , P
r
.
(iii) Ever proper submodule N _ M has a primary decomposition. If P is a
minimal element of Ass(M/N), then the P-primary component of N is

1
P
(N
P
), where
P
: M M
P
is the canonical map (in particular the
P-primary component is uniquely determined by M, N, P).
14. Lecture 14
14.1. Proof of Theorem 13.2.3. (i): Assume N is not primary. Then by The-
orem 13.1.1 there exist P
1
,= P
2
in Ass(M/N). So we can nd submodules K
i

M/N where K
i

= A/P
i
, for i = 1, 2. But then K
1
K
2
= 0 (since any 0 ,= x K
i
has ann(x) = P
i
). This shows that N is reducible.
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 33
(ii): WLOG N = 0, and 0 = N
1
N
r
. Since M
i
M/N
i
, we have
Ass(M) P
1
, . . . , P
r
.
We want to prove that P
1
Ass(M) (the same argument applies to any other
P
i
). As N
2
N
r
,= 0, we may choose 0 ,= x N
2
N
r
, so that
ann(x) = (0 : x) = (N
1
: x). But (N
1
: M) = ann(M/N
1
) is a primary ideal with
_
(N
1
: M) = P
1
, so P

1
M N
1
for some 1. Therefore P

1
x N
1
and thus
P

1
x = 0 for some 1. Choose 0 such that
P

1
x ,= 0, P
+1
1
x = 0.
Let y be any non-zero element of P

1
x, so that y satises
P
1
y = 0, and so P
1
ann(y);
y N
2
N
r
, and so y / N
1
.
Since N
1
is primary, we see that a ann(y) = a
_
ann(M/N
1
) = P
1
. Thus
in fact P
1
= ann(y), proving that P
1
Ass(M), as desired.
(iii): Every proper submodule N has an irreducible decomposition, hence a primary
decomposition (by (i)). Let N = N
1
N
r
be a shortest primary decomposition.
We want to prove that if, say, P
1
is minimal in Ass(M/N), then N
1
is determined
by M, N, P
1
.
Write P for P
1
from now on. Localizing, we get N
P
= (N
1
)
P
(N
r
)
P
. Also,
there is a > 0 such that, for each i > 1, we have P

i
ann(M/N
i
). Since we are
assuming P is minimal in Ass(M/N), we have P
i
_ P for i > 1. From these two
remarks we see that (M/N
i
)
P
= 0 that is, (N
i
)
P
= M
P
, for i > 1 (check this!).
It follows that N
P
= (N
1
)
P
, and so
1
P
(N
P
) =
1
P
((N
1
)
P
) = N
1
, as desired.
Let us check the non-trivial inclusion of this last equality more carefully. If
m
1
P
((N
1
)
P
), then we may write
m
1
=
n1
s
for some n
1
N
1
and s A P.
There is then a t / P such that tm N
1
. Let m M/N
1
denote the image of m.
We see that t / P such that tm = 0. If m ,= 0, then the fact that N
1
is P-primary
means that for a A, a ann(m) = a
_
ann(M/N
1
) = P. Applying this
implication to a = t, we would have t P, a contradiction. It follows that m = 0,
i.e., m N
1
. This shows
1
P
((N
1
)
P
) N
1
, as desired.
14.2. Examples and applications. The next corollary follows immediately from
Theorem 13.2.3.
Corollary 14.2.1. If A is a Noetherian ring, then every proper ideal I has a
shortest primary decomposition I = q
1
q
r
. The set of ideals p
1
, , p
r
to
which the q
i
s belong is uniquely determined by I. The q
i
s belonging to minimal
p
i
s are uniquely determined by I.
The following example shows that an ideal may have two (or more) distinct
shortest primary decompositions.
Example. In k[X, Y ], let I = (X
2
, XY ) = (X) (X, Y )
2
= (X) (X
2
, Y ). Note
that (X, Y )
2
and (X
2
, Y ) both have as radical the maximal ideal (X, Y ), hence both
are (X, Y )-primary. The ideal (X) is prime, hence primary. So, we have two distinct
shortest primary decompositions for I. Note that Ass(A/I) = (X), (X, Y ), so
that (X) is isolated, and (X, Y ) is embedded.
We can also use primary decompostions to prove the unique factorization of
ideals in a Dedekind domain. We will prove in the next lecture the following
proposition/denition which characterizes Dedekind domains.
34 THOMAS J. HAINES
Proposition 14.2.2. Let A be a Noetherian domain with dimension 1. Then the
following statements are equivalent.
(1) A is normal.
(2) Every primary ideal in A is a power of a prime ideal.
(3) Every localization A
p
, for p ,= 0 a prime ideal, is a DVR.
If A satises these properties, we call it a Dedekind domain.
Using this, we get the aforementioned unique factorization of ideals in A.
Corollary 14.2.3. Suppose A is a Dedekind domain and I is a proper, non-zero
ideal. Then I = p
a1
1
p
ar
r
for a uniquely determined set of non-zero prime ideals
p
i
, and positive integers a
i
, for i = 1, . . . , r.
Proof. By (2) above, the shortest primary decomposition takes the form I = p
a1
1

p
ar
r
for distinct non-zero prime ideals p
i
and positive integers a
i
. Since the
dimension of A is 1, the p
i
s are in fact maximal ideals, hence are pairwise coprime:
p
i
+ p
j
= A, if i ,= j. Furthermore, this implies that p
ai
i
+ p
aj
j
= A. From this
it follows that the intersection is actually a product: I = p
a1
1
p
ar
r
(for a proof,
see Atiyah-Macdonald, Prop. 1.10). The uniqueness of this expression follows from
the uniqueness statement in Corollary 14.2.1, since each p
i
is a minimal prime in
Ass(A/I) (in fact since dim(A) = 1, and I ,= 0, all primes containing I are minimal
primes containing I).
15. Lecture 15
Characterizations of DVRs, and applications to Dedekind domains (proofs).
Characterization of normal rings.
15.1. Characterizations of DVRs. Let A denote a DVR with fraction eld K,
with valuation v : K

Z. As usual set v(0) = . Note the following two facts:

The only non-zero ideals of A are the sets of the form m
k
:= y [ v(y) k.
(Check this! Use that any ideal a ,= 0 possesses an element y with minimal
valuation.) Thus, A Noetherian, as every ascending chain of ideals taken
from the set m
1
m
2
is stationary.
The maximal ideal is m = m
1
. We have m = (x) for any element x satisfying
v(x) = 1. In that case, we also have m
k
= (x
k
), for all k 1. So the only
prime ideals are m, (0); and so dim(A) = 1.
Thus, any DVR is a Noetherian local domain of dimension 1, in which every ideal
is principal. In fact this characterizes DVRs among all Noetherian local domains
of dimension 1.
The following is Proposition 9.2 from Atiyah-Macdonald.
Proposition 15.1.1. Let (A, m) be a Noetherian local domain of dimension 1, with
residue eld k := A/m. Then TFAE
4
:
(i) A is a DVR.
(ii) A is normal.
(iii) m is principal.
(iv) dim
k
(m/m
2
) = 1.
(v) Every non-zero ideal is m
k
, for some k 0.
4
The Following Are Equivalent
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 35
(vi) x A such that every non-zero ideal is (x
k
), for some k 0.
Proof. Note that any ideal a ,= (0), (1) is m-primary, hence a m
n
for some n 1.
To see this, use that

a = m, since m is the only prime ideal containing a.

(i) = (ii): Every valuation ring is normal (Lemma ????).
(ii) = (iii): Assume 0 ,= a m. Then n 1 such that m
n
(a), but
m
n1
_ (a). Choose b m
n1
(a), and set x = a/b K. We have x
1
/ A (since
b / (a)), hence x
1
is not integral over A. Hence x
1
m _ m (if x
1
m m, then
m would be a faithful A[x
1
]-module, f.g. as an A-module, and thus x
1
would be
integral). But x
1
m A, hence x
1
m = A, and m = (x).
(iii) = (iv): Assume m = (x). Clearly m/m
2
is generated by the image of x,
hence its dim
k
is 1. If the dimension is zero, then m = m
2
and NAK implies
m = 0, a contradiction.
(iv) = (v): dim
k
(m/m
2
) = 1 and NAK imply m = (x) for some x. For an ideal
a ,= (0), (1), we can choose an integer k > 0 such that m a m
k
= (x
k
).
Consider a A/(x
k
) =: A (an Artinian ring). WLOG a ,= (x
k
). Since x is
nilpotent, there exists r with a (x
r
) and a _ (x
r+1
). Take y a with y / (x
r+1
).
Note
y = u x
r
u A

x
r
a
a = (x
r
)
a = m
r
= (x
r
).
This completes the proof.
(v) = (vi): m ,= m
2
implies x m m
2
. By hypothesis (x) = m
r
for some
r 1. But then we must have r = 1, so that m = (x) and m
k
= (x
k
) for all k 1.
(vi) = (i): Write m = (x); note (x
k
) ,= (x
k+1
) for all k 0. Take a A 0,
and write (a) = (x
k
), some k 0. then a A

x
k
.
Claim: K

kZ
(A

x
k
).
Proof: Given
a
b
K

, write a = ux
k
and b = vx
m
, for u, v A

. Then
a
b
=
uv
1
x
km
.
Now, we can dene a function v : K

Z by setting v(ux
k
) = k. It is easy
to see that v is a discrete valuation of K, with valuation ring A. Hence A is a
DVR.
15.2. Proof of Proposition 14.2.2. (1) (3): Use Proposition 15.1.1 above
and the fact that normality is a local property.
(2) (3): Use Proposition 15.1.1 and the fact, proved in Atiyah-Macdonald 4.8,
that contraction of ideals gives a bijective correspondence
primary ideas in S
1
A contracted primary ideals in A,
and a similar one, where the word primary is replaced with prime.
15.3. Improvement on (iii) = (vi) in Proposition 15.1.1. For later pur-
poses, we need to give a proof of the implication (iii) = (vi), without the
dimension 1 hypothesis.
Proposition 15.3.1. Suppose (A, m) is a Noetherian local domain in which m is
principal and non-zero. Then A is a PID (hence of dimension 1, hence a DVR).
36 THOMAS J. HAINES
Proof. Write m = (x). Consider the family T = a A [ a is not principal. We
will assume T ,= , and derive a contradiction.
If T ,= , it contains a maximal element, say a. So a ,= (0), (1), m.
We will need the notion of invertible ideal. For any ideal I A, dene I
1
:=
x K [ xI A, otherwise known by the symbol (A : I). By denition we have
I I
1
A. We say I is invertible if I I
1
= A. Note that every principal ideal is
invertible.
Claim: a is not invertible.
Proof: If a a
1
= A, then a
i
a and b
i
a
1
such that

i
a
i
b
i
= 1. At least one
summand a
i
b
i
A

, and then
a = a
i
b
i
a a
i
A a,
so a = (a
i
), a contradiction. The claim is proved.
Now we know that a m. Thus m
1
a _ m
1
m = A (if m
1
a = A, then
a
1
= m
1
and a is invertible).
On the other hand, we know that a = m
1
ma m
1
a. This leaves us only two
options.
Case 1: a = m
1
a. Then ma = a, hence by NAK a = (0), a contradiction.
Case 2: a _ m
1
a. Then by choice of a as a maximal element of T, we know
m
1
a is principal, equal to (y), for some y A. But then a = m(y) = (xy), a
contradiction.
So, T ,= leads to a contradiction in every case.
15.4. Characterization of normal domains. The rst step is the following
proposition.
Proposition 15.4.1. Let A be a Noetherian domain, and P ,= 0 a prime ideal.
Then if P is invertible, then ht(P) = 1, and A
P
is a DVR.
Proof. P invertible = PA
P
invertible = (by proof of claim appearing in
Proposition 15.3.1 above) PA
P
is principal = (by Proposition 15.3.1 itself) A
P
has dim 1 and is a DVR, and ht(P) = 1.
Proposition 15.4.2. Let A be a normal Noetherian domain. Then
(i) For all P Ass(A/(a)), ((a) ,= (0)), we have ht(P) = 1, hence all such
Ps are isolated primes.
(ii) A =

ht(P)=1
A
P
.
Proof. (i): Fix a ,= 0. If P Ass(A/(a)), we can write P = (aA : b), for some
b A. Then
m := PA
P
= (aA
P
: b) = (A
P
: ba
1
),
and thus ba
1
m A
P
and ba
1
/ A
P
.
If ba
1
m m, then ba
1
is integral over A
P
, contradicting the normality of A
P
.
Hence ba
1
m = A
P
, and so m
1
m = A
P
. By Proposition 15.4.1, ht(m) = ht(P) =
1.
(ii): Its ETS the following statement: if a, b A, a ,= 0 and b aA
P
for all P of
ht 1, then b aA.
Consider a shortest primary decomposition aA = q
1
q
r
, where P
i
:=

q
i
for each i.
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 37
By (i), each P
i
has ht 1. Therefore each P
i
is minimal, and by the uniqueness
statement in Corollary 14.2.1, each q
i
is uniquely determined. In fact, we have
q
i
= A aA
Pi
.
Since b belongs to the intersection of all the terms on the RHS by hypothesis, it
also belongs to
i
q
i
= aA, as desired.
Note that we had to use the full strength of the uniqueness of shortest primary
decompositions to prove this statement.
We conclude this subsection with a characterization of the Noetherian domains
which are normal.
Theorem 15.4.3. Let A be a Noetherian domain. Then A is normal if and only
if the following two statments hold:
(a) If P is a ht 1 prime ideal, then A
P
is a DVR.
(b) If a ,= 0, every P Ass(A/(a)) has ht 1.
Proof. First assume A is normal. Then (a) holds, since A
P
is a Noetherian local
domain of dimension 1 which is normal, hence a DVR by Proposition 15.1.1. Also,
Proposition 15.4.2 ensures that (b) holds.
Conversely, suppose (a) and (b) hold. By the proof of (ii) in Proposition 15.4.2,
(b) implies that A =

ht(P)=1
A
P
. By (a), each A
P
appearing in this intersection is
normal, and thus A is normal too.
16. Lecture 16
Beginning of completions. Basic questions arising for

G. On exactness of G

G,
and completeness of

G.
16.1. Completions of abelian topological groups. Suppose (G, +) is an abelian
topological group. This means that G is an abelian group and also a topological
space, such that the group operations + : G G G and inv : G G are
continuous functions (where in the rst case, GG has the product topology).
Note that G is not necessarily Hausdor. In fact, G is Hausdor if and only if
0 is a closed set. One direction is immediate: if G is Hausdor, then any point
is a closed set; in particular 0 is closed. Conversely,, suppose 0 is a closed set,
and consider the continuous map
d : GG G
given by d(x, y) = x y. Then clearly d
1
0 = GG, where denotes the
diagonal subset. So is a closed subset, and it follows that G is Hausdor. (In
fact, a space X is Hausdor i the diagonal X X is closed.)
We dene the completion

G to be the set of all equivalence classes of Cauchy
sequences in G. Recall that a sequence x
n
is Cauchy if x
n
x
m
0 as n, m
(I leave it to you to make this precise). Also, two Cauchy sequences x
n
and y
n

are equivalent provided that x

n
y
n
0 as n .
Clearly, we may add or subtract Cauchy sequences term-by-term, and this gives
well-dened operations of +, on

G. It is easy to check that

G is itself an abelian
group, and that the map G

G given by taking g G to the constant Cauchy
sequence g, is a group homomorphism.
38 THOMAS J. HAINES
Lemma 16.1.1. G is Hausdor if and only if G

G.
Proof. Let H := 0

, the closure of the subgroup 0. Clearly H is a subgroup

of G.
Claim: 0

0U
U.
Proof:
x

0U
U 0 x U, U 0
x 0

,
the last equivalence holding because the open sets xU form a neighborhood basis
of open sets containing x, as U varies over all open subsets containing 0.
Finally, H = ker(G

G). So we are done.
Now we assume the topology on G is such that there is a countable sequence of
subgroups
G = G
0
G
1
G
2
G
n

which form a basis of opens sets around 0 G. This means that a subset V G is
a neighborhood of 0 i it contains contains some G
n
. In particular, each G
n
is an
open and therefore a closed subgroup of G.
We have projections
n+1
: G/G
n+1
G/G
n
, so we can form the inverse limit
lim

G/G
n

G/G
n
,
the subset of the direct product consisting of tuples (x
n
)
n

G/G
n
such that for
all n 0,
n+1
x
n+1
= x
n
.
Unless otherwise mentioned,

G will denote the completion of G with respect to
the topology determined by the ltration G G
n
.
Proposition 16.1.2. There is a canonical isomorphism of abelian groups

G = lim

G/G
n
.
Proof. If
n
is a Cauchy sequence, then
N
is ultimately constant in G/G
n
. So
we can dene a map

G lim

G/G
n
by sending
n
(x
n
)
n
, where
x
n

N
mod G
n
, N >> 0.
To dene the inverse map, let
n
G be an arbitrary lift of x
n
G/G
n
. Then

n
is a Cauchy sequence whose equivalence class is independent of the choice of
lifts.
Each G/G
n
is a discrete abelian group, and the product topology on

G/G
n
makes the latter a topological abelian group. Then

G = lim

G/G
n
is also a topo-
logical abelian group (give it the subspace topology from

G/G
n
).
Questions: In what sense is G

G functorial? Is

G

=

G? Does the topology on

G depend on the choice of subgroups G

n
?
We shall answer these questions (at least for the concrete cases we need) in the
next few sections.
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 39
16.2. Functoriality of G

G. So far, it is pretty obvious that G

G gives
us a functor Top.Ab Ab. In fact any continuous homomorphism f : G
1
G
2
determines a homomorphism

f :

G
1


G
2
: if
n
G
1
is Cauchy, then f(
n
)
G
2
is also Cauchy.
Exercise 16.2.1. Assume that the topologies on G
1
and G
2
are dened by countable
neighborhood bases of subgroups. Show that for the topologies on

G
1
and

G
2
dened
above, the map

f :

G
1


G
2
just dened is continuous.
Our next goal is to show that the functor G

G is exact, in a certain sense. To
state the proposition suppose we are given an exact sequence of topological abelian
groups
0

G

G
p

0
where we are assuming G has topology given by the ltration G
n
, G

G has the
subspace topology (therefore has basis G

n
:= G

G
n
around 0) and G

has the
quotient topology (therefore has basis G

n
:= p(G
n
) around 0).
Proposition 16.2.2 (Mittag-Leer lemma). The sequence
0

G

G
p

G

0
is exact, where the completions

G

and

G

are dened using the ltrations G

n
and
G

n
respectively.
Proof. Its ETS that
0

lim

/G

n

lim

G/G
n
p

lim

/G

n

0
is exact. More generally, suppose we have an exact sequence of inverse systems of
abelian groups
0

A

0.
Then we will prove
(1)
0

lim

A
n

lim

B
n
p

lim

C
n
is exact;
(2) the map p : lim

B
n
lim

C
n
is surjective, if we assume
n+1
: A
n+1
A
n
is surjective for all n 0.
Let A :=

A
n
, and dene d
A
: A A by (a
n
)
n
(a
n

n+1
a
n+1
)
n
. Note that
ker d
A
= lim

A
n
. Similarly dene B, C, d
B
, d
C
. We have the commutative diagram
with exact rows
0

A

d
A

B

d
B

C

d
C

0
0

A

B

C

0.
The snake lemma gives the short exact sequence
0

ker d
A

ker d
B

ker d
C
,
ie., statement (1).
40 THOMAS J. HAINES
For (2), again by the snake lemma its ETS that coker d
A
= 0, i.e that d
A
is
surjective. So, given (a
n
)
n
A, we must nd (a

n
)
n
A such that
a
n
= a

n+1
a

n+1
,
for all n 0. Since
n+1
is surjective, we can solve for the a

n
s recursively.
Corollary 16.2.3. For all n, we have inclusions

G
n


G.
Now we can dene a topology on

G by declaring that the sequence of subgroups

G

G
1


G
2

denes a basis of open subsets around 0

G.
Exercise 16.2.4. Show that this topology on

G agrees with the one dened by giving
lim

G/G
n
the subspace topology from the product

G/G
n
.
Proposition 16.2.5. There is a canonical isomorphism

G

G. In particular,

G
is Hausdor and complete (every Cauchy sequence in

G converges).
Proof. The exact sequence
0

G
n

G

G/G
n

0
gives us the exact sequence
0

G
n

G

G/G
n

0.
Since G/G
n
is discrete, we have

G/G
n
= G/G
n
, and hence a canonical isomorphism
G/G
n


G/

G
n
.
So

G = lim

G/G
n
lim

G/

G
n
=

G.

16.3. Examples: I-adic completions of rings and modules. In the following

examples, A denotes a ring with ideal I A, and M is an A-module.
Examples
(a) Let G = A, G
n
= I
n
. Note that A is a topological ring WRT
5
the I-adic
topology given by
A I I
2
.
(b) Let G = M, G
n
= I
n
M. Then M is a topological A-module when both A
and M are given the I-adic topologies. Furthermore

A is a topological ring,
and

M is a topological

A-module, i.e. the natural action map

A

M

M
is continuous. This follows from the fact that

A I
n

M is mapped by the
action map into I
n

M.
(c) M is Hausdor for the I-adic topology i [ker M

M] =

n=1
I
n
M = (0).
(d) If f : M N is A-linear, it is automatically continuous for the I-adic
topologies (check this!). Thus M

M is a functor from A-modules to

A-modules.
5
With Respect To
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 41
(e) Letting A = Z, and I = (p), we get the p-adic ring

A = Z
p
. Similarly,
letting A = k[X] and I = (X), we get

A = k[[X]].
17. Lecture 17
Applications to I-adic completions of A and M. (Stable) I-ltrations, and proof
of Artin-Rees lemma. Application: I-adic completion is an exact functor on cate-
gory of f.g. modules over a Noeth. ring. More applications.
*************
17.1. Basic notions related to I-ltrations. Let A be Noetherian, I A an
ideal, and M a f.g. A-module.
Goal: If 0 M

M M

0 is exact, then 0

M

M

M

0 is
exact, where each completion is dened using the I-adic topology.
N.B.: This does not follow immediately from Proposition 16.2.2: it is not at all
obvious that the I-adic topology on M

is the subspace topology M

inherits from
M. We need to prove this.
Theorem 17.1.1. Let A, I, M be as above, and let M

M be a submodule. The
the ltrations I
n
M

and M

I
n
M have bounded dierence, hence dene the same
topology on M

(and hence the same completion).

Let us rst recall some basic facts and terminology. We say M = M
0
M
1

M
2
is an I-ltration if IM
n
M
n+1
for all n 0. We say it is a stable
I-ltration if also IM
n
= M
n+1
for all n >> 0. For instance, I
n
M is a stable
I-ltration.
Lemma 17.1.2. Any stable I-ltrations M
n
, M

n
have bounded dierence: there
exists n
0
such that M
n+n0
M

n
and M

n+n0
M
n
for all n 0.
Proof. WLOG M

n
= I
n
M. Since M
n
is an I-ltration, I
n
M M
n
, n, hence
I
n+n0
M M
n
, n, n
0
.
Since M
n
is stable, n
0
such that I
n
M
n0
= M
n+n0
, n, hence M
n+n0
= I
n
M
n0

I
n
M, n.
By the lemma, its ETS that M

I
n
M is a stable I-ltration on M

. We need
to take a detour through graded rings/modules.
17.2. Graded rings and modules. Let A =

n=0
A
n
be a graded ring: A
0
A
is a subring, and A
n
A
m
A
n+m
, n, m. In particular, A
+
:=

n=1
A
n
is an ideal
in A.
Let M =

n=0
M
n
be a graded A-module: A
n
M
m
M
n+m
, n, m. In
particular, each M
n
is an A
0
-module..
Let N also be a graded A-module. An A-module morphism f : M N is
graded if f(M
n
) N
n
, n.
Lemma 17.2.1. TFAE:
(1) A is Noetherian.
(2) A
0
is Noetherian and A is a f.g. A
0
-algebra.
Proof. (2) (1): Use Hilberts Basis Theorem.
(1) (2): A
0
= A/A
+
, hence is Noetherian. The ideal A
+
is f.g.. Suppose
A
+
= (y
1
, . . . , y
r
), where WLOG y
i
A
ki
, for k
i
> 0. Let A

:= A
0
[y
1
, . . . , y
r
].
We will prove that A

= A; its ETS that A

n
A

for all n, which we prove by

42 THOMAS J. HAINES
induction on n. If y A
n
, write y =

i
a
i
y
i
for some a
i
A
nki
(take a
i
= 0 if
n < k
i
). We are done because the induction hypothesis gives a
i
A

for all i.
Now let A be a ring, and let M be an A-module equipped with an I-stable
ltration M
n
. We will apply the above considerations to the graded ring A

:=

n=0
I
n
and its graded module M

:=

n=0
M
n
. (Check that these are indeed
graded rings/modules.)
Lemma 17.2.2. With the notation above, assume A is Noetherian and M is f.g.
Then TFAE:
(i) M

is a f.g. A

-module.
(ii) M
n
is stable.
Proof. Since M
n
is a f.g. A-module, so is Q
n
:=
n
r=0
M
r
M

. It is easy to see
(check this!) that Q
n
generates the A

-submodule
M

n
:= M
0
M
n
IM
n
I
2
M
n
.
Note that M

n
is f.g. as an A

-module (since Q
n
is f.g. as an A-module). Since
A

is Noetherian (Lemma 17.2.1), M

is f.g. as an A

-module i the following

ascending chain which exhausts M

is stationary:
M

n
M

n+1
,
which holds i
M

= M

n
for some n M
n+n0
= I
n
M
n0
, for some n
0
and every n
M
n
is stable.

This gives us the next very useful result.

Proposition 17.2.3 (Artin-Rees Lemma). Suppose A is Noetherian, I A is an
ideal, M is a f.g. A-module, M
n
is a stable I-ltration in M, and M

M is a
submodule. Then M

M
n
is a stable I-ltration on M

.
Proof. We have I(M

M
n
) M

IM
n
M

M
n+1
, so M

M
n
is an I-
ltration. Applying Lemma 17.2.2 to both M

and
n
M

M
n
, we see rst that
M

is a f.g. A

-module, and further that

n
M

M
n
is a graded A

-submodule of M

it is a f.g. A

-module
M

M
n
is stable.

17.3. Some consequences of Artin-Rees. There is an obvious map

A
A
M

M (here completions are the I-adic ones).

Proposition 17.3.1. (1) If M is f.g, then

A
A
M

M.
(2) If A is Noetherian and M is f.g., then

A
A
M

M.
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 43
Proof. (1): There is an exact sequence 0 N A
n
M 0. This gives a
commutative diagram with exact upper row

A
A
M

A
n

A
A
M

0
0

N

(A
n
)


M

0.
(We could have asserted that the bottom row is also exact, had we assumed A
Noetherian; in any case the map is surjective, by the Mittag-Leer lemma.) Now
it follows that the right vertical map is surjective, and (1) holds.
(2): If A is Noetherian, then the bottom row is exact, and N is f.g., so we have
the surjectivity of the left vertical arrow from part (1). But now the snake lemma
implies that the right vertical arrow is an isomorphism.
Corollary 17.3.2. If A is Noetherian, then any I-adic completion

A is a at
A-module.
Proof. The implication M

M

A
A
M

A
A
M holds when M

, M
are f.g. modules, by the left-exactness of the functor on f.g. modules. But this
is sucient to prove this implication for arbitrary M

, M, by Atiyah-Macdonald,
Prop. 2.19.
18. Lecture 18
Further consequences of Artin-Rees, such as Krulls theorem. Associated graded
rings/modules. Proof that ANoeth

ANoeth. Geom. meaning of G(A): tangent
cone. Hensels lemma.
************************
18.1. Further consuequences of Artin-Rees.
Lemma 18.1.1. Suppose A is Noetherian, I A is an ideal, and denotes I-adic
completion.
(0) If M is a f.g. A-module, and M

M is a submodule, then k Z
0
such
that I
nk
(I
k
M M

) = I
n
M M

, for all n k.
(1)

I =

AI

=

A
A
I.
(2)

I
n
=

I
n
.
(3) I
n
/I
n+1
=

I
n
/

I
n+1
. Similarly, A/I
n
=

A/

I
n
.
(4)

I Jac. rad. of

A.
Proof. (0): This is a reformulation of the fact that I
n
MM

is a stable I-ltration
on M

, a consequence of the Artin-Rees lemma.

(1): We have established the natural isomorphism

A
A
I

I. Note that the image
of this map is just

AI.
(2): By (1), we have

I
n
=

AI
n
= (

AI)
n
=

I
n
.
(3): By (2) and exactness of , we have

I
n
/

I
n+1
=

I
n
/

I
n+1
=

I
n
/I
n+1
. This is
just I
n
/I
n+1
, since the latter is discrete.
(4): Note that since

I
n
=

I
n
, the ring

A is complete for the

I-adic topology. But
then for any

I, we have the convergent geometric series
(1 )
1
= 1 + +
2
+

A,
44 THOMAS J. HAINES
and so 1

A

. It follows that belongs to the Jacobson radical of

A.
Lemma 18.1.2. If (A, m) is a Noetherian local ring, and

A is the m-adic comple-
tion, the (

A, m) is local.
Proof. By (3),

A/ m = A/m, and so m is a maximal ideal in

A. By (4), m is the
Jacobson radical, and hence is the only maximal ideal.
Theorem 18.1.3 (Krulls Theorem). Suppose A is Noetherian, I A is an ideal,
M is a f.g. A-module, and

M is its I-adic completion. Let E :=

n=0
I
n
M, the
kernel of the natural map M

M. Then
E = x M [ (1 )x = 0 for some I.
Proof. Note that the subspace topology E inherits from M is the trivial one: the
only open subsets are and E itself. By the Artin-Rees lemma, the I-adic topology
on E is also trivial, which means that E = IE. If E = Ax
1
+ +Ax
n
, then we may
write x
i
=

j

ij
x
j
, for some
ij
I. Then the element det(
ij

ij
) 1 + I
kills E (by the Cramers Rule Trick).
Conversely, (1 )x = 0, I = x = x =
2
x =

n=1
I
n
M = E.
Corollary 18.1.4. If A is a Noetherian domain, and I ,= (1), then

n=1
I
n
= 0.
Proof. Note that 1 +I has no zero-divisors.
Corollary 18.1.5. If A is Noetherian and the ideal I belongs to the Jacobson
radical of A, and M is a f.g. A-module, then

n=1
I
n
M = 0 (and thus M is
Hausdor WRT the I-adic topology). In particular, if (A, m) is Noetherian local,
and M is f.g., then

n=1
m
n
M = 0.
Proof. Note that 1 +I A

.
Corollary 18.1.6. If B is Noetherian, and p Spec(B), then ker(B B
p
) is the
intersection of the p-primary ideals of B.
Proof. See Atiyah-Macdonald, 10.21.
18.2. A Noetherian implies

A Noetherian. Our goal is to prove that

A is
Noetherian if A is. We need to study associated graded rings/modules.
Suppose I is an ideal in a ring A, and M is an A-module equipped with an
I-ltration M
n
. Dene the associated graded ring G
I
(A) :=

n=0
I
n
/I
n+1
.
Similarly, dene the associated graded module G
I
(M) :=

n=0
M
n
/M
n+1
.
It is clear that G
I
(A) is a graded ring, with multiplication dened by x
n
x
m
=
x
n
x
m
(where this notation has the obvious meaning). Also, G
I
(M) is a graded
module over G
I
(A).
Proposition 18.2.1. Suppose A is Noetherian.
(1) G
I
(A) is Noetherian.
(2) G

I
(

A) = G
I
(A).
(3) If M is a f.g. A-module, and M
n
is stable, then G(M) is f.g. G(A)-module.
Proof. (1): Write I = (x
1
, . . . , x
r
), and I/I
2
= ( x
1
, . . . , x
r
). Then G(A) =
A/I[ x
1
, . . . , x
r
], hence is Noetherian by the Hilbert basis theorem.
(2): This is clear from Lemma 18.1.1.
(3): Suppose M
n0+n
= I
n
M
n0
for n 0. Then G(M) is generated over G(A) by

n0
n=0
M
n
/M
n+1
. Each M
n
/M
n+1
is f.g. as an A/I-module, hence we are done.
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 45
We need a kind of converse to the implication in (3) above. The following lemma
is a crucial tool.
Lemma 18.2.2. Suppose : A B is a homomorphism of ltered abelian groups,
i.e., (A
n
) B
n
for all n. Clearly induces maps G() : G(A) G(B) and

:

A

B. Then:
(1) G() injective =

is injective.
(2) G() surjective =

is surjective.
Proof. See Atiyah-Macdonald, 10.23.
Lemma 18.2.3. Suppose A is I-adically complete, and M has a separated I-
ltration M
n
(i.e.
n
M
n
= 0). Then G(M) f.g. over G(A) = M is f.g.
over A.
Proof. Choose a nite set of homogeneous generators x
i
M
n(i)
for G(M) over
G(A). Dene F
i
= A endowed with the stable I-ltration F
i
k
:= I
kn(i)
(where we
set I

= A if < 0). Dene

: F :=
i
F
i
M
by e
i
x
i
M
n(i)
, where e
i
is the ith standard basis vector. This map is ltered
in the sense of the previous lemma. By denition G() is surjective, and so by
Lemma 18.2.2,

is surjective. Now consider the commutative diagram
F

M.
The left vertical arrow is an isomorphism because A =

A. Since

is surjective, we
see that the right vertical arrow is surjective. But then the right vertical arrow is
an isomorphism (it is injective since
n
M
n
= 0). Now going around the diagram,
we see that is surjective, and so M is f.g. as an A-module.
Corollary 18.2.4. Let A, M be as above. Then G(M) is a Noetherian G(A)-
module = M is a Noetherian A-module.
Proof. Let M

M be a submodule; we want to show M

is f.g. as an A-module.
Consider the separated I-ltration M

n
:= M

M
n
. We have M

n
/M

n+1

M
n
/M
n+1
, which implies that G() : G(M

) G(M). Thus G(M

) is a f.g.
G(A)-module, and then by Lemma 18.2.3, we see M

if a f.g. A-module.
Theorem 18.2.5. A Noetherian =

A Noetherian.
Proof. Note that G
I
(A) = G

I
(

A) so the latter is Noetherian. Also,

A is

I-adically
complete (Lemma 18.1.1), and so the

I-adic topology is separated:
n

I
n
= 0.
Applying the above corollary to M =

A, we get that

A is Noetherian.
Remark 18.2.6. If k is a eld, then k[[X
1
, . . . , X
n
]] is Noetherian, since it
is the (X
1
, . . . , X
n
)-adic completion of the Noetherian ring k[X
1
, . . . , X
n
].
Let m Spec
m
(A). Then the following two m-adic completions are canoni-
cally isomorphic:

A =

A
m
. Why? Use the identities A
m
/m
n
A
m
= (A/m
n
)
m
=
A/m
n
.
46 THOMAS J. HAINES
18.3. Geometric meaning of the associated graded ring: the tangent cone.
The following discussion is taken from Mumfords book The Red Book of Varieties
and Schemes, III, 3. Suppose k = k. Given a closed point x V (I) for I
k[X
1
, . . . , X
n
] a radical ideal, let A = k[X
1
, . . . , X
n
]/I and let m Spec
m
(A) be
the maximal ideal corresponding to x. Then we dene the tangent cone to V (I) at
x to be the ane scheme Spec(G
m
(A)).
How can we compute this scheme? Mumford explains it, and here we just
give the answer: WLOG x is the origin in the ambient ane space A
n
. For any
element 0 ,= f k[X
1
, . . . , X
n
], write it in the form f = f
s
+ f
s+1
+ + f
s+r
,
where f
s
,= 0 and each summand f
k
is the degree k homogeneous part of f; denote
the lowest degree term f
s
simply by f

. Let I

be the ideal generated by all f

,
where f ranges over all elements 0 ,= f I. Then
G
m
(A) = k[X
1
, . . . , X
n
]/I

.
From this, we can see why Spec(G
m
(A)) is called the tangent cone. Lets consider
two examples:
Curve with cusp: Let V (I) be the plane curve given by Y
2
X
3
. Then f

= Y
2
,
and so the tangent cone at (0, 0) V (I) is the spectrum of the ring k[X, Y ]/Y
2
, in
other words, the X-axis with multiplicity two. Note that the tangent cone is an
ane scheme but is not an ane variety (the ring is not reduced), even though the
curve we started out with was a variety. This is another example of how schemes
enter into the study of varieties.
Nodal curve: Let V (I) be the place curve given by X
2
Y
2
+ X
3
. Then f

=
(X Y )(X +Y ), and the tangent cone at (0, 0) is the union of the lines X = Y .
18.4. Hensels Lemma. This is another application of complete local rings. For
motivation, consider the equation X
2
+ 1 = 0 in Z[X]. It has no solutions X Z.
However, modulo 5, this equation has two distinct solutions, namely, X = 2, 3. We
cant lift these solutions to Z, but we can lift them to the completion of Z at the
prime ideal 5Z, namely the 5-adic numbers Z
5
. Indeed that is a very special case
of the following result.
Proposition 18.4.1 (Hensels Lemma). Let (A, m) be a local ring such that A is
m-adically complete. Let k := A/m. Suppose F(X) A[X] is monic, and write
F(X) k[X] for its reduction modulo m. If F = gh in k[X], where (g, h) = 1 and
g, h are monic, then there exist monic G, H A[X] such that F = GH and G = g,
H = h.
Proof. Choose arbitrary monics G
1
, H
1
A[X] which lift g, h respectively. Then
F G
1
H
1
mod m[X].
By induction, suppose we have constructed monics G
n
, H
n
A[X] with G
n
= g,
H
n
= h, and F G
n
H
n
mod m
n
[X]. Then we can write
F G
n
H
n
=

i
U
i
(X)
where
i
m
n
and deg(U
i
) < deg(F), for all i.
Since (g, h) = 1, there exist v
i
, w
i
k[X] such that U
i
= gv
i
+ hw
i
. WLOG
deg(v
i
) < deg(h) (if necessary, replace v
i
with its remainder mod h, absorbing the
dierence into w
i
).
Then deg(hw
i
) = deg(U
i
gv
i
) < deg(F), hence deg(w
i
) < deg(g).
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 47
Choose V
i
, W
i
A[X] such that V
i
= v
i
, W
i
= w
i
, and with deg(V
i
) = deg(v
i
)
and deg(W
i
) = deg(w
i
). Set
G
n+1
= G
n
+

i
W
i
H
n+1
= H
n
+

i
V
i
.
Note that F G
n+1
H
n+1
mod m
n+1
[X] (check this!). We then set
lim
n
G
n
= G
lim
n
H
n
= H.
(By construction and the completeness of A, these limits exist.) It is easy to check
that G, H have the desired properties.
19. Lecture 19
Hilbert functions, etc: Motivation comparing dimA
0
[X
1
, . . . , X
s
] = s with ord
t=1
P(A, t)
and deg
n

A0
A
n
...
Rationality and explicit expression for P(M, t). Lemma for I-stable lt M
n
of
graded A-module M: (M/M
n
) = g(n) for n >> 0.
*******************
19.1. Combinatorics and a motivating example. We consider the polynomial
ring A = A
0
[X
1
, . . . , X
s
], where A
0
is an Artin ring (that is, a Noetherian ring
of dimension 0). This is a graded ring A =
n0
A
n
, where A
n
is the free A
0
-
module generated by the set of monomials of form X
m1
1
X
ms
s
, where m
i
0
and

i
m
i
= n. It is not hard to count these monomials: arrange n + s 1 dots
in a row, and cross out s 1 of them. We get s ordered clumps of dots, with say
m
i
dots in the ith clump, and the total number of remaining dots is n. Clearly
such arrangements correspond bijectively to the monomials we are counting. On
the other hand, the number of possible such arrangements is simply
_
n+s1
s1
_
.
As a consequence, setting all X
i
to t in the obvious formula
(19.1.1)
s

i=1
(1 X
i
)
1
=

n0
(

|m|=n
X
m1
1
X
ms
s
)
yields
(19.1.2) (1 t)
s
=

n0
_
n +s 1
s 1
_
t
n
=

n0
rank
A0
(A
n
)t
n
,
where rank
A0
(A
n
) denotes the rank of the A
0
-module A
n
. We have
rank
A0
(A
n
) =
A0
(A
n
)
A0
(A
0
)
1
,
where
A0
(A
n
) denotes the length of the A
0
-module A
n
.
We will see later that dim A = s + dim A
0
= s (we already know this when A
0
is a eld)
6
. Thus, we see that the dimension of A, which is s, is also the order of
6
Actually, we wont have time for this. Here are the basic facts. If A is any ring and B = A[X],
then dim A + 1 dim B 2dim A + 1. If A is Noetherian, then dim A + 1 = dim B. A good
reference for this is [Serre], III.D.1.
48 THOMAS J. HAINES
the pole at t = 1 in the power series

n0

A0
(A
n
)t
n
. We will generalize this fact
to other rings in the next sections.
19.2. Hilbert functions. Let A =
n
A
n
be a Noetherian graded ring (so that A
0
is Noetherian, and there is a nite set of homogeneous elements x
i
A
ki
, k
i
> 0,
such that A = A
0
[x
1
, . . . , x
s
]). Let M =
n
M
n
be a f.g. graded A-module
(so that each M
n
is a f.g. A
0
-module: indeed, if M is generated over A by ho-
mogeous m
1
, . . . , m
t
of degree r
1
, . . . , r
t
, then M
n
is generated over A
0
by terms
g(x
1
, . . . , x
s
)m
j
where r
j
n and g is a monomial of degree n r
j
).
Let be a Z-valued additive function on the category of nite-length A
0
-modules.
This means that for a short exact sequence of such A
0
-modules
0 M

M M

0
we have (M) = (M

) +(M

). It follows that if 0 K
0
K
1
K
l
0
is exact in this category, then

i
(1)
i
(K
i
) = 0.
We dene the Poincare series of M (WRT ) to be the formal power series
P(M, t) :=

n=0
(M
n
)t
n
Z[[t]].
Theorem 19.2.1. We have P(M, t) =
f(t)

s
i=1
(1 t
ki
)
, for some f(t) Z[t].
Proof. Induction on s. If s = 0, then A = A
0
and M
n
= 0 for n >> 0. Thus
P(M, t) Z[t] in this case.
Assume the theorem holds for graded rings generated over A
0
by s 1 elements.
We have an exact sequence for every n
(19.2.1) 0

K
n

M
n
xs

M
n+ks

L
n+ks

0.
Let K :=
n
K
n
and L :=
n
L
n
(for the latter, the initial terms for n < k
s
are not
dened simply set them equal to 0). These are f.g. graded A-modules, killed by
x
s
(check!), so are f.g. graded A
0
[x
1
, . . . , x
s1
]-modules.
Multiplying
(K
n
) (M
n
) +(M
n+ks
) (L
n+ks
)
by t
ks
and summing over n, we get the equality
(1 t
ks
)P(M, t) = t
ks
P(K, t) +P(L, t) +g(t),
for some g(t) Z[t]. Using the induction hypothesis applied to K, L, the result
follows.
Now we dene
d(M) := ord
t=1
P(M, t),
the order of the pole at t = 1. The number d(M) is an integer s.
Corollary 19.2.2. If k
i
= 1 for all i, then (M
n
), as a function of n, belongs to
[n] for suciently large n, having degree d(M) 1.
Proof. Let d := d(M). WLOG P(M, t) =
f(t)
(1t)
d
, where f(1) ,= 0. Write f(t) =

N
k=0
a
k
t
k
. Now (M
n
) is the coecient of t
n
in the product
N

k=0
a
k
t
k
(1 t)
d
,
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 49
which by using (19.1.2) with s = d is
(M
n
) =
N

k=0
a
k
_
n k +d 1
d 1
_
as long as n N. Viewed as a polynomial in n (with -coecients), this has
leading term
(
N

k=0
a
k
)n
d1
/(d 1)!,
proving the corollary.
The following related result is also useful.
Proposition 19.2.3. Suppose x A
k
is not a zero-divisor for M. Then d(M/xM) =
d(M) 1.
Proof. In (19.2.1) we have K
n
= 0, and the exact sequence becomes
0

M
n
x

M
n+k

L
n+k

0.
From the previous argument, we see that (1t
k
)P(M, t) = P(M/xM, t) +g(t) (for
some polynomial g) and this implies the result.
Proposition 19.2.4. Let (A, m) be a Noetherian local ring, and I an m-primary
ideal. Let M be a f.g. A-module, with a stable I-ltration M
n
. Then
(i)
A
(M/M
n
) < .
(ii) If I is generated by s elements x
1
, . . . , x
s
, then (M/M
n
) = g(n) for n >>
0, where g [n] is a polynomial in n with degree s.
(iii) The degree and leading coecient of g(n) depend on M and I, but not the
choice of I-stable ltration M
n
.
Proof. (i): Its ETS (M
n
/M
n+1
) < . Now M
n
/M
n+1
is a f.g. A/I-module,
and as A/I is Artin (being Noetherian and dim 0 check this, using that I is
m-primary), this means that (M
n
/M
n+1
) =
A/I
(M
n
/M
n+1
) < .
(ii): The associated graded ring G(A) = A/I[x
1
, . . . , x
s
] is Noetherian, and G(M) =

n
M
n
/M
n+1
is f.g. as a G(A)-module (since the ltration is stable). Therefore,
by Corollary 19.2.2, (M
n
/M
n+1
) = f(n) is a polynomial in [n] for n n
0
(some
n
0
), having degree s 1. Now the equality
(M/M
n+1
) = (M/M
n0
) +(M
n0
/M
n0+1
) + +(M
n1
/M
n
) +(M
n
/M
n+1
),
shows that for n n
0
, (M/M
n
) is given by a polynomial g(n) [n] having
degree s.
(iii): Let

M
n
denote another I-stable ltration on M, and g(n) [n] be the
corresponding polynomial giving (M/

M
n
) for n >> 0. Since M
n
and

M
n
have
bounded dierence, n
0
such that M
n+n0


M
n
and

M
n+n0
M
n
for all n 0.
It follows that g(n + n
0
) g(n) and g(n + n
0
) g(n) for all n >> 0 This implies
that g and g have the same degree and leading coecient.
50 THOMAS J. HAINES
19.3. Characteristic polynomial of a primary ideal. We dene
M
I
(n) :=
g(n) = (M/M
n
). In the special case M = A, we call
I
(n) :=
A
I
(n) the charac-
teristic polynomial of the m-primary ideal I.
Lemma 19.3.1. Let (A, m), I be as above. Then deg
I
(n) = deg
m
(n), so that
deg
I
(n) is independent of the choice of I.
Proof. There exists an integer r 1 such that m I m
r
, so that m
n
I
n
m
rn
,
for all n 0.
This implies that
m
(n)
I
(n)
m
(rn) for all n 0, which implies the
result.
Thus we may dene the quantity d(A) := deg
I
(n), where I is any m-primary
ideal. Note that
d(A) = d(G
m
(A)),
where the RHS is dened as in the beginning of this section. From Corollary 19.2.2
and the equality
deg
n
(A/m
n
) = deg
n
(m
n
/m
n+1
) + 1
we see that d(A) = d(G
m
(A)).
20. Lecture 20
Proof that d(A) = dim(A) = (A), for Noetherian local rings A.
**********************
20.1. The equality of three characterizations of Krull dimension. Let (A, m)
be a Noetherian local ring. We have not yet shown that the dimension of A is nite.
We will next prove the stronger fact that the following three numbers are equal:
dim A
(A) := the minimal number of generators of an m-primary ideal
d(A) = deg
n

I
(n), for any m-primary ideal I.
Well show that (A) d(A) dim A (A). We have already proved (A)
d(A). Indeed, this follows from Proposition 19.2.4, (ii).
Next, we will show d(A) dim A, by induction on d(A). If d(A) = 0 then
(A/m
n
) is constant for large n, and hence m
n
= m
n+1
for large n. By NAK this
implies that m
n
= 0 and so A

= A/m
n
. Thus A is Artin, and so dim A = 0.
Now assume that d(A) 1, and that the inequality holds for rings A

with
d(A

) < d(A).
Claim: dim A 1.
Proof: If dim A = 0, then m is the unique prime ideal, and so (0) is m-primary. So
using Lemma 19.3.1 we see
I
(n) is constant and d(A) = 0, a contradiction.
Now write r 1 for dim A. Choose a chain of prime ideals p
0
_ p
1
_ _ p
r
in A.
Claim: r d(A).
Proof: Choose x p
1
, x / p
0
. Let x

= x A/p
0
=: A

.
Note that d(A

) d(A): m

:= m A

is maximal, and A/m

n
A

/(m

)
n
. This
implies (A

/(m

)
n
) (A/m
n
), and so d(A

) d(A).
Using this together with Lemma 20.1.1 below (applied to B = A

and y = x

), we
see d(A

/(x

)) d(A

) 1 d(A) 1, and hence the induction hypothesis applies

to the ring A

/(x

). Thus dim A

/(x

) d(A

/(x

)).
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 51
But modulo (x

), we have the chain p

1
_ _ p
r
in Spec(A

/(x

)), and so the

above remarks give us r 1 d(A) 1, or dim A d(A), as desired. The claim is
proved, modulo the lemma below.
Lemma 20.1.1. Let (B, n) be a Noetherian local ring with n-primary ideal I, and
let M be a f.g. B-module. If y B is not a zero-divisor in M, and if M := M/yM,
then deg
M
I
deg
M
I
1. In particular, taking M = B, we have d(B/yB)
d(B) 1.
Proof. We have N := yM

= M as a B-module. Consider the exact sequence
0 N/(N I
n
M) M/I
n
M M/I
n
M 0.
We have
(M/I
n
M) = (N/(N I
n
M)) +(M/I
n
M).
By Artin-Rees and N

= M (and using Proposition 19.2.4, (iii)), the rst summand
on the RHS has the same degree and leading coecient as the LHS. This implies
the desired bound on the degree of the remaining term on the RHS.
It remains to prove the dim A (A). Write d = dim A. Its enough to construct
a sequence of elements
x
1
, . . . , x
i
, . . . , x
d
such that for each i the following holds:
(20.1.1) For any prime p (x
1
, . . . , x
i
), we have ht(p) i.
Indeed, then for any prime p (x
1
, . . . , x
d
) we would have ht(p) d, which
can only happen if p = m. Thus (x
1
, . . . , x
d
) is m-primary, and is generated by d
elements. Thus, d (A).
We will construct the sequence x
1
, . . . , x
d
as in (20.1.1) by induction. Suppose
i 1 and we have already constructed x
1
, . . . , x
i1
satisfying (20.1.1). Let p
j
, for
j = 1, . . . , l, be the minimal primes containing (x
1
, . . . , x
i1
) having height i 1
(there might be no such primes).
Claim: m _
l
j=1
p
j
.
Proof: If m
j
p
j
, then by Atiyah-Macdonald 1.11, we would have m p
j
for
some j. This would imply m = p
j
, and so ht(m) = i 1 < d, a contradiction.
So we may choose an element x
i
m, x
i
/
j
p
j
. We claim that x
1
, . . . , x
i
satises (20.1.1).
Suppose that we have a prime q (x
1
, . . . , x
i
). We need to show that ht(q) i.
Clearly q p, where the latter is some minimal prime containing (x
1
, . . . , x
i
).
Case 1: p = p
j
, for some j. Then as q ,= p
j
(by choice of x
i
), we have ht(q)
ht(p) + 1 = i.
Case 2: p ,= p
j
for all j. Then ht(q) ht(p) i.
In summary, we have now proved the following fundamental theorem.
Theorem 20.1.2. Let (A, m) be a Noetherian local ring. Then dim A = d(A) =
(A). In particular, dim A < .
52 THOMAS J. HAINES
20.2. Consequences of the dimension theorem for Noetherian local rings.
Recall that for any prime ideal in any ring A, we have ht(p) = dim A
p
.
Corollary 20.2.1. If A is Noetherian, and p is a prime ideal, then ht(p) < ,
and therefore prime ideals in A satisfy the d.c.c. (descending chain condition: all
descending chains are eventually stationary).
Example: It is clear that the Noetherian hypothesis is necessary: consider A =
k[X
1
, X
2
, X
3
, . . . ] the polynomial ring in innitely many variables. Then the chain
of prime ideals
(X
1
, X
2
, X
3
, . . . ) _ (X
2
, X
3
, X
4
, . . . ) _ (X
3
, X
4
, X
5
, . . . )
is not stationary.
Corollary 20.2.2. If (A, m) is Noetherian local, and k := A/m, then dim A
dim
k
m/m
2
.
Proof. Indeed, NAK implies that m is generated by at most dim
k
m/m
2
elements.

Corollary 20.2.3. If A is Noetherian, then any minimal prime p (x

1
, . . . , x
r
)
has ht(p) r.
Proof. The ideal (x
1
, . . . , x
r
) becomes pA
p
-primary in A
p
. So the result follows
from the inequality dim A
p
r.
Corollary 20.2.4 (Krulls Hauptidealsatz). Let A be a Noetherian ring, and x A
neither a unit nor a zero-divisor. Then every minimal prime p (x) has ht(p) = 1.
Proof. By the preceding corollary, ht(p) 1. If ht(p) = 0, then p belongs to 0, and
thus every element in p is a zero-divisor (Lemma 12.1.1, (b)), a contradiction.
Corollary 20.2.5. If (A, m) is Noetherian local, and x m is not a zero-divisor,
then dim A/(x) = dim A1.
Proof. We have already proved, just a few paragraphs ago, that d(A/(x)) d(A)
1, and so holds. To prove the opposite inequality, write d = dim A/(x), and sup-
pose x
1
, . . . , x
d
generate an m/(x)-primary ideal. It follows that I := (x, x
1
, . . . , x
d
)
is m-primary: if the only prime ideal containing I/(x) is m/(x), then the only prime
ideal containing I is m. Hence d + 1 dim A, as desired.
Corollary 20.2.6. Let (

A, m) denote the m-adic completion of (A, m). Then

dim A = dim

A.
Proof. Since A/m
n
=

A/ m
n
for all n 0, we have (A) = (

A).
Example: We have dim k[[X
1
, . . . , X
n
]] = n. In the corollary above, take A =
k[X
1
, . . . , X
n
]
m
, where m = (X
1
, . . . , X
n
), and observe that k[[X
1
, . . . , X
n
]] is the
m-adic completion of A.
21. Lecture 21
When a Noetherian domain is a UFD. Denition of regular local rings.
**************************
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 53
21.1. Applications, in particular of Krulls Hauptidealsatz. Recall: for A a
domain, we dene
a ,= 0 is irreducible if a / A

and a is not the product of two non-units.

A is a UFD if each a ,= 0 is the pruduct of a unit and nitely many
irreducible elements, uniquely up to units and reordering.
Lemma 21.1.1. A Noetherian domain A in which every irreducible element gen-
erates a prime ideal is a UFD.
Proof. First we need to show the existence of factorizations.
Consider the family of (principal) ideals
o := 0 ,= (a) [ a is not the product of nitely many irred. elements.
If o ,= , then o has a maximal element (a). The element a is neither a unit nor
irreducible. So a = a
1
a
2
, where a
i
is a non-unit. We have (a) _ (a
i
), for i = 1, 2,
and by maximality both a
i
are products of nitely many irreducible elements. Hence
so is a, a contradiction.
Therefore, given a, we may write a = ua
1
a
r
, where u A

and each a
i
is
irreducible. To prove uniqueness, assume
ua
1
a
r
= vb
1
b
s
where v A

and each b
j
is irreducible. Since vb
1
b
s
(a
1
) (a prime ideal by
hypothesis), WLOG b
1
(a
1
). Thus
(b
1
) (a
1
);
this is an inclusion of prime ideals by hypothesis, and such primes are ht 1, by
Corollary 20.2.4. Thus (b
1
) = (a
1
).
7
Thus WLOG b
1
= a
1
(absorbing a unit into
v, say), and then we may cancel these from both sides. Continuing, the uniqueness
statement follows.
Theorem 21.1.2. A Noetherian domain A is a UFD i every prime of ht 1 is
principal.
Proof. (): Suppose A is irreducible. By the preceding lemma, its ETS ()
is prime. Let p () be a minimal prime ideal. Corollary 20.2.4 implies ht(p) = 1,
and so p is principal by hypothesis, say p = (a). Then (as noted in the footnote
above) () = (a) and so () is indeed prime.
(): Suppose ht(p) = 1. Choose x ,= 0, x p. WLOG (factor x), we can assume
x is irreducible. Note that because A is a UFD, (x) is then a prime ideal (check
this!).
Then (0) _ (x) p, and (x) is prime, so ht(p) = 1 = (x) = p, and p is
principal, as desired.
The following fact is important in number theory.
Corollary 21.1.3. Let A be a Dedekind domain (i.e. a Noetherian normal domain
of dimension 1). Then TFAE:
(1) A is a UFD.
(2) Every non-zero prime ideal in A is principal.
7
We can also argue as follows: note that if b is irreducible, then the ideal (b) is a maximal
element in the collection of all proper principal ideals; this shows (b
1
) = (a
1
), without invoking
Krulls Hauptidealsatz (pointed out in class by Moshe Adrian).
54 THOMAS J. HAINES
(3) A is a PID.
Proof. (1) (2): Note that a prime ideal is non-zero i it has ht 1.
(2) (3): Every proper non-zero ideal a is product of non-zero prime ideals a =
p
a1
1
p
ar
r
. Hence every such a is principal if and only if every non-zero prime ideal
is principal.
Aside: The geometric meaning of the Theorem (see Hartshorne II, Prop. 6.2): Let
A be a Noetherian domain. and X = Spec(A). Then A is a UFD i A is normal
and Div(X)/(principal divisors) = 0.
21.2. Denition of regular local ring. Throughout this subsection, let (A, m)
denote a Noetherian local ring, with residue eld k = A/m. Let d = dim A.
Theorem 21.2.1. TFAE:
(i) G
m
(A)

= k[t
1
, . . . , t
d
] as graded k-algebras (the t
i
s are indeterminates).
(ii) dim
k
m/m
2
= d.
(iii) m can be generated by d elements.
We call a Noetherian local ring A a regular local ring if the equivalent condi-
tions (i)-(iii) hold.
For the proof, note that (i) (ii) is clear, and (ii) (iii) follows by NAK.
Before giving the proof of (iii) (i), we need some preliminaries. First, we call
any set of d elements x
1
, . . . , x
d
which generate an m-primary ideal a system of
parameters. Choose such a system and set I = (x
1
, . . . , x
d
).
Lemma 21.2.2. Let x
1
, . . . , x
d
I/I
2
. Let f(t
1
, . . . , t
d
) A[t
1
, . . . , t
d
] be homo-
geneous of degree s. If f(x
1
, . . . , x
d
) I
s+1
, then f m[t
1
, . . . , t
d
].
Proof. Let f(t
1
, . . . , t
d
) A/I[t
1
, . . . , t
d
] denote the image of f modulo I. There is
a surjective graded A/I-algebra homomorphism
: A/I[t
1
, . . . , t
d
] G
I
(A)
given by t
i
x
i
, for i = 1, . . . , d. By assumption f ker().
Assume that not all coecients of f belong to m; then some coecient of f is a
unit, and so (check this!), f is not a zero-divisor in A/I[t
1
, . . . , t
d
]. Then we have
d = d(G
I
(A))

d(A/I[t
1
, . . . , t
d
]/(f))
= d(A/I[t
1
, . . . , t
d
]) 1
= d 1,
which is a contradiction. We are done, modulo the inequality

above. This
follows from the following general lemma.
Lemma 21.2.3. Let M =
n
M
n
N =
n
N
n
be a surjective homomorphism of
graded modules over the graded ring A/I[t
1
, . . . , t
d
]. Then
A/I
(N
n
)
A/I
(M
n
),
and thus d(N) d(M).
Proof. Note that d(N) 1 = deg
n
(N
n
), and similarly for M replacing N.
Proof of Theorem 21.2.1, (iii) (i): Suppose m = (x
1
, . . . , x
d
). Dene the graded
A/m-algebra surjective homomorphism
A/m[t
1
, . . . , t
d
] G
m
(A)
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 55
by t
i
x
i
m/m
2
. By Lemma 21.2.2, this is injective, hence is an isomorphism.

Corollary 21.2.4. Suppose (

A, m) is the m-adic completion of the Noetherian local

ring (A, m). Then A is regular i

A is regular.
Proof. Indeed, we have G
m
(A) = G
m
(

A), and so the latter is a polynomial ring

over k in d = dim(A) = dim(

A) variables.
As a corollary of Lemma 21.2.2 above, we have the following result we will use
later.
Corollary 21.2.5. Suppose A contains a eld k mapping isomorphically onto A/m.
Then any system of parameters x
1
, . . . , x
d
is algebraically independent over k.
Proof. Let I be the ideal generated by the elements x
1
, . . . , x
d
. Suppose (x
1
, . . . , x
d
)
is a zero of 0 ,= f(t
1
, . . . , t
d
) k[t
1
, . . . , t
d
]. Write f = f
s
+ (deg > s terms), where
f
s
,= 0 is homogeneous of degree s. Then f
s
(x
1
, . . . , x
d
) I
s+1
, and so by Lemma
21.2.2, all coecients of f
s
belong to k m = 0, a contradiction.
21.3. Regular local rings are domains, and a consequence.
Proposition 21.3.1. If A is a regular local ring, then A is a domain.
This follows from the next lemma.
Lemma 21.3.2. Let A be any ring and I is an ideal such that
n1
I
n
= 0. Assume
that G
I
(A) is a domain. Then A is a domain.
Proof. If x, y ,= 0, then there exist non-negative integers n, m with x I
n
I
n+1
and y I
m
I
m+1
. So 0 ,= x I
n
/I
n+1
and 0 ,= y I
m
/I
m+1
. Since G
I
(A) is a
domain, it follows that 0 ,= xy I
n+m
/I
n+m+1
. Thus xy ,= 0.
Here is a nice consequence:
Corollary 21.3.3. The dimension 1 regular local Noetherian rings are precisely
the dimension 1 local Noetherian domains such that m is principal, ie., the DVRs.
By a curve we mean a 1-dimensional variety over a eld k. We usually assume
k = k, though this is not always necessary. We say a curve X is regular if every
local ring O
x
is regular, for every closed point x. The above corollary means that
an irreducible curve X is regular i it is normal (meaning each O
x
is normal).
22. Lecture 22
More dimension theory. Comparison of regular local rings with non-singular
points on alg. var.
******************
22.1. More dimension theory. We can now give the proof of a fact we mentioned
earlier (Proposition 7.6.1).
if: Since f is irreducible and k[X
1
, . . . , X
n
] is a UFD, the ideal (f) is prime, call
it P. Because any maximal chain of prime ideals in k[X
1
, . . . , X
n
] has length n, we
know that
ht(P) + dim(A/P) = n.
By Krulls Hauptidealsatz, ht(P) = 1, and so dim(Y ) = dim(A/P) = n 1.
56 THOMAS J. HAINES
only if: Since Y is closed and irreducible, Y = V (P) for some prime ideal P. Since
dim(Y ) = n 1, the above reasoning shows that ht(P) = 1. Since k[X
1
, . . . , X
n
] is
a UFD, P is principal, say P = (f). Since P is prime, f must be irreducible.
Our next goal is to prove a related result, where k[X
1
, . . . , X
n
] is replaced with
an arbitrary f.g. k-algebra which is a domain.
Proposition 22.1.1. Suppose k = k. Let A be a f.g. k-algebra, which is a domain.
Suppose f A, f / A

, f ,= 0. Let d = dim(A). Then V (f) = Spec(A/(f)) is

pure of dimension d 1.
To say that V (f) is pure of dimension d 1 means, by denition, that all the
irreducible components of Spec(A/(f)) have dimension d 1.
Proof. Let Z
1
, . . . , Z
l
be the irreducible components of V (f) = Spec(A/(f)). Then
Z
i
= V (P
i
), where P
i
ranges over the nite set of minimal primes P
i
(f). By
Krulls Hauptidealsatz, ht(P
i
) = 1 for each i, and from the equality
ht(P
i
) + dim(A/P
i
) = dim(A) = d
we get dim(Z
i
) = dim(A/P
i
) = d 1, for each i.
Here is another useful result.
Proposition 22.1.2. Suppose A = k[X
1
, . . . , X
n
]/(f
1
, . . . , f
t
), and (f
1
, . . . , f
t
) ,=
(1). Then dim(A) n t.
Proof. An irreducible component of Spec(A) is of the formV (P) Spec(k[X
1
, . . . , X
n
])
where P (f
1
, . . . , f
t
) is minimal. We have the equality
ht(P) + dim(V (P)) = n.
On the other hand, we know that ht(P) t. Putting these together, we nd
dim(V (P)) n t.
Since dim(A) = sup
P
dim(V (P)), the desired inequality follows.
We say a closed irreducible subset Y A
n
k
as above is a local complete in-
tersection if there exist f
1
, . . . , f
t
k[X
1
, . . . , X
n
] such that I(Y ) = (f
1
, . . . , f
t
)
and
dim Y = n t.
22.2. Regularity vs non-singularity. The following comparison between regu-
larity of local rings and non-singularity of points on varieties is taken from Hartshorne,
I 5.
Fix k = k, and a closed subvariety Y A
n
k
(i.e. a Zariski-closed subset of
Spec(k[X
1
, . . . , X
n
]), which is reduced: we can write Y = V (I(Y )) where I(Y )
k[X
1
, . . . , X
n
] is a radical ideal; then Y

= Spec(k[X
1
, . . . , X
n
]/I(Y )) and the ring
of regular functions on Y , namely O
Y
:= k[X
1
, . . . , X
n
]/I(Y ) has no non-zero
nilpotents). Suppose in addition that Y is irreducible: this means I(Y ) is a prime
ideal.
Write I(Y ) = (f
1
, . . . , f
t
).
Suppose the closed point P = (a
1
, . . . , a
n
) belongs to Y , i.e. (X
1
a
1
, . . . , X
n

a
n
) I(Y ). Then we say Y is non-singular at P if
(22.2.1) rank J(P) = n dim(Y ),
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 57
where J denotes the Jacobian matrix
J =
_
f
i
X
j
_
ij
a t n matrix with entries which belong to k[X
1
, . . . , X
n
], so may be evaluated at
the point P, yielding J(P).
Remark 22.2.1. In order for the denition to make sense, we must have
n dim Y mint, n. But Y ,= = (f
1
, . . . , f
t
) ,= (1), and then
n dim Y t follows from Proposition 22.1.2.
The denition appears to depend on the choice of the generators f
1
, . . . , f
t
for I(Y ). We shall see below that in fact it does not.
We shall see below that we always have the inequality rank J(P) n
dim(Y ), so saying the P is a non-singular point of Y is saying that the
rank of J(P) is as large as possible.
Let us denote the maximal ideal (X
1
a
1
, . . . , X
n
a
n
) corresponding to P by
a
P
Spec
m
k[X
1
, . . . , X
n
]. Let m Spec
m
(O
Y
) denote the image of a
P
under the
projection k[X
1
, . . . , X
n
] O
Y
. We use the same symbol m to denote the maximal
ideal in the local ring A := O
Y,m
.
Theorem 22.2.2. Y is non-singular at P i A is regular.
Proof. For brevity write k[X] = k[X
1
, . . . , X
n
].
Claim 1: There is a k-linear isomorphism

: a
P
/a
2
P
k
n
.
Proof: Dene : k[X] k
n
by
(f) =
_
f
X
1
(P), . . . ,
f
X
n
(P)
_
.
It is clear that (a
2
P
) = 0. Also, the image of the set X
i
a
i

i
gives a k-basis
for a
P
/a
2
P
, which is taken by to the standard basis of k
n
. Hence induces the
isomorphism

.
Claim 2: rank J(P) = dim
k
(I(Y )) = dim
k
I(Y ) + a
2
P
a
2
P
.
Proof: Any h I(Y ) can be written in the form h = g
1
f
1
+ + g
t
f
t
, for some
g
i
k[X]. We have (h) = g
1
(P)(f
1
) + g
t
(P)(f
t
), which shows that the rows
of J(P) span (I(Y )). The rst equality follows.
For the second equality, note that (I(Y )) =
I(Y )
I(Y )a
2
P

=
I(Y )+a
2
P
a
2
P
.
Claim 3: m/m
2
= a
P
/(I(Y ) + a
2
P
), as k-vector spaces.
Proof: We have
m = a
P a
P
/I(Y )
a
P
m
2
= (a
2
P
+I(Y ))
a
P
/I(Y )
a
P
, and thus
m/m
2
= (a
P
/(a
2
P
+I(Y )))
a
P
,
which is just a
P
/(a
2
P
+I(Y )).
Putting Claims 1-3 together, we get
(22.2.2) dim
k
m/m
2
+ rank J(P) = n.
This equation implies the theorem. Indeed, recall that since O
Y
is a f.g. k-
algebra and a domain, we have dim Y = dim A. Write r for this dimension. Then
58 THOMAS J. HAINES
the ring A is regular i dim
k
m/m
2
= r which by (22.2.2) holds i rank J(P) =
n r.
We also see that since we always have the inequality dim Y dim
k
m/m
2
, equa-
tion (22.2.2) implies that we always have the inequality
(22.2.3) rank J(P) n dim Y.
In particular, the point P is singular (ie. not non-singular) if and only
rank J(P) < n dim Y.
Examples:
If Y A
n
k
is cut out by a single non-zero non-unit element f k[X], then
P Y is non-singular i J(P) k
n
is not the zero vector.
For the curves cut out by X
2
Y
3
and X
2
Y
2
+X
3
in the plane A
2
, the
only singularity in each case is P = (0, 0), as is easily checked.
Exercise 22.2.3. Let Y be an irreducible (ane) variety, that is, in the ane
case Y = Spec(A) where A is a f.g. domain over an algebraically eld k. Show
that the set Sing(Y ) of singular points is a proper Zariski-closed subset of Y . [See
Hartshorne, Algebraic Geometry, II, 8, Cor. 8.16.]
Exercise 22.2.4. Let Y be an irreducible hypersurface in A
n
k
, i.e. Y = V (f), where
f k[X
1
, . . . , X
n
] is an irreducible element. Show that Sing(Y ) = V (f,
f
X1
, . . . ,
f
Xn
).
Show that the singularities are isolated, meaning that dim Sing(Y ) = 0, if and only
the k-vector space k[X
1
, . . . , X
n
]/(f,
f
X1
, . . . ,
f
Xn
) is nite-dimensional.
As a nal remark, let (x
1
, . . . , x
d
) = m denote a system of parameters in a d-
dimensional regular local ring (A, m), where A is also a k := A/m-algebra (for
example, A could be a localization at a maximal ideal of a f.g. k-algebra). Then
we can dene a k-algebra homomorphism
: k[t
1
, . . . , t
d
] A
by sending t
i
x
i
, for i = 1, . . . , d. This induces an isomorphism
G() : k[t
1
, . . . , t
d
] G
m
(A).
Hence by Lemma 18.2.2, the induced map

on completions is also an isomorphism:

: k[[t
1
, . . . , t
d
]]

A.
This proves the following result.
Proposition 22.2.5. If P Y is any non-singular point on a d-dimensional irre-
ducible variety Y , there is an isomorphism
k[[t
1
, . . . , t
d
]]

=

O
Y,P
.
22.3. Some deeper connections of regularity with geometry. If X = Spec(A),
where A is a Noetherian domain, we say X is regular if A
m
is a regular local ring
for every maximal ideal m Spec
m
(A). Similarly, we say X is normal if A
m
is
normal for all m. (We already know that this is equivalent to saying A is normal,
even without the assumption that A be Noetherian.) Similar terminology applies
to reduced and irreducible schemes.
Fact 1: Assume (A, m) is a regular local ring (hence a domain). Then A is normal.
Thus, a regular scheme is normal.
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 59
In fact, a much stronger result is true: any regular local ring is a UFD (the
Auslander-Buchsbaum theorem). The proof of this is beyond the scope of this
course. We will prove Fact 1 in the next lecture.
Fact 2: Any normal scheme is regular in codimension 1.
What does regular in codim 1 mean? Let us discuss this in the ane case, i.e.
where A is a Noetherian domain. Then X = Spec(A) is regular in codimension
1 if for any height 1 prime ideal P A, the ring A
P
is regular. Now the proof of
Fact 2 is almost trivial: the ring A
P
is a Noetherian local domain of dimension 1,
and if A is normal, A
P
is also normal; thus, A
P
is a DVR, hence is regular.
Fact 2 can be used to prove the following proposition.
Proposition 22.3.1 (Consequence of Fact 2). If X is a normal variety, and
Sing(X) X is the (Zariski-closed) set of singular points, then codim(Sing(X))
2. In particular, any normal curve is non-singular, and any normal surface has
only isolated singularities.
A key ingredient of the proof is a theorem of Serre which states that if A is
regular, then so is A
p
for every p Spec(A) (see [Serre], IV.D.Prop.23). We will
not prove this, but it is quite important.
Many interesting singular varieties are normal: for example Schubert varieties
are usually singular, but they are at least always normal. We shall give some
concrete examples next lecture.
23. Lecture 23
Proof that A regular implies A is normal (using almost integral extensions).
Example of k[X, Z]/(Z
2
f). Denition of module of relative dierentials: con-
struction of the universal derivation.
**********************
23.1. Proof that regular implies normal. We will now prove Fact 1 from
the previous lecture. We will follow the treatment in [Mat1],17.D, p.119.
Proposition 23.1.1. If A is a regular local ring, then A is normal.
To prove this, we need some preliminaries. Temporarily, we let A denote any
domain, with K = Frac(A).
We say u K

is almost integral if 0 ,= a A such that au

n
A, n > 0.
We abbreviate almost integral by a.i. Note that u, v a.i. implies that uv and
u v are also a.i.
Lemma 23.1.2. (i) If u is integral, it is a.i.
(ii) The converse holds if A is Noetherian.
Proof. (i): Write u = a/b, for a, b A0. Consider a relation
_
a
b
_
n
+
n1
_
a
b
_
n1
+ +
0
= 0,
where
i
A for all i. Then it is easy to see that b
n1
_
a
b
_
k
A, for all k > 0.
Hence u is a.i.
(ii): If u is a.i. with au
n
A for all n > 0, we have an inclusion A[u] a
1
A. Then
since A is Noetherian, the A[u] is a nite A-module. It follows that u is integral
over A.
60 THOMAS J. HAINES
Now to prove the proposition above, it is enough to establish the implication
G
m
(A) normal domain = A normal domain. We will prove something more
general.
Proposition 23.1.3. Suppose A is a Noetherian ring, with I A an ideal belong-
ing to the Jacobson radical. Then
G
I
(A) is a normal domain = A is a normal domain.
Proof. As a consequence of Krulls theorem, we know that

n=0
I
n
= 0. So for
0 ,= a A there is a unique non-negative integer n with a I
n
I
n+1
. Denote
ord(a) = n, and let a

be the image of a in I
n
/I
n+1
. By convention, set 0

= 0
G
I
(A).
We already know that A is a domain; let K = Frac(A). Suppose a/b K

is
integral. We want to show that a bA. Since A/bA is Hausdor in the I-adic
topology, we have
bA =

n=0
(bA+I
n
).
Hence, its ETS the implication
(23.1.1) a bA+I
n1
= a bA+I
n
.
Write a = br + a

, where r A and a

I
n1
. This gives a

/b = a/b r,
which is integral over A. Hence by replacing a with a

we may assume WLOG that

a I
n1
.
Now a/b a.i. means that 0 ,= c A such that ca
m
b
m
A, for all m > 0.
Since G
I
(A) is a domain, the map x x

is multiplicative, so we have c

(a

)
m

(b

)
m
G
I
(A). Hence a

/b

G
I
(A) is a.i. Since G
I
(A) is Noetherian, this shows
that a

/b

is integral. But then since G

I
(A) is normal, we have a

G
I
(A).
Write a

= b

, for some element d A (check: we can do this). Then we have

a

= (bd)

, and so
n 1 ord(a) < ord(a bd),
which implies that a bd +I
n
, as desired. This completes the proof.
23.2. The varieties cut out by Z
2
f. Again let k[X] = k[X
1
, . . . , X
n
]. Suppose
f k[X] is not a square, so that Z
2
f k[X, Z] is irreducible. Some time ago
we asked: when is the domain A := k[X, Z]/(Z
2
f) normal? Here is the answer:
Proposition 23.2.1. Assume char k ,= 2. Then A is normal i f is square-free.
Proof. () : Suppose f is not-square free. We will produce an element in the
fraction eld of A but not in A, which is a.i. (hence integral over A). This will
show that A is not normal.
Write f(X) = h(X)g(X)
2
. Then we claim that Z/g(X) is a.i. Indeed, note that
for every n 0 we have the following two equalities
g(X)
_
Z
g(X)
_
2n
= g(X)h(X)
n
g(X)
_
Z
g(X)
_
2n+1
= Zh(X)
n
.
(): For this direction, we prove something more general.
Lemma 23.2.2. Let R be a UFD in which 2 is a unit. Suppose f R is square-free.
Then R[Z]/(Z
2
f) = R[

F] is normal.
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 61
Proof. Write = Z =

f. Then K := Frac R Frac R[], and is an element of

the latter.
If K, then R (since R is normal), and this contradicts the assumption
that f is not a square. So / K, and so K() = K[] = K +K.
Suppose = x +y K() (x, y K) is integral over A[]. We will show that
x, y R, showing that R[] is normal.
First, WLOG y ,= 0. Indeed, if y = 0, then x is integral over R, hence x R.
Now the minimal polynomial for over K takes the form
min
K
() = X
2
2xX + (x
2
y
2
f).
Now is integral over R[] (thus over R) 2x, x
2
y
2
f R, which holds
x, y
2
f R. So the latter holds.
We claim that y R. If is an irreducible element of R and divides the
denominator of y, then y
2
f R implies that
2
divides f, a contradiction of f
being square-free. Thus y R and the lemma is proved.
The lemma clearly implies the direction () of the proposition.
The following exercise gives some very concrete examples of varieties which are
normal and singular.
Exercise 23.2.3. Suppose that f(X
1
, . . . , X
n
) = (X
1
a
1
)(X
2
a
2
), where a
1
,=
a
2
. Let Y = Spec(k[X, Z]/(Z
2
f)). By the above proposition, Y is a nor-
mal irreducible variety. Show that Y has a singularity at any point of the form
(a
1
, a
2
, x
3
, . . . , x
n
, 0). What is the dimension of the singular locus? What is its
codimension in Y ?
23.3. Derivations and the module of Kahler dierentials. We will develop
the algebraic theory of dierential 1-forms. We actually develop a relative version,
i.e. we x a ring k, and a k-algebra A (that is, a ring A with a ring homomorphism
k A). We will dene an A-module
A/k
=
1
A/k
, which is uniquely determined
in a certain sense.
To do so, we rst x an A-module M and dene the set Der
k
(A, M) of k-
derivations A M. What is a k-derivation? It is a k-linear map D : A M
satisfying the Leibniz rule:
D(ab) = aDb +bDa.
The set Der
k
(A, M) is naturally an A-module: (aD)(b) := a(Db). Also, an A-linear
map : M M

gives rise to Der

k
(A, M) Der
k
(A, M

) by D D.
Theorem 23.3.1. The covariant functor M Der
k
(A, M) is represented by a
unique pair (M
0
, d):
- d : A M
0
is a k-derivation;
- For every k-derivation D : A M, there exists a unique A-homomorphism
: M
0
M such that D = d.
The pair (M
0
, d) is unique up to a unique isomorphism. We denote M
0
by
A/k
and call it the module of relative dierentials. The theorem gives a (functorial
in M) isomorphism
(23.3.1) Der
k
(A, M) = Hom
A
(
A/k
, M).
62 THOMAS J. HAINES
Construction of (M
0
, d): Consider the A-algebra homomorphism
: A
k
A A
dened by x y xy. Let I := ker(). Let
A/k
:= I/I
2
. The exact sequence
0

I/I
2

A
k
A/I
2

A

0
splits in A-Mod in two ways:
1
(a) := 1 a, and
2
(a) := a 1 both determine
sections of

. Therefore the dierence

1

2
has image in I/I
2
, and since
i
is
a k-algebra map, the map
d :=
1

2
: A I/I
2
is a k-derivation. Indeed, I/I
2
has A-module structure given by multiplication by
either 1 a or a 1, so that (
1

2
)(ab) =
1
(a)
1
(b)
2
(a)
2
(b) is the sum
the following two expressions:
a(
1

2
)(b) =
1
(a)(
1
(b)
2
(b))
b(
1

2
)(a) =
2
(b)(
1
(a)
2
(a)).
The proof gives us a principle which we will use repeatedly:
Lemma 23.3.2. Suppose
1
,
2
: A B are k-algebra homomorphisms, and as-
sume
1

2
takes values in an A-submodule N B whose A-module structure is
given by multiplication by
1
(a) and assume
1
(a)
2
(a) acts by zero on N. Then

2
: A N is a k-derivation.
We will complete the proof that (
A/k
:= I/I
2
, d) satises the universal property
in the next lecture.
24. Lecture 24
24.1. Universal property of (
A/k
, d). We need a preliminary construction. If
M A-Mod, we dene a k-algebra A M by setting A M = A M and by
dening multiplication by
(a, m)(a

, m

) := (aa

, am

+a

m).
Clearly A M is a k-algebra with unit (1, 0). The exact sequence
0 M A M M 0
splits. The inclusion of M is given by m (0, m) and the projection onto A is
(a, m) a. The latter has the obvious section a (a, 0). Note also that M
2
= 0
in A M.
Now given D Der
k
(A, M) dene : A
k
A A M by
(x y) = (xy, xDy).
It is easy to check the following statements:
- is a k-algebra homomorphism;
-

i
x
i
y
i
I = (

i
x
i
y
i
) = (0,

i
x
i
Dy
i
) M. Therefore
: I M A M;
- Since M
2
= 0 in A M, the map determines : I/I
2
M.
- We have (da) = (1 a a 1) = (0, Da) (since D1 = 0), and thus
d = D;
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 63
- is A-linear: a(

i
x
i
y
i
) =

i
ax
i
y
i
(0,

i
ax
i
Dy
i
) = a(

i
x
i

y
i
).
We have now proved the existence of the factoring d = D.
It remains to prove that is the unique A-linear map with the property d = D.
This will follow from the fact that
A/k
is generated over A by the set da, a A.
Why is this true? Observe that
a a

= (a 1)(1 a a 1) +aa

1.
So =

i
x
i
y
i
I =

i
x
i
dy
i
in I/I
2
. This completes the proof of the
universal property of (
A/k
, d).
24.2. Examples.
Let A be a k-algebra, generated as an algebra by a subset U A. Then

A/k
is generated over A by da, a U. To prove this, note that an element
in A can be written in the form a = f(a
1
, . . . a
n
) for some a
i
U and
f k[X
1
, . . . , X
n
]. Our claim results from the following exercise.
Exercise 24.2.1. Show that Leibniz rule implies
(24.2.1) da =

i
f
X
i
(a
1
, . . . , a
n
) da
i
.
In particular, if A = k[X
1
, . . . , X
n
], then
A/k
= AdX
1
+ + AdX
n
.
Moreover, we have
A/k

= A
n
, i.e., the dX
i
s are linearly independent over
A. To prove this, for each i let D
i
Der
k
(A, A) denote the k-derivation
D
i
=

Xi
. This corresponds to the A-linear map
i
:
A/k
A, such that

i
d = D
i
. Now if there is a relation, a
1
dX
1
+ +a
n
dX
n
= 0, applying

i
gives a
i
= 0.
24.3. 0-smooth, 0-unramied, and 0-etale homomorphisms. Let k be a ring,
and let k A be a k-algebra. We say A is 0-smooth (over k) if for every k-algebra
C and ideal I C such that I
2
= 0, if we are given a k-algebra map u : A C/I,
then there is a lift of u to a k-algebra map v : A C. In other words, given a
commutative square below, there is a map v making the triangles commute:
A
u

C/I
k

We say A is 0-unramied if given u there is at most one such map v. We say

A is 0-etale if given u there is exactly one such v. Thus, 0-etale = 0-smooth +
0-unramied,
Lemma 24.3.1. A/k is 0-unramied i
A/k
= 0.
Proof. (): Suppose v
1
, v
2
are two lifts of u; give the ideal I C the structure
of an A-module by multiplication by v
1
(a) (or v
2
(a) : since v
1
(a) v
2
(a) I and
I
2
= 0, the two structures coincide). Then by Lemma 23.3.2, v
1
v
2
: A I is a
k-derivation. Since
A/k
= 0, the only k-derivation A I is zero, and so v
1
= v
2
.
This shows A/k is unramied.
64 THOMAS J. HAINES
(): Consider the diagram
A
i

A
k
A/I
k

A
k
A/I
2
.

Since the maps

i
: A A
k
A/I agree, by hypothesis so do the maps
i
. This
means that the derivation d :=
1

2
: A
A/k
is zero. Since
A/k
is generated
over A by dA, we get
A/k
= 0, as desired.
Lemma 24.3.2. Let S A be a multiplicative subset. Then A
S
:= S
1
A is 0-etale
over A.
Proof. Consider the diagram
A
S
u

C/I
A
p

C.

We want to produce the unique morphism A

S
C making the diagram commute.
But up sends S into (C/I)

, so q sends S into C

(note that c C

i its image
c (C/I)

, since I
2
= 0). Thus, q factorizes uniquely through A
S
, as desired.
24.4. The First Fundamental Exact Sequence. The next result is the First
Fundamental Exact sequence.
Theorem 24.4.1. Let k
f

A
g

B be two ring homomorphisms.

(1) There is an exact sequence of B-modules
(24.4.1)
A/k

A
B

B/k

B/A

0,
where (d
A/k
a b) = bd
B/k
g(a), and (d
B/k
b) = d
B/A
b
(2) If B is 0-smooth over A, then
(24.4.2) 0


A/k

A
B

B/k

B/A

0
is split exact.
Proof. (1): By a standard argument (see Atiyah-Macdonald, 2.9), N

N N

is exact in B-Mod i for every T B-Mod, Hom

B
(N

.T) Hom
B
(N, T)
Hom
B
(N

, T) is exact. Therefore, its ETS: for every M B-Mod, the following

diagram is commutative with exact rst row:
Der
k
(A, M)
=

Der
k
(B, M)
=

rest.

Der
A
(B, M)
=

incl.

Hom
B
(
A/k

A
B, M) Hom
B
(
B/k
, M)

Hom
B
(
B/A
, M)

0.

(The exactness of the rst row is easy; check the diagram commutes, where ,
are dened as in the statement of (1)!)
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 65
(2): Suppose B is 0-smooth over A. Fix T B-Mod, and D Der
k
(A, T). Consider
the diagram
B
id

B
A
g

B T,

where is dened by (a) := (g(a), Da). By hypothesis, the factoring h exists.

Write h(b) = (b, D

b), for a k-derivation D

: B T. We have D = D

g.
We can write D

d
B/k
, for a unique B-linear map ; :
B/k
T.
Now in the above diagram, take T =
A/k

A
B and D = d
A/k
1. Then
the map

we get is a B-linear map

:
B/k

A/k

A
B, and the equality
D = D

g implies that d
A/k
1 =

d
B/k
g, and thus

= id

A/k

A
B
.
Therefore the sequence splits.
25. Lecture 25
25.1. The Second Fundamental Exact Sequence.
Theorem 25.1.1. Consider a diagram k
f

A
g

B = A/J , where g is sur-

jective and J = ker(g). Then we have:
(1) The following sequence is exact
(25.1.1) J/J
2

A/k

A
B

B/k

0,
where (x) = d
A/k
(x) 1, and is dened as in the First Fundamental
Sequence.
(2) If B is 0-smooth over k, then
(25.1.2) 0

J/J
2


A/k

A
B


B/k

0
is split exact.
Proof. (1): For T B-Mod, consider the diagram
Hom
B
(J/J
2
, T) Der
k
(A, T)

Der
k
(B, T)

0.

Note that

is simply the restriction to J map. So

(D) = 0 i D vanishes on
J i D comes from Der
k
(B, T). Hence this sequence is exact T, and hence (1)
follows.
(2): Suppose B is 0-smooth over k. Then we have the factoring map s in the
diagram
B
id

B
k

A/J
2
.
g

Thus, s gives a splitting in k-Mod of the exact sequence

0

J/J
2

A/J
2
g

B

0.
Now sg : A/J
2
A/J
2
is a k-algebra homomorphism, trivial on J/J
2
, and
g(id sg) = 0.
66 THOMAS J. HAINES
Hence (by Lemma 23.3.2), D := id sg : A/J
2
J/J
2
is a k-derivation.
Now x T B-Mod as in the proof of (1). We want to show

is surjective by
constructing a section of

. In fact, the map taking Hom

B
(J/J
2
, T) to the
composition D

A

A/J
2
D

J/J
2

T
is such a section: if x J, and x x mod J
2
, then

(x) = D(x) = (x sg(x)) = (x).

Taking T = J/J
2
now, we see the sequence in (2) is split exact.
Example: Let A = k[X
1
, . . . , X
n
], and B = k[X
1
, . . . , X
n
]/(f
1
, . . . , f
m
) = k[x
1
, . . . , x
n
].
Then

B/k
= (
A/k

A
B)/

i
B df
i
= F/R,
where F is the free B-module with basis dX
1
, . . . , dX
n
, and R is the B-submodule
generated by df
i
=

j
fi
Xj
dX
j
.
For instance, if k is a eld with char(k) ,= 2, then for B = k[X, Y ]/(X
2
+Y
2
) =
k[x, y] we have

B/k
= Bdx +Bdy,
where the only relation is xdx +ydy = 0.
If char(k) = 2, then
B/k

= B
2
.
25.2. On tangent spaces and cotangent spaces. Next, we want to esh out the
analogy with dierential geometry. We start by dening the tangent and cotangent
spaces to a variety (or scheme) at a closed point. Then we discuss vector elds.
Let k = k be an algebraically closed eld. Let X be a k-variety, or more generally
a nite-type separated scheme over k (to be more concrete, for our purposes, we will
assume X = Spec(A), where A is a f.g. k-algebra. However, we will use notation
that indicates that everything holds also in the non-ane case). We dont need
to assume A is reduced or a domain for this discussion to be valid. By Hilberts
Nullstellensatz, a closed point x X corresponds to a maximal ideal m
x
A. In
fact, we have
x X closed point m
x
Spec(A)
k-alg. map x : A k
k-alg. map x : O
x
k.
In the last line, O
x
denotes the stalk at x of the structure sheaf O
X
. Recall that
the sheaf O
X
on X has global sections O
X
(X) = A, in a canonical way. The stalk
O
x
can be identied with the localization A
mx
.
Now apply the second fundamental exact sequence to k

O
x
x

k . Since
the composition of these maps is the identity, we nd that the second fundamental
sequence gives a canonical isomorphism
m
x
/m
2
x
=
Ox/k

Ox
k.
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 67
We call Hom
k
(m
x
/m
2
x
, k) the tangent space of X at the point x. We call the
dual k-vector space
Ox/k

Ox
k, the cotangent space of X at x. Note that the
tangent space is also the k-vector space Der
k
(O
x
, k):
Hom
k
(m
x
/m
2
x
, k) = Hom
Ox
(
Ox/k
, k)
= Der
k
(O
x
, k).
Note also that by using a problem in Homework 2, we can identify the cotangent
space as

Ox/k

Ox
k =
A/k
/m
x

A/k
.
Exercise 25.2.1. Assume k = k. Let Y = V (f
1
, . . . , f
t
) A
n
k
denote a closed
irreducible subset. Let P = (P
1
, . . . , P
n
) denote a closed point which lies in Y ;
denote the maximal ideals corresponding to P by a
P
= (X
1
P
1
, . . . , X
n
P
n
), and
m = a
P
A := k[X
1
, . . . , X
n
]/(f
1
, . . . , f
t
). Show that Hom
k
(m/m
2
, k), the tangent
space Y at P, can be identied with the kernel of J(P) =
_
fi
Xj
(P)
_
ij
, where this
t N matrix is viewed as a k-linear map k
N
k
t
.
Here is another important way to think about the tangent space. Recall that
in dierential geometry, a tangent vector at a point x is an equivalence class of
germs of curves going through x. In algebraic geometry, the role of equivalence
class of curve is played by a map of k-schemes Spec(k[]/(
2
)) X. Saying it
goes through x means the following. Denote := k[]/(
2
), and consider the
canonical k-algebra homomorphism p : k dened by 0. Then a map
f

: Spec() X is given by a k-algebra homomorphism f : O

x
. It turns
out that saying f

goes through x is the same as saying that pf = x, as maps

O
x
k. In fact, we have the following result making this precise.
Lemma 25.2.2. There is a canonical bijection
k-alg. maps f : O
x
[ pf = x = Hom
k
(m
x
/m
2
x
, k).
Proof. Consider the exact sequence
0

m
x
/m
2
x

O
x
/m
2
x

O
x
/m
x

0.
The map x : O
x
/m
x
k O
x
/m
2
x
gives a splitting of the above exact sequence,
and thus an identication (which depends on x : O
x
k) of k-algebras
O
x
/m
2
x
= O
x
/m
x
m
x
/m
2
x
.
We also have a canonical identication
= k k.
Note that any k-algebra homomorphism f : O
x
such that pf = x necessarily
takes m
x
into k, hence factors through O
x
/m
2
x
, and is uniquely determined by its
restriction to m
x
/m
2
x
. Thus giving such a homomorphism f is the same as giving a
k-linear map m
x
/m
2
x
k, in other words, an element of Hom
k
(m
x
/m
2
x
, k).
68 THOMAS J. HAINES
25.3. Vector elds. We push the dierential geometry analogy a little further by
dening vector elds. In dierential geometry, a vector eld is a rule assigning
to each point x an element in the tangent space at x. We thus want to dene
something that gives us, for each closed point x X, a derivation in Der
k
(O
x
, k).
We call any element D Der
k
(O
X
, O
X
), a vector eld. We claim it gives rise to
D
x
Der
k
(O
x
, O
x
), for each x X. Indeed, D determines a family of k-derivations
D(U) : O
X
(U) O
X
(U),
for U ranging over the open subsets of X which contain x. Taking direct limits, we
get a k-derivation
D
x
: O
x
O
x
.
Now viewing x X as the k-algebra homomorphism x : O
x
k, D then
determines a k-derivation xD
x
Der
k
(O
x
, k), for each x. Thus, D really deserves
to be called a vector eld.
Fix a tangent vector t Der
k
(O
x
, k). We say D takes value t at x if xD
x
= t.
Notation: if f O
x
, its image x(f) O
x
/m
x
= k is often denoted by f(x).
Note that if X = Spec(A) and D Der
k
(A, A), f A, we have the formula
x D
x
(f) = D(f)(x).
25.4. A vector eld criterion for regularity. In the following statement, X
is any nite-type separated k-scheme. Since the statement is local around x, we
might as well assume X = Spec(A) where A is a nitely-generated k-algebra, and
x corresponds to the maximal ideal m
x
A.
Proposition 25.4.1. Suppose char(k) = 0. Let x X be a closed point. Sup-
pose that X has n = dim
k
m
x
/m
2
x
vector elds which are linearly independent at x
(meaning that the values they take at x are linearly independent tangent vectors).
Then x is a regular point of X, i.e., O
x
is a regular local ring.
Proof. Here we give a sketch of the proof, leaving you to ll in the details.
Let x
1
, . . . , x
n
be elements of m
x
whose images give a k-basis for m
x
/m
2
x
. WLOG
there are derivations D
1
, . . . , D
n
Der
k
(O
x
, O
x
) such that
D
i
(x
j
)(x) =
ij
,
in other words,
D
i
(x
j
)
ij
mod m
x
.
For each p 1, we clearly have D
i
(m
p
x
) m
p1
x
, and hence D
i
extends by continuity
to give a uniquely determined k-derivation D
i
Der
k
(

O
x
,

O
x
), where

O
x
denotes
the m
x
-adic completion of O
x
.
Dene a k-algebra homomorphism : k[[t
1
, . . . , t
n
]]

O
x
by t
i
x
i
.
Dene :

O
x
k[[t
1
, . . . , t
n
]] by
(f) =

=(1,...,n)Z
n
0
(D

f)(x)
!
t

,
where by denition D

:= D
1
1
D
n
n
, :=
1
!
n
!, and t

:= t
1
1
t
n
n
.
Leibniz rule (or rather, the generalized form D
m
fg =

m
k=0
_
m
k
_
D
k
f D
mk
g)
and an argument by induction on n shows that is a continuous k-algebra homo-
morphism.
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 69
Now is surjective since its image contains the x
i
s and

O
x
is complete. Since
(x
i
) t
i
mod (t
1
, . . . , t
n
)
2
, the elements (x
i
) generate the ideal (t
1
, . . . , t
n
), and
hence is also surjective. For both cases, use Lemma 18.2.2.
Then the composition is a surjective ring endomorphism of the Noetherian
ring k[[t
1
, . . . , t
n
]], hence is an automorphism. Thus is an isomorphism, and this
shows

O
x
is a regular local ring. It follows that O
x
is also regular.
26. Lecture 26
26.1. Application of vector eld criterion for regularity: group schemes.
A good reference for group schemes and Hopf algebras is W.C. Waterhouse, Intro-
duction to Ane Group Schemes, Springer-Verlag, 1979.
Let k be any eld, and assume A is a f.g. k-algebra. Suppose G = Spec(A) is
a k-group scheme. This is the same thing as saying that A is k-Hopf algebra.
By denition this means that there are comultiplication, counit, and coinverse
homomorphisms
: A A
k
A
: A k
S : A A
which are compatible in a certain sense with each other (you can recover the com-
patibilities certain commutative diagrams by writing down the commutative
diagrams encapsulating the group axioms for G, and then taking the dual com-
mutative diagrams with respect to the anti-equivalence of categories A Spec(A)).
The following is an important application of the vector-eld criterion for regu-
larity, Proposition 25.4.1.
Theorem 26.1.1. If char(k) = 0, then any k-Hopf algebra A is a regular ring (that
is, each localization A
m
is regular, where m ranges over all maximal ideals m A).
Thus, any k-group scheme is regular, and hence is reduced and non-singular as a
variety.
This is far from true when char(k) = p > 0. Indeed, the ring A = F
p
[X]/(X
p
)
is a Hopf-algebra over F
p
whose corresponding group scheme Spec(A) is the group
subscheme
p
G
a
whose R-points for a F
p
-algebra R is the additive group r
R [ r
p
= 0. Note that the ring A is not even reduced here.
Proof. For simplicity, let us assume k = k. For any closed point x G, we want to
check that the local ring O
x
is regular. By translating x back to the origin e G
using the group action, it is enough to check this for x = e. To apply Proposition
25.4.1, we need to check that there are dim
k
(m
e
/m
2
e
) vector elds dened near e
which give a linearly independent set of values at e. To construct these vector elds,
the key fact about Hopf algebras we use is that there is an isomorphism

A/k

= A
k
m
e
/m
2
e
.
(See Theorem 11.3 in Waterhouse.) Using this, we see that
Der
k
(A, A) = Hom
A
(
A/k
, A)
= Hom
A
(A
k
m
e
/m
2
e
, A)
= Hom
k
(m
e
/m
2
e
, A).
70 THOMAS J. HAINES
Now composing the derivations with the homomorphism e : A k shows that the
derivations on the LHS take as values at e precisely the set
Hom
k
(m
e
/m
2
e
, k),
which is what we wanted to prove.
26.2. Separability: various notions. Let k be a eld, and A a k-algebra. We
say A is separable over k if for every extension eld k

k, the ring A

:= A
k
k

is reduced.
Facts (easy exercises):
- Any subalgebra of a separable algebra is separable.
- A is separable i every f.g. k-subalgebra of A is separable.
- A
k
k

is reduced for every f.g. extension eld k

k = A is separable.
- A is separable over k = A
k
k

is separable over k

.
We want to better understand this notion of separable, when Ais nite-dimensional.
So, assume dim
k
(A) < , and x a k-basis
1
, . . . ,
n
for A. Dene the discrim-
inant
disc
A/k
= det[tr(
i

j
)].
[For a A, recall that tr(a) is the trace of the k-linear map A A given by
multiplication by a.] Note that d := disc
A/k
is a well-dened element of k/(k

)
2
:
If

1
, . . . ,

n
is another k-basis, write

l
=

i
c
li

i
, and note that
det[tr(

j
)] = det(c
ij
)
2
det[tr(
i

j
)].
Proposition 26.2.1. A is separable over k i d ,= 0.
Proof. (): Let k

k and A

= A
k
k

. Suppose N := rad(A

) ,= 0. Let

1
, . . . ,
n
be a k

-base for A

such that
1
, . . . ,
r
is a k

-base for N. Since every

element of N is nilpotent, we see that
i

j
is nilpotent for i or j r. This implies
that tr(
i

j
) = 0 for such i, j. Hence d = det[tr(
i

j
)] = 0.
(): Let K denote an algebraic closure of k. The ring A
k
K is reduced and
Artinian, so if p
1
, . . . , p
n
are the prime (= maximal) ideals of A
k
K, we get
A
k
K = A
k
K/
i
p
i
=

i
A
k
K/p
i
.
Since A
k
K/p
i
is a nite eld extension of K, it is

= K, and so A
k
K

= K
n
.
Choose a basis of idempotents e
i
, so that e
i
e
j
=
ij
. Then d = det[tr(e
i
e
j
)] = 1 ,=
0.
Now change notation: assume A := K is a eld extension of k. Suppose K/k is
an algebraic extension (that is, an integral extension). Recall what it means to say
K is separable in the usual sense over k: this is the case i the minimal
polynomial f k[X] which satises has (f, f

) = 1. If is not separable in the

usual sense, then it is easy to see that char(k) = p > 0, and f(X) = g(X
p
), for
some polynomial g k[X].
Proposition 26.2.2. Suppose K/k is an algebraic eld extension. Then K/k is
separable in the usual sense it is separable.
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 71
Proof. (): If K/k is not separable in the usual sense, then K such that the
minimal polynomial f k[X] of has (f, f

) ,= 1. In particular, f does not have

distinct roots in k

= k, and the subalgebra k() K has

k()
k
k

= k

[X]/(f),
a ring with non-zero nilpotents. Hence K/k is not separable.
(): Assume K/k is separable in the usual sense. WLOG K is f.g. as a eld
extension over k; being algebraic, this means it is f.g. as a k-algebra, hence is a
nite extension of k. Then, since K/k is separable in the usual sense and is now
also nite, K = k(), for some K. Let f k[X] be the minimal polynomial of
. Let k

k, and factor f in k

[X] as
f = f
1
f
r
,
where the f
i
are distinct irreducible elements of k

[X]. By the Chinese remainder

theorem,
K
k
k

= k

[X]/(f) =
r

i=1
k

[X]/(f
i
).
This is a product of elds, hence is reduced. This shows that K/k is separable.
We say a eld extension K/k is separably generated if K has a transcendence
basis such that K/k() is a separable algebraic extension.
Lemma 26.2.3. Any separably generated extension is separable.
Proof. Suppose is the aforementioned transcendence basis. Let k

k be any
eld extension.
The natural map k()
k
k

() is an isomorphism (this follows easily, using

that it restricts to give the obvious isomorphism k[]
k
k

[]).
Thus K
k
k

= K
k()
(k()
k
k

) = K
k()
k

(). Since K/k() is separable,

the latter is reduced, and thus so is K
k
k

.
For the next proposition, assume char(k) = p > 0, and dene k
1/p
:= x
k [ x
p
k. Note that k
1/p
is an extension eld of k.
Proposition 26.2.4. Suppose char(k) = p, and K is a f.g. extension eld of k.
Then TFAE:
(1) K is separable over k.
(2) K
k
k
1/p
is reduced.
(3) K is separably generated over k.
Proof. The implication (1) (2) is trivial, and in the above lemma we proved the
implication (3) (1).
Let us prove (2) (3). Write K = k(x
1
, . . . , x
n
). WLOG x
1
, . . . , x
r
comprise
a transcendence basis for K/k. Lets assume that k(x
1
, . . . , x
r
, . . . , x
q
) is separable
over k(x
1
, . . . , x
r
), but y = x
q+1
is not separable over k(x
1
, . . . , x
r
).
Let f(Y
p
) be the minimal polynomial of y over k(x
1
, . . . , x
r
). Clearing denomi-
nators, get an irreducible polynomial F(X
1
, . . . , X
r
, Y
p
) with F(x, y
p
) = 0.
If F/X
i
= 0 for all 1 i r, then F(X, Y
p
) = G(X, Y )
p
, for some G(X, Y )
k
1/p
[X, Y ]. But then k[x
1
, . . . , x
r
, y]
k
k
1/p
= k[X, Y ]/(F(X, Y
p
))
k
k
1/p
, which
is also k
1/p
[X, Y ]/(G(X, Y )
p
) K
k
k
1/p
. So K
k
k
1/p
is not reduced.
72 THOMAS J. HAINES
Therefore, we can assume WLOG that F/X
1
,= 0. Then x
1
is separable al-
gebraic over k(x
2
, . . . , x
r
, y), hence so are the elements x
r+1
, . . . , x
q
(check this!).
Thus, exchanging x
1
y = x
q+1
, we nd x
r+1
, . . . , x
q+1
are separable algebraic
over k(x
1
, . . . , x
r
). So by induction on q, we conclude that, after possibly rearrang-
ing and relabeling the elements x
1
, . . . , x
n
repeatedly, K is separable algebraic over
k(x
1
, . . . , x
r
), as desired.
26.3. Perfect elds. We say a eld k is perfect if every algebraic extension K/k
is separable. For example, every characteristic zero eld is perfect, since K/k is
clearly separable in the usual sense.
Lemma 26.3.1. If k is perfect then
(1) every extension K/k is separable;
(2) a k-algebra is separable i it is reduced.
Proof. (1): If char(k) = 0, then K/k is separably generated (once one checks it has
a transcendence basis), and thus separable. If char(k) = p, then k = k
1/p
: note
that if k
1/p
,= k, then k
1/p
is an algebraic extension of k which is not separable in
the usual sense (check!).
From k = k
1/p
it follows from the preceding proposition that every f.g. subex-
tension of K/k is separable. Hence K/k is separable.
(2): We need to show that if A is a reduced k-algebra, then it is separable (the
converse being immediate). WLOG A is f.g. over k, so is Noetherian and reduced.
In that case, the exercise below asserts that the total ring of fractions of A,
namely the localization A := S
1
A where S is the set of all non-zero divisors in
A, is a product of elds. Write A = K
1
K
r
. Each K
i
/k is separable by
(1), and so A is also separable. Since A A, we see A is separable as well.
Exercise 26.3.2. Show that is A is a reduced Noetherian ring, then the total ring
of fractions A is a product of elds.
Exercise 26.3.3. Suppose k has char(k) = p and k
1/p
= k. Show that k is perfect.
Thus, perfect elds are precisely those satisfying one of the following two properties:
(1) char(k) = 0, or
(2) char(k) = p and k
1/p
= k.
Remark 26.3.4. Are non-perfect elds important? Yes, they arise very natu-
rally, especially in algebraic geometry and number theory. For example, the non-
Archimedean local eld F
p
((t)) is non-perfect, as is the global function eld F
p
(t)
(= the eld of meromorphic functions on the curve P
1
over the eld F
p
).
27. Lecture 27
27.1. Regularity via the structure of
B/k
. Let K/k be a f.g. extension of
elds. In the following subsection we will prove that
(27.1.1) dim
K

K/k
tr.deg
k
K
with equality i K/k is separably generated (we will actually prove something more
general). Let us assume this for now, and derive some consequeces.
In this subsection, we assume B is the localization at a maximal ideal m of a
f.g. k-algebra A. In the next statement, we use the symbol m also to denote the
maximal ideal of the local ring B. The proof is deferred to the next lecture.
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 73
Proposition 27.1.1. Assume k is perfect and that A/m = k (e.g. k could be
any algebraically closed eld). The local ring (B, m) is regular i
B/k
is a free
B-module of rank dim(B).
Why is this important? Returning to our algebra A above, which we now assume
is a domain, we can now prove that A
m
is regular, for generic m. Let K :=
Frac(B) = Frac(A); this is a f.g. eld extension of k.
Theorem 27.1.2 (Comp. Hartshorne, II Cor. 8.16, and Exercise 22.2.3 of these
notes.). Assume k = k. Let X be an irreducible variety over k (ie a nite-type,
separated, reduced and irreducible k-scheme). Then there is an open dense set of
X which is non-singular.
Proof. Because non-singularity is a local property, we may assume X = Spec(A),
where A is a k-algebra and a domain. We need to nd a non-empty open subset
D(f) X such that for each maximal ideal m D(f), the local ring A
m
is regular.
By the proposition above, this amounts to showing that for such ms, the A
m
-module

Am/k
=
A/k

A
A
m
is a free A
m
-module of rank dim(A
m
) = dim(A) = tr.deg
k
(A)
(we used Theorem 7.3.1 for these last equalities). Let us write M =
A/k
; the
equality M
m
=
Am/k
cited above is a consequence of the general equality

S
1
C/A
=
C/A

C
S
1
C
for an A-algebra C, assigned in Homework 2.
Let K = Frac(A). Now, since k = k is perfect, the extension K/k is automat-
ically separably generated, and hence by (27.1.1) we have dim
K

K/k
= dim(A).
Also, by the Homework exercise just cited above, we have
K/k
= M
A
K.
Now we apply the following general argument to complete the proof that A
m
is
generically regular, which completes the proof of the Theorem.
Lemma 27.1.3. Let A be a Noetherian domain, with fraction eld K and let M
be a f.g. A-module. Assume that M
A
K = K
n
. Then there exists f A0 such
that M
f
= M
A
A
f

= A
n
f
. Thus, for m D(f), we have M
m

= A
n
m
.
Proof. We may choose a K-basis of M
K
:= M
A
K having the form x
1
1, . . . , x
n

1, where all x
i
M. Sending e
i
x
i
denes an A-module map A
n
M which
becomes an isomorphism upon tensoring with K. Consider the exact sequence
0 Ker A
n
M Cok 0.
The A-modules Ker and Cok are f.g., and have Ker
K
= Cok
K
= 0. Hence there
exists f A 0 which annihilates both Ker and Cok. This f has the required
properties.
27.2. Relating
L/k
and
K/k
. Consider the following general set-up: L K k
are eld extensions, and L/K is a f.g. eld extension. Dene r(L) = dim
L

L/k
and r(K) = dim
K

K/k
. We want to nd the relation between the numbers r(L)
and r(K).
By induction, we reduce to the case L = K(t), where t L. Then there are
essentially four cases to consider:
-(1) t is transcendental over K.
-(2) t is a separable algebraic element.
-(3) L = K[X]/(X
p
a), where a K and d
K/k
a = 0.
-(4) L as above, but d
K/k
a ,= 0.
74 THOMAS J. HAINES
Case (1): For psychological reasons, write t = X. Then since K[X]/K is 0-smooth
(check this!), the rst fundamental sequence for k K K[X] is split exact.
Thus,

K[X]/k
= (
K/k

K
K[X])
K[X]/K
.
Applying
K[X]
L and recalling
K[X]/K
= K[X] dX, we see

L/k
= (
K/k

K
L) LdX,
and thus r(L) = r(K) + 1.
Case (2): We will prove in the lemma below that L/K is 0-etale, and hence the
rst fundamental sequcence associated to k K L is split exact, and moreover
the third member has
L/K
= 0 (since L/k is 0-unramied). Thus we see
L/k
=

K/k

K
L, and so r(L) = r(K).
Lemma 27.2.1. If L/K is a separable algebraic extension, it is 0-etale.
Proof. It is not hard to reduce to the case where L/K is a nite separable extension
(by uniqueness the tower of lifts glue to dene one on L), which is all we need in
Case 2 anyway.
Write L = K(), where has minimal polynomial f K[X]. So L = K[X]/(f).
Consider a diagram
L
v

C/J
K

where J C is an ideal such that J

2
= 0, and the map u is given such that the
square commutes. We want to show that a unique v exists making the triangles
commute.
Existence of v: Its enough to nd an element y C such that f(y) = 0 and
y mod J = u(); then we can dene v by sending the image of X in L to y C.
To nd y, let y

C be any lift of u(). Note that f(y

) J, since the image of

f(y

) in C/J is f(u()) = 0. Since J

2
= 0, for any J we have
f(y

+) = f(y

) +f

(y

).
As f

() L

(since (f, f

) = 1), we see that f

(y

) C

. Then by taking
:=
f(y

)
f

(y

)
,
an element of J, we get f(y

+) = 0. So we can set y = y

+.
Uniqueness of v: If y, y + are two lifts in C of u() C/J, then J, and we
have
f(y +) = f(y) +f

(y).
If in addition we have f(y +) = f(y) = 0, then because f

(y) C

, we must have
= 0. This shows that v is unique, proving the lemma.
For the remaining two cases, we may assume L/K is a purely inseparable exten-
sion of form L = K[X]/(X
p
a), where a K. Write f(X) := X
p
a.
Claim:
L/k
= ((
K/k

K
L) LdX)/Lf, where f := df(t) +f

(t) dX.
Proof: Here, the symbol df
K/k

K
K[X] is the element given by applying d
K/k
to the coecients of f(X), and df(t) is the reduction modulo (f) of that element,
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 75
i.e. its image in
K/k

K
L. Also, f

(t) is the reduction modulo (f) of f

(X), so
that f

(t) dX LdX.
To prove the claim, rst apply the second fundamental sequence to k K[X]
L to get the exact sequence
(27.2.1)
(f)
(f)
2

K[X]/k

K[X]
L
L/k
0,
where the rst map sends f d
K[X]/k
f 1.
Also, since K[X]/K is 0-smooth, the rst fundamental exact sequence for k
K K[X] gives a split exact sequence
(27.2.2) 0
K/k

K
K[X]
K[X]/k

K[X]/K
0
where the splitting is given by d
K[X]/k
g(X) dg, a left-inverse of the map
K/k

K
K[X]
K[X]/k
(check it is a left-inverse!).
Now substituting (27.2.2) into (27.2.1) proves the claim.
Case (3): We have (X
p
a) = 0, and so the claim shows that
L/k
= (
K/k

K
L) LdX, and hence r(L) = r(K) + 1.
Case (4): We have (X
p
a) ,= 0, and so r(L) = r(K).
In summary, we have the following formulas:
Case (1): r(L) = r(K) + 1;
Case (2): r(L) = r(K);
Case (3): r(L) = r(K) + 1;
Case (4): r(L) = r(K).
This immediately implies the rst parts of the following theorem.
Theorem 27.2.2. Suppose L K k are extension of elds, and suppose L/K
is a f.g. eld extension. Then
(i) dim
L

L/k
dim
K

K/k
+ tr.deg
K
L;
(ii) Equality holds if L/K is separably generated.
(iii) If L/k is f.g., then dim
L

L/k
tr.deg
k
L, and equality holds i L/k is
separably generated. In particular,
L/k
= 0 L/k is a separable algebraic
exension.
Proof. (i,ii): By induction on the number of generators of the eld extension L/K,
we may assume L = K(t), and then these two statements follow by a consideration
of Cases (1-4) above.
(iii): Take K = k to get the inequality . Next, assume
L/k
= 0. So r(L) = 0,
and for every eld K with L K k we have r(K) = 0 as well. Only Case (2)
above can occur for L/K/k, and so we see that L/k is separable and algebraic.
Next, assume r(L) = tr.deg
k
L =: r. Choose x
1
, . . . , x
r
L such that dx
1
, . . . , dx
r
form an L-base of
L/k
. It is easy to show that the elements x
1
, . . . , x
r
are alge-
braically independent over k.
Let k(x) := k(x
1
, . . . , x
r
) L. We claim L/k(x) is separable and algebraic. The
rst fundamental exact sequence applied to k k(x) L gives an exact sequence

k(x)/k

k(x)
L
L/k

L/k(x)
0
in which the left-most arrow is surjective (by choice of the x
i
). Thus
L/k(x)
= 0,
and so by the statement proved above, L/k(x) is separable algebraic, as desired.
76 THOMAS J. HAINES
28. Lecture 28
28.1. Proof of Proposition 27.1.1. Recall we have assumed k is perfect, and
B/m = k. Note that we have not assumed A (or B) is a domain for this proposition.
First suppose
B/k
is a free B-module of rank dim(B). Then the second funda-
mental exact sequence for k B B/m yields
m/m
2

B/k

B
B/m.
So dim
k
(m/m
2
) = rank
B

B/k
= dim(B), and so B is regular.
Conversely, assume (B, m) is regular. Recalling that B is then automatically a
domain, we set K := Frac(B). Using the argument above, we get from dim
k
(m/m
2
) =
dim(B) that dim
k

B/k

B
k = dim(B) =: r. Since k is perfect, the extension K/k
is separably generated, and thus we have
dim
K

B/k

B
K = tr.deg
k
K = r.
Now the B-module
B/k
has
dim
F

B/k

B
F = r
for F = K and F = k. It follows from this that
B/k

= B
r
, and we are done.
We used the following general lemma (see Hartshorne, II, Lemma 8.9).
Lemma 28.1.1. Suppose (A, m) is a Noetherian (this is not needed) local domain
with K := Frac(A) and k := A/m. Suppose M is a f.g. A-module such that
dim
k
M
A
K = dim
k
M
A
k = r. Then M is free of rank r.
28.2. Formal smoothness. Here is our motivation. If (A, m, K) is a local ring
(K := A/m), then we say A has a coecient eld if there is a subeld K

A
such that the composition K

A K is an isomorphism. Obviously in order

for A to have a coecient eld, it must contain some eld. An important question
goes in the opposite direction: supposing A contains some eld, does it then have
a coecient eld?
Theorem 28.2.1 (I.S. Cohen). If (A, m, K) is a complete Noetherian local ring
and A contains a eld k, then A has a coecient eld. If K/k is separable, then
there is a coecient eld containing k.
Corollary 28.2.2 (Cohen Structure Theorem). If (A, m, K) is a complete regular
local ring containing a eld, then A

= K[[X
1
, . . . , X
d
]], where d = dim(A).
Here is the idea behind the proof of the theorem (well give the details later).
Its ETS that there is a map u : K A such that pu = id
K
, where p : A
A/m = K is the projection. Since A = lim

A/m
i
, its ETS that for each successive
lift u
i
: K A/m
i
of u
1
= id
K
, we can lift one step further, i.e. nd u
i+1
making
the following commute:
K
ui

ui+1

A/m
i
A/m
i+1
.

If possible, we lift u
1
to get a compatible family of lifts u
1
, u
2
, u
3
, . . . , and these
determine the desired map u : K lim

A/m
i
= A.
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 77
Thus, what is required of the map k K (where is k A is the given subeld)
is that it be 0-smooth. We will study a circle of ideas related to proving that in
many cases K/k is 0-smooth. Along the way it is convenient to introduce a notion
of smoothness wherein the topology plays a role. This notion is called formal
smoothness.
To dene it we need some preliminary denitions. Suppose A is a topological
ring. We say I A is an ideal of denition if I
n
is a basis of open neighbor-
hoods around 0 A. We say a topological A-module M is discrete if IM = (0) for
some open ideal I A. If A is a local or semi-local ring and J A is the Jacobson
radical of A, unless otherwise mentioned we always give A the J-adic topology.
Suppose g : k A is a continuous map of topological rings. We say g is formally
smooth, or fs, if for every discrete ring C, and ideal N C with N
2
= 0, if we
are given continuous maps u, v making the following square commute, there is a lift
v

: A C of v making the triangles commute:

A
v

C/N
k

C.

Remark 28.2.3. (1) The map v

is automatically continuous: there is an open

ideal I A such that v(I) = 0. i.e. v

(I) N and so v

(I
2
) = 0. Since I
2
is an open ideal, this shows that v

is continuous.
(2) In the denition of fs, we can replace N
2
= 0 with N is nilpotent.
Indeed, suppose N
m
= 0 and that we can perform lifting for ideals whose
squares are zero. Then lift A C/N rst to A C/N
2
, and then to
A C/N
3
. Continuing, we eventually lift to A C/N
m
= C.
(3) If C is a complete and Hausdor with ideal of denition N (so that C =
lim

C/N
i
), then we can use the above argument to show that we can lift
v : A C/N to A lim

C/N
i
= C.
If A is fs over k for the discrete topologies on k, A, then we say A is smooth
over k. This is the same as our earlier notion of 0-smooth.
Thus, k A smooth implies k A is fs for any adic topologies on k, A such
that k A is continuous.
The following lemma explains to some extent why we use the terminology for-
mally smooth (since completions are connected to Grothendiecks theory of for-
mal schemes). It also highlights the importance of the continuity hypotheses in
the denition of formal smoothness.
Lemma 28.2.4. Let

A denote the I-adic completion of A, a Noetherian k-algebra
(where k is any ring). Then A is fs over k i

A is fs over k.
Proof. Suppose given a continuous v : A C/N making the following diagram
commute:
A
v

C/N
k

C.

78 THOMAS J. HAINES
Since v is continuous, it factors through a map v : A/I
m
C/N. Clearly if v
lifts to k-algebra map v

: A/I
m
C, then v lifts to a k-algebra map v

: A
C. Conversely, if v lifts to v

, then for some suciently large m, v lifts to a v

.
Thus, k A is fs i given any such diagram, for a suciently large m, the map
v : A/I
m
C/N lifts to a map v

: A/I
m
C.
The same argument applies to

A

I replacing A I. Also, recall that for every
integer m, A/I
m
=

A/

I
m
. It is now clear that A is fs over k i

A is.
Examples
(1) A = k[. . . , X

, . . . ] is smooth over k, for any ring k and any family of

indeterminates X

.
(2) If k denotes a Noetherian ring endowed with the discrete topology, then
A = k[[X
1
, . . . , X
n
]] is fs over k. (This follows from the lemma above, since
A is the (X
1
, . . . , X
n
)-adic completion of the fs (even smooth) k-algebra
k[X
1
, . . . , X
n
].
28.3. Some properties of formally smooth morphisms. The following prop-
erties are analogous to properties of smooth morphisms in the categories of varieties
or schemes.
Transitivity: If B is a fs A-algebra and A is a fs k-algebra, then B is a fs k-algebra.
Proof. Given the morphisms u, v making the outer quadrilateral commute, we rst
lift to nd w (using A/k is fs), and then lift w to nd v

(using B/A is fs).

B
v

C/N
A

k.

Localization: If S A is a multiplicative set, then S

1
A is smooth over A. (In
fact, we showed that S
1
A is 0-etale over A, which is stronger.)
Base Change: Suppose k, A, k

are topological rings, and k A and k k

are
continuous ring homomorphisms. Let A

= A
k
k

be endowed with the topology

of the tensor product. This means that if I
n
A and J
m
k

are families of
ideals dening the topologies on A and k

, then we dene the topology on A

to be
the one dened by the family I
n
A

+J
m
A

n,m
.
If A is fs over k, then A

is fs over k

.
Proof. Let p : A A

and k

be the canonical maps; they are continuous.

Given C, N, u, v in the diagram below making the rightmost square commutative
A
p

C/N
k

use the fs of A/k to nd the lifting w : A C of vp : A C/N. Then dene

A

C by a k

w(a)u(k

). This is a lift of v, as desired.

MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 79
28.4. Separability and smoothness for eld extensions. The following fact
is fundamental, and will easily imply the theorems of Cohen.
Theorem 28.4.1. A eld extension K/k is smooth i it is separable.
We refer to this as the fundamental fact.
28.5. Proofs of Cohens theorems, modulo the fundamental fact. We now
prove Theorem 28.2.1 and Corollary 28.2.2, modulo Theorem 28.4.1.
Proof of Theorem 28.2.1: If K/k is separable, then by the fundamental fact, K/k
is smooth. Hence we can lift id
K
: K A/m to a k-algebra homomorphism
K lim

A/m
i
= A. In general, let k
0
k be the prime eld. Then K/k
0
is
separable since k
0
is perfect, hence the above argument applies to produce the
coecient eld.
Corollary 28.5.1. Let (A, m, K) be a complete Hausdor local ring containing a
eld. Then if m is a f.g. ideal, the ring A is Noetherian.
Proof. Suppose m = (x
1
, . . . , x
n
), and let K

A be the coecient eld produced

by Theorem 28.2.1. Sending X
i
x
i
denes a surjective K

-algebra homomor-
phism
K

[[X
1
, . . . , X
n
]] A.
The denition makes sense since A is complete. Also, note that it is surjective on
the associated graded level (check this!), and so surjective by Lemma 18.2.2. Hence
A is Noetherian, being a quotient of a Noetherian ring.
Proof of Corollary 28.2.2: Note that since (A, m, K) is regular of dimension d, we
have in this case m = (x
1
, . . . , x
d
). Now from the proof of the corollary above, we
have
K[[X
1
, . . . , X
d
]]/P = A,
where P is a prime ideal (recall that A is a domain, being regular). But then di-
mension considerations show that P = 0 (otherwise the LHS would have dimension
< d = dim(A)). We are done.
28.6. Formal smoothness implies regularity.
Proposition 28.6.1. let (A, m, K) be a Noetherian local ring containing a eld k.
If A/k is fs, then A is regular.
Proof. Let k
0
k denote the prime eld. Note that k/k
0
is separable, hence
smooth, hence fs. Also, A is fs over k by hypothesis. Hence by transitivity, A/k
0
is fs. Thus, WLOG k is perfect.
Let K

denote a coecient eld of the complete local ring A/m

2
, containing k
(use Theorem 28.2.1). Let x
1
, . . . , x
d
m be a set of elements whose reductions
determine a K

-basis of m/m
2
.
There is an isomorphism of k-algebras
v
1
: A/m
2
K

[X
1
, . . . , X
d
]/J
2
,
where J := (X
1
, . . . , X
d
). (To see this, use that the obvious map from the RHS to
the LHS is an isomorphism on the associated graded level, hence is an isomorphism
since both sides are complete.)
80 THOMAS J. HAINES
Dene v : A K

[X]/J
2
as the composition of v
1
with the projection A
A/m
2
. Now, using that A/k is fs, lift v to k-algebra maps v

n
: A K

[X]/J
n+1
,
for n = 2, 3, . . . .
Since the elements v(x
1
), . . . , v(x
d
) generate J/J
2
= J/J
2
(where J := J/J
n+1
),
the elements v

n
(x
1
), . . . , v

n
(x
d
) generate J (by NAK).
It follows that
K

[X]/J
n+1
= v

n
(A) +J
2
= v

n
(A) +

i
v

n
(x
i
)(v

n
(A) +J
2
)
= v

n
(A) +J
3
=
= v

n
(A) +J
n+1
= v

n
(A).
Thus v

n
: A K

[X]/J
n+1
. Therefore
dim(A) = deg (A/m
n+1
) deg (K

[X]/J
n+1
) = d.
Since m is generated by d elements, this shows that A is regular, as desired.
29. Lecture 29
29.1. How liftings lead to 2-cocycles. Here we give the rst steps of our goal:
a homological criterion for smoothness over a eld k. This will be a key ingredient
in our proof of the fundamental fact (Theorem 28.4.1).
Assume for the time being that k is a eld. Consider the usual diagram
A
v

C/N
k

C.
q

From this we dene a k-subalgebra E AC by

E := (a, c) [ v(a) = q(c).
This is part of an extension of A by N:
0 N E A 0,
where N E is n (0, n) and E A is (a, c) a. The following result is
fundamental, but is easy and is left to the reader.
Lemma 29.1.1. The map v lifts to a k-algebra map v

: A C i 0 N E
A 0 splits in the category of k-algebras (meaning the splitting map A E is a
k-algebra homomorphism).
Proof. Exercise.
Now, since k is a eld, and everthing in sight is a k-vector space, the extension E
always splits in the category of k-vector spaces. We may therefore write E = AN,
and then express the multiplication in terms of a symmetric 2-cocycle
f : AA N.
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 81
That is, the multiplication in E can always be expressed as
(a
1
, n
1
) (a
2
, n
2
) = (a
1
a
2
, a
1
n
2
+a
2
n
1
+f(a
1
, a
2
)),
where f is symmetric, bilinear and satises (by associativity in E) the relation
af(b, c) f(ab, c) +f(a, bc) f(a, b)c = 0,
for all a, b, c A.
Such extensions are called Hochschild extensions. We will dene this for-
mally in the next subsection. Note that it is already clear that the splitting of the
extension in the category k-Alg is detected by whether the 2-cocycle f is trivial
or not. Thus, the smoothness of k A is going to be related to the vanishing of a
certain H
2
cohomology group.
In the next few subsections, we will explain this more formally.
29.2. Extensions. Here we continue to work towards a cohomological criterion for
smoothness of k A, where k is a eld. The same discussion goes over word-for-
word when we only assume k is a ring and A is projective as a k-module.
Given a k-algebra and ideal C N, with N
2
= 0, write C

:= C/N. Then
N is naturally a C

-module. Conversely, suppose C

is a ring and let N be any

C

-module. An extension of C

by N is a triple (C, , i) such that

- C is a ring;
- : C C

, and ker()
2
= 0;
- i : N ker() is an isomorphism of C

-modules.
We represent the extension with an exact sequence
0

N
i

0.
Given a C

-module N, we always have the trivial extension C

N = C

N,
where the multiplication is dened by
(a, x) (b, y) := (ab, ay +bx).
An isomorphism between (C, , i) and (C
1
,
1
, i
1
) is a ring homomorphism f : C
C
1
such that the following commutes:
0

N
i

C
f

0
0

N
i1

C
1
1

0.
Such an f is automatically an isomorphism (the snake lemma or the 5-lemma), and
is unique (check this!).
Exercise 29.2.1. Show that (C, , i)

= C

N a ring homomorphism section

s : C

C for such that s = id

C
.
29.3. Hochschild extensions. We say (C, , i) is a Hochschild extension if the
exact sequence
0

N
i

0
splits in Z-mod: there exists an additive map s : C

C such that s = id
C
.
In that case, C = C

N as an abelian group, and the multiplication in C is

given by
(a, x) (b, y) = (ab, ay +bx +f(a, b)),
82 THOMAS J. HAINES
for a function f : C

N. Why? Write (a, 0) = s(a). Note that (s(a)s(b)

s(ab)) = 0 implies that the function f is given by
f(a, b) := s(a)s(b) s(ab) N.
Note that f is symmetric (since C is commutative), bilinear, and satises the
following cocycle relation (a rephrasing of C is associative):
(29.3.1) af(b, c) f(ab, c) +f(a, bc) f(a, b)c = 0,
for a, b, c C

. Such a function f : C

N is called a symmetric 2-cocycle.

Without the hypothesis of symmetry, f is called simply a 2-cocycle.
Conversely, any such f gives rise to a Hochschild extension of C

by N. The
extension is isomorphic to the trivial extension C

N i g : C

N such that
(29.3.2) f(a, b) = ag(b) g(ab) +g(a)b.
In this case we say f is a 2-coboundary. More generally, two Hochschild extensions
determined by f
1
, f
2
: C

N are isomorphism i f
1
f
2
is a 2-coboundary
(check this!).
The quotient of symmetric 2-cocycles modulo 2-coboundaries is denoted H
2
(C

, N)
sym
.
We can also formulate all of the above in the category of k-modules: then an
extension is Hochschild if it splits in the category k-Mod. In this case the k-module
quotient of symmetric 2-cocycles modulo 2-coboundaries is denoted H
2
k
(C

, N)
sym
.
We can summarize the above discussion as follows.
Lemma 29.3.1. Given a k-algebra C

and a C

-module N, there is a canonical

bijection
_
Hochschild extns of C

by N
_
/

= H
2
k
(C

, N)
sym
.
29.4. Relation of Hochschild extensions to smoothness. Assume A is pro-
jective over the ring k (e.g. k could be a eld). Let N denote an A
k
A-bimodule
(i.e. a A
k
A
op
-module). In the next subsection we are going to dene Hochschild
(co)homology groups H
n
k
(A, N) (resp. H
k
n
(A, N)) for all n 0.
Here is the connection with the notion of smoothness. Consider a diagram
(29.4.1) A
v

C/N
k

C,
q

where N
2
= 0. This gives rise to a Hochschild extension E := (a, c) A
C [ v(a) = q(c) in k-alg. (We used A is projective over k.) We represent the
extension as
(29.4.2) 0

N
i

A

0,
where i(n) = (0, n) and (a, c) = a.
Note that in (29.4.1), v lifts to a k-algebra map v

: A C i the extension E in
(29.4.2) is trivial as a k-algebra extension: there exists a k-algebra map s : A E
such that s = id
A
.
Therefore,
A/k is smooth every extension (29.4.2) splits in k-alg
H
2
k
(A, N)
sym
= 0 for every N arising from (29.4.1).
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 83
In summary, we have
Proposition 29.4.1. Let A be a projective k-algebra. Then A/k is smooth i
H
2
k
(A, N)
sym
= 0 for all A-modules N.
Proof. Any N as in (29.4.1) is a C/N-module hence (via v) is an A-module. Con-
versely, given an A-module N, let C = AN, which contains N as an ideal such that
N
2
= 0. Thus N appears in a diagram of the form (29.4.1). Now, the proposition
follows from our discussion above.
29.5. Hochschild (co)homology. In this subsection k denotes a ring, and A
denotes a k-algebra (not necessarily commutative!). Let M denote an A
k
A-
bimodule, that is, a left A
e
:= A
k
A
op
-module. The ring A is itself an A
e
-module,
via the homomorphism : A
k
A A given by a b ab.
For n 0, we dene the Hochschild cohomology by
H
n
k
(A, M) := Ext
n
A
e(A, M),
and the Hochschild homology by
H
k
n
(A, M) := Tor
A
e
n
(A, M).
Recall that if 0 A P
0
P
1
is an A
e
-free resolution of A, then
Ext
n
A
e(A, M) = H
n
(Hom
A
e(P

, M))
Tor
A
e
n
(A, M) = H
n
(P

A
e M).
This is useful, as we can construct a very simple and explicit resolution A P

as follows. For simplicity, at this point we assume A is k-free. Let us dene

X
0
= A
k
A = A
e
X
1
= A
k
A
k
A

X
n
= A

k
n+2

= A
e

k
X
n2
.
Note that X
n
is an A
e
-module by (a b) (x
0
x
n+1
) = ax
0
x
n+1
b.
As A
e
-modules we have
X
n

= A
e

k
X
n2
given by a x b (a b) x. Therefore, since X
n2
is k-free, we see that X
n
is A
e
-free, for all n. Thus, we can dene an A
e
-free resolution A P

by
0 A

X
0

X
1
d1

X
2
d2

d3

where d
n
: X
n
= A

k
n+2
A

k
n+1
= X
n1
is given by
d
n
(x
0
x
n+1
) =
n

i=0
(1)
i
x
0
x
i
x
i+1
x
n+1
.
It is clear that and each d
n
is A
e
-linear. Moreover, it is easy to see d
n
d
n+1
= 0
for all n 0 (by convention, d
0
= ). Why is the sequence exact? This follows
from the existence of contracting homomorphisms
A
s1

X
0
s0

X
1
s1

84 THOMAS J. HAINES
such that
s
1
= id
A
d
1
s
0
+s
1
= id
X0
d
n+1
s
n
+s
n1
d
n
= id
Xn
,
the latter holding for all n 1. For each n 1, set
s
n
(x
0
x
n+1
) := 1 x
0
x
n+1
.
Now we want to use this explicit resolution to identify H
n
k
(A, M) more concretely.
As stated above, we know that
H
n
k
(A, M) = H
n
_
0 Hom
A
e(X
0
, M) Hom
A
e(X
1
, M)
_
.
Now
Hom
A
e(X
n
, M) = Hom
A
e(X
n2

k
A
e
, M)
= Hom
k
(A

k
n
, M)
= C
n
(A, M),
where C
n
(A, M) denotes the additive group of all k-bilinear maps A
n
M.
Dene : C
n
(A, M) C
n+1
(A, M) by setting f(x
1
, . . . , x
n+1
) to be
x
1
f(x
2
, . . . , x
n+1
)+
n

i=1
(1)
i
f(x
1
, . . . , x
i1
, x
i
x
i+1
, x
i+2
, . . . , x
n+1
)+(1)
n+1
f(x
1
, . . . , x
n
)x
n+1
.
The commutativity of the diagram (check it!)
Hom
A
e(X
n
, M)

Hom
A
e(X
n+1
, M)

C
n
(A, M)

C
n+1
(A, M)
yields
Theorem 29.5.1.
H
n
k
(A, M) =
ker( : C
n
(A, M) C
n+1
(A, M))
im( : C
n1
(A, M) C
n
(A, M))
.
Let us see explicitly what elements in H
2
k
(A.M) look like with the present de-
nition. A function f : AA M satisifes (f(a, b, c) = 0 i
af(b, c) f(ab, c) +f(a, bc) f(a, b)c = 0,
that is, i f is a 2-cocycle in our earlier terminology.
Further, given a function g : A M, we have
g(a, b) = ag(b) g(ab) +g(a)b.
Thus, g is precisely a 2-coboundary in our earlier terminology. Hence, we conclude
that H
2
k
(A, M) as dened in this subsection agrees with the denition given in
subsection 29.3.
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 85
29.6. Proof of the fundamental fact, Theorem 28.4.1. We now prove that
K/k is smooth is it separable.
Proof.
(): We will use Proposition 29.4.1. Hence, we must prove that for a K-module
N, we have H
2
k
(K, N)
sym
= 0.
We may write K =
i
L
i
, where each L
i
/k is a nitely generated and separable
(hence separably generated) eld extension.
Lemma 29.6.1. Any separably generated eld extension L/k is smooth.
Proof. Pure transcendental extensions are smooth (why?). Separable algebraic ex-
tensions are 0-etale (Lemma 27.2.1), hence smooth. The result now follows by the
transitivity property of smoothness.
Hence by Proposition 29.4.1, we have H
2
k
(L
i
, N)
sym
= 0 for all i. From this, it
follows that H
2
k
(K, N)
sym
= 0.
Lets check this last statement in the case where K has countable transcendence
degree over k (for the general case, see [Mat1]). In this case, we can write K =
i
L
i
as a countable directed union. That is, we may assume
L
i
L
i+1
.
Let f be a symmetric 2-cocycle, f : KK N. By hypothesis, f[
Li
= g
i
, where
g
i
: L
i
N, for each i. We want to glue the g
i
s to get a function g : K N
such that f = g. The obvious problem is, g
i+1
[
Li
might not be g
i
. The idea is to
alter g
i+1
so that this is true (without disturbing the property g = f).
Note that (g
i
g
i+1
[
Li
) = 0, so that g
i
g
i+1
[
Li
: L
i
N is a k-derivation.
Since L
i+1
/k is a f.g. separably generated extension, so is L
i+1
/L
i
. (Why? Its
enough to check that L
i+1

Li
L

is reduced for any eld L

L
i
. But this embeds
into (L
i+1

k
L
i
)
Li
L

= L
i+1

k
L

, which is reduced since L

i+1
/k is separable.)
Since L
i+1
/L
i
is f.g. separably generated, hence 0-smooth, the splitting of the rst
fundamental sequence for k L
i
L
i+1
shows that the natural restriction map
Der
k
(L
i+1
, N) Der
k
(L
i
, N)
is surjective. Thus, we can extend g
i
g
i+1
[
Li
to a k-derivation

g
i
g
i+1
[
Li
:
L
i+1
N. Now replace g
i+1
with g
i+1
+

(g
i
g
i+1
[
Li
) : L
i+1
N. This new g
i+1
has
f[
Li+1
= g
i+1
g
i+1
[
Li
= g
i
.
Continuing in this way, we can glue the g
i
s together to get g : K N such that
f = g. Thus, H
2
k
(K, N)
sym
= 0, as desired. This completes the proof of ().
(): Let k

k be a eld extension. We need to show that K

k
k

is reduced. It
is enough to prove this in the case where k

/k is nite. Why? First, it is clearly

enough to consider the case where k

/k is a f.g. eld extension. Then let k

be
a transcendence basis for k

/k and note that

K
k
k

= (K
k
k())
k()
k

.
Now our reduction to the case k

/k is nite follows: K
k
k() is a smooth k()-
algebra, and the algebraic extension k

/k() is a union of nite algebraic extensions.

86 THOMAS J. HAINES
Thus, we henceforth assume k

/k is nite. Then K
k
k

is a nite dimensional
K-vector space, hence is an Artinian ring. By Atiyah-Macdonald Theorem 8.7,
K
k
k

= A
1
A
r
,
where each A
i
is an Artinian local ring, and a nite-dimensional K-algebra. Now
K
k
k

is smooth over k

implies (exercise) that each A

i
is smooth over k

. Therefore
by Proposition 28.6.1, each A
i
is a regular local ring. But regular local rings are
domains, and thus each A
i
is actually a eld. But then K
k
k

is reduced, as
desired. This completes the proof of ().
29.7. Geometric regularity, and nal remarks.
Theorem 29.7.1. Let (A, m, K) be a Noetherian local ring, containing a eld k.
Let

A denote the m-adic completion of A. Suppose K/k is separable. Then TFAE:
(1) A is regular.
(2)

A

= K[[X
1
, . . . , X
d
]] as K-algebras and as k-algebras too (where d =
dim(A)).
(3)

A is fs over k.
(4) A is fs over k.
Proof. (1) (2): Since

A is complete and regular, and contains k, (2) follows from
the Cohen Stucture theorem (Corollary 28.2.2).
(2) (3): Clear since then

A fs over K and K fs over k (since K/k separable; use
Theorem 28.4.1).
(3) (4): Lemma 28.2.4.
(4) (1): Proposition 28.6.1.
For now on, assume (A, m) is a Noetherian local ring, and contains a eld k.
Lemma 29.7.2. If B is a nite A-module, then B is semi-local.
Proof. Note that B/mB is a nite A/m-module, hence is Artin, and thus has nitely
many maximal ideals. The maximal ideals of B all lie over m (by the Going-Up
theorem), so B has only nitely many maximal ideals. Thus B is semi-local.
In particular, for every nite extension k

k, the ring A

:= A
k
k

is semi-local.
Recall that we say such a ring is regular provided all of its localizations at maximal
ideals are regular. We say A is geometrically regular over k if A

:= A
k
k

is
regular, for every nite extension k

/k.
Lemma 29.7.3. If A/m is separable over k, then
A is regular A is fs over k
=A

is fs over k

=A

is regular.
Proof. Only the nal implication needs explanation. It does not follow immediately
from Proposition 28.6.1 because A

is not local, but only semi-local. Nevertheless,

if n A

is a maximal ideal, then A

n
is fs over k

(recall A

n
/A

is 0-etale), and then

Proposition 28.6.1 yields A

n
is regular for each n. Thus A

is regular.
Thus, in case A/m is separable over k, we have regular geometrically
regular. In general, we can say the following.
MATH 603: INTRODUCTION TO COMMUTATIVE ALGEBRA 87
Proposition 29.7.4. Suppose (A, m, K) be Noetherian local, containing a eld k.
Then A is fs over k i A is geometrically regular over k.
Proof. (): If A is fs over k, then A

is fs over k

, and then the proof of the Lemma

above shows that A

is regular. Thus A is geometrically regular over k.

(): (Sketch; we consider only the case where K/k is a f.g. eld extension.) Take
a radiciel extension k

k such that K(k

) is separable over k

. Then A

:= A
k
k

is regular, and has residue eld K(k

). So A

is fs over k

. We conclude that A is
fs over k by invoking the following lemma.
Lemma 29.7.5. Let A be a topological ring containing a eld k. Let k

k be a
k-algebra endowed with the discrete topology. Then A is fs over k i A

:= A
k
k

is fs over k

.
Proof. We assume A

is fs over k

; we need to prove A is fs over k. Consider the

usual diagram
A
v

C/N
k

C.

Tensoring the diagram with

k
k

yields a diagram
A

/N

where C

= C
k
k

, N

= N
k
k

, and v

= v
k
id
k
. The lifting w exists since
A

/k

is fs.
Now choose a k-submodule V of k

such that k

= kV as k-vector spaces. Note

that C

= C (C V ), and C V is a C-submodule of C

. Write
w(a) = u(a) +r(a),
where u(a) C and r(a) C V , for a A. Since the image of w(a) modulo
N

is v(a) C/N, we see that r(a) N V , for all a A. This implies that
r(a)r(b) = 0, for a, b A. Thus u : A C is a k-algebra homomorphism, lifting v.
This shows that A is fs over k, as desired.
References
[Mat1] H. Matsumura, Commutative Algebra.
[Mat2] H. Matsumura, Commutative Ring Theory, Camb. stud. in adv. Math. 8, Cambridge Univ.
Press, 1992, 320 pp. + xiii.
[Serre] J.-P. Serre, Local Algebra, Springer Monographs in Math., 2000.