Noetherian Algebras

1. Basic Theory

1.1. R ring with a 1, not necessarily commutative.

Definition. A left R-module is an abelian group M (under +) together with a

multiplication
RM M
(r, m) 7 r.m
satisfying
r(m + n) = rm + rn
(r + s)m = rm + sm
(rs)m = r(sm)
1m = m
for all m, n M and r, s R.
Right R-modules are defined similarly.

Examples. (1) R itself is both a left R-module and a right R-module.

(2) Any abelian group M is a left (and right) Z-module.
(3) If R = k is a field, an R-module is just a k-vector space.

1.2. Definition. The left R-module M is said to be cyclic if it can be generated

by a single element: M = Rx for some x M . M is finitely generated if it can be
written as a finite sum of cyclic submodules M = Rx1 + Rx2 + . . . + Rxn .

Lemma. Any cyclic left R-module M = Rx is isomorphic to a quotient of R.

Proof. The map R M given by r 7 rx is a left R-module homomorphism which

is onto and has kernel ann(x) = {r R : rx = 0}. This is a left ideal of R, known
as the annihilator of x. Hence M
= R/ ann(x). 

1.3. Lemma. Let M be a left R-module. The following are equivalent:

(i) Every submodule of M is finitely generated
(ii) Ascending chain condition: There does not exist an infinite strictly as-
cending chain of submodules of M
(iii) Maximum condition: Every non-empty subset of submodules of M contains
at least one maximal element. (If S is a set of submodules, then N S is a
maximal element if and only if N 0 S, N N 0 implies N = N 0 ).

Proof. (i) (ii). Suppose M1 ( M2 ( . . .. Let N = Mn . Then N is a submodule

of M so N is finitely generated by m1 , . . . , mr say. If mi Mni , then it follows
that N = Mn where n = max ni , a contradiction.
(ii) (iii) If S is a nonempty subset with no maximal element, pick M1
S. Since S has no maximal element, we can find M2 S such that M1 ( M2 .
1
2

Continuing like this gives a strictly ascending infinite chain M1 ( M2 ( . . ., a

contradiction.
(iii) (i) Let N be a submodule of M and let S be the set of submodules of
N which are finitely generated. Since 0 S, S has a maximal element L, say. Let
x N . Since L + Rx is a finitely generated submodule of N and L is maximal in
S, L + Rx = L so x L. Hence N = L is itself finitely generated. 

Dually, we have the descending chain condition and the minimum condition;
these are equivalent to each other.

1.4. Definition. An R-module satisfying (i), (ii), (iii) of Lemma 1.3 is Noetherian.
The ring R is left Noetherian if it is Noetherian as a left R-module.
An R-module satisfying the descending chain condition is said to be Artinian.
The ring R is left Artinian if it is Artinian as a left R-module.
We have similar definitions on the right hand side. Note that if the ring is
commutative, there is no difference between left and right.

Examples. (1) Any division ring D is right and left Noetherian (as well as
right and left Artinian)
(2) Any principal ideal domain R is Noetherian - for example, R = Z or R =
Q[x].

1.5. Proposition. Let N be a submodule of M , a left R-module. Then M is

Noetherian if and only if both N and M/N are Noetherian.

Proof. () Since any submodule of N is a submodule of M , N is left Noetherian.

If Q1 /N < Q2 /N < . . . is a strictly ascending chain in M/N then Q1 < Q2 < . . . is
a strictly ascending chain in M , so M/N is left Noetherian.
() Suppose N and M/N are Noetherian and let L1 L2 . . . be an ascending
chain in M . Then (L1 + N )/N (L2 + N )/N . . . is an ascending chain in
M/N which terminates since M/N is Noetherian. Hence there exists k such that
Li + N = Lk + N for all i k. Since N is Noetherian, there exists m such that
Li N = Lm N for all i m.
We claim that Li = Ln for all i n := max(k, m). Clearly Ln Li ; let x Li .
Then x Li Ln + N so there exists y Ln and z N such that x = y + z. Now
z = x y Li N = Ln N so x Ln also, as required. 

1.6. Lemma. Suppose M can be expressed as a sum M = M1 + M2 + . . . + Mn .

Then M is Noetherian if and only if each Mi is Noetherian.

Proof. Suppose each Mi is Noetherian. By an easy induction using (1.5), the di-
rect sum N = M1 M2 Mn is Noetherian. Now there is an R-module
homomorphism f : N M such that f (m1 , . . . , mn ) = m1 + . . . + mn . Since
M = M1 + M2 + . . . + Mn , f is onto so M
= N/ ker f is Noetherian by (1.5).
 3

Conversely, since each Mi is a submodule of M , if M is Noetherian then each

Mi is Noetherian, again by (1.5). 

Corollary. Let R be a left Noetherian ring. Then any finitely generated left R-
module is left Noetherian.

Proof. Any finitely generated module is a sum of cyclic modules. The result follows
from (1.2) and (1.6). 

1.7. Proposition. Let R be a left Noetherian ring and let S be a ring containing
R such that S is a finitely generated left R-module under left multiplication by R.
Then S is Noetherian.

Proof. By Corollary 1.6, S is Noetherian as a left R module. Since any left ideal
of S is necessarily an R-submodule, S satisfies the ACC on left ideals. Hence S is
Noetherian. 

1.8. Examples of Noetherian rings.

(1) Rings which are finite dimensional vector spaces over a field k, for example
Mn (k) and group algebras kG of finite groups (see (1.12)). Use Proposition
1.7 with R = k.
(2) Rings of integers in algebraic number field, Mn (Z). These are both finitely
generated Z-modules.
(3) k[[X]], the power series ring in one variable over a field k (this is a PID).
(4) Let p be a fixed prime and let Z(p) = {q = m/n Q : (m, n) = 1 and p - n}.
This is an example of a localisation of Z. The localisation of a Noetherian
ring is Noetherian - see later.
(5) R = k[X1 , . . . , Xd ], k a field. This is Hilberts Basis Theorem - see
Theorem 1.14.

1.9. Free algebras. Let k be a field. A k-algebra is a ring R containing k as a

central subfield.

Definition. The free associative algebra on n generators khx1 , . . . , xn i is the k-

vector space with basis given by all possible products y1 ym where y1 , . . . , ym
{x1 , . . . , xn }. Multiplication is given by concatenation on basis elements and is
extended by k-linearity to the whole of khx1 , . . . , xn i.

Note that khx1 , . . . , xn i is not finite dimensional over k.

For example, if n = 1 then khxi has {1, x, x2 , . . .} as a basis. In fact khxi = k[x],
the polynomial algebra.
Similarly, khx, yi has as a k-basis the set {1, x, y, x2 , xy, yx, y 2 , x3 , x2 y, . . .}. It
can be shown (Exercise) that khx, yi is not Noetherian. For n 3 there is a ring
homomorphism khx1 , . . . , xn i khx1 , x2 i which sends xi to xi if i = 1 or 2 and to
0 if i 3. This map is surjective, so khx1 , . . . , xn i is not Noetherian for n 2.
4

Let V be the vector space spanned by {x1 , x2 , . . . , xn }. Then

T (V ) := k V (V V ) (V V V )
is the tensor algebra on V . Multiplication is given on homogeneous components
by concatenation and is extended by linearity to T (V ). Check that T (V ) =
khx1 , . . . , xn i where n = dim V , as k-algebras.

1.10. Enveloping algebras.

Definition. Let k be a field. A Lie algebra over k is a k-vector space g, equipped

with a Lie bracket [.] : g g g satisfying
(1) (x, y) 7 [x, y] is bilinear
(2) [x, x] = 0 for all x g and hence [y, z] = [z, y] for all y, z g
(3) [x, [y, z]] + [y, [z, x]] + [z, [y, x]] = 0 for all x, y, z g.
Note that this bracket is not associative.

Examples. (1) Any (associative) k-algebra R becomes a Lie algebra under the
commutator bracket [x, y] = xy yx.
(2) gln (k), the set of all n n matrices over k with the commutator bracket.
(3) sln (k), the set of traceless n n matrices over k with commutator bracket.
(4) If V is any vector space, we can define the trivial bracket [x, y] = 0 for all
x, y V . This is the abelian Lie algebra.

Definition. The universal enveloping algebra U(g) of the Lie algebra g is defined
to be
U(g) := khx1 , . . . xn i/I
where {x1 , x2 , . . . , xn } is a basis for g and I is the ideal of khx1 , . . . xn i generated
by the set {xi xj xj xi [xi , xj ], 1 i, j n}.

For example, if g is abelian, then U(g) is just the polynomial algebra k[x1 , . . . , xn ].

Proposition. U(g) is left (and right) Noetherian for any finite dimensional Lie
algebra g.

Proof. See Corollary 7.8. 

1.11. Weyl algebras. We begin with a motivational result.

Lemma. Let A = k[x] and consider the k-linear maps x : A A and x
b : A A,
where x b is multiplication by x. Then

[ ,x b] = 1.
x
Proof. By the product rule,

[ ,x b](f ) = (xf )0 xf 0 = f
x
for all f A. 
 5

Now consider the polynomial algebra A = k[x1 , . . . , xn ] and the k-linear maps

xbi : A A and xi : A A
f
f 7 xi f f 7 xi
,

for 1 i n. These maps are examples of differential operators on A. It can be

verified that all these operators commute, except for x i
and xbi , which satisfy the
relation

[ , xbi ] = 1.
xi
Definition. Let k be a field. The n-th Weyl algebra An (k) over k is defined to be

An (k) := khx1 , . . . , xn , y1 , . . . yn i/I

where I is the ideal of khx1 , . . . , xn , y1 , . . . yn i generated by

xi xj xj xi 1 i, j n,
y i yj y j yi 1 i, j n,
yi xi xi yi 1 1 i n,
xi yj yj xi i 6= j.

When n = 1, we obtain

A1 (k) = khx, yi/hyx xy 1i.

There is a surjection An (k)  S where S is the k-subalgebra of Endk (A) gen-

f f f
erated by {cx1 , . . . , x
cn , x1
, . . . , x n
}, mapping xi to xbi and yi to xi
. In fact,

An (k) = S if and only if the characteristic of k is zero.

Proposition. The Weyl algebras are left (and right) Noetherian for all n 1.

Proof. See Corollary 7.8. 

1.12. Group Algebras.

Definition. Let G be a group and let R be a ring. The group algebra RG consists
of formal linear combinations
X
rg g,
gG

where rg R for all g G and all but finitely many rg are zero. Addition and
multiplication is given by
X X X
( rg g) + ( sg g) = (rg + sg )g
gG gG gG

X X X X
( rh h)( sk k) = ( rh sk )g.
hG kG gG h,kG
hk=g
6

When k is a field and V is a k-vector space, recall that a representation of G on

V is a group homomorphism
: G Autk (V ).
Such a homomorphism extends uniquely to a k-algebra homomorphism
: kG Endk (V )
and V may then be regarded as a left kG-module, via
x.v = (x)(v)
for all x kG.
Conversely, if V is a left kG-module, there is a representation : G Autk (V )
given by
(g)(v) = g.v
for all v V .
Thus there is a 1-1 correspondence between representations of G and left kG-
modules.

1.13. Group algebras of poly-(cyclic or finite) groups.

Definition. The group G is said to be poly-(cyclic or finite) if there is a chain

1 = G0 / G1 / . . . / Gn1 / Gn = G
of subgroups of G such that each Gi /Gi1 is cyclic or finite for each i = 1, . . . , n.

Examples. (1) Finite groups G.

(2) Infinite cyclic G = hxi = Z. In this case, kG = k[x1 , x], the ring of
Laurent polynomials.
n 1 1
(3) Free
abelian G =hx1 , . . . , xn i = Z : kG = k[x1 , x1 , . . . , xn , xn ].
1 Z Z
(4) G = 0 1 Z. Here we have the chain

0 0 1
1 / G1 / G2 / G3 = G

1 0 Z 1 Z Z
where G1 = 0 1 0 and G2 = 0 1 0 .

0 0 1 0 0 1

Proposition. Let R be a left(right) Noetherian ring and let G be a poly-(cyclic or

finite) group. Then RG is left(right) Noetherian.

Proof. After the proof of Hilberts Basis Theorem. 

Question. Suppose ZG is left Noetherian. Must G be poly-(cyclic or finite)?

Theorem (Auslander-Swan). Let G be a poly-(cyclic or finite) group. Then

 7

(1) G is isomorphic to a subgroup of GLn (Z) for some n.

(2) G has a subgroup H of finite index isomorphic to a subgroup of Trn (O),
the group of upper-triangular matrices in GLn (O), where O is the ring of
integers of an algebraic number field.

Proof. Omitted. 

1.14. Hilberts Basis Theorem. We will prove the following noncommutative

version of Hilberts Basis Theorem:

Theorem (McConnell, 1968). Let S be a ring, R a left Noetherian subring and let
x S.
(1) If R + xR = R + Rx and S = hR, xi, then S is left Noetherian.
(2) Suppose Aut(R) is such that rx = xr for all r R. If S = hR, xi,
then S is left Noetherian.
(3) Suppose x is a unit in S such that x1 Rx = R. If S = hR, x, x1 i, then S
is left Noetherian.

Proof. (1). R + Rx + . . . + Rxn = R + xR + . . . + xn R:

This follows by from R + xR = R + Rx. To see this, use induction to show that
x R R + Rx + . . . + Rxn and Rxn R + xR + . . . xn R for all n 1.
n

Consequences:
(a) The set of all elements of S of the form
r0 + xr1 + . . . + xn rn , n0 ()
forms a subring of S. Since it contains both R and x and S = hR, xi, we see
that S is the ring of all such polynomials. Note that elements of S need not
be uniquely expressible in the form ().
(b) The set of polynomials of degree n, namely R + Rx + . . . + Rxn , is both a
left and a right R-submodule of S.
(c) For each r R and n 0 there exists r0 R such that r0 xn = xn r + s where
deg s < n.
Now, let I be a left ideal in S. We will show that I is finitely generated. Let
In = {rn R : there exists s I such that s = r0 + xr1 + . . . + xn rn }.
Its clear that In is closed under addition. Let r R. By part (c) above, we can
find r0 R such that r0 xn xn r has degree < n. Since I is a left ideal, r0 s I, and
r0 s r0 xn rn xn (rrn )
modulo terms of degree < n. Hence rrn In so In is a left ideal of R.
Pn Pn+1
Next, if s = i=0 xi ri I, then xs = i=1 xi ri1 I so rn In+1 . Hence
In In+1 for all n 0.
Since R is left Noetherian, the increasing chain
I0 I1 . . . In . . .
8

must terminate. Say Im = Im+1 = . . .. For i = 0, . . . , m let {rij } be finitely many

elements of R generating Ii as a left ideal of R. Choose sij = xi rij + lower degree
terms I.
Claim: X = {sij : 0 i m, all j} generates I as a left ideal.
Let s = r0 + xr1 + . . . + xn rn I, so that rn In ; well show that s RX.
Proceed by induction on n, the case n = 0 being trivial.
aj rmj for some aj R. Choose a0j R such
P
If n m then rn Im so rn =
that aj x = x aj + lower degree terms. Then s a0j xnm smj I and modulo
0 n n
P

terms of degree < n,

X X X
s a0j xnm smj xn rn a0j xn rmj xn rn xn aj rmj = 0.
So s a0j xnm smj has smaller degree than s and we can apply induction.
P

aj rnj for some aj R, so for suitable a0j R, s

P
If n m then rn =
P 0
aj snj I also has smaller degree than s. By induction, these smaller degree
elements of I lie RX, as required.
(2) If rx = xr for all r R, then Rx = xR so R + xR = R + Rx. Apply (1).
(3) Let T = hR, xi. Then T is left Noetherian by (2). Let I be a left ideal of S.
Pn
Now, I T is a left ideal of T and is hence finitely generated: I T = i=1 T si ,
Pn
say. If s S, then xm s I T for some m 0, so s = i=1 xm ai si for some
ai T . Hence the si s generate I as a left ideal of S.


Proof of Proposition 1.13. Choosing a chain of subgroups

1 = G0 / G1 / . . . / Gn1 / Gn = G
we see that its sufficient to show that if RGi1 is left Noetherian then so is RGi
for all i = 1, . . . , n.
Suppose first that Gi /Gi1 is finite. Choose a finite transversal X for Gi1 in
Gi , so that Gi = xX Gi1 x. Then
X
RGi = RGi1 x
xX

is a finitely generated left RGi1 -module, so RGi is left Noetherian by Proposition

1.7. Now, if Gi /Gi1 is infinite cyclic, choose a generator xGi1 ; then RGi is gen-
erated by RGi1 ,x and x1 . Since Gi1 / Gi , RGi1 is invariant under conjugation
by x, so RGi is left Noetherian by Theorem 1.14(3). 
 9

2. Ideal structure

Throughout this chapter, R denotes an arbitrary ring, unless stated otherwise.

2.1. Simple modules.

Definition. An R-module is M is simple or irreducible if M 6= 0 and the only

submodules of M are 0 and M .

Suppose M is simple. Choose 0 6= x M ; then M = Rx so M = R/I where

I = ann(x) is the point annihilator of x, by Lemma 1.2. Note that ann(x) need
not be equal to ann(y) if x, y are distinct nonzero elements of M , unless R is
commutative.
Note that M = Rx is simple if and only if ann(x) is a maximal left ideal of R.

2.2. Zorns Lemma.

Definition. A poset is a set equipped with a binary relation which is reflexive,

transitive and antisymmetric. A chain in a poset S is a countable subset C =
{x1 , x2 , . . .} of S such that
x1 x2 . . . .
An upper bound for a subset C of S is an element u S such that x u for all
x C. We say that x S is a maximal element if x y with y S forces x = y.

Theorem (Zorns Lemma). Let S be a nonempty poset. Suppose every chain in S

has an upper bound. Then S has a maximal element.

This is equivalent to the Axiom of Choice, which we will always assume.

2.3. Lemma. Suppose L is a proper left ideal of R. Then L is contained in a

maximal ideal I of R. Equivalently, every cyclic module has a simple quotient.

Proof. Since L is proper, 1 / L. Let S = {K /l R : L K, 1 / K}. Since L S,

this set is nonempty. S is partially ordered by inclusion. If K1 K2 . . . is a chain
in S, then Kn also contains L and doesnt contain 1, i.e. Kn S. Hence every
chain in S has an upper bound in S. By Zorns Lemma, S has a maximal element
I. Its clear that I is now a maximal left ideal of R containing L as required. 

2.4. Primitive ideals. Recall that by an ideal of R we mean a two-sided ideal.

Definition. Let I be a two-sided ideal of R. Then I is primitive if I is the anni-

hilator of a simple left R-module M :
\
I = AnnR (M ) = {x R : xM = 0} = ann(x).
xM

This definition is not symmetrical, and it is known that left primitivity does not
imply right primitivity. Nonetheless, the attribute left is usually omitted. Note
that the annihilator I of any module M is always an ideal of R.
10

Lemma. Let M = Rx be a simple left R-module. Then I = AnnR (M ) is the

largest two-sided ideal contained in L = ann(x).

Proof. Note that this largest two-sided ideal K exists, since the sum of all two-sided
ideals contained in L is itself a two-sided ideal contained in L. Certainly I L, so
I K. Now KM = KRx Kx Lx = 0 since K is two-sided, so K I. 

Corollary. Every maximal ideal of R is primitive. Moreover, if R is commutative,

every primitive ideal is maximal.

2.5. Jacobson radical.

Definition. The Jacobson radical J(R) of R is defined to be the intersection of all

primitive ideals of R. The ring R is said to be semiprimitive if J(R) = 0.

It will be shown in Corollary 2.7 that this definition is left-right symmetric. Note
that J(R) is the set of elements of R which annihilate every simple left R-module.

Lemma. J(R) is equal to the intersection K of all maximal left ideals of R.

Proof. Let I be a maximal left ideal. Then P = AnnR (R/I) is primitive, so J(R)
P I by Lemma 2.4. Hence J(R) K.
Now let P = AnnR (M ) be a primitive ideal, where M is a simple R-module.
Note that P = 06=xM ann(x) is an intersection of maximal left ideals, so K P .
It follows that K J(R) as required. 

2.6. Nakayamas Lemma.

Lemma. Let M be a finitely generated nonzero left R-module and let J = J(R).
Then JM is strictly contained in M .

Proof. Since M is finitely generated, we can choose a cyclic quotient N of M , which

has a simple quotient M/K by Lemma 2.3. Then J.(M/K) = 0 so JM K which
is strictly contained in M . 

Note that the condition that M is finitely generated is necessary here: if p is a

prime, R = Z(p) and M = Q then J = pR and JM = pQ = M .

2.7. Recall that an element x R is a unit if there exists y R such that

xy = yx = 1.

Proposition.
J(R) = {x R : 1 axb is a unit for all a, b R} =: K.

Proof. Let x K, let I be a maximal left ideal of R and suppose that x

/ I. Since
I is maximal, I + Rx = R, so 1 ax I for some a R. Since x K, 1 ax
is a unit, a contradicting the fact that I is proper. Hence x I so K I for all
maximal left ideals I of R. By Lemma 2.5, K J(R).
 11

Now let x J(R). Since J(R) is a two-sided ideal, to show that x K its
sufficient to show 1 x is a unit. Now, if R(1 x) is a proper left ideal, we can find
a maximal left ideal L containing it by Lemma 2.3. By Lemma 2.5, x J(R) L
and 1 x L so 1 L, a contradiction. Hence there exists y R such that
y(1 x) = 1.
Now, 1 y = yx J(R), so by the above argument applied to 1 y, we can find
z R such that
z(1 (1 y)) = zy = 1.
Hence zy(1 x) = 1 x = z so zy = 1 and yz = 1, meaning that z = 1 x is a
unit. as required. 

This result shows that J(R) is the largest ideal A of R such that 1 A consists
entirely of units of R.

Corollary. The Jacobson radical is left-right symmetric. It follows that the inter-
section of all maximal left ideals of R is equal to the intersection of all maximal
right ideals.

2.8. Prime ideals.

Definition. The ideal P of R is said to be prime if P 6= R and whenever A, B are

ideals of R such that AB P , either A or B is contained in P . The ring R is
said to be prime if 0 is a prime ideal. The ideal N of R is nilpotent if there exists
n 0 such that N n = 0.

Note that if R is commutative, then P is prime if and only if R/P is an integral

domain, which agrees with the old definition.

Lemma. (1) Let P be prime and let N be nilpotent. Then N P .

(2) Any primitive ideal P = AnnR (M ) is prime.

Proof. (1) Since N n = 0 P , either N P or N n1 P . Continue.

(2) Suppose A, B / R are such that AB P . Then ABM = 0. If BM = 0 then
B P . Otherwise, BM = M since M is simple, so AM = 0 and A P . 

2.9. Minimal primes.

Definition. Let I / R. A prime P of R is a minimal prime over I if P I and

I Q P with Q prime forces Q = P . P is a minimal prime of R if it is a
minimal prime over 0.

Proposition. Let R be a left Noetherian ring and let I / R be a proper ideal. Then
(1) There exist primes P1 , . . . , Pn containing I such that P1 Pn I.
(2) The set of minimal primes over I is equal to the set of minimal primes in
{P1 , . . . , Pn }.
12

Proof. Suppose that (1) is false. Since R is left Noetherian, we can choose a maximal
counterexample I. Thus I contains no finite product of prime ideals containing I,
and I is maximal with respect to this property.
Claim: I is prime.
If I is not prime, we can find A, B / R such that AB I but A 6 I and B 6 I.
By maximality of I, I + A contains the product of primes P1 , . . . , Pn containing
I + A, and similarly Q1 Qm I + B for some primes Q1 , . . . , Qm containing
I + B. Hence
P1 Pn Q1 Qm (I + A)(I + B) I 2 + AI + IB + AB I,
so I itself contains a finite product of primes containing it. This contradicts the
definition of I, so in fact I is prime.
Thus we have a contradiction, and (1) follows.
Hence we have a finite set of primes P1 , . . . , Pn containing I such that P1 Pn
I. Let {X1 , . . . Xm } be the distinct minimal primes of {P1 , . . . , Pn }. Thus each Pj
contains some Xij so I contains some product of the Xk s, possibly with repetition:
Xi1 Xin P1 Pn I.
Now, suppose Q is any prime containing I. Then Xi1 Xi2 Xin I Q which
forces Xij Q for some j. If Q is a minimal prime over I, Q must equal Xij .
Finally, we show that each Xk is a minimal prime over I. If I Q Xk then
Xj Q Xk for some j by the above. But the Xs are minimal in {P1 , . . . , Pn },
so Xj = Q = Xk and (2) follows. 

2.10. Prime radical.

Definition. The prime radical N (R) of R is the intersection of all prime ideals of
R. R is semiprime if N (R) = 0. An ideal I / R is semiprime if its an intersection
of some collection of prime ideals, or equivalently, if N (R/I) = 0.

Note that by Lemma 2.8(1), N (R) contains every nilpotent ideal of R.

Proposition. Let R be a left Noetherian ring with nilradical N = N (R) and let
P1 , . . . , Pn be the minimal primes of R. Then
(1) N = P1 . . . Pn .
(2) N is nilpotent.

Proof. Part (1) is clear. By the proof of Proposition 2.9, 0 contains a product of k
of the Pi s, possibly with repetition, so N k = 0, as required for (2). 

2.11. When R is commutative, there is a nice characterisation of the nilradical.

Recall that an element x is nilpotent if xn = 0 for some n 0.

Proposition. Let R be a commutative ring. Then the set K of all nilpotent ele-
ments of R is an ideal and equals N (R).
 13

Proof. Let x, y K, so that xn = y m = 0 for some integers n, m 0. Clearly

(xy)n+m = 0, so xy K. Now, each term in the binomial expansion of (x + y)n+m
is a multiple of either xn or y m so (x + y)n+m = 0 and x + y K. Hence K is an
ideal.
Now, if x K, xR is nilpotent so xR N (R). Hence K N (R). Suppose
therefore that x / K. Then x is not nilpotent, so the set C = {1, x, x2 , . . .} doesnt
contain 0. Let S = {I / R : I C = }. Since 0 S, S is nonempty and is clearly
closed under unions. By Zorns Lemma, S has a maximal element P . We will show
that P is prime.
Suppose A, B / R are such that AB P but A, B are not contained in P . Then
x P +A and xm P +B for some integers n, m, so xn+m (P +A)(P +B) P ,
n

contradicting P S.
Hence x
/ K implies x / P for some prime ideal P , whence x / N (R). 

There is a similar characterisation for N (R) when R is arbitrary in terms of

strongly nilpotent elements - see McConnell and Robson.
14

3. Artinian rings

3.1. Recall that a right Artinian ring R is one which is satisfies the DCC as a
right module.
Proposition. Suppose S be a ring containing a right Artinian ring R such that S
is a finitely generated right R-module under right multiplication by R. Then S is
right Artinian.
Proof. This can be proved in exactly the same way as Proposition 1.7. In fact, the
Artinian analogues of Propositions 1.5 and 1.6 are also valid. 
Examples. The following rings are all right Artinian:
(1) The group algebra kG of a finite group G over a field k.
(2) The matrix ring Mn (D), where D is a division ring.
(3) Any finite ring, for example Z/mZ, m 6= 0.
(4) k[x]/(x2 ), k a field.
The structure of a right Artinian ring is quite well understood. The following is
a summary of whats known:
Theorem. Let J = J(R) be the Jacobson radical of the right Artinian ring R.
(1) (Artin-Wedderburn) R/J is isomorphic to a finite direct sum of matrix
rings over division rings D1 , . . . , Dk :
R/J
= Mn1 (D1 ) Mn2 (D2 ) . . . Mnk (Dk )
(2) J is nilpotent and J = N (R).
(3) (Hopkins) R is right Noetherian.
3.2. Semisimple modules.
Definition. A right R-module is said to be semisimple or completely reducible if
M is a direct sum of simple submodules.
Lemma. Let M be a semisimple R-module. The following are equivalent:
(1) M is Noetherian.
(2) M is Artinian.
(3) M is a direct sum of finitely many simple modules.
Proof. By Proposition 1.6 and its Artinian analogue, (3) implies (1) and (2). If M
is not a finite direct sum, we can find a submodule N of M which is a countably
infinite direct sum N = M1 M2 . . . of nonzero submodules Mi . Now the chains
M1 < M1 M2 < M1 M2 M3 < . . .
and
M1 M2 M3 > M2 M3 > . . .
show that M is not Noetherian nor Artinian. 
 15

3.3. Lemma. Any submodule N of a semisimple Artinian module M is itself

semisimple Artinian.

Proof. Let M = M1 M2 . . . Mn be a direct sum of simple submodules Mi .

Proceed by induction on n, the case n = 1 being obvious.
If N Mi = 0 for some i, then N , M1 . . . Mi1 Mi+1 . . . Mn , so N
is semisimple by induction. Hence we may assume that N Mi 6= 0 for all i. But
Mi is simple, so Mi N for all i, whence N = M is semisimple. 

3.4. Proposition. Suppose R is semiprimitive and right Artinian. Then the right
R-module RR is semisimple.

Proof. Let S consist of all finite intersections I1 . . . In of maximal right ideals

Ij . Since R is right Artinian, S has a minimal element L = I1 . . . In , say. Now
if I is a maximal right ideal, L I S, so L I = L by minimality of L. Hence
L I for all maximal I /r R, whence L J(R) = 0 by Lemma 2.5.
Now, the natural map R nj=1 (R/Ij ) has kernel I1 I2 . . . In = 0, so RR
is a submodule of the semisimple Artinian module nj=1 (R/Ij ). The result follows
from Lemma 3.3. 

Because of this result, a semiprimitive right Artinian ring is more commonly

known as semisimple Artinian. This is left-right symmetric by Wedderburns the-
orem - see comments in (3.7).

3.5. Some linear algebra.

Proposition. Let M = M1 M2 . . . Mn be a direct sum of right R-modules.

Then EndR (M ) = HomR (M, M ) is isomorphic to the matrix ring

Hom(M1 , M1 ) Hom(M2 , M1 ) Hom(Mn , M1 )
Hom(M1 , M2 ) Hom(M2 , M2 ) . . . Hom(Mn , M2 )


S=




Hom(M1 , Mn ) Hom(M2 , Mn ) Hom(Mn , Mn )
with the obvious multiplication.

Proof. This is best seen by writing elements of M as column vectors and thinking
of R-module endomorphisms acting by matrix multiplication on the left of these
column vectors.
Let j : Mj , M and i : M  Mi be the canonical injections and projections.
Formally, we can define a map : EndR (M ) S by setting the (i, j) element of
(f ) to be the composition
j f
Mj M M i Mi ;
16

thus (f )ij = i f j . We can also define : S EndR (M ) by

X
(T ) = j Tji i .
i,j

Check that and are mutually inverse ring homomorphisms. 

3.6. Schurs Lemma. Let M and N be simple right R-modules. Then

(1) EndR (M ) is a division ring.
(2) If M  N then HomR (M, N ) = 0.

Proof. Let : M N be a nonzero R-module homomorphism. Then ker() < M

and Im() > 0. The simplicity of M and N forces ker() = 0 and Im() = N , so
is an isomorphism.
Part (2) follows immediately, whereas if N = M , the above argument shows that
every nonzero element of EndR (M ) is left and right invertible. Therefore EndR (M )
is a division ring, proving (1). 

3.7. Artin-Wedderburn.
Theorem. Let R be a semiprimitive right Artinian ring. Then R is isomorphic to
a direct sum of matrix rings over some division rings D1 , . . . , Dk :
R
= Mn1 (D1 ) Mn2 (D2 ) . . . Mnk (Dk )
Proof. Consider : R HomR (RR , RR ) given by (x)(y) = xy. This is a ring
homomorphism. If (x) = 0, (x)(1) = x = 0, so is an injection. Also, if f : R
R is a right module map, then f (r) = f (1)r = (f (1))(r) so that f = (f (1)) and
is an isomorphism.
By Proposition 3.4, we can write R as a direct sum of simple right R-modules.
Grouping these together, we may write
RR n
= An1 1 An2 2 . . . Ak k
where the Ai are pairwise nonisomorphic simple modules. Applying Proposition
3.5, we obtain

EndR (An1 1 ) 0 0
0 EndR (An2 2 ) 0


R= EndR (RR )

=




nk
0 0 EndR
(Ak )
Mn1 (D1 ) 0 0
0 Mn2 (D2 ) 0




=




0 0 Mnk (Dk )
where each Di = EndR (Ai ) is a division ring, by Schurs Lemma (3.6). 
 17

As an exercise, the reader should check directly that a ring of the form
Mn1 (D1 ) Mn2 (D2 ) . . . Mnk (Dk )
is semisimple Artinian.
Note that if we were dealing with left R-modules, we would obtain that the op-
posite ring Rop is isomorphic to the direct sum of matrix rings. Since Mn (Dop )op
=
Mn (D) and the opposite ring of a division ring is also a division ring, the theorem
holds with right replaced by left. This justifies the term semisimple Artinian
ring, without reference to side.

3.8. Before we can prove Hopkins Theorem, we will need some results on semisim-
ple modules.

Proposition. Let R be an arbitrary ring and let M be an R-module. Suppose M

is a sum of simple submodules. Then M is semisimple.
P
Proof. Write M = A M for some simple submodules M of M . Let
X
S = {B A : the sum M is direct}.
B

Since S contains singletons {} for all A, S is nonempty. Its easy to check that
unions of chains in S are again in S, so by Zorns Lemma, S contains a maximal
element A0 .
Now, if X = A0 M 6= M , we can find M such that M ( X. Hence
M X = 0 and the sum X + M is direct. Therefore A0 {} S, contradicting
the maximality of A0 . Hence X = M is a direct sum of simple modules, so M is
semisimple. 

3.9. Lemma. Suppose the ring R is semisimple Artinian. Then every R-module
is semisimple.

Proof. R is a sum of simple modules by Proposition 3.4. Any cyclic module is a

homomorphic image of R and hence is a sum of simple modules. Now, an arbitrary
module is a sum of cyclic submodules, and hence also sum of simple modules. The
result follows, again from Proposition 3.8. 

3.10. Proposition. The Jacobson radical J of a right Artinian ring R is nilpotent.

Proof. The descending chain J J 2 J 3 . . . must terminate since R is right

Artinian. Hence J n = J n+1 = . . . for some n 0. Let X = lann(J n ) = {x R :
xJ n = 0}, this is a two-sided ideal of R.
Suppose for a contradiction that X < R. Then R/X has a minimal nonzero sub-
module Y /X, being Artinian. Its easy to see that Y /X is simple. Now (Y /X).J = 0
so Y J X. It follows that Y J n = Y J n+1 XJ n = 0, so Y lann(J n ) = X,
contradicting Y /X 6= 0.
Hence X = R so J n = RJ n = XJ n = 0 as required. 
18

Corollary. If R is right Artinian, then J(R) = N (R).

Proof. Since every primitive ideal is prime by Lemma 2.8(2), N (R) J(R) in
any ring. On the other hand if R is right Artinian then J(R) is nilpotent by
Proposition 3.10, so J(R) P for any prime ideal P by Lemma 2.8(1). Hence
J(R) N (R). 

3.11. Hopkins Theorem. Let R be a right Artinian ring. Then R is right Noe-
therian.

Proof. By Proposition 3.10, J n = 0 where J is the Jacobson radical of R. Now,

each J m /J m+1 is an R/J-module and R/J is semisimple Artinian. By Lemma 3.9,
each J m /J m+1 is a semisimple Artinian R/J-module, and as such is Noetherian by
Lemma 3.2. The result follows from Proposition 1.5. 

Its tempting to think that the left-right symmetry is so strong that every right
Artinian ring is left Artinian. This is not the case, however.
 19

4. Commutative Rings

R is commutative with 1 throughout this chapter.

4.1. The Nullstellensatz. We saw in Corollary 3.10 that if R is right Artinian,

then N (R) = J(R).

Theorem. Let k be a field and R be a finitely generated kalgebra (equivalently,

a quotient of the polynomial ring k[X1 , . . . , Xn ]). Then

N (R) = J(R).

This is an unusual way of stating the Nullstellensatz, see (4.3).

4.2. Weak Nullstellensatz.

Theorem. Suppose k is a field and R is a finitely generated kalgebra which is

also a field. Then
(1) R is algebraic over k.
(2) If k is algebraically closed, then R
= k.

Corollary. If k = k then the maximal ideals of k[X1 , . . . , Xn ] are of the form

ma = (X1 a1 , . . . , Xn an ) where a = (a1 , . . . , an ) k n .

Proof. Its clear that each ma is a maximal ideal, being of codimension 1. Now if I
is maximal, then the R/I = k so Xi + I 7 ai for some ai k. Hence Xi ai I for
all i = 1, . . . , n so ma I. Since the former ideal is maximal, the result follows. 

4.3. Geometric viewpoint. Recall that by Proposition 2.11, I / R is semiprime

if and only if f n I f I.

Definition. If R is commutative, an ideal is radical if it is semiprime.

In the first lecture, we defined an algebraic set to be a subset of k n of the form

Z(A) = {(a1 , . . . , an ) k n : f (a1 , . . . , an ) = 0 for all f A}

where A is an ideal of k[X1 , . . . , Xn ]. We also defined, for X k n

I(X) = {f k[X1 , . . . , Xn ] : f (a1 , . . . , an ) = 0 for all (a1 , . . . , an ) X}

this is an ideal in k[X1 , . . . , Xn ].

Its easy to check that Z(I(X)) = X for any algebraic set X. The Nullstellensatz
is usually stated as follows:

Theorem. Suppose k = k. If A is a radical ideal of k[X1 , . . . , Xn ], then

I(Z(A)) = A.

So we have a bijection between radical ideals and algebraic sets.

20

Proof. f I(Z(A)) f (a) = 0 for all a Z(A) f ma for all a Z(A)

\ \
f ma = ma
aZ(A) ma A

since a Z(A) iff A(a) = 0 iff A ma .

By Corollary 4.2, all maximal ideals of R = k[X1 , . . . , Xn ]/A have the form
ma /A for some a k n , so f + A J(R) = 0 by Theorem 4.1 and f A.
Finally, if f A then f (Z(A)) = 0 so f I(Z(A)). 

4.4. Proof of Theorem 4.2. We will only deal with the case when k is an un-
countable field in this course. For the general case, see Atiyah and Macdonald, (7.8)
and (7.9).
Suppose R is not algebraic. Then we can find R such that k() = k(t),
the field of fractions of the polynomial ring k[t]. Choose an uncountable subset
{ci : i I} of k and suppose 1 , . . . , m k are such that
1 2 m
+ + ... + = 0.
ci1 ci2 cim
Clearing denominators gives
1 ( ci2 )( ci3 ) ( cim ) + . . . + m ( ci1 )( ci2 ) ( cim1 ) = 0.
Substituting 1 = ci1 into this equation gives 1 (ci1 ci2 )(ci1 ci3 ) (ci1 cim ) = 0
so 1 = 0 and similarly j = 0 for all j = 1, . . . , m.
1
Hence { c i
k() : i I} is an uncountable linearly independent subset of
R. But R is a quotient of the polynomial ring k[X1 , . . . , Xn ] and as such has a
countable spanning set, consisting of the images of the monomials in R. This is a
contradiction, proving (1).
Now, a finitely generated field extension of k which is algebraic must necessarily
be finite dimensional. If k = k then R is a finite field extension of k so R = k as
required for (2).

4.5. Lemma. Let k be a field and let R be a finite dimensional kalgebra which
is a domain. Then R is a field.

Proof. If 0 6= x R then multiplication by x is an injective klinear map from R

to itself. Since dimk R < this map must be surjective. Hence there exists y R
such that xy = 1 and the result follows. 

4.6. Proof of Theorem 4.1. As we have observed in the proof of Corollary 3.10,
N (R) J(R) for any ring R. Suppose a / N (R) so a is not nilpotent. By Sheet 1
Exercise 6(2), 1 ax is not a unit in the polynomial ring R[x]. By Lemma 2.3 we
can find a maximal ideal M of R[x] containing 1 ax.
Since R is a finitely generated kalgebra, so is R[x]/M . By Theorem 4.2, R[x]/M
is algebraic over k and is hence finite dimensional over k. Now, we have a klinear
injection R/(R M ) , R[x]/M , so R/(R M ) is a finite dimensional kalgebra
 21

which is a domain. Hence R M is a maximal ideal of R by Lemma 4.5 and

a
/ R M as otherwise 1 M . Hence a
/ J(R) as required.

4.7. Associated primes. From now on well be dealing with a commutative Noe-
therian ring R and finitely generated Rmodules M .

Definition. A prime P of R is said to be an associated prime of M if there exists

x M such that ann(x) = P , or equivalently, if R/P , M . The set of all
associated primes of M is denoted by Ass(M ).

Example. Let R = Z and let M be a finite abelian group. Then M has an element
of order p for all primes p dividing |M |, so

Ass(M ) = {pZ : p | |M |}.

4.8. Lemma. If M 6= 0 then Ass(M ) 6= .

Proof. Let S = {ann(x) / R : 0 6= x M }. Since R is Noetherian, we can pick a

maximal element P = ann(x) S. Suppose a / P and b
/ P but ab P . Then
bx 6= 0 so ann(bx) S. Since P ann(bx), P = ann(bx) by maximality of P . Now
abx = 0 so a P , a contradiction. Hence P is prime and so P Ass(M ). 

4.9. Proposition. Suppose M 6= 0.

(1) If P / R is a prime then Ass(R/P ) = {P }.
(2) Let N be a submodule of M . Then

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

Proof. Let L = R/P and 0 6= y L. Clearly, P ann(y). Now if ry = 0 then

r P since R/P is a domain. Hence P = ann(y) and (1) follows.
Next, the first inclusion of (2) is obvious. Let P = ann(x) Ass(M ). If
Rx N = 0 then R/P = Rx , M/N and P Ass(M/N ), so suppose Rx N 6= 0.
Choose y Rx N ; then by the above ann(y) = P so P Ass(N ) as required. 

4.10. Theorem. Suppose M 6= 0. Then

(1) There exists a chain of submodules

0 = M0 < M1 < . . . < Mn = M

of M such that Mi /Mi1

= R/Pi for some prime ideals P1 , . . . , Pn of R.
(2) Ass(M ) is finite.

Proof. (1) By Lemma 4.8, Ass(M ) 6= . Let M1 = x1 R for some x1 M such

that ann(x1 ) = P1 Ass(M ). Then M1 /M0 = R/P1 . If M1 = M we are done.
Otherwise, M/M1 is nonzero so we can find P2 = ann(x2 ) Ass(M/M1 ) for some
x2 M/M1 . Let M2 be defined by M2 /M1 = x2 R = R/P2 . This process must
stop since M is Noetherian.
22

(2) By Proposition 4.9,

Ass(M ) Ass(M1 ) Ass(M2 /M1 ) . . . Ass(Mn /Mn1 ) = {P1 , . . . , Pn }

so Ass(M ) is finite. 

This result shows that any finitely generated Rmodule M can be thought of
as being built up from finitely many modules of the form R/Pi for some primes Pi .

4.11. Proposition. Every minimal prime of R is in Ass(R).

Proof. Let {P1 , . . . , Pn } be the distinct minimal primes of R. By Proposition 2.9,

we can find integers s1 , . . . , sn 1 such that

P1s1 P2s2 Pnsn = 0.

Its enough to show that P1 Ass(R). Let M = P2s2 Pnsn if n > 1 and M = R
otherwise.
By Theorem 4.10(1) we can choose a chain

0 = M0 < M1 < . . . < Mt = M

where Mi /Mi1 = Qi for some primes Q1 , . . . , Qt . Note that P1s1 kills each
Mi /Mi1 so P1s1 Qi and hence P1 Qi for all i = 1, . . . , t by the primality
of Qi .
Also, Q1 Q2 Qt M = 0 implies Q1 Q2 Qt P2s2 Pnsn P1 . Since P1 is prime
and Pi ( P1 for all i 2, Qk P1 for some k, so in fact Qk = P1 for some k.
Pick k 1 least such that Qk P1 . Hence if k > 1 then Q1 Qk1 ( P1 . If
k = 1 let r = 1 and if k > 1 choose r Q1 Qk1 \P1 . Since Ann(Mk /Mk1 ) =
Qk = P1 and r / P1 we can find x Mk such that rx / Mk1 .
Claim: P1 = ann(rx). Note that rP1 x Q1 Q2 Qk1 Qk x = 0 so P1
ann(rx). If srx = 0 then s(rx + Mk1 ) = 0. We chose x so that rx / Mk1 , and
hence s P1 because Mk /Mk1 = R/Q k =
R/P 1 is a domain. Hence ann(rx) =
P1 Ass(M ) as required. 

4.12. Primary ideals.

Definition. Let P be a prime. An ideal Q / R is P primary if Ass(R/Q) = {P },

and primary if its P -primary for some prime P .

One should think of primary ideals as generalisations of prime powers in Z.

Lemma. (1) If Q is P primary, then P = Q.

(2) If P is maximal and Q / R has P = Q, then Q is P primary.

Proof. Since any minimal prime over Q must be in Ass(R/Q) by Proposition 4.11,
we see that R/Q has precisely one minimal prime, namely P/Q. Hence N (R/Q) =

Q/Q = P/Q so P = Q.
 23

For (2), note that P/Q is the prime radical of R/Q and hence is the only prime
ideal of R/Q, being maximal. Hence Ass(R/Q) must be {P }, being nonempty by
Lemma 4.8, so Q is P primary. 

Examples. (1) The primary ideals of Z are the prime powers and 0.

(2) Let R = k[x, y], Q = (x, y 2 ). Then Q = (x, y) =: P is maximal so Q is
primary. However, Q is not a prime power since P 2 ( Q ( P .

Warning: not all prime powers are primary!

4.13. Primary decomposition.

Definition. A primary decomposition of an ideal I /R is an expression of the form

I = Q1 Q2 . . . Qk
for some primary ideals Qi . The decomposition is minimal if
The primes Pi s are pairwise distinct, where Qi is Pi primary
T
For all i, Qi ) j6=i Qj .

Example. In Z, a nonzero ideal I has a minimal primary decomposition of the

form
I = (ps11 ) (ps22 ) (psnn )
where I = (x) and x = ps11 psnn is a factorization of x into prime powers.

4.14. Theorem.
(1) (Existence) Every proper ideal has a minimal primary decomposition.
(2) (Uniqueness) Let I = Q1 . . . Qk be a minimal primary decomposition
of I, with Qi a Pi primary ideal. Then
Ass(R/I) = {P1 , . . . , Pn }.

Proof. Omitted. 
24

5. Commutative localisation

R is a commutative ring throughout this chapter.

5.1. Motivation. Let S be a subset of R which is multiplicatively closed (m.c. for

short). We will stick to the convention that any such S necessarily contains 1.
We will construct a ring RS , called the localisation of R at S by inverting the
elements of S - thus the elements of S are units in RS .
This generalises the process of constructing the field of fractions of a commutative
integral domain R - there the set S = R\{0} of nonzero elements is m.c., S consists
of units in the field of fractions F of R, and moreover every element of F can be
written in the form rs1 for some r R and s S.
What if S contains zero divisors? If we want to invert S, then the elements
of R which kill an element of S must be zero in RS because ab = 0 and bc = 1
together imply a = abc = 0.
Definition. The assassinator of S is defined to be
[
ass(S) = {x R : xs = 0 for some s S} = ann(s).
sS

Lemma. ass(S) is an ideal of R.

Proof. Let x, y ass(S). So there exist s, t S such that xs = yt = 0. Then
(x + y)st = 0 and st S so x + y ass(S). Also, if z R then (zx)s = 0 so
zx ass(S). 

5.2. Commutative localisation.

Definition. Let S R be a m.c. subset. A localisation of R at S is a ring RS
together with a homomorphism : R RS such that
(a) (s) is a unit in RS for all s S.
(b) Every element of RS can be written in the form (r)(s)1 for some r R
and s S.
(c) ker = ass(S).
Note that is not in general injective.
Lemma. If RS exists, its unique upto isomorphism.
Proof. Suppose : R T is a ring homomorphism satisfying properties (a),(b),(c).
Define : RS T by ((r)(s)1 ) = (r)(s)1 . If (r)(s)1 = (u)(v)1
then rv us ker() = ass(S) so there exists t S such that (rv us)t = 0. Hence
rvt = ust so (r)(v)(t) = (u)(s)(t) and (r)(s)1 = (u)(v)1 . Therefore
is well defined and is easily checked to be a ring homomorphism.
The conditions on force to be onto, whereas if (r)(s)1 = 0 then r
ker() = ass(S) = ker() so (r)(s)1 = 0 and is injective. Hence is a ring
isomorphism. 
 25

5.3. Construction.

Definition. An element x R is regular if ann(x) = 0. Equivalently, x is not a

zero divisor.

Assume first that S consists of regular elements, so that ass(S) = 0.

Define a relation on R S by setting

(r, s) (u, v)

if and only if
rv = us.
This is clearly reflexive and symmetric. If (r, s) (u, v) and (u, v) (a, b) then
rv = us and ub = av, so (rb as)v = rvb asv = ubs avs = 0 so rb as
ass(S) = 0 and (r, s) (a, b). Hence is an equivalence relation.
Let RS = R S/ as a set. Write r/s for the equivalence class of (r, s) in RS
and define addition and multiplication on RS by
r/s + u/v = (rv + us)/(sv)
r/s u/v = (ru)/(sv)

Exercise. Check the following:

(1) Addition is well defined.
(2) Multiplication is well defined.
(3) RS forms a commutative ring with zero element 0/1 and identity 1/1.
(4) The map : R RS given by (r) = r/1 is a ring homomorphism.
(5) Conditions (a), (b) and (c) in Definition 5.2 hold.

We check that addition is well-defined: Suppose r/s = r0 /s0 so rs0 = r0 s. Now,

r/s + u/v = (rv + us)/(sv) and (r0 /s0 ) + (u/v) = (r0 v + us0 )/(s0 v),

but (rv + us)(s0 v) = (rs0 )v 2 + uss0 v = (r0 s)v 2 + us0 sv = (r0 v + us0 )(sv), so

r/s + u/v = r0 /s0 + u/v = r0 /s0 + u0 /v 0

if u/v = u0 /v 0 . Hence addition is well-defined.

Hence, the localisation at S exists and is unique.

5.4. The general case. If ass(S) is not necessarily zero, let R = R/ ass(S) and
let S denote the image of S in R. Then ass(S) = 0 so we can form the localisation
RS , together with a ring homomorphism : R RS as constructed above. Let

: R RS

be defined by = where is the natural projection of R onto R. So we have

Theorem. The localisation RS exists.

26

Proof. Since ass(S) = 0, ker() = 0 so ker() = ass(S). Thus : R RS satisfies

all the conditions of Definition 5.2. 

Note that (r)(s)1 = (u)(v)1 iff (rv us) = 0 iff rvt = ust for some
t S. So we could have constructed RS by imposing the equivalence relation on
R S instead, where

(r, s) (u, v) if and only if rvt = ust for some t S.

5.5. Examples.
(1) If R is an integral domain then S = R\{0} is a m.s. set. Then RS is just
the field of fractions of R. This is a special case of (2).
(2) If R is arbitrary and P is a prime ideal, then S = R\P is m.c. The
localisation RS is usually denoted by RP and is called

the localisation at the prime ideal P.

The ring Z(p) from Example 1.8 (4) is a special case of this construction.
(3) For any x R, the set {1, x, x2 , . . .} is m.c., and the localisation is usually
denoted by Rx .
(4) If 0 S then ass(S) = R so RS = 0.
(5) If I / R then S = 1 + I is m.c.
Of these, Example (2) is the most important.

5.6. Localisation of modules.

Definition. Let S be a m.c. subset of R and let M be an Rmodule. The locali-

sation of M at S is defined to be the set of equivalence classes

MS = {m/s : m M, s S}

in M S under the equivalence relation given by

(m, s) (n, t) if and only if mtu = nsu for some u S.

This is an RS module, with the addition and RS action given by

m/s + n/t = (mt + ns)/(st)
m/s r/u = (mr)/(su)
for m/s, n/t MS and r/u RS . The S-torsion submodule of M is defined to be

assM (S) = {m M : ms = 0 for some s S}.

Exercise. Check that

(1) The operations are well defined and turn MS into an RS module.
(2) N = assM (S) is a submodule of M .
(3) assM/N (S) = 0.
(4) The map : M MS given by (m) = m/1 has kernel precisely assM (S).
 27

(5) MS = 0 if and only if assM (S) = M .

In fact, MS is naturally isomorphic to M R RS .

5.7. Exact sequences.

Definition. Let A, B, C be Rmodules and suppose we have Rmodule maps

f : A B and g : B C. The sequence
f g
ABC

is said to be exact at B if Im(f ) = ker(g). The sequence

0 f g 0
0ABC0

is called a short exact sequence if it is exact at A, B and C, or equivalently, if f

is injective, Im(f ) = ker(g) and g is surjective.

Note that whenever N is a submodule of N , we have a natural short exact

sequence

0 N M M/N 0
where is the inclusion of N into M and is the quotient map.

5.8. Exactness of localisation. One of the most important properties of locali-

sation is that it is exact, meaning that it takes exact sequences to exact sequences.
Suppose S is a m.c. set in R and f : M N is a map of Rmodules. Then we
can define a map fS : MS NS by setting

fS (m/s) = f (m)/s for all m M, s S.

Exercise. Check that

(1) fS is well defined.
(2) fS is an RS module homomorphism.

Proposition. Localisation is exact. In other words, whenever

f g
ABC

is a short exact sequence of Rmodules,

fS gS
AS BS CS

is a short exact sequence RS modules.

Proof. Since Im(f ) = ker(g), g f = 0. Hence

gS (fS (a/s)) = gS (f (a)/s) = g(f (a))/s = 0

for any a A and s S, so gS fS = 0 and Im(fS ) ker(gS ).

28

Suppose b/s ker(gS ). Then gS (b/s) = g(b)/s = 0 so there exists t S such

that g(b)t = g(bt) = 0. Hence bt ker(g) = Im(f ) since the first sequence is exact,
so we can find a A such that bt = f (a). Now
fS (a/(st)) = f (a)/(st) = bt/(st) = b/s
so b/s Im(fS ) and Im(fS ) = ker(gS ) as required. 

5.9. Localisation at a prime ideal. Let P be a prime ideal of R and set S =

R\P . A commutative ring is called local if it has a unique maximal ideal.
Proposition. (1) ass(S) P .
(2) RP is a local ring with maximal ideal PP .
Proof. If s
/ P and xs = 0 then xs P which is prime so x P , proving (1).
The natural short exact sequence
0 P R R/P 0
gives rise to the short exact sequence of RP modules
0 PP RP (R/P )P 0
by Proposition 5.8. This means that PP is an ideal of RP with quotient isomorphic
to (R/P )P .
If x/s RP \PP then x / P so x S and x/s is a unit in RP . Conversely, if x/s
is a unit then x/1 is a unit also, so x/1.y/t = 1 for some y/t RP . Hence xyu = t
for some t S, meaning that x / P.
Hence PP is the set of all nonunits of RP . If y + PP RP /PP is nonzero then
y / PP so y is invertible in RP and hence in RP /PP . Hence RP /PP is a field so
PP is maximal. Moreover, any maximal ideal consists of nonunits and must hence
be contained in PP , proving (2). 
Exercise. Let F denote the field of fractions of R = R/P . Check that there is
a well defined ring isomorphism : (R/P )P F given by (r/s) = r/s. Hence
RP /PP = F.
5.10. Theorem. Suppose R is Noetherian and S is a m.c. subset. Then RS is
Noetherian.

Proof. Since R = R/ ass(S) is Noetherian by Proposition 1.5 and RS = RS , we

may assume that ass(S) = 0. Hence R , RS .
Now, if I / RS then (I R)RS I. If x/s I then x = x/s.s I R so
x/s = x.(1/s) (I R)RS . Hence I = (I R)RS for any I / RS .
Suppose I1 I2 . . . is an increasing chain of ideals in RS . Since R is Noe-
therian, the chain I1 R I2 R . . . terminates, so there exists k such that
Ik R = Im R for all m k. But then Ik = (Ik R)RS = (Im R)RS = Im for
all m k, so RS is Noetherian as required. 
 29

6. Noncommutative localisation

Now we drop the requirement that R be commutative and try to generalize the
results established so far about localisation.

6.1. We start with some definitions.

Definition. Let S be a m.c. subset of R. Its assassinator is defined to be

[
ass(S) = {x R : xs = 0 for some s S} = lann(s).
sS

As before, ass(S) has to be zero in any ring in which S consists of units, but we
have to be careful about what kind of annihilators we take.

Definition. Let S R be a m.c. subset. A right localisation of R at S is a ring

RS together with a homomorphism : R RS such that
(a) (s) is a unit in RS for all s S.
(b) Every element of RS can be written in the form (r)(s)1 for some r R
and s S.
(c) ker = ass(S).

We are interested in right fractions, i.e. those of the form rs1 . Hence the term
right localisation. One can play this game on the left, too. Note that if RS exists,
then ass(S) has to be a two-sided ideal - this is not true in general.

6.2. The Ore condition. Suppose a right localisation RS of R at S exists. Then

for all r R and s S, the element (s)1 (r) must lie in RS , so by condition (2)
we can find a R and b S such that

(s)1 (r) = (a)(b)1 .

Hence (rb sa) = 0 so there exists t S such that r(bt) = s(at). Note that bt S
as S is m.c.

Definition. A m.c. set S is a right Ore set, or satisfies the right Ore condition if
and only if

for all r R, s S there exist r0 R, s0 R such that rs0 = sr0 .

This is a condition on the m.c. subset S which is automatically satisfied in

commutative rings. The above discussion shows that if the right localisation RS
exists, then S must be a right Ore set.

Lemma. If S is a right Ore set, then ass(S) is a two-sided ideal of R.

Proof. Let x, y ass(S) be such that xs = yt = 0 for some s, t S. The right

Ore condition gives s0 R and t0 S such that st0 = ts0 . Now (x + y)(st0 ) =
(xs)t0 + (yt)s0 = 0 and st0 S as S is m.c. Hence x + y ass(S).
30

If z R then (zx)s = 0 so zx ass S. Also the Ore condition gives z 0 S

and s00 S such that zs00 = sz 0 . Hence (xz)s00 = (xs)z 0 = 0 and s00 S so
xz ass S. 

6.3. Ores Theorem. We must first refine Definition 5.3.

Definition. An element x R is said to be left regular if lann(x) = 0 and right

regular if rann(x) = 0. x R is regular if its both right and left regular, and thus
not a zero-divisor.

Assuming S is a right Ore set, we may factor out by the two-sided ideal ass(S)
and obtain a ring R such that the image S of S in R consists of left regular elements.
If the right localisation RS exists, then R is a subring of RS and (S) consists of
units. Hence every element of S must actually be regular (and not just left regular).
This is another necessary condition on the set S for RS to exist.

Theorem (Ore, 1930). Let S be a m.c. subset of the ring R. Then the right
localisation RS exists if and only if
(1) S is a right Ore set, and
(2) S consists of regular elements in R = R/ ass(S).

Proof. We have shown in the above discussions the necessity of these conditions.
Assuming (1) and (2), it remains to construct the ring RS satisfying the conditions
of Definition 6.1.
By Lemma 6.2, ass(S) is a two-sided ideal in R. Its easy to check that the
image S also satisfies the right Ore condition. As before, by passing to the quotient
R = R/ ass S and using condition (2), we may assume that S is a right Ore set in
R consisting of regular elements.
We can now construct RS as a set of equivalence classes in R S under a
certain equivalence relation, in a very similar spirit to the construction of RS in the
commutative case (5.3). The construction is very tedious and is non-examinable.
See handout. 

Definition. A m.c. set S is said to be a right divisor set if conditions (1), (2) are
satisfied.

Note that if R is commutative, or slightly more generally, if S consists of central

elements (i.e. sr = rs for all s S, r R), then the two conditions are auto-
matically satisfied and the localisation RS exists. One can view this Theorem as a
generalisation of Theorem 5.4.

6.4. Ore domains.

Definition. A ring R is a domain if all nonzero elements are regular, or equiva-

lently if it has no zero divisors. R is a right Ore domain if S = R\0 is a right Ore
set in R.
 31

By Theorem 6.3, its clear that if R is a right Ore domain, then the localisation
RS exists. Moreover, every nonzero element r/s RS is invertible, so RS is a
division ring, known as the division ring of fractions of R.

Theorem. Let R be a right Noetherian domain. Then R is an Ore domain and

hence R has a division ring of fractions.

Proof. Let x, y S and suppose that xR yR = 0. Then xR yR , R. Now,

x S is regular so xR
= R as right Rmodules. Hence
x(xR yR) = x2 R xyR , xR,
so x2 R xyR yR , R. Repeating this trick, we see that
xn R xn1 yR xn2 yR xyR yR , R
for any n 1. Hence the sum yR + xyR + x2 yR + . . . is direct, contradicting the
fact that R is right Noetherian.
Hence we can find a, b R such that xa = yb 6= 0. Since x, y are regular, a, b
are nonzero so S is a right Ore set. 

This proof might be reminiscent of Sheet 1, Exercise 12.

Examples. Let k be a field. The following rings are all right Ore domains and
hence have division rings of fractions:
(1) The Weyl algebras An (k).
(2) kG, where G is a torsionfree polycyclic group.
(3) U(g) where g is a f.d. Lie algebra over k.
(4) The Iwasawa algebra G if G is a torsionfree compact padic Lie group.

6.5. Modules.

Definition. Let S be a right divisor set in R and let M be a right Rmodule. The
localisation of M at S is defined to be the set of equivalence classes
MS = {m/s : m M, s S}
in M S under the equivalence relation given by
(m, s) (n, t) if and only if mt0 u = ns0 u for some u S,
where s0 , t0 R are such that st0 = ts0 S. The S-torsion submodule of M is
defined to be
assM (S) = {m M : ms = 0 for some s S}.

Proposition. (1) MS has the structure of a right RS module.

(2) N = assM (S) is an Rsubmodule of M .
(3) assM/N (S) = 0.
(4) The map : M MS given by (m) = m/1 has kernel precisely assM (S).
32

(5) MS = 0 if and only if assM (S) = M .

(6) Localisation is exact.

Proof. Exercise - mimic proofs in (5.6), (5.8) and the handout. 

Again, an alternative way of constructing MS is given by the isomorphism

MS
= M R RS .
6.6. Goldies Theorem.

Definition. The classical right ring of quotients Q(R) is the localisation of R at

the m.c. subset S consisting of all regular elements of R, if it exists.

Theorem (Goldie, 1958, 1960). The ring R has a classical right ring of quotients
which is semisimple Artinian if and only if
(1) N (R) = 0,
(2) R contains no infinite direct sum of right ideals, and
(3) R satisfies the ACC on right annihilators.

Corollary. If R is semiprime Noetherian, then Q(R) exists and is isomorphic to

Mn1 (D1 ) Mn2 (D2 ) . . . Mnk (Dk )
for some division rings D1 , . . . , Dk .

Proof. Follows directly from Theorem 3.7 and Theorem 6.6. 

We will only prove (), the other direction being easier and less interesting. For
the remainder of this chapter, assume that R is a ring satisfying conditions (1), (2)
and (3) of Theorem 6.6. Let S denote the set of all regular elements of R.

6.7. Essential right ideals.

Definition. Let M be an Rmodule. Then N M is essential if N K 6= 0 for

any nonzero K M .

There is a connection between right regular elements and essential right ideals.
Let a = rann(a) for any a R. Note that x = 0 iff x is right regular and x = R
iff x = 0.

Proposition. T /r R is essential if and only if T S 6= .

Proof. () Suppose x T is right regular. Its sufficient to prove that xR is

essential since xR T .
Suppose 0 6= I /r R and I xR = 0. Consider the sum
L = I + xI + x2 I + . . .
Pk i k
If i=0 x ri L, then r0 = xr1 . . . x rk I xR = 0 so r0 = 0 and
xr1 + x2 r2 + . . . + xk rk = 0. Since x = 0, r1 + xr2 + . . . + xk1 rk = 0. Continuing
 33

like this, r0 = r1 = . . . = rk = 0 so the sum is direct, contradicting (2). Hence xR

is essential.
() See (6.11). 

6.8. Proof of Goldies Theorem - Step 1: RS exists.

Proof. By Theorem 6.3 its sufficient to show that S is a right Ore set. Let r R
and s S; since s is regular, sR is essential in R by Proposition 6.7(). Let
T = {a R : ra sR}.
T is a right ideal in R; well show that T is essential. Suppose I /r R is nonzero. If
rI = 0 then I T so I T 6= 0. Otherwise sR rI 6= 0 since sR is essential. If
0 6= rx sR rI then 0 6= x I T so I T 6= 0 as claimed.
By Proposition 6.7(), we can find a T S. Then ra = sb for some b R
and S is a right Ore set. 

6.9. Lemma. Let x R be right regular. Then x is regular.

Proof. Suppose yx = 0 but y 6= 0. Then y 6= R. Now xR y so y is essential

by Proposition 6.7(). Let z ( R be a maximal annihilator containing y - this
exists by assumption (3).
Now, if zRz = 0 then (RzR)2 = 0, so z = 0 by assumption (1) and z = R, a
contradiction. Hence there exists b R such that zbz 6= 0.
Hence z (zbz) 6= R so z = (zbz) by maximality of z . If u bzR z then
u = bzv and zu = 0 so zbzv = 0. Hence v z and u = bzv = 0. This shows that
bzR z = 0 and bz 6= 0, so z is not essential, a contradiction.
Hence y = 0 and x is regular. 

6.10. Nil right ideals.

Definition. X R is nil if x is nilpotent for all x X.

Lemma. Let J /r R.
(a) If J is nil then J = 0.
(b) If J 6= 0 then there exists y J such that yR y = 0.

Proof. (a) Suppose for a contradiction that there exists nonzero a J. Then
aR J is nil, and hence Ra is also nil, because (ax)n = 0 (xa)n+1 = 0.
Let S = {b : 0 6= b Ra}. By (3), we can pick 0 6= b Ra so that b is maximal
in S. Since N (R) = 0 by (1), bRb 6= 0, so we can find x R such that bxb 6= 0.
Since Ra is nil and 0 6= xb Ra, there exists m 2 such that (xb)m1 6= 0 but
(xb)m = 0. Now ((xb)m1 ) S and b ((xb)m1 ) as by = 0 (xb)m1 y = 0.
By maximality of b , xb ((xb)m1 ) = b so bxb = 0, a contradiction.
(b) By (a), J is not nil. Choose z J which is not nilpotent. Now, the chain
z (z 2 ) (z 3 )
34

stops by (1), so (z n ) = (z n+1 ) = . . . for some n. Let y = z n so that (y 2 ) = y .

Note that y 6= 0 since z is not nilpotent.
Suppose that u = yv yR y . Then yu = y 2 v = 0 so v (y 2 ) = y and
u = yv = 0. Hence yR y = 0 as required. 

6.11. Proof of Proposition 6.7().

Proof. Suppose for a contradiction that T S = . We will construct an infinite

direct sum inside T . More precisely, we will construct a sequence x1 , x2 , . . . T
such that for all m 1
Im 6= 0 where Im := {x1 , . . . , xm } .
The sum x1 R + x2 R + . . . + xm R + (Im T ) is direct.
Proceed by induction on m.
Base case m = 1: Pick 0 6= x1 T such that x1 R x1 = 0 using Lemma 6.10(b).
Since x1 is not regular, its not right regular by Lemma 6.9, so I1 = x1 6= 0. The
sum x1 R + x1 is direct by construction, so x1 R + (I1 T ) is also direct.
Inductive step: Since Im 6= 0 by induction and T is essential, Im T 6= 0. Choose
xm+1 Im T such that xm+1 xm+1 R = 0 using Lemma 6.10(b).
Note that since xm+1 Im T , the sum
x1 R + x2 R + . . . + xm+1 R
is direct, being contained in x1 R + x2 R . . . + (Im T ) which is direct by induction.
Hence Im+1 = {x1 , . . . , xm+1 } = (x1 + . . . + xm+1 ) . Since x1 + . . . + xm+1 T
and T contains no regular elements, Im+1 6= 0.
It remains to show that x1 R + x2 R + . . . + xm+1 R + (Im+1 T ) is direct. Now,
xm+1 R Im T
by construction, so if Z = x1 R + . . . + xm R then
(Z + xm+1 R) (Im T ) = (Z Im T ) + xm+1 R = xm+1 R
by the modular law, because Z Im = 0 by induction. Since Im+1 = Im xm+1 ,
(Z + xm+1 R) (Im+1 T ) = (Z + xm+1 R) (Im T ) xm+1
xm+1 R xm+1 = 0,
as required. Thus the sum x1 R + x2 R + . . . is direct contradicting (2), so T must
contain a regular element. 

6.12. Closed right ideals.

Definition. J /r R is closed if for any I /r R such that I * J there exists 0 6= L/r R

with L I and L J = 0.

Draw a picture!

Lemma. R has the minimum condition on closed right ideals.

 35

Proof. Suppose J1 ) J2 ) J3 ) . . . is a strictly decreasing chain of closed right

ideals. Since J1 * J2 , there exists 0 6= L1 /r R such that L1 J1 and L1 J2 = 0.
Similarly, since Jn * Jn+1 , there exists 0 6= Ln /r R such that Ln Jn and
Ln Jn+1 = 0. Then we get an infinite direct sum
L1 L2 L3 . . . ,
contradicting (2). 

6.13. Proof of Goldies Theorem - Step 2: RS is semisimple Artinian.

Proof. Let K /r RS . By adapting the proof of Theorem 5.10, K = (K R).RS . We

claim that J = K R is closed.
For, suppose I * J. Pick z I\J and let T = {r R : zr J}, a right ideal.
If s T S then zs J so z J.RS = K and z K R = J, a contradiction.
Hence T S = , so T is not essential by Proposition 6.7 (). Hence there exists
0 6= V /r R such that V T = 0.
It follows that zV I and zV J = 0 as required.
Now, if K1 ) K2 ) K3 ) . . . is a strictly descending chain of right ideals in RS ,
then K1 R ) K2 R ) K3 R ) . . . is a strictly descending chain of closed right
ideals in R, contradicting Lemma 6.12. Hence RS is right Artinian.
Finally, Let J be the Jacobson radical of RS . By Proposition 3.10, J is nilpotent
and hence so is J R. As R is semiprime, J R = 0. But now J = (J R).RS = 0
so RS is semisimple Artinian, as required. 
36

7. Filtrations, associated graded rings and completions

Throughout this chapter, k denotes a field.

7.1. Filtered rings.

Definition. A (Z)filtration on a ring R is a set of additive subgroups (Ri )iZ

such that
Ri Ri+1 for all i Z,
Ri .Rj Ri+j for all i, j Z,
1 R0 , and
iZ Ri = R.
If R has a filtration, R is a filtered ring. The filtration on R is said to be
positive if Ri = 0 for all i < 0,
negative if Ri = R for all i 0,
separated if iZ Ri = 0.

Note that the axioms imply that R0 is a subring of R and that each Ri is a left
and right R0 module. Note also that iZ Ri is always an ideal in R.

Examples. (1) R = Q, Ri = pi Z(p) where p is a fixed prime.

(2) R = k[x, x1 ], Ri = xi k[x1 ] = k{xi , xi1 , xi2 , . . .}.
(3) Suppose R is a finitely generated kalgebra with generating set {x1 , . . . , xn }.
If i 0 let Ri be the ksubspace of R spanned by words in the xj s of length
at most i, let R0 = k and let Ri = 0 whenever i < 0. This is a positive
filtration on R.
(4) R = Z, Ri = pi Z for i 0 and Ri = R otherwise. This is a negative
filtration. More generally,
(5) Let R be any ring and let I be an ideal of R. Then Ri = I i is a negative
filtration, called the Iadic filtration.

7.2. Graded rings.

Definition. A (Z)graded ring is a ring S which can be written as

M
S= Si
iZ

for some additive subgroups Si S, satisfying Si .Sj Si+j for all i, j Z and
1 S0 . Si is called the ith homogeneous component of S, and an element s S
is homogeneous iff it lies in some Si .
A graded right ideal J /r S is a right ideal of the form iZ Ji with Ji Si .

Lemma. If J is a finitely generated graded right ideal of a graded ring S, then J

has a finite generating set consisting of homogeneous elements.
 37

Proof. Since J is graded, note that J Si = Ji for each i Z. Let {x1 , . . . , xn } be

P
a generating set for J. If xj = iZ aij for some aij Ji , we see that each aij lies
in J and is homogeneous. Clearly, {aij : aij 6= 0} is now a finite generating set for
J consisting of homogeneous elements. 

7.3. Associated graded ring.

Definition. Let R be a filtered ring with filtration (Ri )iZ . Define the abelian group
M
gr R = Ri /Ri1 .
iZ

Equip gr R with multiplication, which is given on homogeneous components by

Ri /Ri1 Rj /Rj1 Ri+j /Ri+j1
r + Ri1 , s + Rj1 7 rs + Ri+j1
and on the whole of gr R by bilinear extension. Then gr R becomes a ring called the
associated graded ring of R.

Note that the multiplication is well-defined because Ri Rj Ri+j , Ri1 Rj

Ri+j1 and Ri Rj1 Ri+j1 . One should think of gr R as an approximation
to the ring R which is often easier to understand but nonetheless contains useful
information about the ring R itself.

Notation. If r Ri \Ri1 then the symbol of r in gr R is the homogeneous element

(r) = r + Ri1 gr R.

7.4. Almost commutative algebras.

Definition. An almost commutative kalgebra is a finitely generated kalgebra

with generators {x1 , . . . , xn } such that

xi xj xj xi k{1, x1 , . . . , xn }

for all i, j = 1, . . . , n.

Examples. (1) The commutative polynomial kalgebra k[x1 , . . . , xn ].

(2) The Weyl algebra An (k).
(3) The universal enveloping algebra U(g) of a kLie algebra g, dimk g < .
(4) Any homomorphic image of U(g).

Proposition. Let R be an almost commutative algebra generated by {x1 , . . . , xn }.

Equip R with the positive filtration as in Example 7.1(3). Then there exists a
surjective homomorphism of kalgebras

: k[X1 , . . . , Xn ]  gr R

given by (Xi ) = (xi ), i = 1, . . . , n.

38

Proof. Note that xi R1 for all i. If xi , xj

/ k, we have

(xi ) = xi + R0 R1 /R0 and (xj ) = xj + R0 R1 /R0 .

Since R is almost commutative,

(xi )(xj ) = xi xj + R1 = xj xi + R1 = (xj )(xi ),

meaning that (xi ) and (xj ) commute. If one of xi , xj lies in k then this is also
true. Hence the kalgebra map exists.
To show that is surjective, its sufficient to show that u + Rt1 lies in Im for
any u Rt \Rt1 . Write u as a sum of words of length at most t in the generators
{x1 , . . . , xn }. Since were interested in u + Rt1 , we can assume that each word
actually has length t.
Pm (j) (j)
Writing u = j=1 j xi(j) xi(j) xi(j) for some i1 , . . . , it {1, . . . , n}, we have
1 2 t

m
X
u + Rt1 = j (Xi(j) )(Xi(j) ) (Xi(j) ) Im .
1 2 t
j=1

7.5. Proposition. Any almost commutative kalgebra R is a quotient of U(g) for

some finite dimensional kLie algebra g.

Proof. Let {x1 , . . . , xn } be the generating subset of R and let g = k{1, x1 , . . . , xn }.

R is a kLie algebra under the commutator bracket [a, b] = ab ba, and the
condition that R is almost commutative says precisely that g is a Lie subalgebra of
R and hence a finite dimensional Lie algebra itself.
Now we get a ring homomorphism : U(g) R extending the natural injection
g , R of kLie algebras. Since {x1 , . . . , xn } generates R as a kalgebra, is onto,
as required. 

Example. Let h2n+1 be the Heisenberg Lie algebra of dimension 2n + 1 with gen-
erators {x1 , . . . , xn , y1 , . . . , yn , z} and relations [yi , xi ] = z for i = 1, . . . , n, all the
other brackets being zero. Then there is a surjective map of kalgebras

: U(h2n+1 )  An (k)

given by (xi ) = xi , (yi ) = yi and (z) = 1.

Note that one can also represent h2n+1 by (n + 2) (n + 2) matrices:

0 k k k
0 0 0 k

h2n+1

= gln+2 (k).

0 0 0 k

0 0 0 0
 39

7.6. Filtered and graded modules.

Definition. Let R be a filtered ring with filtration (Ri )iZ and let M be a right
Rmodule. A filtration on M is a set (Mi )iZ of additive subgroups of M satisfying
Mi Mi+1 for all i Z,
Mi .Rj Mi+j for all i, j Z,
iZ Mi = M .
Positive, negative and separated filtrations are defined analogously to (7.1). Fil-
tered left modules are defined similarly.

Example. Let M be a finitely generated right Rmodule with generating set A.

Then Mi = ARi for all i Z gives a filtration of M , known as the standard
filtration.

Definition. Let S = iZ Si be a graded ring. A graded right Smodule is a right

Smodule V of the form
M
V = Vi
iZ
such that Vi Sj Vi+j for all i, j Z.

7.7. Associated graded modules.

Definition. Let R be a filtered ring and let M be a filtered right Rmodule with
filtration (Mi )iZ . Define the abelian group
M
gr M = Mi /Mi1 .
iZ

Equip gr M with a gr Raction, which is given on homogeneous components by

Mi /Mi1 Rj /Rj1 Mi+j /Mi+j1
m + Mi1 , r + Rj1 7 mr + Mi+j1
and on the whole of gr M by bilinear extension. Then gr M becomes a graded
gr Rmodule, called the associated graded module of M .
If N is a submodule of M , define the subspace filtration (Ni )iZ on N by Ni =
N Mi . Also, define the quotient filtration ((M/N )i )iZ on M/N by (M/N )i =
(Mi + N )/N .

Proposition. Let R be a filtered ring, let M be a filtered right Rmodule with

filtration (Mi )iZ and let N be a submodule of M . Equip N with the subspace
filtration and M/N with the quotient filtration. Then
(1) There exists an injection : gr N gr M of right gr Rsubmodules.
(2) Identifying gr N with its image in gr M , we have
gr(M/N )
= gr M/ gr N
as right gr Rmodules.
40

Proof. The natural composition of maps Ni , Mi and Mi  Mi /Mi1 has kernel

Ni Mi1 = N Mi1 = Ni1 . So we have an injection of abelian groups
i : Ni /Ni1 , Mi /Mi1
for all i Z. Putting these together we get an injection
= i : gr N gr M.
Exercise: check that is a right gr Rmodule homomorphism.
This proves (1).
We have the quotient filtration ((Mi + N )/N )iZ of M/N . Consider the compo-
sition
u Mi + N vi (Mi + N )/N
i : Mi /Mi1 i
Mi1 + N (Mi1 + N )/N
where ui (m + Mi1 ) = m + Mi1 + N and vi is the natural isomorphism.
Note that ui is onto, whereas
Mi (Mi1 + N ) Mi1 + (Mi N ) Mi1 + Ni
ker(ui ) = = = = Im(i )
Mi1 Mi1 Mi1
by the modular law. Since vi is an isomorphism, i is onto and ker(i ) = Im(i )
for all i Z. Letting
= i : gr M gr(M/N )
we see that is onto and ker() = Im().
Exercise: check that is a right gr Rmodule homomorphism.
Hence induces the required isomorphism of right gr Rmodules
gr M/(gr N ) = gr M/ ker()
= Im() = gr(M/N ).


7.8. Lifting the Noetherian property - I.

Theorem. Suppose R is a positively filtered ring such that gr R is right Noetherian.

Then R is right Noetherian.

Proof. If I J are right ideals of R, equip I and J with the subspace filtrations
and J/I with the quotient filtration. Proposition 7.7 shows that gr I and gr J are
right ideals of gr R, and moreover that gr I gr J and gr(J/I)
= gr J/ gr I.
Suppose
I1 I2 I3 . . .
is an increasing chain of right ideals of R. Applying gr we obtain an increasing
chain of right ideals of gr R
gr I1 gr I2 gr I3 . . .
which stops because gr R is right Noetherian. So its sufficient to prove that if
gr I = gr J for right ideals I J then I = J.
 41

Let (Ri )iZ be the filtration on R. If I < J then I Ri < J Ri for some i.
Since the filtration on R is positive, we can choose a least such i. Choose some
x (J Ri )\I. Then x / Ri1 as I Ri1 = J Ri1 by minimality of i. Now

(x) = x + Ri1 gr J = gr I

so there exists y I such that (x) = (y) = y + Ri1 . Hence x y Ri1 . But
x y J since y I J, so x y J Ri1 = I Ri1 . Hence x y I so
x I, a contradiction. The result follows. 

Of course, the same result holds with right replaced by left.

Corollary. Any almost commutative algebra R is right and left Noetherian. This
includes An (k) and U(g).

Proof. By Proposition 7.4, gr R is a quotient of a polynomial algebra k[X1 , . . . , xn ]

for some n, which is right and left Noetherian by Theorem 1.14. Hence R is right
and left Noetherian by Theorem 7.8. 

7.9. Topologies. Assume until the end of this chapter that R is a negatively fil-
tered ring and M is a negatively filtered right Rmodule with filtration (Mi ).
We can define a topology on M called the filtration topology, by choosing the open
subsets to be unions of sets of the form m + Mi . Then as required, the collection
of open sets is closed under arbitrary unions, and also under finite intersections -
since the intersection of m1 + Mi(1) , . . . , mn + Mi(n) is a union of sets of the form
m + Mi where i = min{i(1), . . . , i(n)}. Another way of phrasing this is to say that
the cosets m + Mi form a basis for the filtration topology.
If the filtration is separated, we can define a metric on M : fix a real number
c > 1 and set
d(x, y) = inf{ck : x y Mk }.
Note that we need Mi = 0 to ensure that d(x, y) = 0 x = y. It can be checked
that the topology we get on M from the metric is the filtration topology, so the
choice of the constant c is irrelevant.
Two filtrations (Mi ) and (Mi0 ) give the same filtration topology on M if and
0
only if for all i there exist s(i) such that Ms(i) Mi and t(i) such that Mt(i) Mi0 .
The filtrations are then said to be topologically equivalent.

7.10. Cauchy sequences and completions.

Definition. A Cauchy sequence on M is a sequence (xn ) n=0 such that for any
open set U containing 0, there exists c(U ) N such that xn xm U whenever
n, m c(U ). Two Cauchy sequences (xn ) and (yn ) are said to be equivalent if
xn yn 0. Equivalently, given any open set U containing 0, there exists N (U ) Z
such that xn yn U whenever n N (U ).
42

Exercise. Check that this gives an equivalence relation on the set of all Cauchy
sequences on M .

Let [xn ] denote the equivalence class of the Cauchy sequence (xn ) under this
equivalence relation, and let M
c denote the set of all such equivalence classes. The
operations
[xn ] + [yn ] = [xn + yn ]
[xn ] . [yn ] = [xn yn ] [xn ], [yn ] R
b

turn R
b into an ring, and

[xn ] + [yn ] = [xn + yn ] [xn ], [yn ] M

c
[xn ] . [rn ] = [xn rn ] [xn ] M , [rn ] R
c b

turn M
c into a right Rmodule.
b There is a ring homomorphism
R R b
r 7 [r]
where [r] is the equivalence class of the constant sequence with value r. The kernel
of this map is Ri . This also enables us to view M c as a right Rmodule, and we
have an analogous map of Rmodules
M M c
m 7 [m]

Definition. We say that Rb (Mc) is the completion of R (M , respectively). We call

R (M ) complete if the natural map R R b (M M c) is an isomorphism.

Note also that we can define a filtration on M

c by setting

ci = {[xn ] : ci N
M such that xn Mi whenever n ci }

the set of equivalence classes of Cauchy sequences which end up in Mi eventually.

Thus R b and M c have naturally defined topologies, and the natural maps M M c
and R R b are continuous with respect to these topologies.
Note that a complete module M is necessarily separated, since the kernel of the
natural map M M c is Mi .

Exercise. Check all the details above!

7.11. Inverse limits. There is another way of approaching completions. Let r < s
and let rs : M/Mr  M/Ms be the natural projection. Define

Y
lim M/Mi = {(yn +Mn )nZ M/Mn : rs (yr +Mr ) = ys +Ms whenever r < s.}

iZ

This becomes a right Rmodule with pointwise addition and scalar multiplica-
tion. When M = R, these operations turn lim R/Ri into a ring.

 43

Lemma. M
c is isomorphic to lim M/Mi as Rmodules.

Proof. If (xn ) is a Cauchy sequence in M , then for each i Z there exists c(Mi ) N
such that xn xm mod Mi for all n, m c(Mi ). We can choose the c(Mi ) to
satisfy

c(M0 ) c(M1 ) c(M2 ) . . . .

Now if r < s then rs (xc(Mr ) + Mr ) = xc(Mr ) + Ms = xc(Ms ) + Ms since c(Mr )

c(Ms ), which means that (xc(Mi ) + Mi )iZ lim M/Mi .

Define
: M c lim M/Mi

[xn ] 7 (xc(Mi ) + Mi )iZ .

Note that this doesnt depend on the choice of the numbers c(Mi ): if d(Mi ) are
such that xn xm mod Mi whenever n, m d(Mi ), then

xc(Mi ) xmax(c(Mi ),d(Mi )) xd(Mi ) mod Mi

for all i Z.
Now, if [xn ] = [yn ] M
c, then xn yn 0 as n . Hence for all i Z there
exists t(Mi ) such that xn yn Mi whenever n t(Mi ). Let c(Mi ), d(Mi ) be the
integers corresponding to [xn ] and [yn ]. Letting e(Mi ) = max(c(Mi ), d(Mi ), t(Mi )),
we have

xc(Mi ) xe(Mi ) ye(Mi ) yd(Mi ) mod Mi

for all i Z. Hence ([xn ]) doesnt depend on the choice of Cauchy sequence inside
the equivalence class [xn ], so is well defined.

Exercise. Check that is a map of right Rmodules.

It remains to show that is an isomorphism. If [xn ] ker(), then xc(Mi ) Mi

and hence xn Mi for all n c(Mi ). Hence xn 0 and [xn ] = 0, so is injective.
If y = (yi + Mi )iZ lim M/Mi , let xn = yn M for all n N. Then (xn ) n=0

is a Cauchy sequence and that ([xn ]) = y. Hence is surjective. 

Q
Equip each M/Mi with the discrete topology and iZ M/Mi with the (Ty-
chonov) product topology. Then the map constructed above is actually a home-
omorphism, where we give M c the filtration topology and lim M/Mi Q
iZ M/Mi
the subspace topology.
Also, in the case when M = R, is a ring isomorphism.
The advantage of inverse limits is that the elements are easy to manipulate; the
disadvantage is that everything is dependent on the filtration.
44

7.12. Examples.
(1) Zp , the ring of padic integers is the completion of R = Z with respect to
the padic filtration Ri = pi Z for i 0. Thus Zp = lim Z/pi Z.

(2) More generally, if R is any ring and I / R is a two-sided ideal, we can
complete R with respect to the Iadic filtration:
b = lim R/I i .
R

b
(3) If R = k[X1 , . . . , Xn ] and I = (X1 , . . . , Xn ) / R then R = k[[X1 , . . . , Xn ]],
the ring of power series. Here we identify a power series with the sequence
of partial sums. For example, if n = 1,

a0 + a1 X + a2 X 2 + . . . (a0 + I, a0 + a1 X + I 2 , . . .)

This works similarly for arbitrary n - the notation is cumbersome.

(4) Consider the group of all n n invertible matrices with coefficients in Zp .
This is GLn (Zp ). Let

K = {M GLn (Zp ) : M Inn mod p} = ker(GLn (Zp ) GLn (Fp )).

K is an example of a torsionfree compact padic Lie group.

Let R = Zp [K] be the group algebra of K over Zp . Let I be the mod p
augmentation ideal of R: I = ker() where

:R Fp
P P
gK g g 7 gK g

where Fp is the reduction of Zp modulo p. Then R,

b the completion
of R with respect to the Iadic filtration is the Iwasawa algebra of K,
written K .

7.13. Proposition. Let R be a ring and I / R an ideal such that R/I is a division
ring. Then the completion R b of R with respect to the Iadic filtration is a local
ring, that is, it has a unique maximal left and right ideal.

Proof. Let x Ib = {(xn + I n ) lim R/I n : x1 I}. Then

y = 1 + x + x + x + . . . = (1 + I, 1 + x2 + I 2 , 1 + x3 + x23 + I 3 , . . .) R
2 3 b

and y(1 x) = (1 x)y = 1, so 1 x is invertible in R b whenever x I.b Now if

u = (un + I n ) b then u1
/ I, / I. Since R/I is a division ring, we can find v R
such that u1 v vu1 1 mod I. Hence uv 1 Ib so uv and vu are invertible by
the above. Hence u is invertible whenever u / Ib so Ib is the unique maximal left
and right ideal of Rb as required. 

This shows that Z(p) , Zp for any prime p.

 45

7.14. Lifting the Noetherian property - II.

Theorem. Let R be a complete negatively filtered ring such that gr R is right Noe-
therian. Then R is right Noetherian.

Proof. Let I /r R be a right ideal. Then gr I is a graded right ideal of gr R by

Proposition 7.7. Since gr R is right Noetherian, gr I is finitely generated. We can
choose a homogeneous generating set, by Lemma 7.2:
m
X
gr I = si gr R
i=1

where s1 , . . . , sm are all homogeneous. Each si is the symbol of some xi I:

si = (xi ) Rn(i) /Rn(i)1 , say.

We will show that

m
X
I= xi R.
i=1
Since each xi I and I is a right ideal, the reverse inclusion is obvious. Let y I
Pm
and suppose that y Rj \Rj1 (if y Rj then y = 0 i=1 xi R since the
filtration on R is separated.)
Pm
Now (y) = y + Rj1 gr I = i=1 si gr R so we can find ti gr R such that
m
X
(y) = si ti .
i=1

Since (y) and the si are homogeneous, we may assume that ti Rjn(i) /Rjn(i)1 ,
i = 1, . . . , m. If ti = 0 set ri1 = 0, otherwise choose ri1 Rjn(i) \Rjn(i)1 such
that ti = ri1 + Rjn(i)1 . Hence
m
X
y xi ri1 mod Rj1 .
i=1
Pm
Now let y1 = y i=1 si ri1 I Rj1 . If y1 = 0, stop; if not, repeat the
above with y1 in place of y. This gives a sequence (ri1 , ri2 , ri3 , . . .) such that rik
Rjn(i)k+1 and
Xm Xk
yk = y xi ril I Rjk
i=1 l=1
Pk
for every k 1. But now ( l=1 ril )
k=1 is a Cauchy sequence in R which converges
to some ri R because R is complete. Also, yk 0 since yk Rjk for all k.
P
Hence, letting ri = l=1 ril , we obtain
m
X m
X
y= xi ri xi R
i=1 i=1

as required. 
46

Examples. (1) Giving Zp the padic filtration, gr Zp = Fp [t] which is Noe-

therian, and of course Zp is complete with respect to this filtration. Hence
Zp is Noetherian.
(2) Suppose K is a compact padic Lie group of a special type, namely a uni-
form pro-p group, of dimension d. One example is the K appearing in
Example 7.12(4), it has dimension n2 . It can be shown that there is a fil-
tration of K which is topologically equivalent to the Iadic filtration of
K , such that
gr K
= Fp [X0 , X1 , . . . , Xd ],
the polynomial algebra over Fp in d + 1 variables. Hence K is right and
left Noetherian.
 47

8. Weyl algebras

Throughout this chapter, k denotes a field of characteristic 0, and all modules

will be left modules.

8.1. Recall. The n-th Weyl algebra An has generators {x1 , . . . , xn , y1 , . . . , yn } as

a kalgebra and relations [yi , xi ] = 1 for i = 1, . . . n.
Let A = k[X1 , . . . , Xn ] be the polynomial algebra. From (1.11), we know that
there is a natural kalgebra homomorphism

: An Endk (A)
xi 7 (f 7 Xi f )
f
yi 7 (f 7 X i
).
We can therefore think of A as a left An module, via

r.f = (r)(f ) for all r An , f A.

8.2. Notation.
When = (1 , . . . , n ) Nn , write x = x n 1
1 xn and y = y1 yn .
1 n

The degree of the monomial x is defined to be || = 1 + . . . + n . We

also write ! = 1 ! n !.
((An )m ) is the positive filtration on An defined in Example 7.1(3): (An )m
consists of the span of all words in the generators {x1 , . . . , xn , y1 , . . . , yn }
of length at most m. This is also known as the Bernstein filtration on An .
The degree of a nonzero element r An is the integer m = deg r such that
r (An )m \(An )m1 .

8.3. Proposition. The set {x y : , Nn } is a basis for An .

Proof. Any word in the generators can be brought to a linear combination of the
x y s by pushing xs to the left, using the relations yi xi = xi yi + 1. Hence
{x y } spans An .
Next, suppose r = ,Nn x y An is zero but some coefficient is
P

nonzero. Choose Nn such that 6= 0 for some Nn , but = 0 whenever

|| < ||.
Now, if || || then either = or i > i for some i. But in the latter case,
i
yi .Xii = 0 so y .X = 0, whereas y .X = !. Hence

r.X =
P P P
||<|| x y .X + ||||,6= x y .X + Nn ! X
= ! Nn X 6= 0
P

since char(k) 6= 0 and 6= 0. This contradicts r = 0 and the result follows.



In fact, this is true even when char(k) > 0.

48

Corollary. Equip An with the positive filtration (An )m given in Example 7.1(3).
Then
gr An
= k[Z1 , . . . , Z2n ].

Proof. Let : k[Z1 , . . . , Z2n ]  gr An be the surjective kalgebra homomorphism

given by (Zi ) = (xi ), (Zi+n ) = (yi ), i = 1, . . . , n, as in Proposition 7.4.
Note that sends homogeneous polynomials of degree m into the mth graded
part of gr An . Hence if f ker() then we may assume f to be homogeneous of
degree m say: X 1 n
f= Z11 Znn Zn+1 Z2n .
,Nn
||+||=m
Now (f ) = 0 implies that
X
x y (An )m1 .
,Nn
||+||=m

Since the {x y } are linearly independent over k by Proposition 8.3, = 0 for

all relevant , and hence f = 0. Hence is an isomorphism. 

8.4. Zero divisors.

Proposition. Suppose R is a filtered ring with a separated filtration (Ri )iZ , such
that gr R is a domain. Then R is a domain also.

Proof. Suppose for a contradiction that r, s R are nonzero but rs = 0. Since the
filtration is separated, there exist i, j Z such that r Ri \Ri1 and s Rj \Rj1 .
But now (r) = r + Ri1 and (s) = s + Rj1 are nonzero in gr R and
(r).(s) = rs + Ri+j1 = 0,
contradicting the assumption that gr R is a domain. 

Corollary. (1) An is a domain.

(2) If G is a uniform pro-p group, then the Iwasawa algebra G is a domain.

8.5. PBW Theorem. Let g be a finite dimensional kLie algebra with basis
{x1 , . . . , xn }. The enveloping algebra U(g) contains the standard monomials
x = x1 2 n
1 x2 xn

for all Nn .

Theorem (Poincare-Birkhoff-Witt). {x : Nn } form a basis for U(g).

Proof. Omitted. 

We can deduce exactly as in Corollary 8.3 that gr U(g) is a polynomial algebra in

n variables, and it follows from Proposition 8.4 that U(g) is a domain. This works
also when char(k) > 0.
 49

8.6. Proposition.
(1) Z(An ) = k.
(2) An is a simple ring.

Proof. First, note that if a, b, c are elements of some ring R then

[ab, c] = a[b, c] + [a, c]b and [a, bc] = b[a, c] + [a, b]c.

Next, we prove that a, b satisfy [a, b] = 1 then [am , b] = mam1 . This is true when
m = 1 so assume inductively that m > 1 and [am1 , b] = (m 1)am2 . Then

[am1 , b] = a[am1 , b] + [a, b]am1 = (m 1)aam2 + am1 = mam1

as claimed. In particular, since [yi , xi ] = 1, [yii , xi ] = i yii 1 for all i N.

Now let ei Nn be the element with 1 in ith position and 0s elsewhere.
Because xi commutes with every generator apart from yi , the above shows that
inside An ,
[x y , xi ] = i x y ei .
i 1
Similarly, since [xi , yi ] = 1 we see that [x
i , yi ] = i xi
i
so

[x y , yi ] = i xei y .

x y Z(An ) then r must commute with each generator.

P
Now if r = ,Nn
Hence
X
[r, xi ] = i x y ei = 0
,Nn

so i = 0 for every possible i, , by Proposition 8.3. By considering [r, yi ] we

obtain i = 0 for all i, , . As char(k) = 0, 6= 0 i = i = 0 for all i, so
the only possible nonzero coefficient in r is 0,0 . Hence r k and part (1) follows.
Next, suppose that I is a nonzero two-sided ideal of An . Pick a nonzero element
r I of least degree m, say. The above equation for [r, xi ] shows that

deg[r, xi ] deg r 1 < m

and similarly deg[r, yi ] deg r 1 for all i. Since I is a two-sided ideal [r, xi ] and
[r, yi ] must lie in I for all i. By minimality of m, these must all be zero and hence
r Z(An ) = k. Since r 6= 0 and k is a field, 1 I and hence I = An . 

Warning: this result is false in positive characteristic!

8.7. Rings of differential operators. Let R be a commutative kalgebra. The

ring of differential operators of R is a subring of Endk (R). We construct it induc-
tively. Note that we have an injection a 7 (a : r 7 ar) of R into Endk (R). Thus
we can identify R with a subring of Endk (R). Define

D0 (R) = {D Endk (R) : [D, a] = 0 for all a R}

D1 (R) = {D Endk (R) : [D, a] D0 (R) for all a R}

50

D2 (R) = {D Endk (R) : [D, a] D1 (R) for all a R}

and so on. These are kvector spaces D0 (R) D1 (R) D2 (R) . . . Endk (R).

Lemma. Let P Dn (R) and Q Dm (R). Then P Q Dm+n (R).

Proof. Induct on n + m. If n + m = 0, then n = m = 0 so P, Q commute with R

and hence P Q also commutes with R, meaning that P Q D0 (R).
Suppose now that this is true whenever m+n < l and consider the case m+n = l.
Fix a R. By definition, [Q, a] Dm1 (R) and [P, a] Dn1 (R), so by inductive
hypothesis P [Q, a] Dm+n1 (R) and [P, a]Q Dm+n1 (R). Hence

[P Q, a] = P [Q, a] + [P, a]Q Dm+n1 (R)

and hence P Q Dm+n (R) as required. 

Definition. The ring of differential operators D(R) of R is the subring n

n=0 D (R)
of Endk (R).

Note that by Lemma 8.7, D(R) is a subring, and moreover, (Di (R))iZ is a
positive filtration on D(R), provided we assume Dk (R) = 0 for negative k.

8.8. Derivations.

Definition. A derivation of a commutative kalgebra R is a klinear map D :

R R satisfying the Leibnitz rule

D(ab) = aD(b) + D(a)b for all a, b R.

The kvector space of all derivations of R is denoted by Der(R).

Note also that if x R then xD : R R is another derivation, since

(xD)(ab) = x(aD(b) + D(a)b) = a(xD)(b) + (xD)(a)b

Hence Der(R) is a left Rmodule.

For small values of n, we have alternative descriptions of Dn (R).

Proposition. Let R be a commutative kalgebra. Then

(1) D0 (R) = {a : a R}, and
(2) D1 (R) = D0 (R) + Der(R).

Proof. (1) Since R is commutative, a D0 (R) for all a R. On the other hand,
if D D0 (R) then D(a) = D(a.1) = (D.a)(1) = (a.D)(1) = a(D(1)) = D(1)(a).
[
[ as required.
Hence D = D(1)
(2) If D Der(R) and a R, then

[D, a](b) = (Da)(b) (aD)(b) = D(ab) aD(b)

[
= D(a)b = D(a)(b).
 51

[ D0 (R) by (1) for any a R, showing that Der(R) D1 (R).

Thus [a, D] = D(a)
Hence D0 (R) + Der(R) D1 (R).
On the other hand, let Q D1 (R) and set P = Q Q(1) [ D1 (R), so that
1 0
P (1) = 0. Now P D (R) so [P, a] D (R) for any a R and hence [[P, a], b] = 0
for any a, b R. Applying this to 1 R we get
0 = [P, a]b(1) b[P, a](1) = (P a aP )(b) b(P a aP )(1) =
= P (ab) aP (b) bP (a) + baP (1)

[ Der(R)+D0 (R)
so P (ab) = aP (b)+bP (a) and P Der(R). Hence Q = P + Q(1)
as required. 

8.9. The case R = k[X1 , . . . , Xn ].

Proposition. Let R = k[X1 , . . . , Xn ] and let i = Xi . Then
n
X
Der(R) = Ri .
i=1
Pn
Proof. Its clear that every map D : R R of the form D = i=1 fi i is a
derivation. Conversely, let D Der(R). Then D(Xim ) = mXim1 D(Xi ) by an easy
induction. Hence
Xn
(D D(Xi )i )(X11 Xnn ) = 0
i=1
Pn Pn
so D = i=1 D(Xi )i i=1 Ri , since the monomials form a kbasis for R. 

By this result and Proposition 8.8(2), we see that

D1 (R) = R + R1 + . . . + Rn .

This is a part of the Weyl algebra. Indeed, we have

R
P
Theorem. D(R) = Nn = An .

Proof. Omitted. 

Warning: this is false in characteristic p > 0!

8.10. An modules.

Proposition. (1) The natural An module A = k[X1 , . . . , Xn ] is simple.

Pn
(2) A = An / i=1 An yi as left An modules.
(3) No nonzero left An module M is finite dimensional over k.

Proof. (1) Let 0 6= f A. Let X = X11 Xnn be a monomial of maximal

degree in f with nonzero coefficient , say. Then

y .f = y .X = ! 6= 0
52

since char(k) = 0 so 1 An f . Hence any nonzero element of A generates A, so A

is simple.
(2) We have A = An .1 is a cyclic module, so A = An /I where I = ann(1) /r An .
Pn
Since yi .1 = 0 for all i = 1, . . . , n, I contains i=1 An yi . Suppose r I. By
Proposition 8.3, we can write r = a + b where a = a(x1 , . . . , xn ) k[x1 , . . . , xn ] and
Pn Pn
b i=1 An yi . Now r.1 = a.1 = a(X1 , . . . , Xn ) = 0 forces a = 0 so r i=1 An yi
as required.
(3) Suppose M is a finite dimensional An module of dimension t. The action
of An gives a ring homomorphism
: An Endk (M )
= Mt (k).
Now (y1 )(x1 ) (x1 )(y1 ) = Itt . Since tr(U V ) = tr(V U ) for any matrices
U, V , taking traces we conclude that t = 0, so M = 0 as required. 

Part (3) is a special case of Bernsteins Inequality. (1) and (3) are false in positive
characteristic, and only (2) remains valid.

8.11. Solutions of PDEs. Suppose we have a finite set of elements P1 , . . . , Pm

An . Then we can consider the set of simultaneous partial differential equations
() P1 .f = P2 .f = . . . = Pm .f = 0
where f A = k[X1 , . . . , Xn ]. The An module associated with this system of
Pm
equations is MP1 ,...,Pm = An / i=1 An Pi .

Proposition. The kvector space of polynomial solutions of () is isomorphic to

HomAn (MP1 ,...,Pm , A).

Proof. We can define a left An module map f : An A by f (1) = f . If f A

is a solution, then Pi ker(f ) for all i, so we obtain a map f : MP1 ,...,Pm A.
Conversely, if : MP1 ,...,Pm A is a map of left An modules, then (1) A is
a solution of the system (). This gives a bijective correspondence. 

One can also look for solutions in other spaces B, which are An modules - the
solution space will be HomAn (MP1 ,...,Pm , B).
For example, when k = R, we can take B = C (U ), the set of all infinitely
differentiable functions f : U R, for some open set U Rn . This is a left
An module, with the action given by
(xi .f )(a1 , . . . , an ) = ai f (a1 , . . . , an ) and
f
yi .f = .
ai
 53

9. Dimensions

Throughout, k denotes a field. Again, modules will be on the left.

9.1. Hilbert polynomials. We start by considering finitely generated graded

modules
M
N= Ni
i=0

for the commutative polynomial algebra S = k[X1 , . . . , Xn ] = i=0 Si . Here Si is

the space of all homogeneous polynomials of degree i.
Pick a homogeneous generating set {u1 , . . . , um } for N , where deg uj = kj , say.
Then
Xm
M Xm Xm
N= uj S = uj Sikj so Ni = uj Sikj
j=1 i=0 j=1 j=1

for all i 0. Since each graded component Si of S is finite dimensional over k, we

see that each Ni must itself be finite dimensional over k.

Definition. The Poincare series for N is the power series

X
P (N, t) = (dim Ni ) ti Z[[t]].
i=0

Theorem (Hilbert,Serre). Let S and N be as above. Then

(1) There exists f (t) Z[t] such that
f (t)
P (N, t) = .
(1 t)n
(2) Let d N be the order of the pole at t = 1 of P (N, t). Then there exists
(t) Q[t] of degree d 1, such that for large enough i, dim Ni = (i).
(3) There exists (t) Q[t] of degree d, such that for large enough i,
i
X
dim Nj = (i).
j=0

Proof. (1) Proceed by induction on n. When n = 0, N is a finitely generated

kmodule, and hence dimk N < . Hence Ni = 0 for large enough i, so P (N, t) is
clearly a polynomial.
Now, assume n > 1 and consider the map : N N given by (v) = Xn v.
Since R is commutative, this is a map of Rmodules. Since Xn has degree 1,
maps Ni to Ni+1 . Let Ki = ker(|Ni ) and let Li+1 = coker(|Ni ) = Ni+1 /Xn Ni .
Let K =
i=0 Ki and L = i=0 Li where we set L0 = N0 . Then K = ker()
and L = coker(). Note that both K and L are finitely generated Smodules since
S is Noetherian and N is finitely generated. Also, both are killed by Xn , and are
54

hence finitely generated graded k[X1 , . . . , Xn1 ]modules. By induction, P (K, t)

and P (L, t) have the required form:
a(t) b(t)
P (K, t) = and P (L, t) =
(1 t)n1 (1 t)n1
for some a(t), b(t) Z[t]. Since Ni /Ki
= Xn Ni as kvector spaces, we see that
dim Ni dim Ki = dim Ni+1 dim Li+1 .
Multiply this equation by ti+1 and sum over i to get
tP (N, t) tP (K, t) = P (N, t) P (L, t),
since P (N, 0) = P (L, 0). Hence
P (L, t) tP (K, t) b(t) ta(t)
P (N, t) = =
1t (1 t)n
as required.
(2) By cancelling powers of (1 t) if necessary, we may assume that
f (t) = a0 + a1 t + . . . + as ts Z[t]
satisfies f (1) 6= 0, so that P (N, t) = f (t)(1 t)d with d the order of the pole of
P (N, t) at t = 1.
Now, (1 t)1 = 1 + t + t2 + . . .. Repeated differentiation gives
 
d
X d+m1 m
(1 t) = t , so
m=0
d1

s   min(s,i)  
X X d + m 1 X X d + i j 1
P (N, t) = aj tj tm = aj ti .
j=0 m=0
d 1 i=0 j=0
d 1
We deduce that for all i = s, s + 1, . . .,
     
d+i1 d+i2 d+is1
() dim Ni = a0 + a1 + . . . + as .
d1 d1 d1
The right hand side of this expression can be rearranged as (i) for some polynomial
(t) Q[t]. Note also that
f (1) d1
(t) = t + lower degree terms.
(d 1)!
Since f (1) 6= 0, we see that (t) has degree d 1.
(3) Since ab = a1 a1




b1 + b , we see that
i    
X d+j1 d+i
= .
j=0
d1 d

Replacing i by j in () and summing from j = 0 to i, we obtain

i      
X d+i d+i1 d+is
dim Nj = a0 + a1 + . . . + as = (i)
j=0
d d d
 55

for some polynomial (t) Q[t] of degree d. This is valid whenever i s. 

9.2. Definition. Let S = k[X1 , . . . , Xn ] and let N =

i=0 Ni be a finitely gener-
ated graded Smodule. The objects appearing in Theorem 9.1 have names:
The Hilbert polynomial of N is (t) Q[t].
The Samuel polynomial of N is (t) Q[t].
The dimension of N is the integer d = d(N ).
The multiplicity of N is the integer m(N ) = d!a where a is the leading
coefficient of (t).
Our next goal will be to define dimension and multiplicity for finitely generated
modules over almost commutative algebras. For this, we first need to make a
digression to study Rees rings and good filtrations.

9.3. Rees rings and modules. Let R be a filtered ring with filtration (Ri )iZ ,
and let M be a filtered left Rmodule with filtration (Mi )iZ .

Definition. The Rees ring R

e of R is a subring of the ring of Laurent polynomials
R[t, t1 ]:
M M
Re= R i ti Rti = R[t, t1 ].
iZ iZ

The Rees module M

f of M is the abelian group
M
Mf= Mi t i
iZ

where the action of R

e is given by on homogeneous components by

R i ti M j tj Mi+j ti+j
r i ti , mj tj 7 mi rj ti+j .

Note that t R
e is a central regular element, since 1 R1 always. There is a
certain amount of interplay between the Rees ring of R and the associated graded
ring gr R.

Lemma. Let R and M be as above. Then

(1) R/t
e R e= gr R as rings,
(2) M /tM
f f = gr M as left gr Rmodules,
e 1)R
(3) R/(t e= R as rings,
(4) M /(t 1)M
f f = M as left Rmodules.

Proof. We will only prove the result for the rings, leaving the modules as an exercise.
(1). We have an isomorphism of abelian groups
i
L
e = L iZ Ri t = gr R.
M
R/t
e R
i
Ri /Ri1 =
iZ Ri1 t iZ

It can be checked that this is also a ring isomorphism.

56

(2). Define a ring homomorphism : R e R by (P ri ti ) = P ri . Since

(Ri ti ) = Ri we see that is onto. Clearly t 1 ker(). Check that in fact
ker() = (t 1)R.
e The result follows. 

So R e is a ring which has both R and gr R as epimorphic images. Therefore,

sometimes gr R is thought of as a deformation of R. Also note that if R
e is right
(or left) Noetherian, then so are both R and gr R.
Thanks are due to Lorenzo Di-Biagio for pointing out that an earlier version of
the following Lemma was incorrect.

9.4. Lemma. Suppose R is an almost commutative algebra with finite generating

set {x1 , . . . , xn }. Equip R with the positive filtration as in Example 7.1(3). Then
e = R 0 R 1 t R 2 t2
R
is right and left Noetherian.

Proof. We observe first that k[t] is a subring of R e and that R e is generated as

a k[t]algebra by {tx1 , tx2 , . . . , txn }. Note that we can widen the definition of
almost commutative kalgebra in 7.4 to the case when k is any commutative
ring, and that Proposition 7.4 will still be valid.
Since we can write [txi , txj ] as a k[t]linear combination of {tx1 , . . . , txn }, we
see that R e is an almost commutative k[t]algebra. Hence gr R e is a quotient of
k[t][X1 , . . . , Xn ] by Proposition 7.4, and therefore is Noetherian by Theorem 1.14.
Hence R e is right and left Noetherian, by Corollary 7.8. 

This result will be useful to us in what follows.

9.5. Good filtrations.

Definition. Let R be a filtered ring and let M be a left Rmodule.

A filtration (Mi ) on M is said to be good iff the Rees module M
f is finitely
generated over R.e
Two filtrations (Mi ) and (Mi0 ) on M are algebraically equivalent (or just
equivalent) if there exist c, d Z such that
Mi0 Mi+c and 0
Mj Mj+d for all i, j Z.

Note that if (Mi ) is a good filtration, then gr M

=Mf/tM f is finitely generated
over gr R and M =M f/(t 1)Mf is finitely generated over R, by Proposition 9.3.

Proposition. Let R be a filtered ring and let M be a left Rmodule.

(1) A filtration (Mi ) on M is good if and only if there exist k1 , k2 , . . . , ks Z
and m1 Mk1 , m2 Mk2 , . . . , ms Mks such that
Mi = Rik1 m1 + Rik2 m2 + + Riks ms for all i Z.
(2) All good filtrations on M are equivalent.
 57

Proof. (1) If the graded module M f is finitely generated, it has a finite homogeneous
generating set {tk1 m1 , . . . , tks ms } say, with mj Mkj . Then the ith homogeneous
component of M f is

ti Mi = Rik1 tik1 (tk1 m1 ) + + Riks tiks (tks ms ), so

Mi = Rik1 m1 + Rik2 m2 + + Riks ms .

Conversely, any filtration of this form is good, since {tk1 m1 , . . . , tks ms } is then
a generating set for M
f.
(2) Take two good filtrations (Mi ) and (Mi0 ). Then we have
Mi = Rik1 m1 + + Riks ms for all i
Mj0 = Rjl1 m01 + + Rjlr m0r for all j.
We can find c Z such that m0t Mlt +c for all t = 1, . . . , r. Then Mi0 Mi+c for
0
all i, and similarly there exists d Z such that Mi Mi+d for all i. 

One consequence of this result is that any finitely generated module has some
good filtration: just take a generating set X = {x1 , . . . , xs } and set the integers ki
to be 0, so that Mi = Ri x1 + . . . + Ri xs = Ri X (the standard filtration on M ) is
good.

9.6. Dimension theory for almost commutative algebras. Let R be an al-

most commutative algebra with generating set {x1 , . . . , xn }. Let (Ri ) be the usual
filtration on R, so that gr R is a quotient of the polynomial algebra k[X1 , . . . , Xn ]
by Proposition 7.4. Let M be a finitely generated left Rmodule and let (Mi ) be
some good filtration on M .
Note that each Mi is finite dimensional over k by Proposition 9.5(1), since each
Ri is. Also, gr M is a finitely generated graded gr Rmodule by Lemma 9.3 and
hence a finitely generated graded k[X1 , . . . , Xn ]module. As such, it has a Samuel
polynomial (t) defined in (9.2). Note that
i
X
(i) = dim(Mj /Mj1 ) = dim Mi dim M1
j=0

for sufficiently large i.

Definition. The dimension d(M ) of M is the degree d of (t). The multiplicity

m(M ) is d!a where a is the leading coefficient of (t). Thus d(M ) = d(gr M ) and
m(M ) = m(gr M ).

At first sight, this depends on the choice of good filtration (Mi ) on M . However,
let (Mi0 ) be another good filtration on M and let 0 (t) be the corresponding Samuel
polynomial. By Proposition 9.5(2), there exists c N such that
0 0
Mjc Mj Mj+c for all j Z,
58

0 0
so dim Mjc dim Mj Mj+c . Hence, for large enough j,

0 (j c) (j) 0 (j + c).

Since the behaviour of a polynomial for large enough j is determined by its leading
term, we see that dimension and multiplicity is well defined.

9.7. Examples.
(1) Let R = k[X1 , . . . , Xn ] and let (Ri ) be the usual positive filtration on R,
given by Ri = k.{X : || i}. Then dim Ri is the number of monomials
in n variables of length at most i which is well known to be
in
 
n+i (i + n)(i + n 1) (i + 1)
dim Ri = = = + o(in1 ).
n n! n!
So if we view R as an Rmodule by multiplication, we can read off d(R) = n
and m(N ) = 1.
(2) Let R = An with the usual filtration. Viewing R as a filtered module
over itself with the same filtration, we see that gr R = k[X1 , . . . , X2n ] by
Corollary 8.3, so d(R) = 2n and m(R) = 1.
(3) Let R = An and let M = k[X1 , . . . , Xn ] be the natural An module with
the usual filtration. Then d(M ) = n and m(M ) = 1.
(4) Let R be any almost commutative kalgebra and M a finitely generated
Rmodule. Then d(M ) = 0 if and only if dim M < .

9.8. Theorem. Let R be an almost commutative kalgebra and let M be a finitely

generated Rmodule. Suppose N is a submodule of M . Then
(1) d(M ) = max{d(N ), d(M/N )}.
(2) If d(N ) = d(M/N ) then m(M ) = m(N ) + m(M/N ).

Proof. Let (Mi ) be a good filtration on M . Give N the subspace filtration (Ni )
and M/N the quotient filtration (M/N )i . First, we prove that these filtrations are
good.
Note that Ni Mi and Mi /Ni = Mi /(N Mi ) = (Mi + N )/N . A computation
similar to the proof of Proposition 7.7 shows that
g
M f e
N M and
e f = M /N
N
as left Rmodules.
e
Now R e is left Noetherian by Lemma 9.4, so both N ^ are finitely gener-
e and M/N
ated Rmodules,
e as (Mi ) is good. Hence the filtrations on N and M/N are good
and we can use them to compute dimensions and multiplicities.
Next, we have
Nj (M/N )j Mj
dim + dim = dim
Nj1 (M/N )j1 Mj1
 59

so summing over j, we see that for large enough i,

(*) N (i) + M/N (i) = M (i).

Hence N + M/N = M as polynomials, since () holds for infinitely many i.

Since the leading coefficients of the Samuel polynomials are positive, we cannot
have cancellation, so (1) follows. Moreover, if d(N ) = d(M/N ) then the leading
coefficients add, and we obtain (2). 

9.9. Proposition. Let R be an almost commutative algebra with d(R) = n.

(1) If M is a finitely generated left Rmodule, then d(M ) n.
(2) If R is a domain and 0 6= I /r R then d(R/I) < n.

Proof. From Theorem 9.8(1), we see that d(Rm ) = d(R) = n for any m 1.
Since M is finitely generated, its a left Rmodule quotient of Rm for some m, so
d(M ) d(Rm ) = n, proving (1).
Next, pick 0 6= x I. Since R/I  R/Rx, we see that d(R/I) d(R/Rx) by
Theorem 9.8(1). Suppose for a contradiction that d(R/Rx) = d(R) = n. Since
R is a domain, Rx = R as left Rmodules, so d(Rx) = d(R/Rx). But now by
Theorem 9.8(2), m(Rx) + m(R/Rx) = m(R) so m(R/Rx) = 0, a contradiction.
Hence d(R/Rx) < n as required. 

9.10. Bernsteins Inequality.

Theorem (Bernstein, 1972). Suppose char(k) = 0 and let M be a nonzero finitely

generated left An (k)module. Then d(M ) n.

Proof. (Joseph)
Pick a finite generating set X and let (Mi ) be the standard filtration on M
given by Mi = (An )i X. This filtration is good and positive. Let (t) be the
Samuel polynomial of gr M with respect to this filtration. Then for large enough i,
(i) = dim Mi by the remarks in (9.6).
Claim: dim Ri dim Homk (Mi , M2i ) = (dim Mi )(dim M2i ).
Assuming the claim is true, for large enough i, dim Ri (i)(2i). Since
dim Ri = i+2n


2n is a polynomial in i of degree 2n, (t)(2t) must be a polyno-
mial of degree 2n. But it has degree 2d(M ), so d(M ) n as required.
To prove the claim, consider the klinear map

i : R i Homk (Mi , M2i )

r (m rm)

Its sufficient to show that i is injective. Equivalently, we must show that if

0 6= r Ri then rMi 6= 0. Proceed by induction on i. When i = 0, r k\0 is a
unit so rM0 = M0 = kX 6= 0 since M 6= 0.
60

Now assume i > 0 and suppose for a contradiction that 0 6= r Ri is such

that rMi = 0. Then r / k = Z(An ) by Proposition 8.6(1), so [r, u] 6= 0 for some
u {x1 , . . . , xn , y1 , . . . , yn }. Weve seen in the proof of Proposition 8.6 that
[r, u] Ri1 ,
so [r, u]Mi1 6= 0 by induction. But uMi1 Mi and rMi = 0 so
[r, u]Mi1 = (ru ur)Mi1 = 0,
a contradiction. The result follows. 

Note that when n = 1, this says that any nonzero finitely generated An module
is infinite dimensional over k, by Example 9.7(4). This is also the content of Propo-
sition 8.10(3).